Add files using upload-large-folder tool

preprint/preprint__b31d84793df4971083539bc948a969ff9f9a2b3016824fa91f01a5480ede8c80/images_list.json

@@ -0,0 +1,93 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "glial RNAi",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 32
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "injured 5 dpi",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        427,
+        390,
+        536
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "D",
+    "footnote": [],
+    "bbox": [
+      [
+        520,
+        427,
+        740,
+        536
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "F",
+    "footnote": [],
+    "bbox": [
+      [
+        172,
+        590,
+        384,
+        707
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "VGlut>QUAS-mCD8::GFP, repo>UAS-Cas9 remaining axon debris 5 dpi",
+    "footnote": [],
+    "bbox": [
+      [
+        511,
+        590,
+        725,
+        707
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "H",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        755,
+        440,
+        864
+      ]
+    ],
+    "page_idx": 36
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "B",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 36
+  }
+]

preprint/preprint__b31d84793df4971083539bc948a969ff9f9a2b3016824fa91f01a5480ede8c80/preprint__b31d84793df4971083539bc948a969ff9f9a2b3016824fa91f01a5480ede8c80_det.mmd

@@ -0,0 +1,770 @@
+<|ref|>title<|/ref|><|det|>[[44, 106, 821, 177]]<|/det|>
+# LC3-associated phagocytosis promotes glial degradation of axon debris after injury  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 912, 240]]<|/det|>
+Aron Szabo Biological Research Center, Eötvös Loránd Research Network (ELKH) https://orcid.org/0000- 0003- 3619- 5092  
+
+<|ref|>text<|/ref|><|det|>[[44, 245, 911, 288]]<|/det|>
+Viraq Vincze Biological Research Center, Eötvös Loránd Research Network (ELKH)  
+
+<|ref|>text<|/ref|><|det|>[[44, 293, 911, 336]]<|/det|>
+Sarolta Bognar Biological Research Center, Eötvös Loránd Research Network (ELKH) https://orcid.org/0000- 0002- 6397- 3235  
+
+<|ref|>text<|/ref|><|det|>[[44, 341, 911, 383]]<|/det|>
+Katalin Varga Biological Research Center, Eötvös Loránd Research Network (ELKH)  
+
+<|ref|>text<|/ref|><|det|>[[44, 389, 650, 431]]<|/det|>
+Poulami Banik Biological Research Center, Eötvös Loránd Research Network (ELKH)  
+
+<|ref|>text<|/ref|><|det|>[[44, 437, 650, 479]]<|/det|>
+Andras Jipa Biological Research Center, Eötvös Loránd Research Network (ELKH)  
+
+<|ref|>text<|/ref|><|det|>[[44, 484, 650, 526]]<|/det|>
+Lukas Neukomm University of Lausanne https://orcid.org/0000- 0002- 5007- 3959  
+
+<|ref|>text<|/ref|><|det|>[[44, 531, 616, 572]]<|/det|>
+Gabor Juhasz ( juhasz.gabor@brc.hu ) Biological Research Center, Eötvös Loránd Research Network (ELKH)  
+
+<|ref|>text<|/ref|><|det|>[[44, 575, 650, 599]]<|/det|>
+Gabor Juhasz ( juhasz.gabor@brc.hu ) Biological Research Center, Eötvös Loránd Research Network (ELKH)  
+
+<|ref|>text<|/ref|><|det|>[[44, 601, 650, 620]]<|/det|>
+Biological Research Center, Eötvös Loránd Research Network (ELKH)  
+
+<|ref|>text<|/ref|><|det|>[[44, 650, 102, 668]]<|/det|>
+Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 688, 135, 706]]<|/det|>
+Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 725, 341, 744]]<|/det|>
+Posted Date: September 7th, 2022  
+
+<|ref|>text<|/ref|><|det|>[[44, 763, 473, 782]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 2009846/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 800, 909, 843]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[128, 88, 872, 168]]<|/det|>
+# LC3-associated phagocytosis promotes glial degradation of axon debris after injury  
+
+<|ref|>text<|/ref|><|det|>[[112, 224, 863, 281]]<|/det|>
+Áron Szabó \(^{1*}\) , Virág Vincze \(^{1}\) , Sarolta Bognár \(^{1}\) , Katalin Eszter Varga \(^{1}\) , Poulami Banik \(^{1}\) , András Jipa \(^{1}\) , Lukas J. Neukomm \(^{2}\) and Gábor Juhász \(^{1,3*}\)  
+
+<|ref|>text<|/ref|><|det|>[[170, 338, 866, 536]]<|/det|>
+1. Biological Research Center, Institute of Genetics, Eötvös Loránd Research Network (ELKH), H-6726 Szeged, Hungary  
+2. Department of Fundamental Neurosciences, University of Lausanne, CH-1005 Lausanne, Switzerland  
+3. Department of Anatomy, Cell and Developmental Biology, Eötvös Loránd University, Budapest, H-1117, Hungary  
+
+<|ref|>text<|/ref|><|det|>[[113, 584, 581, 602]]<|/det|>
+\*Correspondence to: aszabo@brc.hu, juhasz.gabor@brc.hu
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[451, 125, 544, 147]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[110, 166, 888, 680]]<|/det|>
+Glial engulfment of dead neurons and neurites after trauma, during development and in neurodegenerative diseases plays a crucial role in nervous system maintenance. Axon debris generated after traumatic injury is cleared by phagocytic glia via Draper receptor signalling in Drosophila. However, mechanisms governing the efficiency of axon debris phagocytosis and degradation have remained largely unexplored. Here we show that glial LC3- associated phagocytosis (LAP), an engulfment pathway assisted by certain components of the macroautophagy machinery, promotes clearance of degenerating axons in the Drosophila wing nerve. A LAP- specific subset of autophagy- related (Atg) genes is required in glia for efficient debris elimination, which includes members of the Atg8a (LC3) conjugation system and the Vps34 lipid kinase complex subunits UVRAG and Rubicon but not Atg14 or the Atg1 kinase complex. Atg8a and Rubicon are recruited to Rab7- positive phagosomes and Atg8a lipid conjugation is essential for debris- containing phagosome maturation. Finally, Rubicon overexpression in glia accelerates axon debris elimination. Our results reveal the critical role of LAP in glia in the clearance of neuronal debris in vivo, with important implications for the recovery of the injured nervous system.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[431, 125, 564, 147]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[110, 181, 888, 519]]<|/det|>
+Neural injuries, stroke and neurodegenerative diseases generate a significant volume of dead cell- derived material that is cleared by phagocytic glia or macrophages. In mammals, debris generated after harmful events and synapses or neurites formed in excess during development are eliminated by microglia and astrocytes \(^{1 - 3}\) . Major trauma, stroke and neurodegeneration can lead to reactive gliosis, which, if not contained in time, can be maladaptive and could result in inhibition of axon and synapse regeneration and in neuroinflammation \(^{4}\) . However, initial microglial activation and ensuing debris phagocytosis help to resolve the damage in the affected tissue at the lesion site, thereby reducing the risk of inflammation and secondary neurodegeneration. Therefore, establishing the correct balance between glial activation states and in timing of debris clearance seems crucial for nervous system recovery.  
+
+<|ref|>text<|/ref|><|det|>[[110, 548, 886, 710]]<|/det|>
+Studies on Drosophila have elucidated several aspects of the mechanism of axon debris clearance by glia \(^{3,5 - 7}\) . Drosophila glial types are functionally similar to glia in mammals \(^{5}\) . Ensheathing glia are the phagocytic cell type in the adult fly brain \(^{8}\) while in the adult peripheral nervous system (PNS), wrapping and subperineurial glia were found to engulf the majority of axon debris generated after injury \(^{9}\) .  
+
+<|ref|>text<|/ref|><|det|>[[110, 740, 886, 902]]<|/det|>
+Injury of axon tracts of olfactory sensory neurons coupled to genetic screens in glia allowed identification of several components of the debris clearance machinery in flies \(^{6,7}\) . Adult glia engulf dead cell material through the action of Draper (Drpr), a transmembrane receptor with multiple EGF repeats \(^{6,10}\) that recognizes phosphatidylserine \(^{11 - 13}\) . Drpr functions early to activate glial transcription programmes and extension of glial membranes towards debris \(^{10,14 - 16}\) . Drpr initiates
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 87, 886, 283]]<|/det|>
+Rac1 small GTPase activation through guanine nucleotide exchange factor (GEF) complexes Drk/Dos/Sos and Crk/Mbc/dCed- 12<sup>17</sup> that leads to membrane recruitment to debris and thereafter debris internalization. After the elimination of axon fragments, CNS glia retract their processes and return to a resting state<sup>18</sup>. As compared to the body of evidence obtained in the CNS injury model, little is known about the molecular mechanisms of PNS debris removal, except for Drpr being essential during debris engulfment in wrapping and to a lesser extent in subperineurial glia<sup>9</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[110, 311, 886, 610]]<|/det|>
+While the steps of early activation and the extension of glial processes towards debris are relatively well known, whether and how internalization and phagocytosis of axon debris by glia are regulated remains largely unexplored. Injury of olfactory sensory neuron axons leads to a transient increase in glial lysosome acidification in the region of degenerating axons<sup>19</sup>. Glial vesicles that form after injury, most likely phagosomes, are surrounded by Drpr and can reach acidic lysosomes. The Rac1 GEF complex Crk/Mbc/dCed- 12 is not required for glial activation but mediates internalization of axon debris and after Crk and dCed- 12 silencing in glia, no lysosome acidification occurs<sup>19</sup>. However, how phagosome maturation and its fusion with lysosomes is regulated during debris engulfment has not been explored so far.  
+
+<|ref|>text<|/ref|><|det|>[[110, 642, 886, 907]]<|/det|>
+Macroautophagy (hereafter autophagy) is a membrane- contained intracellular degradation pathway that eliminates unwanted cytoplasmic material such as aged or defective organelles, protein aggregates and lipid droplets but also supplies the cells with recycled nutrients upon starvation by degrading more- or- less random portions of the cytoplasm<sup>20</sup>. Autophagy induction in dying cells may limit debris generation and subsequent inflammation. Non- canonical autophagy has been implicated in endocytosis, phagocytosis and secretion among other processes<sup>21</sup>. These functions are based on selective reuse of certain subsets of autophagy proteins (Atg- s). Autophagy has been directly linked to various glial processes including phagocytosis<sup>22,23</sup>.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 85, 888, 423]]<|/det|>
+The core autophagy machinery consists of several modules. The process is headed by an initiation complex (aka. Atg1 kinase complex) that conveys AMPK and Tor signalling- mediated nutritional information to other components followed by a nucleation complex (Class III phosphatidylinositol- 3- kinase, PI3K) with lipid kinase activity<sup>20,24</sup>. This prepares membranes with phosphatidylinositol- 3- phosphate (PI3P) for expansion of cup- like double membrane structures called phagophores that finally close to form autophagosomes containing to- be- degraded cytoplasmic cargoes. Two ubiquitin- like protein conjugation systems promote phagophore expansion and autophagosome closure at the heart of the autophagic process. This involves conjugation of Atg8a to the phagophore/autophagosome membrane on phosphatidylethanolamine (PE). Thereafter, Atg8a also contributes to organizing autophagosome fusion with lysosomes<sup>20,24</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[110, 450, 888, 680]]<|/det|>
+In this work, we aimed to understand the role of autophagy and related pathways in axon debris clearance after injury in the Drosophila wing nerve. We hypothesized that autophagy genes may play a role in three different and non- exclusive scenarios. First, autophagy induction in degenerating axon fragments could accelerate or limit debris production<sup>25</sup>. Second, autophagic vesicles may fuse with phagosomes to promote the clearance of the engulfed material in glia<sup>26,27</sup>. Lastly, an autophagy- related process may be necessary for phagosome maturation and degradation<sup>28</sup>.  
+
+<|ref|>sub_title<|/ref|><|det|>[[457, 712, 538, 735]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 772, 630, 792]]<|/det|>
+## A subset of Atg proteins is required for axon debris clearance  
+
+<|ref|>text<|/ref|><|det|>[[112, 821, 886, 876]]<|/det|>
+Numerous reports have demonstrated that autophagosomes can facilitate the clearance of apoptotic cell corpses<sup>26,27</sup>. In order to investigate whether autophagy participates in the clearance of axon
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 80, 888, 604]]<|/det|>
+debris after injury, we evaluated debris elimination in the \*Drosophila\* L1 vein wing nerve, an established model for Wallerian degeneration in the PNS \(^{9,29,30}\) following unilateral wing transection (Fig. 1A). About 40 wing nerve axons can be labelled by \*OK371-Gal4\*, \*UAS-mCD8::GFP\* that allows monitoring of axon fragmentation and subsequent debris removal (9). \*Atg8a\* is a \*Drosophila\* ortholog of mammalian LC3 proteins \(^{31}\) . The pace of axon fragmentation and debris generation in the viable \*Atg8a\* \(^{44}\) deletion mutant \(^{32}\) was similar to the wild-type at 2 days post-injury (dpi) (Fig. 1B,C). However, at 5 and 10 dpi, accumulated debris lingered in \*Atg8a\* \(^{44}\) wings, unlike the case of wild-type control where debris was gradually cleared (Fig. 1B,C). Axon debris accumulation could be rescued in injured \*Atg8a\* \(^{44}\) flies by expressing 3xmCherry-tagged Atg8a under the control of the \*Atg8a\* promoter (Fig. 1B,C). Contralateral uninjured wings showed no obvious axon morphology defects (Fig. S1A). Atg8a becomes membrane-bound via its covalent conjugation to phosphatidylethanolamine (PE) by the action of an E3-like ligase complex composed of Atg5, Atg12 and Atg16. Flies with a trans-heterozygous \*Atg16\* mutant combination (\*Atg16\* \(^{467/4129}\) ) (Fig. 1B,C) and an \*Atg5\* null mutant (\*Atg5\* \(^{5cc5}\) ) (Fig. 1D,E) also failed to properly clear debris at 5 dpi, similar to \*Atg8a\* \(^{44}\) .  
+
+<|ref|>text<|/ref|><|det|>[[110, 630, 888, 863]]<|/det|>
+Using CRISPR- based genome engineering, we recently created a lipid conjugation- deficient version of \*Atg8a\* where the codon coding for the glycine near the C- terminus (Gly116) to be conjugated to PE is replaced by a stop codon in the endogenous \*Atg8a\* locus (\*Atg8a\* \(^{G116*}\) ). This mutant expresses a form of \*Atg8a\* that cannot undergo lipidation \(^{31}\) . \*Atg8a\* \(^{G116*}\) animals also presented defective axon debris clearance (Fig. 2A,C), phenocopying \*Atg8a\* \(^{44}\) and conjugation system mutants. These results demonstrate that the Atg8a conjugation system and Atg8a lipidation are required for normal clearance of axon debris after injury.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 85, 888, 498]]<|/det|>
+The Atg8a conjugation system plays multiple roles from phagophore expansion, cargo recruitment and membrane closure to autophagosome fusion with lysosomes. Nutrient limitation and other autophagy activating signals are relayed through the upstream- acting Atg1 kinase complex. We thus selected two viable mutants of Atg1 complex members, an Atg101 null (Atg101<sup>46h</sup>) and an Atg17/FIP200 severe hypomorph allelic combination (Atg17<sup>Δ130/MiMC</sup>) to gauge whether debris clearance depends on Atg1 kinase complex activity. To our surprise, neither of these mutants accumulated axon fragments when compared to the wild- type (Fig. 2A- C). Importantly, both mutants abrogated the autophagic process similarly to the conjugation system mutant Atg5<sup>5cc5</sup> as evidenced by severe depletion of vesicle- bound 3xmCherry- Atg8a in uninjured animals, appearing as puncta in the same wing area as used for axon imaging (Fig. 2D,E). This indicates that a process relying on the conjugation system but not requiring the Atg1 complex is at play during debris engulfment.  
+
+<|ref|>text<|/ref|><|det|>[[110, 524, 888, 895]]<|/det|>
+LC3- associated phagocytosis (LAP) is a non- canonical autophagy- related pathway that contributes to phagocytic clearance in certain conditions. During LAP, a Rubicon and UVRAG- containing PI3K complex and the LC3 conjugation system are recruited to early phagosomes that contain cargo such as dead cells, entotic cells, cell remnants, pathogens, and photoreceptor outer segments that have been recognised by cell surface receptors<sup>33- 35</sup>. LC3 lipid conjugation onto the single- membrane phagosome promotes its fusion with the lysosome where the phagosome content is degraded. The function of the initiation complex is dispensable for LAP as it does not rely on nutritional signals. Therefore, we hypothesized that LAP could underlie the processing of internalised axon debris in glia. First, we tested this idea by disruption of a specific domain of Atg16 that is utilized by LAP but not by canonical autophagy. Accumulating evidence suggests that functions of Atg16 domains are separable in that the central coiled- coil domain is essential for
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 85, 886, 283]]<|/det|>
+autophagy while the C- terminal WD40 repeats are only implicated in LAP, both in cultured cells and in vivo<sup>36,37</sup>. We previously characterized a MiMIC transposon insertion (Atg16<sup>MI</sup>) in the same WD40 repeats of Drosophila Atg16, which blocked enteroendocrine cell differentiation by impairing Slit production while autophagy remained intact in posterior midgut cells<sup>38</sup>. Indeed, Atg16<sup>MI</sup> flies showed debris clearance defects, further indicating a potential role for LAP but not for canonical autophagy in debris engulfment (Fig. 2F,G).  
+
+<|ref|>text<|/ref|><|det|>[[110, 310, 886, 547]]<|/det|>
+LAP is initiated by receptor- mediated internalization of extracellular ligands such as apoptotic cells, immune complexes and pathogens. To investigate whether Drpr and the Atg8a conjugation machinery act in the same engulfment process, we combined a drpr heterozygote null mutant (drpr<sup>- 45</sup>) with Atg8a<sup>- 44</sup> and followed the clearance of debris after nerve cut. drpr<sup>- 45</sup> heterozygotes retained axon fragments<sup>39</sup> comparably to Atg8a<sup>- 44</sup> at 5 dpi. As we did not observe a positive additive effect of these two mutations in double mutant animals (Fig. 2H,I), these results indicate that Drpr and Atg8a likely participate in the same pathway underlying debris removal.  
+
+<|ref|>sub_title<|/ref|><|det|>[[112, 625, 560, 644]]<|/det|>
+## Glial function of LAP genes drives debris elimination  
+
+<|ref|>text<|/ref|><|det|>[[110, 674, 886, 905]]<|/det|>
+To corroborate the specific requirements for autophagic complexes in debris engulfment by glia, we drove RNAi against Atg5, Atg16 and Atg1 in glia with repo- Gal4 using dsRNAs targeting two independent regions for both Atg5 and Atg16. RNAi against an unrelated control gene, white (w), was used as a negative control. Wing axons were labelled by an independent binary expression system, the Q system, using VGlut- QF, QUAS- mCD8::GFP. Debris removal defects were evident in all RNAi- s except for Atg1 at 5 dpi (Fig. 3A,B). Similarly, knockdown of the autophagosomal Qa SNARE Syntaxin 17 (Syx17)<sup>40</sup> did not impair debris elimination (Fig. 3C,D). To functionally
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 85, 888, 457]]<|/det|>
+test the efficiency of \(Atg\) RNAi knockdowns, we utilized a classical autophagy assay where aggregation of a selective autophagy cargo, Ref (2)P/p62 mirrors the degree of impediment in its autophagic degradation<sup>41</sup>. We expressed GFP- Ref (2)P in glia, co- expressed the RNAi- s as before and imaged GFP- Ref (2)P puncta in wing glia. All \(Atg\) RNAi- s including \(Atg1\) dramatically accumulated Ref (2)P (Fig. 3E,F). No gross morphological changes were apparent in uninjured wing nerves in either the density or spacing of glia or their interaction with axons upon \(Atg16\) knockdown, arguing for a specific defect in clearance (Fig. S1B). In line with the above data, glial knockdown of \(Atg13\) and \(FIP200/Atg17\) , two other subunits of the \(Atg1\) complex had no effect on axon debris persistance (Fig. S2A,B), despite a significant increase in the number of GFP- Ref (2)P puncta (Fig. S2C,D). Accordingly, a subset of genes that are required for autophagy in wing glia are dispensable for axon debris removal.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 536, 638, 556]]<|/det|>
+## LAP in glia promotes phagosome maturation after engulfment  
+
+<|ref|>text<|/ref|><|det|>[[110, 584, 888, 886]]<|/det|>
+Based on these observations, we postulated that LAP is at play during debris processing through the phagocytic pathway. During LAP, early phagosomes mature into late phagosomes (LAPosomes) through phosphatidylinositol- 3- phosphate synthesis, reactive oxygen species generation and LC3 conjugation to the single membrane that finally culminate in fusion of the phagosome with the lysosome<sup>34</sup>. In the absence of LC3 conjugation, phagolysosome formation should be abrogated. Therefore, we wondered if knockdown of \(Atg16\) would hinder glial phagosome- lysosome fusion after debris engulfment. To explore this, we labelled glial lysosomes with GFP- LAMP1 and axon debris with CD4::tdTomato using the Gal4 and LexA binary expression systems, respectively. At 3 dpi, we observed co- localization between CD4::tdTomato<sup>+</sup>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 87, 886, 213]]<|/det|>
+debris and GFP+ glial lysosomes (Fig. 4A,B), indicating that debris- containing phagosomes fused with lysosomes, while the number of these co- localization events was indeed reduced upon Atg16 silencing in glia. This is consistent with our model that Atg8a conjugation leads to formation of debris- containing phagolysosomes.  
+
+<|ref|>text<|/ref|><|det|>[[110, 238, 886, 860]]<|/det|>
+Phagosome maturation during LAP depends on phosphatidylinositol- 3 phosphate generated by the Class III Vps34 PI3K complex on the phagosome membrane<sup>34</sup>. The Vps34 complex has different subunit compositions depending on the trafficking process in question. In mammals, Atg14L incorporates into Vps34 complex I to promote autophagy in concert with core subunits VPS34, VPS15 and Beclin 1<sup>42</sup>. However, UVRAG can support autophagy, endocytosis and LAP by forming Vps34 complex II through replacement of Atg14L. The Drosophila autophagy nucleation complex similarly lists Vps34, Vps15 and Atg6/Beclin 1 as core subunits and Atg14 as a specific subunit<sup>43</sup>. Drosophila UVRAG is dispensable for autophagy in fat cells<sup>44,45</sup> but is required for endosome maturation and cinophagy<sup>45- 47</sup>. An additional complex II member, Rubicon redirects Vps34 activity towards single- membrane structures and is a defining player in LAP<sup>34</sup>. Rubicon binding to the Vps34 complex inhibits autophagy and endocytosis<sup>48,49</sup>. To dissect which Vps34 complex supports debris clearance by glia in flies, we depleted specific complex components as well as Vps34 in glia and followed axon fragment accumulation after nerve transection. At 5 dpi, knockdown of Vps34, UVRAG and Rubicon all increased remaining debris in wing nerves, whereas Atg14 RNAi had no effect<sup>45,50</sup> (Fig. 5A,B). We confirmed Atg14 and Vps34 RNAi efficiency using the GFP- Ref (2)P accumulation assay for their ability to disrupt autophagy and both RNAi- s accumulated Ref (2)P aggregates (Fig. S3A,B), as expected. These results imply that axon debris clearance depends on the LAP- specific but not on the autophagy- specific PI3K complex in glia.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[112, 88, 883, 143]]<|/det|>
+## Rubicon defines the Vps34 complex that underlies debris processing and associates with debris-containing vesicles  
+
+<|ref|>text<|/ref|><|det|>[[110, 169, 888, 825]]<|/det|>
+Rubicon in mammals serves as the lynchpin of LAPosome formation; besides its role in stimulating Vps34 activity towards the phagosome, it stabilizes the NOX2 complex that generates ROS on LAPosomes, facilitating LC3- II incorporation<sup>51,52</sup>. Little is known about the fly ortholog of Rubicon<sup>53,54</sup>. We thus sought additional evidence for the involvement of Rubicon in glial debris phagocytosis. We created a frameshift mutant of Rubicon (Rubicon<sup>6</sup>) by introducing a small indel into the Rubicon coding region using CRISPR- Cas9 (Fig S4A). Rubicon mRNA level halved in this mutant presumably due to nonsense- mediated decay (Fig S4A). Similar to Rubicon RNAi, Rubicon<sup>6</sup> animals failed to clear axon debris from injured nerves (Fig. 5C,D). We validated this with Rubicon<sup>0/04462</sup> as an independent, hypomorphic mutant allele that also led to a failure of removing axon fragments after injury (Fig. S4B- D). Subsequently, we utilized a tissue- specific CRISPR knockout (tsKO) approach to disrupt Rubicon only in glia. We expressed Cas9 in glia with repo- Gal4, expressed a control (w) versus Rubicon single guide RNA constitutively and labelled axons with VGlut- QF, QUAS- mCD8::GFP. At 5 dpi, remaining debris was apparent in Rubicon but not in control tsKO wing nerves (Fig. 5E,F). Lastly, we overexpressed Rubicon in glia and assessed axon fragment density at a relatively early time point after wing injury. While fragments were still abundant in mtdTomato- overexpressing controls, Rubicon overexpression depleted the pool of undigested axon fragments indicating that Rubicon alone is sufficient to boost debris phagocytosis (Fig. 5G,H). Taken together, Rubicon, a defining factor in LAP, is required for efficient debris clearance.  
+
+<|ref|>text<|/ref|><|det|>[[111, 833, 885, 890]]<|/det|>
+How is then Rubicon regulated after injury to stimulate LAP? We first hypothesized that Rubicon is upregulated in response to axotomy. However, we failed to see changes in Rubicon mRNA levels
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 80, 890, 810]]<|/det|>
+in wings 1 and 3 days post wing transection (Fig. S5A). We then turned our attention to Rubicon localization in glia. Rubicon is mainly localized to vesicles in mammalian cells<sup>48,49</sup>. We expressed Rubicon::mRFP1 in glia and labelled axonal membranes with GFP. In wing glia, Rubicon::mRFP1 was found on vesicles as expected (Fig. 6A). Both 1 and 2 days after wing nerve injury, we observed colocalization of glial Rubicon::mRFP1 vesicles and GFP-labelled axon debris, indicating that engulfed axon fragments are contained in LAPosomes (Fig. 6A). Next, we wanted to elucidate when Rubicon is recruited to the phagosome. Phagocytosis, similarly to endocytosis, relies on a Rab5- Rab7 small GTPase switch that occurs during the transition from early to late phagosomes<sup>55</sup>. We found scarce colocalization of glial Rubicon::mRFP1 with Rab5 (Fig. S5B) or lysosomal LAMP1 (Fig. S5C) on vesicles in either uninjured or injured states. In contrast, robust colocalization of vesicular Rubicon with Rab7 was detected in glia in both conditions (Fig. 6B,C). As phagocytosis is presumably limited in uninjured, uninfected adult flies, the presence of numerous Rubicon- Rab7 double- positive vesicles in intact wings likely reflects the inhibitory role of Rubicon in endocytosis where it binds to Rab<sup>75,56,57</sup>. In contrast, 2 days post injury, a different Rubicon- Rab7 double positive vesicle type appeared with an average size that was tripled compared to vesicles in the uninjured condition (Fig. 6C). These vesicles were also fewer in number and their Rubicon- RFP intensity decreased relative to uninjured flies. The drop in double positive vesicle number was probably driven by limited Rubicon association rather than by failure of Rab7 recruitment as evidenced by the larger decrease in the overall number of Rubicon<sup>+</sup> vesicles (Fig. 6C). These results together with the Rubicon- axon debris colocalization data suggest the presence of maturing LAPosomes that are larger in size as compared to endosomes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 820, 523, 840]]<|/det|>
+## Atg8a is recruited to engulfed axon debris in glia  
+
+<|ref|>text<|/ref|><|det|>[[112, 853, 886, 909]]<|/det|>
+The key event of LAP is LC3 recruitment to the early phagosome. To ascertain that LAP underlies axon debris phagocytosis in glia, we labelled axon fragments with myr::GFP driven by LexA and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 85, 888, 457]]<|/det|>
+expressed mCherry- Atg8a in glia driven by Gal4. We indeed saw punctate co- localization of glial mCherry- Atg8a and GFP- labeled axon debris at 2 dpi, shortly after the complete fragmentation of injured axons (Fig. 7A,B). As GFP is quickly quenched in acidic lysosomes, GFP structures decorated with glial Atg8a correspond to LAPosomes. To perturb the maturation of LAPosomes, we silenced Vps34 in this background. Of note, we saw a robust increase in co- localization between mCherry- Atg8a and \(\mathrm{GFP^{+}}\) phagosomes (Fig. 7A,B). Since Vps34 promotes late steps of phagosome processing<sup>58</sup>, these results are explained by a disruption of phagocytic flux by Vps34 knockdown, leading to upstream stabilization of \(\mathrm{Atg8a^{+}}\) LAPosomes. Hence, our findings provide evidence for Atg8a and Rubicon recruitment to debris- containing phagosomes, thereby identifying LAP as the mechanism that licenses efficient lysosomal degradation of internalized axon debris in glia.  
+
+<|ref|>sub_title<|/ref|><|det|>[[439, 473, 558, 496]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[110, 531, 888, 762]]<|/det|>
+Here we have shown that a subset of autophagy genes is required in glia for proper removal of axon debris generated after nerve injury. Dispensability of the Atg1 complex and dependence on Vps34 complex members Rubicon and UVRAG indicate that we witness glial LC3- associated phagocytosis. This is further strengthened by results obtained with the Atg16 WD40 mutant, which does not abrogate autophagy but still impairs debris clearance. With the use of LAP- specific mutants, we exclude the possibility of autophagosomes being involved in the fusion between phagosomes and lysosomes such as in the case of \(C\) . elegans apoptotic cell corpse clearance<sup>26,27</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[111, 789, 886, 881]]<|/det|>
+LAP has mostly been characterized in macrophages in relation to defence against pathogens and during efferocytosis. Although glial cells are efficient phagocytes, hardly any study has addressed the function of LAP in either microglia, astrocytes or in invertebrate glia<sup>59- 61</sup> and to our knowledge,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 87, 885, 214]]<|/det|>
+a glial role for LAP during uptake of physiologically relevant cargoes in vivo has not been reported. Of note, recently a microglial LC3- mediated endocytic process, LC3- associated endocytosis (LANDO) was discovered where recycling of putative amyloid \(\beta\) receptors to the cell surface is controlled by LAP genes<sup>59,62</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[111, 243, 886, 544]]<|/det|>
+Rubicon binding to the LAPosome stimulates phagocytosis<sup>52</sup>. In glia, a switch from supporting canonical autophagy to supporting LAP is conceivable by the usage of an alternative Rubicon- containing Vps34 complex II after sensing injury signals. In Drosophila glia, it is likely that phagocytic receptor engagement with debris initiates LAP induction by impacting Rubicon association to the Vps34 complex and/or the LAPosome. Rubicon/Vps34 complex and LAPosome assembly could potentially be stimulated by either elevated Rubicon expression, its directed transport or its posttranslational modifications after injury. Although we did not see Rubicon expression changes after injury at the transcript level, Rubicon is recruited to incoming axon debris in glia after injury, indicating formation of LAPosomes.  
+
+<|ref|>text<|/ref|><|det|>[[110, 573, 886, 907]]<|/det|>
+The relationship of Rubicon with Rab GTPases (key players in phagosome maturation) has received little attention, even though Rubicon has a Rab7- binding domain that mediates its inhibitory function in autophagy and endocytosis<sup>56,57</sup>. We find that Rubicon associates with Rab7- but not with Rab5- or LAMP1- positive vesicles, representing late and early endosomes/phagosomes and lysosomes, respectively. Although Rab7/Rubicon double positive vesicles exist in uninjured conditions as well, their size greatly increases after injury, and the drop in their numbers and their fainter Rubicon signal indicates that most of them are phagosomes rather than endosomes. This also points to the intriguing possibility that Rubicon on phagosomes is actually derived from endosomes that are Rubicon- positive in the uninjured state. These endosomes might fuse with late phagosomes supplying Rubicon for LAP, which would explain the size
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 87, 886, 213]]<|/det|>
+increase of vesicles and the dilution of Rubicon, eventually giving a fainter signal. Indeed, phagosome- endosome fusion has been observed as early as 1991 and is a process that is targeted by pathogens \(^{63 - 65}\) . Actually, Rab7 and its effector RILP are required for phagosome- endosome (and phagosome- lysosome) fusion through tubular projections \(^{66}\) .  
+
+<|ref|>text<|/ref|><|det|>[[111, 242, 888, 507]]<|/det|>
+Drpr may signal during engulfment through the GEF complex Crk/Mbc/dCed- 12 to boost debris phagocytosis \(^{19}\) in addition to its function in debris recognition and glial activation. Hinting to a role for Drpr in LAP regulation, drpr knockdown in brain glia was reported to induce accumulation of Atg8a \(^{+}\) puncta that cluster around engulfed cell corpses \(^{67}\) . This effect resembles our results with persisting Atg8a \(^{+}\) phagosomes in the absence of Vps34. We also see that Atg8a and Drpr likely function in the same pathway during debris clearance. Drpr orthologs CED- 1 (C. elegans) and MEGF10/JEDI (mammals) show a conserved function in engulfment of dead cells, and we expect that LAP also would fulfil conserved roles in CNS debris elimination in other species.  
+
+<|ref|>text<|/ref|><|det|>[[110, 536, 888, 906]]<|/det|>
+We predict that glial LAP would support critical functions in the mammalian nervous system based on recent studies showing importance of microglial non- canonical autophagy in disease settings. First, a process reminiscent of LAP was described in myelin debris clearance by microglia in a murine multiple sclerosis (MS) model \(^{68}\) . Microglial deletion of Atg7 but not ULK1 (ortholog of Atg1) leads to defective myelin degradation and impaired recovery from experimental autoimmune encephalomyelitis. Possibly, LAP might also control phagocytosis of myelin and axon debris in various acute and chronic neurodegenerative conditions including traumatic injury and MS. Second, correct developmental wiring of the CNS also depends on phagocytosis of pruned supernumerary synapses. Intriguingly, Atg7- deficient microglia are inefficient in synaptic pruning and therefore mice with microglial deletion of this gene display repetitive behaviours and impaired social interaction \(^{69}\) . Synaptic pruning could accordingly be under the control of LAP as well.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 81, 888, 633]]<|/det|>
+LAP has recently been characterized in multiple phagocytic cell types with studies emphasizing its modulatory role<sup>70</sup>. First discovered in macrophages, LAP was later identified in other immune cells and non- professional phagocytes<sup>35</sup>. LAP in retinal pigment epithelium was shown to mediate breakdown of photoreceptor outer segments (POS, a neural tissue) in a circadian rhythm<sup>71,72</sup>. Beyond its well- established role in defence against pathogens, it is still a largely unanswered question how important LAP is in vivo. What is evident is that knockout of LAP components does not completely abrogate phagocytosis but still has severe long- term consequences such as the development of an autoinflammatory systemic lupus erythematosus (SLE)- like disease in mice<sup>73</sup> although the involvement of LAP in SLE has recently been called into question<sup>74</sup>. Inhibition of LAP also leads to enhanced hepatic inflammation and fibrosis in response to liver injury<sup>75</sup>. Undegraded cell material can turn phagocytes into a proinflammatory state<sup>73</sup> that could also be valid for the nervous system. Neuroinflammation is indeed a major contributor to the long- term corollaries of both traumatic brain injury and neurodegenerative diseases<sup>76</sup>, conditions that generate dying cells in bulk. Our results identify the critical role of LAP in the phagocytic arm of the debris clearance pathway to prevent uncontrolled inflammatory responses to uncleared material in the nervous system.  
+
+<|ref|>text<|/ref|><|det|>[[110, 658, 888, 890]]<|/det|>
+A recent paper suggested that increased neuronal expression of Rubicon during aging decreases autophagy and limits lifespan in multiple models, while loss of Rubicon upregulated autophagy and led to longevity<sup>53</sup>. Since our data demonstrate the importance of Rubicon in removing damaged axon debris by glia after nervous system injury, its expression is likely critical in animals that need to survive in the wild where they are exposed to various injuries, unlike in the case of the artificial environment where laboratory animals are maintained. Indeed, mutations in the human RUBCN gene are associated with a familial form of ataxia with impaired lysosomal degradation<sup>77</sup>. Based
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 87, 886, 177]]<|/det|>
+on our findings with glial Rubicon overexpression, we also expect that enhancing Rubicon function could stimulate LAP in the injured nervous system to facilitate the removal of dead brain cells and debris in acute and chronic neurodegeneration.  
+
+<|ref|>sub_title<|/ref|><|det|>[[450, 243, 545, 266]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 304, 366, 323]]<|/det|>
+## Drosophila rearing and stocks  
+
+<|ref|>text<|/ref|><|det|>[[110, 344, 888, 875]]<|/det|>
+Flies were maintained at \(25^{\circ}\mathrm{C}\) on a cornmeal- yeast- agar- dextrose medium with Nipagin as preservative. Drosophila melanogaster males 4- 8 days after eclosion were used for the experiments. \(w\) ; OK371- Gal4, UAS- mCD8- GFP \(^{78}\) was a kind gift from H. Aberle (Institute of Functional Cell Morphology, Heinrich- Heine- University, Düsseldorf). UAS- GFP- LAMP \(^{79}\) , \(Atg8a^{44 32}\) , \(Atg16^{467 80}\) , \(Atg16^{4129 80}\) , \(Atg55^{cc5 81}\) , \(Atg8a^{G116^{*}31}\) , FIP200 \(^{4130 82}\) , UAS- GFP- ref (2) \(P^{83}\) and \(3x m C h e r r y - A t g8a\) driven by its endogenous promoter \(^{43}\) were previously described. \(Atg101^{46h}\) was a kind gift of Wanzhong Ge (Institute of Genetics, Zhejiang University School of Medicine, Hangzhou, China) \(^{84}\) . The UAS- WT dRubicon::HA stock was described earlier \(^{53,54}\) and was contributed kindly by Mari Suzuki (Tokyo Metropolitan Institute of Medical Science, Japan). The following stocks were obtained from Bloomington Drosophila Stock Center: repo- Gal4 (7415), drpr \(^{45}\) (67033), Mi \(\{Trojan - QF2.2\} /VGlut^{M104979 - TQF2.2}\) (60315), QUAS- mCD8::GFP (30002), FIP200 \(^{M101469}\) (34198), \(Atg16^{M100187}\) (30656), Rubicon \(^{04462}\) (18773), \(w^{GL00094}\) (35573), \(Atg14^{HMS02025}\) (40858), UVRAG \(^{HMS01357}\) (34368), \(Atg5^{JF02703}\) (27551), \(Atg5^{HMS01244}\) (34899), \(Atg1^{GL00047}\) (35177), \(S y x17^{JF01937}\) (25896), \(w\) sgRNA TKO.GS02468 (79543), Rubicon sgRNA TKO.GS04756 (81781), UASp- mCherry- Atg8a (37750), nSyb- lexA.DBD::QF.AD, 13x lexAop2- IVS- myr::GFP (51954), 13xLexAop2- CD4::tdTomato (77139), UAS- IVS- myr::tdTomato (32221),
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 85, 886, 317]]<|/det|>
+UAS- Rab7::GFP (42705) and UAS- GFP::Rab5 (43336) \(^{85}\) . Knockdown of UVRAG by the transgenic RNAi construct UVRAG \(^{HM501357}\) has been validated previously \(^{45,50}\) . We received the following stocks from Vienna Drosophila Resource Center: Rubicon \(^{KK108247}\) , Vps34 \(^{KK107602}\) , Atg13 \(^{KK100340}\) , FIP200 \(^{KK101847}\) , Atg16 \(^{GD10140}\) (v25651) and Atg16 \(^{KK102326}\) . Standard meiotic recombination was used to generate composite transgenes on the same chromosome such as nSyb- lexA.DBD::QF.AD, lexAop2- CD4::tdTomato and Mi{Trojan- QF2.2}Vglut \(^{M104979 - TQF2.2}\) , QUAS- mCD8::GFP.  
+
+<|ref|>text<|/ref|><|det|>[[110, 331, 886, 456]]<|/det|>
+All TRiP lines were outcrossed before use to a \(w\) background (BDSC 5905) to remove the X chromosome which contained a scute mutation in many of them. This mutation in a hemizygous state is presumably responsible for a severe reduction of the number of wing margin neurons that does not occur in outcrossed stocks.  
+
+<|ref|>text<|/ref|><|det|>[[110, 470, 886, 630]]<|/det|>
+To generate the Rubicon \(^{f5}\) mutant, we crossed vas- Cas9 (VK00027, Bloomington # 51324) to a constitutively expressed single guide RNA stock (GS04756, Bloomington # 81781) that targets the Rubicon second coding exon. Single potential indel events were isolated and individual flies were tested for Rubicon indels by PCR, T7 Endonuclease I assay and sequencing. Frameshift mutants were selected and assayed for Rubicon mRNA levels. Line 2/1 was used in further experiments.  
+
+<|ref|>text<|/ref|><|det|>[[110, 666, 886, 850]]<|/det|>
+To obtain the UAS- Rubicon::mRFP1 transgene, a Gateway entry vector (kind gift of Mari Suzuki, Tokyo Metropolitan Institute of Medical Science, Japan) bearing the full length Drosophila Rubicon cDNA coding sequence (pENTR- WT dRubicon) \(^{54}\) was recombined with pTWR from the Carnegie Drosophila Gateway Vector Collection (Murphy lab) in an LR Clonase reaction according to the manufacturer's protocol (Gateway LR Clonase II Enzyme Mix, ThermoFisher
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 108, 884, 171]]<|/det|>
+Scientific). Resulting clones were sequenced and used for Drosophila transgenesis in \(w^{1118}\) background based on standard protocols for P- element transformation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 246, 361, 265]]<|/det|>
+## RNA isolation and RT-qPCR  
+
+<|ref|>text<|/ref|><|det|>[[110, 285, 888, 777]]<|/det|>
+For RNA isolation, 20 adult male carcasses or 40 wings were disrupted with a motor pestle for \(2 \mathrm{x}\) 1min in \(200 \mu \mathrm{l}\) or \(100 \mu \mathrm{l}\) of TRIReagent, respectively and thereafter supplemented with \(400 \mu \mathrm{l}\) TRI Reagent. Total RNA was extracted with the Direct- zol RNA MiniPrep (for carcasses) or Microprep (for wings) (Zymo Research). DNase I digestion was also performed. \(1 \mu \mathrm{g}\) (carcasses) or \(60 \mathrm{ng}\) (wings) total RNA was reverse transcribed in \(10 \mu \mathrm{l}\) reaction volume using the RevertAid First Strand cDNA Synthesis Kit (Thermo Scientific). RT- qPCR was performed in \(20 \mu \mathrm{l}\) reactions in technical triplicates using the PerfeCTa SYBR Green FastMix (Quantabio) with \(1 \mu \mathrm{l}\) cDNA and cycled on a Rotor- Gene Q qPCR machine (Qiagen) with the following program: \(95^{\circ} \mathrm{C}\) , \(3 \mathrm{min}\) ; 45 cycles of \(95^{\circ} \mathrm{C}\) , \(20 \mathrm{s}\) , \(58^{\circ} \mathrm{C}\) , \(20 \mathrm{s}\) and \(72^{\circ} \mathrm{C}\) , \(20 \mathrm{s}\) followed by melting curve analysis. The data was normalised by the \(\Delta \Delta \mathrm{Ct}\) method using Ribosomal protein L32 (RpL32, also known as \(rp49\) ) as an internal control. All primers were designed with Primer- BLAST (https://www.ncbi.nlm.nih.gov/tools/primer- blast) with amplicon length set to \(80 - 150 \mathrm{bp}\) and melting temperature set to \(60^{\circ} \mathrm{C}\) . One primer always spanned an exon- exon junction. The following primers were used for RT- qPCR:  
+
+<|ref|>text<|/ref|><|det|>[[115, 787, 590, 876]]<|/det|>
+rubicon fwd: 5' AGCACAAAGGAACTGGCGAAGG  rubicon rev: 5' ATTGAAGAATGACTGCTCCCTCGTG  RpL32 fwd: 5' TGCTAAGCTGTCGCACAAATGGC
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 87, 562, 106]]<|/det|>
+and \(R p L32\) rev: 5' CGATCCGTAACCGATGTTGGGC.  
+
+<|ref|>text<|/ref|><|det|>[[112, 121, 886, 177]]<|/det|>
+Normalization to \(R p L32\) was used for qPCR datasets. All \(\Delta \mathrm{Ct}\) - derived expression values were multiplied by the same scaling factor so that average of control values would equal 1 or 100.  
+
+<|ref|>sub_title<|/ref|><|det|>[[112, 242, 246, 261]]<|/det|>
+## Injury protocol  
+
+<|ref|>text<|/ref|><|det|>[[111, 283, 886, 444]]<|/det|>
+Wings were unilaterally injured by complete transection with a spring microcissor (Vannas- Tübingen) as described \(^{9,29,30}\) . Contralateral wings were left intact and served as a control for nerve integrity in absence of injury. Transection was administered approximately halfway between the tip and the hinge of the wing that left some labelled axons uninjured. Animals were incubated for the indicated time (days post- injury, dpi) after injury and wings were processed for microscopy.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 517, 377, 536]]<|/det|>
+## Microscopy and image analysis  
+
+<|ref|>text<|/ref|><|det|>[[110, 557, 886, 893]]<|/det|>
+Wings were mounted pairwise as injured and uninjured from the same animal in Halocarbon oil 27 (Sigma- Aldrich, H8773) as described \(^{9,29,30}\) . Samples were imaged immediately after mounting focusing on the proximal wing nerve \(^{9}\) . Structured illumination fluorescence microscopy was performed on an Axio Imager.M2 equipped with an ApoTome.2 structured illumination module (Zeiss) and an ORCA- Flash4.0LT CMOS camera (Hamamatsu). Illumination was provided by a CoolLED pE- 4000 system. For confocal microscopy, we used an LSM800 (Zeiss) inverted laser scanning confocal microscope. The wing nerve was imaged at room temperature using a Zeiss Plan- Apochromat 63x/1.40 NA oil immersion objective with a z- step of 0.25 \(\mu \mathrm{m}\) . The same imaging settings were used for all samples on a given microscope. Co- localization experiments (Figs. 4, 6 and 7, Fig. S5) were performed on the LSM800 confocal microscope, all other samples
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 884, 143]]<|/det|>
+were imaged on the Axio Imager.M2. Maximum intensity z- projections of wing nerve image slices spanning the whole nerve thickness are used for figures.  
+
+<|ref|>text<|/ref|><|det|>[[110, 155, 888, 632]]<|/det|>
+To quantify axon debris abundance in injured wing nerves, we first verified that debris is progressively cleared from day 2 onwards after injury<sup>9</sup>. Some intact axons are always spared after wing transection and serve as internal control for imaging quality. For debris pixel intensity quantification, we used single optical slices. The evaluator was blinded to the identity (genotype, condition) of the image files during quantification. In Fiji (https://fiji.sc), we selected two 400x100 pixel ROI- s covering wholly a section of the fragmented nerve and adjacent, non- GFP<sup>+</sup> region, respectively. We measured integrated density of freehand- selected sub- ROI- s inside the 400x100 pixel ROI of the nerve, which in total completely covered any axon debris in this region but minimally contained GFP<sup>+</sup> background and did not contain uninjured axon fluorescence or autofluorescent cuticle regions. In cases where axon fragments were completely cleared, an ROI of a size similar to previous images was used adjacent to the uninjured axons. We used the exact same selection to measure background fluorescence by measuring integrated density in the second, adjacent non- GFP<sup>+</sup> 400x100 pixel ROI. To obtain normalized pixel intensity, we subtracted the background integrated density from the nerve debris integrated density.  
+
+<|ref|>text<|/ref|><|det|>[[110, 659, 886, 819]]<|/det|>
+To quantify GFP- Ref (2)P and 3xmCherry- Atg8a puncta, single slice images were taken from the same section of the wing nerve where axon debris was quantified. The number of puncta which were clearly distinct from the background and well circumscribed, were counted in Fiji in an ROI of 400x200 pixels with the Cell Counter plugin. The evaluator was blinded to the identity (genotype, condition) of the image files during quantification.  
+
+<|ref|>text<|/ref|><|det|>[[111, 848, 857, 903]]<|/det|>
+To quantify co- localizing puncta, we evaluated either glial GFP- LAMP1 and CD4::tdTomato<sup>+</sup> axon debris co- localization, or glial Rubicon::mRFP1 and mCD8::GFP+ axon debris
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 85, 880, 492]]<|/det|>
+colocalization or myr::GFP<sup>+</sup> axon debris and glial mCherry- Atg8a co- localization in the wing nerve. We selected 5 consecutive single confocal slices where co- localizing puncta were counted in a bounding box of \(40 \times 12 \mu \mathrm{m}\) in Fiji. Co- localization was scored on well- circumscribed structures where dimensions of GFP- LAMP1<sup>+</sup>, Rubicon::mRFP1<sup>+</sup> and mCherry- Atg8a<sup>+</sup> puncta, respectively, were identical to or slightly larger than the dimensions of the axon fragment that showed full co- localization with them. The sum of co- localization events in 5 slices is shown. To evaluate Rab7- GFP colocalization with Rubicon::mRFP1, a single confocal slice was selected where co- localizing puncta were counted and measured for area and mean pixel intensity in a bounding box of \(20 \times 12 \mu \mathrm{m}\) in Fiji. The evaluator was blinded to the identity (genotype, condition) of the image files during quantification. Line plots showing pixel intensity profiles over a distance in color were made with the help of the Fiji plugin RGB Profiler (Christophe Laummonerie, Jerome Mutterer).  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 550, 270, 567]]<|/det|>
+## Statistical analysis  
+
+<|ref|>text<|/ref|><|det|>[[110, 590, 889, 894]]<|/det|>
+Experiments were repeated twice and used \(n \geq 3\) independent biological replicates. Measurements were taken from distinct biological samples for individual data points. Error bars and number of data points (biological replicates) for each experiment are defined in the figures and figure legends. For debris intensity data, truncated violin plots are shown with median and quartiles containing all data points. To test for normal distribution of data, we used the Shapiro- Wilk normality test \((\alpha = 0.05)\) . Normally distributed datasets were compared pairwise with unpaired, two- tailed Student's t- test and samples in which at least one dataset did not show normal distribution were compared pairwise with unpaired, two- tailed Mann- Whitney test. For more than two datasets, Kruskal- Wallis test for non- normally distributed data and one- way ANOVA for normally
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 88, 886, 283]]<|/det|>
+distributed data were applied with post- hoc tests (Dunn's and Fisher's LSD, respectively). \(\alpha = 0.05\) testing level was applied except when correction for multiple comparisons was used. p- values are indicated in the figures. Prism 9.3.1 (GraphPad) was used for statistical analysis and graph generation. No data points were excluded from statistical analysis. Sample size was not predetermined but was similar as in other publications describing axon debris engulfment and autophagy<sup>17,19,31,32,40</sup>.  
+
+<|ref|>sub_title<|/ref|><|det|>[[410, 325, 586, 347]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[111, 384, 803, 430]]<|/det|>
+All data needed to evaluate the conclusions in this paper are present in the paper and its Supplementary Information.  
+
+<|ref|>sub_title<|/ref|><|det|>[[390, 465, 606, 488]]<|/det|>
+## Material availability  
+
+<|ref|>text<|/ref|><|det|>[[111, 523, 856, 560]]<|/det|>
+All materials, \*Drosophila\* stocks and related information are available from the corresponding authors upon reasonable request.  
+
+<|ref|>sub_title<|/ref|><|det|>[[438, 609, 558, 631]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[110, 666, 866, 703]]<|/det|>
+1. Jung, Y.-J. & Chung, W.-S. Phagocytic Roles of Glial Cells in Healthy and Diseased Brains. \*Biomol Ther\* 26, 350-357 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 718, 876, 755]]<|/det|>
+2. Raiders, S. \*et al.\* Engulfed by Glia: Glial Pruning in Development, Function, and Injury across Species. \*J Neurosci\* 41, 823-833 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[110, 771, 878, 808]]<|/det|>
+3. Sapar, M. L. & Han, C. Die in pieces: How Drosophila sheds light on neurite degeneration and clearance. \*J Genet Genomics\* 46, 187-199 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 823, 864, 860]]<|/det|>
+4. Pekny, M. & Pekna, M. Reactive gliosis in the pathogenesis of CNS diseases. \*Biochimica Et Biophysica Acta BBA - Mol Basis Dis\* 1862, 483-491 (2016).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 87, 857, 125]]<|/det|>
+5. Hilu-Dadia, R. & Kurant, E. Glial phagocytosis in developing and mature Drosophila CNS: tight regulation for a healthy brain. Curr Opin Immunol 62, 62–68 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[110, 140, 840, 177]]<|/det|>
+6. Freeman, M. R. Drosophila Central Nervous System Glia. CSH Perspect Biol 7, a020552 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[110, 192, 872, 229]]<|/det|>
+7. Logan, M. A. & Speese, S. D. Axon Degeneration, Methods and Protocols. Methods Mol Biol 2143, 321–338 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[110, 244, 860, 281]]<|/det|>
+8. Doherty, J., Logan, M. A., Taşdemir, Ö. E. & Freeman, M. R. Ensheathing Glia Function as Phagocytes in the Adult Drosophila Brain. J Neurosci 29, 4768–4781 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[110, 297, 883, 351]]<|/det|>
+9. Neukomm, L. J., Burdett, T. C., Gonzalez, M. A., Züchner, S. & Freeman, M. R. Rapid in vivo forward genetic approach for identifying axon death genes in Drosophila. Proc National Acad Sci 111, 9965–9970 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[110, 367, 867, 404]]<|/det|>
+10. MacDonald, J. M. et al. The Drosophila Cell Corpse Engulfment Receptor Draper Mediates Glial Clearance of Severed Axons. Neuron 50, 869–881 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[110, 419, 833, 457]]<|/det|>
+11. Sapar, M. L. et al. Phosphatidylserine Externalization Results from and Causes Neurite Degeneration in Drosophila. Cell Reports 24, 2273–2286 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 472, 875, 509]]<|/det|>
+12. Tung, T. T. et al. Phosphatidylserine recognition and induction of apoptotic cell clearance by Drosophila engulfment receptor Draper. J Biochem 153, 483–491 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[110, 524, 833, 561]]<|/det|>
+13. Williamson, A. P. & Vale, R. D. Spatial Control of Draper Receptor Signaling Initiates Apoptotic Cell Engulfment. J Cell Biol 217, 3977–3992 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 577, 839, 614]]<|/det|>
+14. Lu, T.-Y. et al. Axon degeneration induces glial responses through Draper-TRAF4-JNK signalling. Nat Commun 8, 14355 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[110, 629, 875, 666]]<|/det|>
+15. Doherty, J. et al. PI3K Signaling and Stat92E Converge to Modulate Glial Responsiveness to Axonal Injury. PLoS Biol 12, e1001985 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[110, 682, 875, 736]]<|/det|>
+16. MacDonald, J. M., Doherty, J., Hackett, R. & Freeman, M. R. The c-Jun kinase signaling cascade promotes glial engulfment activity through activation of draper and phagocytic function. Cell Death Differ 20, 1140–1148 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[110, 752, 881, 806]]<|/det|>
+17. Lu, T.-Y., Doherty, J. & Freeman, M. R. DRK/DOS/SOS converge with Crk/Mbc/dCed-12 to activate Rac1 during glial engulfment of axonal debris. Proc. Natl. Acad. Sci. U.S.A. 111, 12544–12549 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[110, 822, 844, 859]]<|/det|>
+18. Logan, M. A. et al. Negative regulation of glial engulfment activity by Draper terminates glial responses to axon injury. Nat Neurosci 15, 722–730 (2012).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[108, 87, 858, 125]]<|/det|>
+19. Ziegenfuss, J. S., Doherty, J. & Freeman, M. R. Distinct molecular pathways mediate glial activation and engulfment of axonal debris after axotomy. Nat Neurosci 15, 979-987 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[110, 140, 852, 177]]<|/det|>
+20. Morishita, H. & Mizushima, N. Diverse Cellular Roles of Autophagy. Annu Rev Cell Dev Biol 35, 1-23 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 192, 875, 229]]<|/det|>
+21. Galluzzi, L. & Green, D. R. Autophagy-Independent Functions of the Autophagy Machinery. Cell 177, 1682-1699 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 244, 844, 281]]<|/det|>
+22. Jülg, J., Strohm, L. & Behrends, C. Canonical and non-canonical autophagy pathways in microglia. Mol Cell Biol 41, e00389-20 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[110, 297, 790, 333]]<|/det|>
+23. Sung, K. & Jimenez-Sanchez, M. Autophagy in Astrocytes and its Implications in Neurodegeneration. J Mol Biol 432, 2605-2621 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[110, 349, 832, 386]]<|/det|>
+24. Gohel, R., Kournoutis, A., Petridi, S. & Nezis, I. P. Molecular mechanisms of selective autophagy in Drosophila. Int Rev Cell Mol Biol 354, 63-105 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[110, 401, 880, 438]]<|/det|>
+25. Zhang, K., Jiang, M. & Fang, Y. The Drama of Wallerian Degeneration: The Cast, Crew, and Script. Annu Rev Genet 55, 1-21 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[110, 454, 875, 508]]<|/det|>
+26. Jenzer, C. et al. Autophagy mediates phosphatidylserine exposure and phagosome degradation during apoptosis through specific functions of GABARAP/LGG-1 and LC3/LGG-2. Autophagy 15, 228-241 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 524, 875, 561]]<|/det|>
+27. Peña-Ramos, O. et al. Autophagosomes fuse to phagosomes and facilitate the degradation of apoptotic cells in Caenorhabditis elegans. eLife 11, e72466 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[110, 577, 870, 613]]<|/det|>
+28. Deretic, V. & Lazarou, M. A guide to membrane atg8ylation and autophagy with reflections on immunity. J Cell Biology 221, e202203083 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[110, 629, 875, 683]]<|/det|>
+29. Fang, Y., Soares, L., Teng, X., Geary, M. & Bonini, N. M. A Novel Drosophila Model of Nerve Injury Reveals an Essential Role of Nmnat in Maintaining Axonal Integrity. Curr Biol 22, 590-595 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[110, 699, 872, 752]]<|/det|>
+30. Paglione, M., Rosell, A. L., Chatton, J.-Y. & Neukomm, L. J. Morphological and Functional Evaluation of Axons and their Synapses during Axon Death in Drosophila melanogaster. J Vis Exp e60865 (2020) doi:10.3791/60865.  
+
+<|ref|>text<|/ref|><|det|>[[110, 769, 863, 805]]<|/det|>
+31. Jipa, A. et al. Analysis of Drosophila Atg8 proteins reveals multiple lipidation-independent roles. Autophagy 17, 2565-2575 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[110, 821, 850, 858]]<|/det|>
+32. Pircs, K. et al. Advantages and Limitations of Different p62-Based Assays for Estimating Autophagic Activity in Drosophila. PLoS One 7, e44214 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[110, 874, 819, 893]]<|/det|>
+33. Romao, S. & Münz, C. LC3-associated phagocytosis. Autophagy 10, 526-528 (2014).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 87, 856, 125]]<|/det|>
+34. Heckmann, B. L. & Green, D. R. LC3-associated phagocytosis at a glance. J Cell Sci 132, jcs231472 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 140, 870, 177]]<|/det|>
+35. Martinez, J. LAP it up, fuzz ball: a short history of LC3-associated phagocytosis. Curr Opin Immunol 55, 54-61 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[111, 192, 860, 247]]<|/det|>
+36. Rai, S. et al. The ATG5-binding and coiled coil domains of ATG16L1 maintain autophagy and tissue homeostasis in mice independently of the WD domain required for LC3-associated phagocytosis. Autophagy 15, 1-14 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[111, 262, 861, 299]]<|/det|>
+37. Fletcher, K. et al. The WD40 domain of ATG16L1 is required for its non-canonical role in lipidation of LC3 at single membranes. EMBO J 37, e97840 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 314, 874, 352]]<|/det|>
+38. Nagy, P. et al. Drosophila Atg16 promotes enteroendocrine cell differentiation via regulation of intestinal Slit/Robo signaling. Development 144, 3990-4001 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[110, 367, 877, 405]]<|/det|>
+39. Ziegenfuss, J. S. et al. Draper-dependent glial phagocytic activity is mediated by Src and Syk family kinase signalling. Nature 453, 935-939 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[110, 419, 852, 457]]<|/det|>
+40. Takats, S. et al. Autophagosomal Syntaxin17-dependent lysosomal degradation maintains neuronal function in Drosophila. J Cell Biol 201, 531-539 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[110, 472, 881, 509]]<|/det|>
+41. Bhukel, A. et al. Autophagy within the mushroom body protects from synapse aging in a non-cell autonomous manner. Nat Commun 10, 1318 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 524, 870, 561]]<|/det|>
+42. Ohashi, Y., Tremel, S. & Williams, R. L. VPS34 complexes from a structural perspective. J Lipid Res 60, 229-241 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 577, 805, 614]]<|/det|>
+43. Hegedüs, K. et al. The Ccz1-Mon1-Rab7 module and Rab5 control distinct steps of autophagy. Mol Biol Cell 27, 3132-3142 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[110, 629, 875, 666]]<|/det|>
+44. Takats, S. et al. Interaction of the HOPS complex with Syntaxin 17 mediates autophagosome clearance in Drosophila. Mol Biol Cell 25, 1338-1354 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[111, 681, 855, 736]]<|/det|>
+45. Lőrincz, P. et al. Atg6/UVRAG/Vps34-Containing Lipid Kinase Complex Is Required for Receptor Downregulation through Endolysosomal Degradation and Epithelial Polarity during Drosophila Wing Development. Biomed Res Int 2014, 1-19 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[110, 752, 863, 789]]<|/det|>
+46. Csizmadia, T. et al. Molecular mechanisms of developmentally programmed crinophagy in Drosophila. J Cell Biol 217, 361-374 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 805, 872, 859]]<|/det|>
+47. Issman-Zecharya, N. & Schuldiner, O. The PI3K Class III Complex Promotes Axon Pruning by Downregulating a Ptc-Derived Signal via Endosome-Lysosomal Degradation. Dev Cell 31, 461-473 (2014).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 87, 879, 125]]<|/det|>
+48. Zhong, Y. et al. Distinct regulation of autophagic activity by Atg14L and Rubicon associated with Beclin 1-phosphatidylinositol-3-kinase complex. Nat Cell Biol 11, 468-476 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[110, 140, 833, 177]]<|/det|>
+49. Matsunaga, K. et al. Two Beclin 1-binding proteins, Atg14L and Rubicon, reciprocally regulate autophagy at different stages. Nat Cell Biol 11, 385-396 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[110, 192, 864, 229]]<|/det|>
+50. Nagy, P., Kovacs, L., Sandor, G. O. & Juhasz, G. Stem-cell-specific endocytic degradation defects lead to intestinal dysplasia in Drosophila. Dis Model Mech 9, 501-512 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[110, 244, 872, 281]]<|/det|>
+51. Wong, S., Sil, P. & Martinez, J. Rubicon: LC3-associated phagocytosis and beyond. FEBS J 285, 1379-1388 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 297, 877, 334]]<|/det|>
+52. Martinez, J. et al. Molecular characterization of LC3-associated phagocytosis reveals distinct roles for Rubicon, NOX2 and autophagy proteins. Nat Cell Biol 17, 893-906 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[110, 350, 854, 387]]<|/det|>
+53. Nakamura, S. et al. Suppression of autophagic activity by Rubicon is a signature of aging. Nat Commun 10, 847 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 402, 875, 440]]<|/det|>
+54. Oba, M., Fukui, K., Sango, K. & Suzuki, M. Dataset on the effect of Rubicon overexpression on polyglutamine-induced locomotor dysfunction in Drosophila. Data Brief 37, 107222 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[110, 455, 860, 491]]<|/det|>
+55. Kinchen, J. M. et al. A pathway for phagosome maturation during engulfment of apoptotic cells. Nat Cell Biol 10, 556-566 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[110, 507, 850, 544]]<|/det|>
+56. Tabata, K. et al. Rubicon and PLEKHM1 Negatively Regulate the Endocytic/Autophagic Pathway via a Novel Rab7-binding Domain. Mol Biol Cell 21, 4162-4172 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[110, 560, 825, 596]]<|/det|>
+57. Sun, Q., Westphal, W., Wong, K. N., Tan, I. & Zhong, Q. Rubicon controls endosome maturation as a Rab7 effector. Proc National Acad Sci 107, 19338-19343 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[110, 612, 872, 666]]<|/det|>
+58. Lu, N. et al. Two PI 3-Kinases and One PI 3-Phosphatase Together Establish the Cyclic Waves of Phagosomal PtDIns(3)P Critical for the Degradation of Apoptotic Cells. PLoS Biol 10, e1001245 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[110, 682, 852, 719]]<|/det|>
+59. Heckmann, B. L. et al. LC3-Associated Endocytosis Facilitates \(\beta\) -Amyloid Clearance and Mitigates Neurodegeneration in Murine Alzheimer's Disease. Cell 178, 536-551.e14 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[110, 735, 841, 789]]<|/det|>
+60. Lee, J.-W. et al. TLR4 (toll-like receptor 4) activation suppresses autophagy through inhibition of FOXO3 and impairs phagocytic capacity of microglia. Autophagy 15, 753-770 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[110, 805, 844, 841]]<|/det|>
+61. Wakida, N. M. et al. Diminished LC3-Associated Phagocytosis by Huntington's Disease Striatal Astrocytes. J Huntington's Dis 11, 25-33 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[110, 857, 872, 894]]<|/det|>
+62. Heckmann, B. L. et al. Noncanonical function of an autophagy protein prevents spontaneous Alzheimer's disease. Sci Adv 6, eabb9036 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 880, 141]]<|/det|>
+63. Desjardins, M., Huber, L., Parton, R. & Griffiths, G. Biogenesis of phagolysosomes proceeds through a sequential series of interactions with the endocytic apparatus. J Cell Biology 124, 677–688 (1994).  
+
+<|ref|>text<|/ref|><|det|>[[111, 158, 857, 211]]<|/det|>
+64. Scianimanico, S. et al. Impaired recruitment of the small GTPase rab7 correlates with the inhibition of phagosome maturation by Leishmania donovani promastigotes. Cell Microbiol 1, 19–32 (1999).  
+
+<|ref|>text<|/ref|><|det|>[[111, 228, 805, 263]]<|/det|>
+65. Mayorga, L. S., Bertini, F. & Stahl, P. D. Fusion of newly formed phagosomes with endosomes in intact cells and in a cell-free system. JBC 266, 6511–6517 (1991).  
+
+<|ref|>text<|/ref|><|det|>[[111, 280, 850, 333]]<|/det|>
+66. Harrison, R. E., Bucci, C., Vieira, O. V., Schroer, T. A. & Grinstein, S. Phagosomes Fuse with Late Endosomes and/or Lysosomes by Extension of Membrane Protrusions along Microtubules: Role of Rab7 and RILP. Mol Cell Biol 23, 6494–6506 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[111, 350, 876, 386]]<|/det|>
+67. Etchegaray, J. I. et al. Defective Phagocytic Corpse Processing Results in Neurodegeneration and Can Be Rescued by TORC1 Activation. J Neurosci 36, 3170–3183 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[111, 403, 857, 438]]<|/det|>
+68. Berglund, R. et al. Microglial autophagy–associated phagocytosis is essential for recovery from neuroinflammation. Sci Immunol 5, eabb5077 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[111, 455, 875, 490]]<|/det|>
+69. Kim, H.-J. et al. Deficient autophagy in microglia impairs synaptic pruning and causes social behavioral defects. Mol Psychiatry 22, 1576–1584 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[111, 507, 805, 543]]<|/det|>
+70. Heckmann, B. L., Boada-Romero, E., Cunha, L. D., Magne, J. & Green, D. R. LC3-Associated Phagocytosis and Inflammation. J Mol Biol 429, 3561–3576 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[111, 560, 850, 595]]<|/det|>
+71. Kim, J.-Y. et al. Noncanonical Autophagy Promotes the Visual Cycle. Cell 154, 365–376 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[111, 613, 878, 666]]<|/det|>
+72. Muniz-Feliciano, L., Doggett, T. A., Zhou, Z. & Ferguson, T. A. RUBCN/rubicon and EGFR regulate lysosomal degradative processes in the retinal pigment epithelium (RPE) of the eye. Autophagy 13, 00–00 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[111, 683, 880, 718]]<|/det|>
+73. Martinez, J. et al. Noncanonical autophagy inhibits the autoinflammatory, lupus-like response to dying cells. Nature 533, 115–119 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[111, 735, 867, 770]]<|/det|>
+74. Gordon, R. A. et al. Rubicon promotes rather than restricts murine lupus and is not required for LC3-associated phagocytosis. Jci Insight 7, (2022).  
+
+<|ref|>text<|/ref|><|det|>[[111, 787, 860, 822]]<|/det|>
+75. Wan, J. et al. LC3-associated phagocytosis protects against inflammation and liver fibrosis via immunoreceptor inhibitory signaling. Sci Transl Med 12, eaaw8523 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[111, 840, 863, 875]]<|/det|>
+76. DiSabato, D. J., Quan, N. & Godbout, J. P. Neuroinflammation: the devil is in the details. J Neurochem 139, 136–153 (2016).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 87, 812, 125]]<|/det|>
+77. Assoum, M. et al. Rundataxin, a novel protein with RUN and diacylglycerol binding domains, is mutant in a new recessive ataxia. Brain 133, 2439–2447 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[112, 140, 876, 195]]<|/det|>
+78. Mahr, A. & Aberle, H. The expression pattern of the Drosophila vesicular glutamate transporter: A marker protein for motoneurons and glutamatergic centers in the brain. Gene Expr Patterns 6, 299–309 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[112, 210, 865, 247]]<|/det|>
+79. Pulipparacharuvil, S. et al. Drosophila Vps16A is required for trafficking to lysosomes and biogenesis of pigment granules. J Cell Sci 118, 3663–3673 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[112, 262, 877, 300]]<|/det|>
+80. Varga, K. et al. Loss of Atg16 delays the alcohol-induced sedation response via regulation of Corazonin neuropeptide production in Drosophila. Sci Rep 6, 34641 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[112, 315, 872, 352]]<|/det|>
+81. Kim, M. et al. Mutation in ATG5 reduces autophagy and leads to ataxia with developmental delay. eLife 5, e12245 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[112, 367, 852, 405]]<|/det|>
+82. Nagy, P. et al. Atg17/FIP200 localizes to perilysosomal Ref(2)P aggregates and promotes autophagy by activation of Atg1 in Drosophila. Autophagy 10, 453–467 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[112, 419, 825, 457]]<|/det|>
+83. Chang, Y.-Y. & Neufeld, T. P. An Atg1/Atg13 Complex with Multiple Roles in TOR-mediated Autophagy Regulation. Mol Biol Cell 20, 2004–2014 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[112, 472, 860, 510]]<|/det|>
+84. Guo, T. et al. The autophagy-related gene Atg101 in Drosophila regulates both neuron and midgut homeostasis. J Biol Chem 294, 5666–5676 (2019).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[393, 125, 604, 148]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[113, 185, 451, 204]]<|/det|>
+Funding: This research was supported by:  
+
+<|ref|>text<|/ref|><|det|>[[140, 219, 888, 672]]<|/det|>
+- the Young Researchers' Excellence Programme of the National Research, Development and Innovation Office (NRDIO) (FK132183) (AS)- the Janos Bolyai Research Scholarship of the Hungarian Academy of Sciences (BO/00078/18) (AS)- the NRDIO New National Excellence Programme (ÚNKP-20-5, ÚNKP-19-4, ÚNKP-18-4) (AS)- the Swiss National Science Foundation SNSF Assistant Professor award (176855) (LJN)- the International Foundation for Research in Paraplegia (P180) (LJN)- SNSF Spark (190919) (LJN)- the NRDIO grant KKP129797 (GJ)- the NRDIO grant GINOP-2.3.2-15-2016-00032 (GJ)- and the Biotechnology National Laboratory program of the National Research, Development and Innovation Office (NKFIH-871-3/2020) (GJ).  
+
+<|ref|>text<|/ref|><|det|>[[111, 718, 888, 844]]<|/det|>
+We thank especially Masaki Oba, Koji Fukui, Kazunori Sango and Mari Suzuki for the UAS- WT dRubicon::HA stock and the Rubicon Gateway entry clone, Anna Galambos, Robert Soltész and Dániel Bócsi for their contribution to preliminary experiments, the Bloomington Stock Center, the Vienna Drosophila Resource Center, Wanzhong Ge, Hermann Aberle for fly stocks and Szilvia
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 884, 141]]<|/det|>
+Bozsó and Ildikó Kresákné Erdődi for technical assistance. We are also grateful to Gábor Csordás, Arindam Bhattacherjee and Tamás Maruzs for fruitful discussions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 209, 294, 225]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[111, 256, 886, 380]]<|/det|>
+Conceptualization: AS and GJ, Investigation: AS, VV, SB, PB, AJ and KEV, Resources: AJ and LN, Results interpretation: AS, VV, SB, KEV, PB, AJ, LN and GJ, Writing: AS and GJ, Manuscript revision: AS, VV, SB, KEV, PB, AJ, LN and GJ, Supervision: AS and GJ, Funding acquisition: AS and GJ.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 432, 283, 450]]<|/det|>
+## Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[112, 482, 571, 500]]<|/det|>
+The authors declare that they have no competing interests.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[148, 131, 796, 789]]<|/det|>
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[112, 88, 884, 142]]<|/det|>
+# Figure 1. The Atg8a conjugation system participates in axon debris clearance in the wing nerve  
+
+<|ref|>text<|/ref|><|det|>[[110, 171, 888, 613]]<|/det|>
+(A) Schematic of the fly wing injury model. Neuronal cell bodies (cb) project axons through the L1 vein that are imaged proximally (blue dashed rectangle for field of view). Axotomy leaves most axons injured that degenerate and are cleared during the following days. (B) Time course analysis of axon debris removal in the Drosophila wing nerve. z-projections of fragmented nerve bundles are shown for the indicated genotypes on OK371> UAS-mCD8::GFP/+ background at 2, 5 and 10 days post wing injury (dpi). (D) z-projections of degenerating wing nerves of the indicated genotypes at 5 dpi. (C,E) Quantification of axon debris abundance in single-slice images of genotypes in (B) and (D), respectively. Truncated violin plots with median and quartiles shown. Independent experiments are separated by a dashed line. (C) One-way ANOVA and Kruskal-Wallis test was used for statistical analysis for 5 and 10 dpi datasets, respectively, and unpaired, two-tailed t-test for the 2 dpi dataset. n=13, 14, 16, 16, 14, 13, 14 and 12. (E) Unpaired, two-tailed Mann-Whitney test. n=10 and 12. Uninjured axons in injured L1 veins are denoted with arrowheads. ns - not significant. Scale bar: 5 μm.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 95, 770, 770]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 137, 884, 193]]<|/det|>
+Figure 2. Atg1 kinase complex is dispensable for axon debris elimination but Atg8a lipidation and a functional Atg16 WD40 domain is required for this process  
+
+<|ref|>text<|/ref|><|det|>[[110, 220, 888, 596]]<|/det|>
+(A,B,F,H) z- projections of degenerating wing nerves of the indicated genotypes on OK371> UAS- mCD8::GFP/+ background at 5 dpi. Scale bar: 5 μm. (C,G,I) Quantification of axon debris abundance in single- slice images of the indicated genotypes at 5 dpi. (C) Kruskal- Wallis test (Atg8a and Atg101 mutants) and unpaired, two- tailed Mann- Whitney test (FIP200/Atg17 mutant), n=8, 11, 11, 11 and 13. (G) unpaired, two- tailed Mann- Whitney test, n= 12 and 11, (I) Kruskal- Wallis test, n=9, 13, 16 and 12. (D) Single- slice images of Atg8a promoter- 3xmCherry- Atg8a- expressing uninjured wing L1 veins of the indicated genotypes. Scale bar: 10 μm. (E) Quantification of the number of 3xmCherry Atg8a<sup>+</sup> puncta in single- slice images of genotypes in (D). Unpaired, two- tailed t- test (Atg101 and Atg17/FIP200 mutants) and Mann- Whitney test (Atg5 mutant) were used for statistics. n= 9, 9, 13, 13, 9 and 9. ns – not significant. (C,E) Independent experiments are separated by a dashed line. Truncated violin plots with median and quartiles shown.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[135, 120, 163, 142]]<|/det|>
+A 
+
+<|ref|>text<|/ref|><|det|>[[164, 160, 441, 191]]<|/det|>
+VGlut>QUAS-mCD8::GFP, repo-Gal4
+injured 5 dpi 
+
+<|ref|>image<|/ref|><|det|>[[122, 191, 477, 330]]<|/det|>
+ 
+
+<|ref|>sub_title<|/ref|><|det|>[[135, 411, 163, 433]]<|/det|>
+C 
+
+<|ref|>text<|/ref|><|det|>[[164, 460, 441, 492]]<|/det|>
+VGlut>QUAS-mCD8::GFP, repo-Gal4
+injured 5 dpi 
+
+<|ref|>image<|/ref|><|det|>[[122, 500, 477, 568]]<|/det|>
+ 
+
+<|ref|>sub_title<|/ref|><|det|>[[135, 636, 163, 658]]<|/det|>
+E 
+
+<|ref|>text<|/ref|><|det|>[[202, 640, 388, 671]]<|/det|>
+repo>UAS-GFP-Ref(2)P
+uninjured 
+
+<|ref|>image<|/ref|><|det|>[[122, 671, 480, 840]]<|/det|>
+ 
+
+<|ref|>sub_title<|/ref|><|det|>[[494, 120, 519, 142]]<|/det|>
+B 
+
+<|ref|>image<|/ref|><|det|>[[511, 160, 808, 375]]<|/det|>
+ 
+
+<|ref|>text<|/ref|><|det|>[[625, 390, 710, 404]]<|/det|>
+glial RNAi 
+
+<|ref|>text<|/ref|><|det|>[[585, 409, 796, 441]]<|/det|>
+VGlut>QUAS-mCD8::GFP, repo-Gal4
+remaining axon debris
+5 dpi 
+
+<|ref|>image<|/ref|><|det|>[[524, 450, 819, 905]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 180, 888, 490]]<|/det|>
+(A,C) z-projections of degenerating wing nerves expressing the indicated RNAi- s in glia, driven by repo- Gal4, on VGlut> QUAS- mCD8::GFP/+ background at 5 dpi. (B,D) Quantification of axon debris abundance in single- slice images of the indicated genotypes in (A) and (C), respectively, Kruskal- Wallis test. (A) n=13, 13, 15, 12, 13 and 11, (C) n=11 and 9. (E) Single- slice images of uninjured wing L1 vein glia expressing repo- Gal4- driven UAS- GFP- ref (2)P and co- expressing the indicated RNAi- s. (F) Quantification of the number of GFP- Ref (2)P puncta in single- slice images of genotypes in (E), one- way ANOVA. n=10, 9, 10, 10, 6 and 11. ns – not significant. w RNAi serves as a negative control. Truncated violin plots with median and quartiles shown. Scale bar: 5 μm.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[137, 92, 794, 580]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 180, 888, 560]]<|/det|>
+(A) Confocal single-slice images of injured wing nerves at 3 dpi with or without the expression of Atg16 RNAi in glia with repo-Gal4, glial lysosomes labelled with GFP-LAMP1, axons with membrane-targeted tdTomato (repo-Gal4>UAS-GFP-LAMP1, nSyb> lexAop-CD4::tdTomato). Magnified images of the area outlined by the dashed rectangle are shown below. Dashed line indicates distance used for intensity profile generation. Pixel intensity plots over the indicated distance are shown below for each channel to estimate colocalization. Arrowheads point to co-localizing puncta. Scale bar: 2 μm. (B) Quantification of the number of different co-localizing GFP-LAMP1 and CD4::tdTomato puncta in 5 consecutive single-slice images of genotypes in (A). Co-localizing puncta were counted in a bounding box of 40 x12 μm. Statistical analysis was performed with unpaired, two-tailed Mann-Whitney test. The graph shows the median with 95% confidence intervals. n=10.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[128, 90, 390, 385]]<|/det|>
+
+
+<|ref|>image<|/ref|><|det|>[[511, 144, 775, 343]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[616, 344, 685, 356]]<|/det|>
+<center>glial RNAi</center> 
+
+<|ref|>image_caption<|/ref|><|det|>[[220, 408, 333, 425]]<|/det|>
+<center>injured 5 dpi</center> 
+
+<|ref|>image<|/ref|><|det|>[[168, 427, 390, 536]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[550, 408, 690, 425]]<|/det|>
+<center>injured 5 dpi</center> 
+
+<|ref|>image<|/ref|><|det|>[[520, 427, 740, 536]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[170, 540, 191, 558]]<|/det|>
+<center>D</center> 
+
+<|ref|>image<|/ref|><|det|>[[172, 590, 384, 707]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[460, 560, 483, 578]]<|/det|>
+<center>F</center> 
+
+<|ref|>image<|/ref|><|det|>[[511, 590, 725, 707]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[560, 540, 719, 565]]<|/det|>
+<center>VGlut>QUAS-mCD8::GFP, repo>UAS-Cas9 remaining axon debris 5 dpi</center> 
+
+<|ref|>image_caption<|/ref|><|det|>[[160, 737, 181, 755]]<|/det|>
+<center>G</center> 
+
+<|ref|>image_caption<|/ref|><|det|>[[270, 737, 370, 753]]<|/det|>
+<center>injured 3 dpi</center> 
+
+<|ref|>image<|/ref|><|det|>[[168, 755, 440, 864]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[460, 737, 483, 755]]<|/det|>
+<center>H</center> 
+
+<|ref|>image<|/ref|><|det|>[[511, 757, 732, 895]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 186, 888, 700]]<|/det|>
+(A) z-projections of degenerating wing nerves at 5 dpi expressing the indicated RNAi-s in glia, driven by repo-Gal4 on VGlut> QUAS-mCD8::GFP/+ background. (B) Quantification of axon debris abundance in single-slice images of the indicated genotypes as in (A) at 5 dpi, Kruskal-Wallis test. n= 13, 11, 13, 16 and 10. (C,E) z-projections of degenerating wing nerves at 5 dpi in either the Rubicon<sup>6</sup> mutant (C) or in a stock expressing Rubicon single guide RNA universally and UAS-Cas9 with repo-Gal4 in glia (E). OK371> UAS-mCD8::GFP (C) and VGlut> QUAS-mCD8::GFP (E) are used for axon labelling. (D, F) Quantification of axon debris abundance in single-slice images of the indicated genotypes as in (C) and (E), respectively at 5 dpi. Unpaired, two-tailed Mann-Whitney test (D), n=14 and 10, and t-test (F), n=12 and 14 were used for statistics. (G) z-projections of degenerating wing nerves at 3 dpi expressing the indicated proteins in glia, driven by repo-Gal4 on VGlut>QUAS-mCD8::GFP/+ background. OE- overexpression. (H) Quantification of axon debris abundance in single-slice images of the indicated genotypes as in (G) at 3 dpi, unpaired, two-tailed Mann-Whitney test. n= 13. ns – not significant. w RNAi and w single guide RNA serve as negative controls. Scale bar: 5 μm. Truncated violin plots with median and quartiles shown.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[108, 90, 803, 830]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 137, 885, 193]]<|/det|>
+Figure 6. Glial Rubicon is recruited to axon debris and colocalizes with large Rab7-positive phagosomes after injury  
+
+<|ref|>text<|/ref|><|det|>[[110, 220, 888, 594]]<|/det|>
+(A) Confocal single-slice images of uninjured and contralateral injured wing nerves at 1 and 2 dpi expressing Rubicon::mRFP1 in glia. Axons are labelled with mCD8::GFP (repo-Gal4> UAS-Rubicon::mRFP1, VGlut> QUAS-mCD8::GFP). (B) Confocal single-slice images of uninjured and injured wing nerves at 2 dpi expressing Rubicon::mRFP1 and Rab7::GFP in glia (repo-Gal4> UAS-Rubicon::mRFP1, >UAS- Rab7::GFP). (A,B) Magnified images of the areas outlined by the dashed rectangles are shown below. Arrowheads point to co-localizing puncta. Scale bar: 5 μm. (C) Single-slice quantification of puncta area, number and mean pixel intensity for the indicated single and double positive vesicles as in (B). GFP<sup>+</sup> RFP<sup>+</sup> puncta, Rubicon-RFP<sup>+</sup> puncta and Rab7-GFP<sup>+</sup> puncta: unpaired, two-tailed t-test, n=11 and 15, mean with standard deviation; GFP<sup>+</sup> RFP<sup>+</sup> punctum size and RFP intensity: two-tailed Mann-Whitney test, n=233 and 162. Truncated violin plots with median and quartiles.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[130, 140, 780, 360]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[140, 371, 164, 393]]<|/det|>
+<center>B</center> 
+
+<|ref|>image<|/ref|><|det|>[[195, 385, 711, 604]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 135, 884, 193]]<|/det|>
+Figure 7. Atg8a is recruited to LAPosomes containing engulfed axon debris, which accumulate upon loss of Vps34  
+
+<|ref|>text<|/ref|><|det|>[[110, 220, 888, 560]]<|/det|>
+(A) Confocal single-slice images of injured wing nerves at 2 dpi without or with Vps34 RNAi in glia with repo-Gal4. Glia also express mCherry-Atg8a. Axons are labelled with myr::GFP (repo-Gal4>UAS-mCherry-Atg8a, nSyb> lexAop-myr::GFP). Magnified images of the areas outlined by the rectangles are shown below. Dashed line indicates distance used for intensity profile generation. Pixel intensity plots over the indicated distance are shown below for each channel to demonstrate colocalization. Arrowheads point to co-localizing puncta. Scale bar: 2 μm. (B) Quantification of the number of different co-localizing mCherry-Atg8a and myr::GFP puncta in 5 consecutive single-slice images of genotypes in (A). Co-localizing puncta were counted in a bounding box of 40 x12 μm. Statistical analysis was performed with unpaired, two-tailed Mann-Whitney test. The graph shows the median with 95% confidence intervals. n=12 and 10.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 523, 150]]<|/det|>
+NatCommSzaboetal.supplementarymaterials.docx
+
+<--- Page Split --->

preprint/preprint__b34277978e1c5aac355e02d55572f47cd483d00a7fe088823698be13a3872daf/images_list.json

@@ -0,0 +1,62 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Synthesis and microstructure of amorphous carbon nanopillar and its nanoporous structures. (a) Schematic diagram of nanopillar fabrication; (b) SEM image of nanopillar array; SEM images of nanopillar and surface structure (c) of fully dense amorphous carbon nanopillar, (d) 46% porosity nanopillar, and (e) 59% porosity nanopillar; HRTEM image of (f) the fully dense amorphous carbon and (g) the 59% porosity nanoporous carbon. The inset shows the selected area diffraction pattern, which includes two halo rings (marked by the white arrows); (h) HRTEM image of the 59% porosity nanoporous carbon. Pore surface is atomically smooth.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        180,
+        875,
+        656
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Mechanical properties of amorphous carbon nanopillar and its nanoporous structures. (a) SEM snapshots of microcompression test of \\(59\\%\\) porosity nanoporous carbon nanopillar (See also Supplementary Video); SEM images of \\(59\\%\\) porosity nanoporous carbon nanopillar (b) before and after plastic deformation, and (c) after shear fracture; (d) Representative engineering stress-strain curves; (e) Young's modulus as a function of porosity. The inset shows the power law relation between Young's modulus and relative density; (f) Fracture strength and yield strength as a function of porosity. The sky-blue color bar indicates the magnitude of E/10. The inset shows the power law relation between strength and relative density; (g) (left) fracture, elastic, and plastic strains as a function of porosity. (right) fracture, elastic, and plastic strains as a function of pore size. These specimens have the \\(51\\%\\) porosity, but different pore sizes: \\(21.83 \\pm 3.32\\) nm (small), \\(39.41 \\pm 3.95\\) nm (medium), and \\(79.42 \\pm 10.00\\) nm (large). The black arrows in both figures show the presence of transition in fracture strain.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        90,
+        876,
+        627
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Deformation mechanisms of amorphous carbon. (a) EELS measurements of the fraction of \\(\\mathrm{sp}^2\\) bonded carbon atoms in micropillar (left graph) and in base (right graph); (b) Raman spectroscopy data before and after compression; (c) Schematic diagram of energy barriers for elastic loading and unloading; (d) Engineering stress-strain curve of cyclic compression test. Black curves show the recoverable hysteresis deformation. Blue curves show the non-linear recovery after plastic deformation. Red broken lines show how the slope of unloading curve changes; (e) Simulated \\(\\mathrm{sp}^2 /\\mathrm{sp}^3\\) fraction of fully dense amorphous carbon. (g) Simulated stress-strain curve of fully dense amorphous carbon and the snapshots of microstructural evolution during loading and unloading (See also Supplementary Video). Red broken lines in the graph show how the slope of unloading curve changes.",
+    "footnote": [],
+    "bbox": [
+      [
+        130,
+        140,
+        876,
+        620
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4. Mechanical properties of amorphous carbon nanopillar and its nanoporous structures. (a) Compressive strength vs. density. This plot includes all published data of nanoporous materials and nanolattices. Their citation is available in Supplementary Materials.; (b) Specific yield strength vs. fracture strain; (c) Specific Young's modulus vs. specific yield strength. The broken line contour shows the magnitude of specific modulus of resilience.",
+    "footnote": [],
+    "bbox": [
+      [
+        116,
+        171,
+        880,
+        707
+      ]
+    ],
+    "page_idx": 23
+  }
+]

preprint/preprint__b34277978e1c5aac355e02d55572f47cd483d00a7fe088823698be13a3872daf/preprint__b34277978e1c5aac355e02d55572f47cd483d00a7fe088823698be13a3872daf.mmd

@@ -0,0 +1,382 @@
+
+# Nanoporous amorphous carbon nanopillars with lightweight, near-theoretical strength, large fracture strain, and high damping capability  
+
+Seok- Woo Lee  
+
+seok- woo.lee@uconn.edu  
+
+University of Connecticut https://orcid.org/0000- 0001- 6752- 5694  
+
+Zhongyuan Li University of Connecticut  
+
+Ayush Bhardwaj University of Massachusetts Amherst  
+
+Jinlong He University of Wisconsin Madison  
+
+Wenxin Zhang California Institute of Technology https://orcid.org/0000- 0002- 6318- 0622  
+
+Thomas Tran California Institute of Technology  
+
+Ying Li University of Wisconsin Madison  
+
+Andrew McClung University of Massachusetts Amherst  
+
+Sravya Nuguri University of Massachusetts Amherst  
+
+James Watkins University of Massachusetts Amherst  
+
+Article  
+
+Keywords:  
+
+Posted Date: December 19th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 3699209/v1
+
+<--- Page Split --->
+
+License: © ① This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Additional Declarations: There is NO Competing Interest.  
+
+Version of Record: A version of this preprint was published at Nature Communications on September 17th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52359-6.
+
+<--- Page Split --->
+
+# Nanoporous amorphous carbon nanopillars with lightweight, near-theoretical strength, large fracture strain, and high damping capability  
+
+Zhongyuan Li \(^{1,*}\) , Ayush Bhardwaj \(^{2,*}\) , Jinlong He \(^{3}\) , Wenxin Zhang \(^{4}\) , Thomas T. Tran \(^{4}\) ,  
+
+Ying Li \(^{3}\) , Andrew McClung \(^{6}\) , Sravya Nuguri \(^{2}\) , James J. Watkins \(^{28}\) , Seok- Woo Lee \(^{18}\)  
+
+1. Department of Materials Science and Engineering & Institute of Materials Science, University of Connecticut, 25 King Hill Road, Storrs CT 06269-3136, United States  
+
+2. Department of Polymer Science and Engineering, University of Massachusetts Amherst, 120 Governors Drive, Amherst, MA 01003, United States  
+
+3. Department of Mechanical Engineering, University of Wisconsin-Madison, 1513 University Ave, Madison, WI 53706, United States  
+
+4. Division of Engineering and Applied Sciences, California Institute of Technology, 1200 E. California Blvd, Pasadena, CA 91125, United States  
+
+5. Department of Electrical and Computer Engineering, University of Massachusetts Amherst, 100 Natural Resources Rd, Amherst, MA 01003, United States  
+
+\*Equal Contribution: Zhongyuan Li, Ayush Bhardwaj \(^{8}\) Corresponding Author: James J. Watkins (watkins@polysci.umass.edu), Seok- Woo Lee (seok- woo.lee@uconn.edu)
+
+<--- Page Split --->
+
+Abstract: Simultaneous achievement of lightweight, high strength, large fracture strain, and high damping capability has been challenging because some of these mechanical properties are mutually exclusive. Here, we have utilized self- assembled polymeric carbon precursor materials in combination with scalable nanoimprinting lithography to produce nanoporous carbon nanopillars. The produced amorphous carbon nanopillars exhibited ultrahigh strength similar to or even higher than one tenth of their Young's modulus \((E)\) , the most widely used approximation of a fundamental upper limit of material breaking strength. Remarkably, nanoporosity induced via sacrificial template significantly reduced the mass density of amorphous carbon to \(0.66 \sim 0.82g / cm^3\) while the strength of \(E / 10\) is still maintained. Moreover, these nanopillars displayed both elastic and plastic behavior with large fracture strain. A reversible part of the \(\mathrm{sp}^2\) - to- \(\mathrm{sp}^3\) transition produces large elastic strain and a high loss factor (up to 0.033) comparable to Ni- Ti shape memory alloys. The irreversible part of the \(\mathrm{sp}^2\) - to- \(\mathrm{sp}^3\) transition enables plastic deformation, leading to a large fracture strain of up to \(35\%\) . These findings have been substantiated using simulation studies. None of the existing structural materials exhibit a comparable combination of density, strength, deformability, and damping capability. Hence, the results of this study illustrate the potential of both dense and nanoporous amorphous carbon materials as superior structural nanomaterials.
+
+<--- Page Split --->
+
+## Introduction:  
+
+Nanoporous materials are structurally classified as materials having pore sizes typically less than \(100~\mathrm{nm}\) , whose properties are governed by matrix material as well as shape, size, and distribution of pores'. Their high surface area and the tunability of pore topologies facilitate unique physical and chemical properties, making them suitable for various applications such as ion exchange, separation, sensors, \(\mathrm{CO_2}\) capture and storage, water purification, catalysis, renewable energy, drug delivery, and tissue engineering?. In addition, their excellent mechanical properties with the combination of lightweight and high strength render them a potential candidate for future structural applications in aerospace, automobile, and military engineering?.  
+
+It is difficult to achieve lightweight and high strength simultaneously using classic microstructural control of bulk monolithic materials because the strength of materials is generally proportional to their mass density'. For instance, it is not easy to create a lightweight polymer material that possesses the strength of heavy steel. Interestingly, the creation of nanoporous structures could enable us to reduce the mass density within the same class of materials without the significant sacrifice of strength because nanoscale ligaments can impart ultrahigh strength due to the mechanical size effects, so- called 'Smaller is Stronger' behavior5- 8. Recent nanomechanics studies have demonstrated that a material exhibits a significant fraction of theoretical strength, the upper limit of material strength, if its dimension approaches the nanoscale. One tenth of Young's modulus (E/10) has been considered the most widely accepted approximation of the maximum theoretical breaking strength'. Nanopillars, nanowires, and nanoparticles often exhibit the strength of E/50- E/30, which is 20- 100 times higher strength than their bulk counterparts because the absence of flaws and defects in such a small volume can improve strength substantially'. It is expected that the nanoscale ligaments in a nanoporous structure could also exhibit ultrahigh
+
+<--- Page Split --->
+
+strength through the mechanical size effect, leading to the high effective strength of nanoporous materials.  
+
+Nanoporous materials have been successfully fabricated by a variety of methods, including dealloying, soft/hard templating, microwave irradiation, additive manufacturing, ion- beam processing, and laser processing<sup>2</sup>. Specifically, nanoporous metals and ceramics are typically fabricated using selective chemical removal of one phase from two- phase alloy mixtures<sup>10</sup> and hard templating with silica colloidal crystals<sup>11</sup>, respectively. Nanoporous metals, ceramics, and their mixtures exhibit reasonably high strength, \(10 \sim 200 \text{MPa}\) , with their high strength- to- weight ratio and a mass density of \(1 \sim 10 \text{g/cm}^3\)<sup>12</sup>. Nevertheless, several factors still inhibit the desired improvement in mechanical properties of nanoporous materials: First, conventional fabrication methods of nanoporous metals or ceramics do not allow the fabrication of nanoporous structures with extremely thin ligament dimensions. Dealloying and templating methods usually limit the ligament thickness down to \(50 \sim 100 \text{nm}^2\) . To exploit the size affected strength more effectively, it is desirable to reduce the ligament thickness down below \(50 \text{nm}\) , which would allow us to maximize the mechanical size effect. Interestingly, the size reduction has also been known to enhance the ductility due to the flaw tolerance effect at the nanoscale. Brittle- to- ductile transition has been observed in metallic glasses<sup>13</sup>, semiconductors<sup>14</sup>, quasicrystals<sup>15</sup>, and nanolattices<sup>16</sup> when their structural length- scale lies in the regime of \(100 \text{nm}\) or smaller. Thus, the reduction of thickness of ligaments could enable the creation of more ductile nanoporous structures. Second, metals and ceramics exhibit relatively low specific strength due to their heavy mass density. To achieve lightweight and ultrahigh strength simultaneously, it is critical to create nanoporous structures from high specific strength materials, such as carbonaceous materials<sup>17,18</sup>.
+
+<--- Page Split --->
+
+Recently, nanoporous amorphous carbon materials have been successfully fabricated in various ways, including sequential chemical dealloying, hard and soft templating<sup>2</sup>. These methods successfully produce ligaments with thicknesses much thinner than 50nm. Most of these materials have been investigated to study the capacitance properties<sup>19</sup>, electrochemical properties<sup>20</sup>, and biocompatibility<sup>21</sup>. Moreover, their mechanical properties are also expected to be excellent, specifically strength and ductility, considering the high breaking strength of C- C bond and the mechanical size effect. Therefore, it is essential to systematically study the mechanical properties of nanoporous amorphous carbon and to compare them with those of other advanced structural materials.  
+
+In this work, therefore, we have developed lightweight nanoporous amorphous carbon nanopillars exhibiting ultrahigh strength, high deformability, and high damping- capability. Nanoporous nanopillar structures were derived by carbonization of the self- assembled nanocomposite of phenol- formaldehyde (PF) resin and bottlebrush block copolymers (BBCP) in combination with low cost and scalable nanoimprinting lithography. The use of BBCP enables precise control over porosity ranging from \(0\%\) to \(59\%\) with remarkably low shrinkage during carbonization (less than \(20\%\) ). These produced specimens possess extremely fine and uniform porous structures with pore diameter \(\sim 50 \mathrm{nm}\) and the ligament thickness in the range of \(20 \mathrm{nm}\) or thinner. Our fully dense amorphous carbon nanopillars and nanoporous structures represent exceptional mechanical properties, including ultrahigh strength comparable to or even higher than \(E / 10\) , a large elastic strain (up to \(13\%\) ), a large fracture strain (up to \(35\%\) ), and high damping capability with the loss factor (up to \(0.033\) ). Due to the lightweight and near- theoretical strength, the specific strength and the modulus of resilience of our materials is higher than any other engineering materials within a similar range of mass density. Micromechanical tests, electron
+
+<--- Page Split --->
+
+energy loss spectroscopy (EELS), Raman spectroscopy, and atomistic simulation revealed that reversible and irreversible portions of the \(\mathrm{sp}^{2} - \mathrm{to} - \mathrm{sp}^{3}\) transition and the nanoscale dimension of ligaments led to large elastic and fracture strains as well as high damping capability. Remarkably, none of the known structural materials achieves this unique combination of mechanical properties in terms of weight, elasticity, plasticity, and damping. Thus, this work demonstrates the preparation of next generation superior structural nanomaterials for aerospace, military, energy, biomedical, filtration, and catalyst applications.  
+
+## Result  
+
+## Fabrication of fully dense and nanoporous nanopillars  
+
+Nanopillars of fully dense amorphous carbon and their nanoporous structures were fabricated by spin- coating the film ( \(\sim 5 \mu \mathrm{m}\) ) of self- assembled BBCP and PF thin films followed by nanoimprinting and thermal annealing (Figures 1(a) and 1(b)) (See also Methods). Nanopillars ( \(\sim 300 \mathrm{nm}\) diameter and \(\sim 1 \mu \mathrm{m}\) height) with the five different porosities (0%, 40%, 46%, 51%, and 59%) were prepared to obtain samples with different mass densities ranging from 0.66 to \(1.6 \mathrm{g} / \mathrm{cm}^{3}\) . The details of the porosity and density measurement are available in Supplementary Materials. Scanning electron microscope (SEM) images (Figures 1(c)- 1(e)) show that fully dense and uniform nanoporous structures in nanopillars were created successfully. High- resolution transmission electron microscope (HRTEM) images of both fully dense and nanoporous carbon show the disordered atomic arrangement (Figure 1(f), 1(g) and 1(h)), confirming the formation of amorphous phase. The two halo rings in the selected area diffraction pattern (SADP) (marked by the white arrows in the inset of Figure 1(f) and 1(g)) correspond to the interplanar spacing of \(\sim 1.15 \mathrm{\AA}\) and \(\sim 1.95 \mathrm{\AA}\) , which are typically observed from fully amorphous carbon structures<sup>22</sup>. Moreover,
+
+<--- Page Split --->
+
+there is no noticeable difference in atomic arrangements of fully dense and nanoporous structure, and hence the mechanical properties of these nanopillars are expected to be solely governed by the difference in their porosity/porous structure. Note that the atomic arrangement of our specimens is different from the other commonly observed structure of crumbled graphite/graphene networks, which possess folded graphene layers in HRTEM images and several halo diffraction rings due to the higher degree of atomic ordering \(^{23 - 25}\) . The different atomic arrangement of our nanopillars may result from the random orientation of the molecular chains of the PF carbon precursor and the carbonization temperature employed, which impedes the easy crystallization of carbon atoms \(^{19}\) .  
+
+It is worthwhile to note that nanoimprinting has a significant advantage over commonly used focused- ion- beam (FIB) milling. FIB milling requires a substantial amount of time to create a single nanopillar \(^{26,27}\) whereas a large number of nanopillars produced using nano imprinting lithography (NIL) quickly and efficiently allows us to test several nanopillars and obtain statistically reliable mechanical properties. Additionally, self- assembled BBCP and PF resin composite as an ink for NIL enabled the formation of nanopillar with homogenous porous structure after carbonization (see Method section). This can be attributed to the relatively very small domain size ( \(\sim 50 \mathrm{nm}\) ) of the nanocomposite as compared to smallest dimension of pillar ( \(\sim 300 \mathrm{nm}\) in diameter), permitting the formation of uniform porous structure in nanopillar. Remarkably, the NIL process and developed ink were also utilized for imprinting various three- dimensional architecture with maintained porous structure demonstrating the versatility of this fabrication technique (See Supplementary Materials).
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Synthesis and microstructure of amorphous carbon nanopillar and its nanoporous structures. (a) Schematic diagram of nanopillar fabrication; (b) SEM image of nanopillar array; SEM images of nanopillar and surface structure (c) of fully dense amorphous carbon nanopillar, (d) 46% porosity nanopillar, and (e) 59% porosity nanopillar; HRTEM image of (f) the fully dense amorphous carbon and (g) the 59% porosity nanoporous carbon. The inset shows the selected area diffraction pattern, which includes two halo rings (marked by the white arrows); (h) HRTEM image of the 59% porosity nanoporous carbon. Pore surface is atomically smooth. </center>
+
+<--- Page Split --->
+
+## In situ micromechanical characterization  
+
+In situ uniaxial compression tests (Figures 2(a)- 2(d)) revealed a significantly large fracture strain from all nanopillars, comprised of substantial elastic recovery and plastic deformation (See also the Supplementary Video). A permanent height change was also observed when the test was stopped right before the fracture (Figure 2(b)). Interestingly, we could not find any signature of local fracture in the nanoporous structures even after a substantial amount of plastic deformation, implying that the plastic strain is produced by mechanisms that cannot be easily observed by SEM. The mechanical loading was also stopped right after a significant displacement jump, which can be regarded as the initiation of fracture, and the SEM image shows that the fracture occurred via shear banding (Figure 2(c)). Nanoindentation was also performed on the thin film area adjacent to nanopillars to measure Young's moduli precisely. The obtained values matched well to that of data derived from micropillar experiments (Figures 2(d) and 2(e)). We also found that there is negligible depth dependence in our Young's modulus data, indicating that nanoindentation data are not influenced by sink- in or pile- up effects. Nanopillar compression and nanoindentation data of all specimens are available in Supplementary Materials. In general, the yield strength \((\sigma_{y})\) , fracture strength \((\sigma_{f})\) , and Young's modulus \((E)\) decreased as the porosity increased (Figures 2(e) and 2(f)). The power law scaling relations \(^{28}E \sim \rho^{- m}\) (the inset of Figure 2(e)) and \(\sigma_{f} \sim \rho^{- n}\) (the inset of Figure 2(f)) with the relative density \((\rho)\) show \(m = 1.42\) and \(n = 2.03\) , respectively. These results imply that the elastic deformation is more bending- dominant \((m< 1.5)\) , but the fracture is more stretching- dominant \((n > 1.5)^{28}\) . This can be attributed to the complex nanoporous structures which may have resulted in different mechanical responses to elasticity and fracture.  
+
+Surprisingly, our mechanical data shows that the fracture and yield strengths of all specimens reached and even surpassed \(E / 10\) (marked by the blue dash line in Figure 2(f)), which
+
+<--- Page Split --->
+
+is the most widely used value to approximate a fundamental upper limit of material breaking strength<sup>9</sup>. Even with the presence of nanoporous structure, the fracture and yield strengths do not fall below \(E / 10\). This result indicates that the creation of additional free surface, which sometimes acts as a source of plastic deformation or fracture, does not play a role while the creation of thin ligaments seem to preserve ultrahigh strength via the nanoscale size effect. It has been reported that nanoscale diamonds with \(\sim 100nm\) in thickness show ultrahigh strength and significantly large fracture strain due to the absence of internal defects and smooth surface in their small volume<sup>8,29</sup>. The ligaments in our nanoporous amorphous carbon are around 20\~50nm, which is thinner than these nanoscale diamonds. Thus, there is an even smaller chance of containing any critical flaws in ligaments of our nanoporous structures. Also, the HRTEM images depicts the formation of the atomically smooth pore surfaces, confirming that the produced nanoporous carbon is free from any critical surface flaws. Therefore, the robust C-C covalent bonding and the absence of defects within nanoscale ligaments and on the pore surfaces should be the main reasons for the ultrahigh strength of our nanopillars. We also fabricated micropillars with \(2\mu m\) in diameter and \(6\mu m\) in height (\~260 times larger volume) using FIB milling. We confirmed that all FIB-ed micropillars show nearly the same mechanical properties as nanoimprinted nanopillars (See also Supplementary Materials), affirming that the mechanical properties remain unchanged up to the micrometer scale. Thus, the sample dimension of our nanopillars is small enough not to contain any detrimental defects, such as abnormally large pores or cracks.  
+
+We also found the fracture strain exhibits an interesting transition with the porosity change (marked by black arrows in Figure 2(g)). In general, the fracture strain of our samples decreases as the porosity increases. However, nanopillar with the highest porosity (59%) shows a higher fracture strain of \(25 \pm 0.11\%\) than that of \(18 \pm 0.45\%\) of the nanopillar with the porosity of 51%.
+
+<--- Page Split --->
+
+These results were repeatable as indicated by the small error in data. This remarkable increase is primarily attributed to the increase in plastic strain limit. In other words, the ductility of the material increases as the porosity of the sample rises from \(51\%\) to \(59\%\) . We also observed a similar transition when the pore size \((21.83 \pm 3.32 \text{nm (small)}, 39.41 \pm 3.95 \text{nm (medium)},\) and \(79.42 \pm 10.00 \text{nm (large)})\) is varied while the porosity \((51\%)\) remains unchanged (Figure 2(g)). As the pore size decreases, the ligament thickness also decreases (Figures 1(d) and 1(e)). This implies that the transition of fracture strain is specifically being governed by the ligament thickness. Enhanced ductility (or fracture strain) has been frequently observed at the nanoscale. Metallic glasses \(^{13}\) , semiconductors \(^{14}\) , and quasicrystals \(^{15}\) are extremely brittle at bulk scale but showed significant improvement in ductility due to the flaw tolerance effect at the nanoscale when their dimensions become smaller than \(100nm\) . Also, some ceramic nanolattices, which could be considered as the highly ordered nanoporous structure, exhibited the improved ductility when the thickness of strut becomes smaller than \(50 \sim 100nm^{16,30,31}\) . The ligament thickness of our highest porosity nanopillar is only around \(20nm\) (Figure 1(e)), which is much smaller than \(100nm\) . Hence, based on the flaw tolerance at the nanoscale, our highest porosity nanopillar (or smaller pore size nanopillar) could also exhibit ductility.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2. Mechanical properties of amorphous carbon nanopillar and its nanoporous structures. (a) SEM snapshots of microcompression test of \(59\%\) porosity nanoporous carbon nanopillar (See also Supplementary Video); SEM images of \(59\%\) porosity nanoporous carbon nanopillar (b) before and after plastic deformation, and (c) after shear fracture; (d) Representative engineering stress-strain curves; (e) Young's modulus as a function of porosity. The inset shows the power law relation between Young's modulus and relative density; (f) Fracture strength and yield strength as a function of porosity. The sky-blue color bar indicates the magnitude of E/10. The inset shows the power law relation between strength and relative density; (g) (left) fracture, elastic, and plastic strains as a function of porosity. (right) fracture, elastic, and plastic strains as a function of pore size. These specimens have the \(51\%\) porosity, but different pore sizes: \(21.83 \pm 3.32\) nm (small), \(39.41 \pm 3.95\) nm (medium), and \(79.42 \pm 10.00\) nm (large). The black arrows in both figures show the presence of transition in fracture strain. </center>
+
+<--- Page Split --->
+
+## Mechanisms of elasticity, plasticity, and fracture  
+
+Mechanisms of elasticity, plasticity, and fractureTo thoroughly investigate the deformation mechanism, we studied deformation- induced changes in atomic arrangement/configuration of amorphous carbon. Here, we utilized fully dense carbon considering the fact that all nanoporous structures irrespective of their porosity are composed of same amorphous carbon as discussed in the previous section. The atomic scale structural change \((\mathrm{sp}^{2} - \mathrm{to} - \mathrm{sp}^{3}\) transition) of a fully dense amorphous carbon nanopillar was analyzed using electron energy loss spectroscopy (EELS), Raman spectroscopy, cyclic loading test, and atomistic simulations. Then, its implication for mechanical properties of nanoporous structures will be discussed in this section.  
+
+## Electron Energy Loss Spectroscopy and Raman Spectroscopy  
+
+EELS analysis was conducted on the pillar at two different locations. The first location was the region compressed to approximately \(35\%\) strain, while the second location was the pillar base, which underwent a negligible amount of plasticity (Figure 3(a)). Two- windows method \(^{32}\) was used to determine the fraction of \(\mathrm{sp}^{2}\) bond by integrating the intensity of \(1\mathrm{s} - \pi^{*}\) and \(1\mathrm{s} - \sigma^{*}\) from 282- \(286eV\) window and 288- \(298eV\) window, respectively. The result shows that the \(\mathrm{sp}^{2}\) fraction \((\sim 94\%)\) of the compressed nanopillar is about \(4\%\) lower to the pillar base \((\sim 98\%)\) .  
+
+Raman spectroscopy analysis was conducted before and after compressing identical nanopillars by \(\sim 35\%\) in strain (Figure 3(b)) to further explore the carbon microstructure changes. In the Raman spectra range of \(800cm^{- 1}\) to \(2000cm^{- 1}\) , two characteristic peaks are typically observed for amorphous carbon materials. The D peak, centered at \(\sim 1380cm^{- 1}\) , corresponds the \(\mathrm{A_{1g}}\) breathing mode only in aromatic rings, while the G peak, centered at \(\sim 1560cm^{- 1}\) , corresponds to the \(\mathrm{E_{2g}}\) stretching mode of \(\mathrm{sp}^{2}\) atoms in both aromatic rings and olefinic chains \(^{33}\) . The Raman spectra shows the increase in D peak intensity and the positive shift of G peak after the compression. The
+
+<--- Page Split --->
+
+positive shift of G peak is related to the increase in \(\mathrm{sp}^{3}\) bonds in \(\mathrm{sp}^{2}\) - dominant amorphous carbon \(^{33}\) . The increase in D peak intensity occurs when the crystal symmetry of graphite or graphene is disrupted by local disorder, resulting from the localized formation of defects or \(\mathrm{sp}^{3}\) bonds \(^{34 - 36}\) . In addition, recent experimental studies on carbon nanotubes showed that the lateral compression of carbon nanotube could increase the D peak \(^{37}\) . We expect that uniaxial compression could induce a similar topological change locally. If a slightly curved hexagonal planar structure is present in a nanopillar, uniaxial compression could fold such structure and increase the local curvature, enhancing the D peak intensity. Although it is difficult to experimentally confirm the local curvature change, our molecular dynamic (MD) simulation captured a permanent folding of hexagonal planar structure after uniaxial compression (The detailed discussion is also available in Supplementary Materials; See also Supplementary Video). Thus, the creation of \(\mathrm{sp}^{3}\) bonds and the local topological change in internal structures could be the reasons for the increase in D peak intensity in our Raman spectra data.  
+
+The combination of EELS and Raman data suggest strongly that a local \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition occurs during compression. In fact, the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition has been widely observed phenomenon in amorphous carbon under hydrostatic \(^{38,39}\) , biaxial \(^{40}\) , and uniaxial compression \(^{41}\) . Hydrostatic compression tests often apply extremely high pressure (10- 40GPa) to induce the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. This raises a question of if the uniaxial fracture strength of our fully dense nanopillars, \(\sim 4\) GPa, may be too low for such transition. However, it is important to consider that the strain is the more critical factor than the stress (or pressure) because the strain is directly related to the atomic displacement, which influences the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. Note that hydrostatic pressure of 20- 40GPa changes the linear dimension of amorphous carbon by less than 10%. But in our case, the fracture strain is around 35%, which is 3 times larger than the hydrostatic pressure case.
+
+<--- Page Split --->
+
+Therefore, it is feasible to induce a significant atomic displacement capable of triggering the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition within our nanopillar. Also, the stress required for the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is strongly dependent on the initial carbon arrangement. The recent density functional theory calculation claimed that over \(10\%\) increase in \(\mathrm{sp}^{3}\) content is possible through the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition in amorphous carbon with \(1.8g / cm^{3}\) in density (similar to our \(1.6g / cm^{3}\) ) only under around 5GPa of hydrostatic pressure<sup>38</sup>. Based on all these results, our amorphous carbon could have an initial carbon arrangement that can be adjusted easily even at relatively low stress level.  
+
+## Cyclic loading test  
+
+One typical feature of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is the mechanical hysteresis<sup>42</sup>. This hysteresis is related to the different energy barriers between the forward \((\mathrm{sp}^{2} \rightarrow \mathrm{sp}^{3})\) and the backward \((\mathrm{sp}^{3} \rightarrow \mathrm{sp}^{2})\) transition (Figure 3(c)). Due to the different energy barrier, the phase transformation during loading \((\mathrm{sp}^{2} \rightarrow \mathrm{sp}^{3})\) and unloading \((\mathrm{sp}^{3} \rightarrow \mathrm{sp}^{2})\) produces different critical stresses for transition under the load- controlled condition, which is our case. As a result, the stress- strain path is different between loading and unloading, i.e., mechanical hysteresis<sup>43</sup>.  
+
+To confirm that the closed hysteresis loops indeed exist in stress- strain data, a cyclic loading test was conducted on the fully dense carbon pillar. The loss factor \((\eta)\) can be calculated with \(\eta = \frac{\Delta W}{\pi W_{max}}\) for the micropillar compression cyclic test<sup>44</sup>, where \(\Delta W\) is the dissipated energy per stress- release cycle, and \(W_{max}\) is the maximum stored energy per unit volume over the cycle. Our fully dense pillar showed the mechanical hysteresis loop (the magenta colored area in Figure 3(d)) with \(\eta = 0.033\) , which corresponds to the high damping materials \((\eta > 0.015)^{45}\) and is close to the loss factor of commercial Nitinol (Ni- Ti) shape memory alloys \((\eta = 0.028 \sim 0.041)\) (Cyclic stress- strain curves and loss factor data of all nanoporous nanopillars are available in
+
+<--- Page Split --->
+
+Supplementary Materials). Based on EELS, Raman spectra, and cyclic loading test, a local \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is likely to be the main mechanism of both elastic and plastic deformation. A reversible part of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition produces large elastic strain and a high loss factor. The irreversible part of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition produces plastic deformation, leading to a large fracture strain of up to \(35\%\) .  
+
+## Atomistic simulations  
+
+Atomistic simulations were used to probe how atomic arrangement is changed under uniaxial loading and unloading. The amorphous carbon structure with the similar mass density \((1.6g / cm^3)\) with the high fraction of \(\mathrm{sp}^{2}\) bonds \((\sim 90\%)\) was constructed, and the fractions of \(\mathrm{sp}^{2}\) and \(\mathrm{sp}^{3}\) bonds were monitored during uniaxial compression up to \(40\%\) strain, which is similar to the experimental fracture strain \((\sim 35\%)\) , and during unloading. The simulation results show that the fraction of \(\mathrm{sp}^{3}\) increases by \(\sim 13\%\) at \(40\%\) strain, but a considerable amount of newly formed \(\mathrm{sp}^{3}\) bonds is transformed back to \(\mathrm{sp}^{2}\) bonds during unloading (Figure 3(e)). This result is similar to the hydrostatic pressure studies showing that most newly formed \(\mathrm{sp}^{3}\) bonds are transformed back to \(\mathrm{sp}^{2}\) bonds during unloading<sup>42</sup>. When unloading is completed, only \(3.7\%\) of \(\mathrm{sp}^{2}\) bonds were permanently reduced by forming the similar amount of \(\mathrm{sp}^{3}\) bonds. This simulation result is similar to our EELS data that showed about \(4\%\) reduction of the \(\mathrm{sp}^{2}\) bonds after unloading. Thus, our simulation data agrees well with our experimental findings and confirms the plastic strain resulted from the irreversible part of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. This atomic scale transformation also explains why the signature of plasticity could not be observed from the high- resolution SEM images even after the large amount of deformation (Figure 2(b)). We also found that the slope of the simulated unloading curve changes nonlinearly when the new \(\mathrm{sp}^{3}\) bonds are transformed back to the \(\mathrm{sp}^{2}\) bonds (two broken red lines in Figure 3(f)). This nonlinear change resembles the shape
+
+<--- Page Split --->
+
+of the experimental unloading curve (two broken red lines in Figure 3(d)), which could be the main reason for mechanical hysteresis. Note that the high stress level of atomistic simulations is typical and unavoidable due to the extremely high strain rate<sup>46,47</sup>. We also studied the strain rate effect (See also Supplementary Information) and confirmed that the high stress level in our simulation data is directly associated with the strain rate. As discussed before, the strain is the primary factor to control the atomic displacement, which is directly related to the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. Thus, our simulation results should be considered valid for studying the fundamental mechanisms of elastic and plastic deformation reasonably in terms of large strain (40%) similar to experimental value (35%).  
+
+Interestingly, plastic strain (Figure 2(g)) and loss factor (See also Supplementary Materials) of nanoporous nanopillars are generally lower than those of fully dense ones. These results can also be understood in terms of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. The \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is driven by compressive stress because the \(\mathrm{sp}^{3}\) bond is more energetically preferred under compression. Thus, both plastic strain and loss factor must be affected by the distribution and magnitude of compressive stress. Because of complex nanoporous structures, ligaments in nanoporous nanopillars should undergo bending stress, which includes both compressive and tensile stresses. The local regions under tensile stress do not undergo the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition, leading to the lower plastic strain and loss factor of nanoporous nanopillars. However, the plastic strain and loss factor could be enhanced if the size of ligaments become extremely thin. As discussed in the previous section, the flaw tolerance effect at the nanoscale could suppress the fracture of ligaments in regions under tension and enable further deformation. This could be the reason for nanopillar with the highest porosity (59%) to exhibit higher fracture strain and higher loss factor than the nanopillar with the porosity of 51%.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3. Deformation mechanisms of amorphous carbon. (a) EELS measurements of the fraction of \(\mathrm{sp}^2\) bonded carbon atoms in micropillar (left graph) and in base (right graph); (b) Raman spectroscopy data before and after compression; (c) Schematic diagram of energy barriers for elastic loading and unloading; (d) Engineering stress-strain curve of cyclic compression test. Black curves show the recoverable hysteresis deformation. Blue curves show the non-linear recovery after plastic deformation. Red broken lines show how the slope of unloading curve changes; (e) Simulated \(\mathrm{sp}^2 /\mathrm{sp}^3\) fraction of fully dense amorphous carbon. (g) Simulated stress-strain curve of fully dense amorphous carbon and the snapshots of microstructural evolution during loading and unloading (See also Supplementary Video). Red broken lines in the graph show how the slope of unloading curve changes. </center>
+
+<--- Page Split --->
+
+## Discussion  
+
+Mechanical characterizations revealed that all nanopillars in this study exhibit remarkably high yield and fracture strengths, which are similar to or even higher than the theoretical breaking strength, \(E / 10\) (Figure 3(f)). In addition, our fully dense specimen has a low mass density \((1.6g / cm^3)\) , and the introduction of nanoscale pores reduces the mass density even further \((0.66 \sim 0.82g / cm^3)\) . As a result, the plot of the fracture strength versus mass density shows that our nanopillars are among the strongest materials for their mass density (Figure 4(a)). These superior mechanical properties could result primarily from the ligament size effects and the sample size effects. The nanoscale thickness of ligaments is too small to contain any significant flaws that would detrimentally affect the strength, and the overall volume of nanopillars may not be large enough to contain flaws. As shown by the FIB- milled micropillar compression tests, the sample size effect seems to be relatively weak because the self- assembly of BBCPs produces uniform and ordered nanoporous structures, which preserves the consistent mechanical properties up to the micrometer scale. The sample size effect would be more observable if the sample dimension goes beyond millimeters and centimeters because the weaker spots will be statistically probable to be present in such large volume. It would be important to study the sample size effect in a larger length scale as a future work.  
+
+In addition, our nanopillars exhibit large fracture strain (Figure 3(g)) and high damping capability (Figure 4(d)). Recent study on the state- of- the- art amorphous carbon nanolattices, which could be considered as a highly ordered porous structure, showed the near theoretical breaking strength \((E / 10)\) , too \(^9\) , but they exhibited extreme brittleness with no measurable plastic strain and no damping capability. In contrast, our nanoporous amorphous carbon nanopillars are much more deformable than that of the 3D architected amorphous carbon nanolattice structures with the
+
+<--- Page Split --->
+
+fracture strain of \(\sim 10\%\) only. The latter requires a more time- consuming and costly fabrication process, not to mention the huge volume shrinkage during pyrolysis process which hampers precise control over the final feature dimensions.  
+
+Due to the ultrahigh strength and low mass density, our materials exhibit the extraordinary combination of specific yield strength and specific Young's modulus. They occupy the white space in the plot of specific yield strength vs. specific Young's modulus (Figure 4(b)). This indicates that our materials exhibit the unprecedentedly high specific modulus of resilience \((\sim 5,000\mathrm{MJ / m^3 / (kg / m^3)})\), which corresponds to the maximum possible elastic energy absorption and release per unit volume and per unit density. The straight contour lines show the magnitude of specific modulus of resilience. The specific modulus of resilience of our materials is certainly the highest within their range of specific yield strength and is comparable with that of elastomers, which show the highest modulus of resilience due to their extremely large elastic deformability. Also, for a given fracture strain, our materials show the highest specific strength among all materials, implying that they are highly deformable even with ultrahigh strength. Some high strength composites show relatively similar specific fracture strength, but their ductility is less than \(1\%\) , which is more than a magnitude much smaller than that of our nanopillars \((17 \sim 35\%)\) .  
+
+As mentioned in the previous section, due to the reversible \(\mathrm{sp}^{2} - \mathrm{to} - \mathrm{sp}^{3}\) transition, our materials can also be classified as highly damping capable materials, which have the loss factor of \(0.015 \sim 0.033\) . Thus, our amorphous carbon nanopillar and its nanoporous structures provide unusual combinations of lightweight, high strength, large fracture strain, and high damping capability. Although there have been similar studies on micromechanical testing of fully dense amorphous carbon<sup>17,48,49</sup>, these systems used different initial conditions with either the crumbled graphite networks or the higher density of \(\mathrm{sp}^{3}\) bonds, and their mass density was not measured
+
+<--- Page Split --->
+
+experimentally. More importantly, our work is unique in aspect that the mass density is tunable by introducing nanoporous structure without sacrificing \(E / 10\) relation of yield strength, a result that no other approach has achieved. In addition, it is also noteworthy to emphasize that our synthesis method is both rapid and scalable. It is possible to create both fully dense and nanoporous structures with even centimeter- scale dimensions in width and hundreds of micrometers in thickness (See also Supplementary Materials). This efficient and scalable fabrication method makes our materials more suitable for potential engineering applications, compared to nanolattices that have limited scalability beyond \(200\mu m\) in dimension. Moreover, there is a large degree of freedom in materials design. The pore distribution can be varied by adding both linear block copolymer (LBCP) and BBCP or by varying the molecular weight of BBCP, so the non- uniform pore distribution, for instance, bimodal distribution, can be easily created<sup>50,51</sup>. Metallic or ceramic nanoparticles with the diameter less than 5nm can be embedded into ligaments of nanoporous structures<sup>52</sup>. A nanoscale ceramic layer can be coated on the pore surfaces using the atomic layer deposition<sup>53</sup>. All these nanoporous carbon and their composite materials will provide a wide range of structure- property- processing controllability. In summary, lightweight, superior mechanical properties, scalable fabrication method, and tunable microstructures of nanoporous amorphous carbon promote the fabrication of not only advanced structural materials for military and aerospace applications but also mechanically robust nanoporous structures for energy, biomedical, filtration, and catalyst applications.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4. Mechanical properties of amorphous carbon nanopillar and its nanoporous structures. (a) Compressive strength vs. density. This plot includes all published data of nanoporous materials and nanolattices. Their citation is available in Supplementary Materials.; (b) Specific yield strength vs. fracture strain; (c) Specific Young's modulus vs. specific yield strength. The broken line contour shows the magnitude of specific modulus of resilience. </center>
+
+<--- Page Split --->
+
+## Reference:  
+
+1. Lu, G. Q. M. & Zhao, X. S. Nanoporous materials: science and engineering. vol. 4 (World Scientific, 2004).  
+2. Hadden, M., Martinez-Martin, D., Yong, K.-T., Ramaswamy, Y. & Singh, G. Recent Advancements in the Fabrication of Functional Nanoporous Materials and Their Biomedical Applications. Materials 15, 2111 (2022).  
+3. Gibson, L. J. & Ashby, M. F. Cellular Solids: Structure and Properties. Cambridge Solid State Science Series (Cambridge University Press, 1997). doi:DOI: 10.1017/CBO9781139878326.  
+4. Ashby, M. F. & CEBON, D. Materials selection in mechanical design. Le Journal de Physique IV 3, C7-1 (1993).  
+5. Zhang, H. et al. Approaching the ideal elastic strain limit in silicon nanowires. Sci Adv 2, e1501382 (2023).  
+6. Greer, J. R. & De Hosson, J. Th. M. Plasticity in small-sized metallic systems: Intrinsic versus extrinsic size effect. Prog Mater Sci 56, 654-724 (2011).  
+7. Zhu, T. & Li, J. Ultra-strength materials. Prog Mater Sci 55, 710-757 (2010).  
+8. Dang, C. et al. Achieving large uniform tensile elasticity in microfabricated diamond. Science (1979) 371, 76-78 (2021).  
+9. Bauer, J., Schroer, A., Schwaiger, R. & Kraft, O. Approaching theoretical strength in glassy carbon nanolattices. Nat Mater 15, 438-443 (2016).  
+10. Biener, J. et al. Size Effects on the Mechanical Behavior of Nanoporous Au. Nano Lett 6, 2379-2382 (2006).  
+11. Saito, Y. et al. Preparation of ordered nanoporous indium tin oxides with large crystallites and individual control over their thermal and electrical conductivities. ACS Appl Mater Interfaces 13, 15373-15382 (2021).  
+12. Yang, W. et al. Light, strong, and stable nanoporous aluminum with native oxide shell. Sci Adv 7, eabb9471 (2023).  
+13. Jang, D. & Greer, J. R. Transition from a strong-yet-brittle to a stronger-and-ductile state by size reduction of metallic glasses. Nat Mater 9, 215-219 (2010).  
+14. Östlund, F. et al. Brittle-to-Ductile Transition in Uniaxial Compression of Silicon Pillars at Room Temperature. Adv Funct Mater 19, 2439-2444 (2009).  
+15. Zou, Y. et al. Superior room-temperature ductility of typically brittle quasicrystals at small sizes. Nat Commun 7, 12261 (2016).  
+16. Lee, S.-W., Jafary-Zadeh, M., Chen, D. Z., Zhang, Y.-W. & Greer, J. R. Size Effect Suppresses Brittle Failure in Hollow Cu60Zr40 Metallic Glass Nanolattices Deformed at Cryogenic Temperatures. Nano Lett 15, 5673-5681 (2015).  
+17. Zhang, X. et al. Theoretical strength and rubber-like behaviour in micro-sized pyrolytic carbon. Nat Nanotechnol 14, 762-769 (2019).  
+18. Hu, M. et al. Compressed glassy carbon: An ultrastrong and elastic interpenetrating graphene network. Sci Adv 3, e1603213 (2017).
+
+<--- Page Split --->
+
+19. Bhardwaj, A. et al. Large-Pore Ordered Mesoporous Turbostratic Carbon Films Prepared Using Rapid Thermal Annealing for High-Performance Micro-pseudocapacitors. ACS Appl Mater Interfaces 13, 61027-61038 (2021).  
+
+20. Yun, Y. S. & Jin, H.-J. Electrochemical performance of heteroatom-enriched amorphous carbon with hierarchical porous structure as anode for lithium-ion batteries. Mater Lett 108, 311-315 (2013).  
+
+21. Jain, S., Sharma, A. & Basu, B. Vertical electric field stimulated neural cell functionality on porous amorphous carbon electrodes. Biomaterials 34, 9252-9263 (2013).  
+
+22. Neidhardt, J., Hultman, L. & Czigány, Zs. Correlated high resolution transmission electron microscopy and X-ray photoelectron spectroscopy studies of structured CNx (0<x<0.25) thin solid films. Carbon N Y 42, 2729-2734 (2004).  
+
+23. Shiell, T. B. et al. Graphitization of glassy carbon after compression at room temperature. Phys Rev Lett 120, 215701 (2018).  
+
+24. Shiell, T. B. et al. Nanocrystalline hexagonal diamond formed from glassy carbon. Sci Rep 6, 37232 (2016).  
+
+25. Wu, Y. et al. Strong amorphous carbon prepared by spark-plasma sintering C60. Journal of the American Ceramic Society 104, 1655-1660 (2021).  
+
+26. Yang, X. et al. Integration of nanoimprint lithography with block copolymer directed self-assembly for fabrication of a sub-20 nm template for bit-patterned media. Nanotechnology 25, 395301 (2014).  
+
+27. Lee, S. et al. FIB-induced dislocations in Al submicron pillars: Annihilation by thermal annealing and effects on deformation behavior. Acta Mater 110, 283-294 (2016).  
+
+28. Bauer, J. et al. Nanolattices: An Emerging Class of Mechanical Metamaterials. Advanced Materials 29, 1701850 (2017).  
+
+29. Banerjee, A. et al. Ultralarge elastic deformation of nanoscale diamond. Science (1979) 360, 300-302 (2018).  
+
+30. Meza, L. R., Das, S. & Greer, J. R. Strong, lightweight, and recoverable three-dimensional ceramic nanolattices. Science (1979) 345, 1322-1326 (2014).  
+
+31. Michelson, A., Flanagan, T. J., Lee, S.-W. & Gang, O. High-strength, lightweight nano-architected silica. Cell Rep Phys Sci 4, (2023).  
+
+32. Bruley, J., Williams, D. B., Cuomo, J. J. & Pappas, D. P. Quantitative near-edge structure analysis of diamond-like carbon in the electron microscope using a two-window method. J Microsc 180, 22-32 (1995).  
+
+33. Ferrari, A. C. & Robertson, J. Interpretation of Raman spectra of disordered and amorphous carbon. Phys Rev B 61, 14095 (2000).  
+
+34. Lucchese, M. M. et al. Quantifying ion-induced defects and Raman relaxation length in graphene. Carbon N Y 48, 1592-1597 (2010).  
+
+35. Eckmann, A. et al. Probing the nature of defects in graphene by Raman spectroscopy. Nano Lett 12, 3925-3930 (2012).  
+
+36. Cancado, L. G. et al. Disentangling contributions of point and line defects in the Raman spectra of graphene-related materials. 2d Mater 4, 025039 (2017).  
+
+37. Picheau, E. et al. Intense Raman D Band without Disorder in Flattened Carbon Nanotubes. ACS Nano 15, 596-603 (2021).
+
+<--- Page Split --->
+
+38. Jiang, X., Ārhammar, C., Liu, P., Zhao, J. & Ahuja, R. The R3-carbon allotrope: a pathway towards glassy carbon under high pressure. Sci Rep 3, 1877 (2013).  
+
+39. Shiell, T. B. et al. In situ analysis of the structural transformation of glassy carbon under compression at room temperature. Phys Rev B 99, 24114 (2019).  
+
+40. McCulloch, D. G., McKenzie, D. R. & Goringe, C. M. Ab initio simulations of the structure of amorphous carbon. Phys Rev B 61, 2349 (2000).  
+
+41. Zhao, Y., Qian, C., Gladkikh, V. & Ding, F. Simulated Pressure-temperature Carbon Structure Map obtained through uniaxial compression of Bulk C60. Carbon N Y 202, 554-560 (2023).  
+
+42. Zeng, Z. et al. Structural transition in cold-compressed glassy carbon. Phys Rev Mater 3, 033608 (2019).  
+
+43. Sypek, J. T. et al. Superelasticity and cryogenic linear shape memory effects of CaFe2As2. Nat Commun 8, 1083 (2017).  
+
+44. Juan, J. S., Nó, M. L. & Schuh, C. A. Nanoscale shape-memory alloys for ultrahigh mechanical damping. Nat Nanotechnol 4, 415-419 (2009).  
+
+45. Zhang, J., Perez, R. J. & Lavernia, E. J. Documentation of damping capacity of metallic, ceramic and metal-matrix composite materials. J Mater Sci 28, 2395-2404 (1993).  
+
+46. Park, H. et al. Toward the constitutive modeling of epoxy matrix: Temperature-accelerated quasi-static molecular simulations consistent with the experimental test. Compos B Eng 142, 131-141 (2018).  
+
+47. Fan, J., Anastassiou, A., Macosko, C. W. & Tadmor, E. B. Molecular dynamics predictions of thermomechanical properties of an epoxy thermosetting polymer. Polymer (Guildf) 196, 122477 (2020).  
+
+48. Yoon, J. et al. In situ tensile and fracture behavior of monolithic ultra-thin amorphous carbon in TEM. Carbon N Y 196, 236-242 (2022).  
+
+49. Liu, C., Lin, Y., Zhou, Z. & Li, K.-Y. Dual phase amorphous carbon ceramic achieves theoretical strength limit and large plasticity. Carbon N Y 122, 276-280 (2017).  
+
+50. Fei, H.-F. et al. Ordered nanoporous carbons with broadly tunable pore size using bottlebrush block copolymer templates. J Am Chem Soc 141, 17006-17014 (2019).  
+
+51. Fei, H.-F. et al. One-step synthesis of hierarchical, bimodal nanoporous carbons via co-templating with bottlebrush and linear block copolymers. Chemistry of Materials 32, 6055-6061 (2020).  
+
+52. Song, D.-P. et al. Millisecond photothermal carbonization for in-situ fabrication of mesoporous graphitic carbon nanocomposite electrode films. Carbon N Y 174, 439-444 (2021).  
+
+53. Bhardwaj, A. et al. Large-Pore Ordered Mesoporous Turbostratic Carbon Films Prepared Using Rapid Thermal Annealing for High-Performance Micro-pseudocapacitors. ACS Appl Mater Interfaces 13, 61027-61038 (2021).
+
+<--- Page Split --->
+
+## METHOD  
+
+## Synthesis of brush block copolymer and phenol formaldehyde resin  
+
+Polydimethylsiloxane- b- poly(ethylene oxide) (PDMS- b- PEO) brush block copolymer was synthesized following ring opening metathesis polymerization (ROMP) using Grubbs generation III catalyst following our previous report<sup>19,50,51</sup>. The molecular weight of brush block copolymer was varied by changing the overall degree of polymerization (DP) which in turn was controlled by changing the feed ratio of monomer and catalyst as described previously<sup>50</sup>. The obtained molecular weight at different DP was calculated to be 250 KDa/mol, 500 KDa/mol and 800 KDa/mol.  
+
+Phenol formaldehyde (PF) resin was synthesized in basic polymerization medium<sup>50</sup>. Phenol is melted at \(42^{\circ}\mathrm{C}\) and 20 wt.\% of NaOH (sodium hydroxide) was added dropwise to it. Formaldehyde was added in the molar ratio of 2 as compared to that of phenol and the entire mixture solution is stirred at \(70^{\circ}\mathrm{C}\) for 1 hr. After cooling down the mixture, the solution pH was neutralized using 0.5 M HCl solution. Water was removed in presence of nitrogen flow overnight, following that ethanol was added to the solution to remove the formed sodium chloride (NaCl). Finally, ethanol was removed by nitrogen and redissolved in the THF to get the desired concentration.  
+
+## Fabrication of nanoimprinted nanopillars  
+
+The carbon precursor film is comprised of PDMS- b- PEO brush block copolymer (BBCP) of different molecular weight as the soft sacrificial template (porogen) and PF resin as the carbon source. The BBCP and PF resin were dissolved in THF separately to achieve a concentration of 40 and \(50\mathrm{mg / mL}\) , respectively. The BBCP and PF resin were added in the different weight ratio namely 1:1.5, 1:2, 1:3 and 1:4.0 and the solvent was evaporated with a nitrogen flow to increase
+
+<--- Page Split --->
+
+the concentration to about \(300\mathrm{mg / ml}\) . The prepared solution was spin- coated at \(1000\mathrm{rpm}\) for 30s on Si wafer substrate. The substrate was ultrasonically cleaned with isopropanol and deionized water for \(10\mathrm{min}\) each, followed by UV- ozone treatment for \(15\mathrm{min}\) prior to coating. Carbon precursor film was patterned via nano imprinting lithography (NIL) using h- pdms stamp. H- pdms stamp was placed on spin coated carbon precursor film followed by heating it at \(170^{\circ}\mathrm{C}\) for 10 minutes under constant pressure of \(180\mathrm{psi}\) . Applied pressure and temperature ensured pattern transfer from the h- pdms stamp to the carbon precursor film with simultaneous crosslinking of the PF resin. Preparation of h- pdms stamp from the Si imprint master was carried out as reported previously<sup>54</sup>. Stamp was removed from the carbon precursor film after cooling down the system to the room temperature. Finally, imprinted carbon precursor film was carbonized in tube furnace at \(750^{\circ}\mathrm{C}\) in nitrogen atmosphere for \(1\mathrm{hr}\) . with the heating rate of \(10^{\circ}\mathrm{C / min}\) . This resulted in the formation of nanostructured mesoporous carbon with film thickness of \(\sim 5\mathrm{um}\) . During the carbonization process BBCP degrades completely and PF resin gets converted into carbon resulting in the formation of porous carbon with the nanostructure transferred using NIL. The residual carbon film from the surface of nanostructured porous carbon was removed using reactive ion etching in \(\mathrm{O_2}\) plasma (HF power - \(10\mathrm{W}\) , pressure - \(12\mathrm{mTorr}\) , \(\mathrm{O_2} - 40\mathrm{scm}\) ) for 90s. Fully dense patterned carbon film was fabricated using the same procedure but without any BBCP in it. We also prepared nanoporous carbon thin film following the procedure described above for porosity and density measurement without performing NIL. Moreover, to demonstrate the versatility of this approach we produced nanoporous carbon with different architecture specially motheye, pyramid and pillar by using their respective desired h- pdms stamp keeping other processes same.
+
+<--- Page Split --->
+
+## Nanomechanical experiments:  
+
+Nanodimentation was performed on the thin film region of the carbon sample at room temperature by using an iNano™ system (Nanomechanics, TN, USA). The indentation was performed utilizing the standard Berkovich tip nanoindentation technique, with an indentation strain rate of \(0.2 \mathrm{s}^{- 1}\) . The indentation depths are set as 1/10 of the thickness of the carbon samples to make sure that the indentation data are not affected by the silicon substrate.  
+
+In situ uniaxial compression tests were performed by the nanoindenter (Nanoflip™, Nanomechanics Inc., TN, USA) with a flat diamond tip, which was installed in a field- emission gun scanning electron microscope (SEM) (JSM- 6335F, JEOL, Japan). The nanopillars were compressed under a constant displacement rate of \(10 \mathrm{nm / s}\) . The stress- strain curves were obtained through the corresponding load- displacement data and were corrected by the Sneddon punch method<sup>55</sup>. The entire deformation process was also recorded to avoid strain measurement errors.  
+
+## Microstructure characterization:  
+
+The microstructure of porous and fully dense carbon pillars was characterized by Titan Themis AC- STEM (ThermoFisher) at the accelerating voltage of \(300\mathrm{kV}\) . The cross- sectional TEM samples were prepared using the lift- out technique with a focused ion beam (FIB) instrument (FEI Helios, ThermaFisher). A protective carbon layer was first deposited on the carbon thin film and the micropillar. Milled graduated trenches were introduced into the bottom Si substrate utilizing \(\mathrm{Ga + }\) ion beams. With the samples mounted on the micromanipulator, careful cutting was carried out on the side and bottom sections. Subsequently, the samples were lifted out and positioned onto a copper grid. Finally, a thorough cleaning and thinning process was performed until achieving the
+
+<--- Page Split --->
+
+desired final thin TEM sample. EELS analysis was also performed in Titan Themis AC- STEM with the accelerating voltage of \(250\mathrm{kV}\) and the EELS curves were analyzed by Two- window method \(^{32}\) to calculate the relative fractions of \(\mathrm{sp}^{2}\) and \(\mathrm{sp}^{3}\) phases.  
+
+The phase transformation upon compression was characterized using Raman spectroscopy with a \(633\mathrm{nm}\) laser (Renishaw Ramascope 2000). The Raman spectra were fitted with Gaussian peaks to quantitatively evaluate the D and G band changes.  
+
+## Molecular Dynamics Simulations:  
+
+The amorphous carbon structures were generated employing two key methods: the liquid- quench method \(^{56,57}\) . The overall process of generating the model carbon structures, based on an experimental density of \(1.6\mathrm{g / cm}^3\) , involves seven steps (more details are described in Supplementary Materials). Based on the generated amorphous configuration, the uniaxial mechanical test was conducted using the Large- scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) package \(^{58}\) . Interaction potentials involving carbon particles are characterized by Adaptive Intermolecular Reactive Empirical Bond Order (AIREBO) Potential \(^{59}\) .  
+
+During the loading process, a relaxation time of \(500\mathrm{ps}\) is first performed under the NPT ensemble at a constant temperature of \(300\mathrm{K}\) and pressure of \(0.0\mathrm{MPa}\) . Following the relaxation period, non- equilibrium Molecular Dynamics (NEMD) simulations are conducted to explore the system's response to loading- unloading with the strain rate, \(10^{9}\mathrm{s}^{- 1}\) , at \(300\mathrm{K}\) and a timestep of \(0.2\mathrm{fs}\) under the \(N\sigma_{ij}\epsilon_{ij}T\) ensemble. During the loading and unloading processes, carbon hybridization (sp, \(\mathrm{sp}^{2}\) and \(\mathrm{sp}^{3}\) ) is also computed based on the number of nearest neighbor carbon atoms within a cutoff radius of \(2.0\mathrm{\AA}\) . We define the bond type of a carbon atom based on a coordination criterion
+
+<--- Page Split --->
+
+as follows: a carbon atom is labeled as sp if it is bonded to 2 other carbon atoms, \(\mathrm{sp}^{2}\) if it forms bonds with 3 other carbon atoms, and \(\mathrm{sp}^{3}\) if it is bonded to 4 other carbon atoms, all within a 2.0 Å cutoff radius.  
+
+## Data availability  
+
+The data presented in the main text and the Supplementary Information are available from the corresponding authors upon reasonable request.  
+
+## Reference:  
+
+54. Einck, V. J. et al. Scalable nanoimprint lithography process for manufacturing visible metasurfaces composed of high aspect ratio TiO2 meta-atoms. ACS Photonics 8, 2400–2409 (2021).  
+55. Lee, S.-W., Han, S. M. & Nix, W. D. Uniaxial compression of fcc Au nanopillars on an MgO substrate: The effects of prestraining and annealing. Acta Mater 57, 4404–4415 (2009).  
+56. Galli, G., Martin, R. M., Car, R. & Parrinello, M. Structural and electronic properties of amorphous carbon. Phys Rev Lett 62, 555 (1989).  
+57. Rosen, J., Warschkow, O., McKenzie, D. R. & Bilek, M. M. M. Amorphous and crystalline phases in thermal quench simulations of alumina. J Chem Phys 126, (2007).  
+58. Plimpton, S. Fast parallel algorithms for short-range molecular dynamics. J Comput Phys 117, 1–19 (1995).  
+59. Stuart, S. J., Tutein, A. B. & Harrison, J. A. A reactive potential for hydrocarbons with intermolecular interactions. J Chem Phys 112, 6472–6486 (2000).
+
+<--- Page Split --->
+
+## Acknowledgement  
+
+A.B., S.N., and J.J.W. acknowledge support from the Office of Naval Research (N00014- 23- 9- 0008) through the American Lightweight Materials Manufacturing Innovation Institute. Z.L. and S.- W.L. acknowledge support from the UConn/Thermo Fischer Scientific Center for Advanced Microscopy and Materials Analysis (CAMMA) for the FIB milling, TEM, and EELS experiment.  
+
+## Author contributions  
+
+Z.L., A.B., J.J.W. and S.- W.L conceived and designed the experiments and analysis. Nanomechanical testing, TEM, EELS, and Raman Spectroscopy experiments were carried out and analyzed by Z.L. and S.- W.L. Synthesis of BBCP and PF, self- assembly, and nanoimprinting experiments were carried out and analyzed by A.B. and J.J.W. W.Z. and T.T.T. helped Z.L. to obtain the in- situ deformation video. Molecular dynamic simulations were carried out by and analyzed by J.H. and Y.L. A.M. and S.N. conducted the porosity measurement. Z.L., A.B., J.J.W. and S.- W.L wrote the paper together. J.J.W. and S.- W.L. supervised and provided support through the paper. All authors have commented and edited the manuscript.  
+
+## Competing interests  
+
+The authors declare no competing interests.
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+SupplementaryMaterials.pdf  SuppleMovie59percentporosityloadingunloading.mp4  SuppleMovieMDLocalStructuralFolding.mp4  SuppleMovieMDLoadingUnloading.mp4
+
+<--- Page Split --->

preprint/preprint__b34277978e1c5aac355e02d55572f47cd483d00a7fe088823698be13a3872daf/preprint__b34277978e1c5aac355e02d55572f47cd483d00a7fe088823698be13a3872daf_det.mmd

@@ -0,0 +1,492 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 933, 210]]<|/det|>
+# Nanoporous amorphous carbon nanopillars with lightweight, near-theoretical strength, large fracture strain, and high damping capability  
+
+<|ref|>text<|/ref|><|det|>[[44, 230, 150, 247]]<|/det|>
+Seok- Woo Lee  
+
+<|ref|>text<|/ref|><|det|>[[52, 258, 300, 273]]<|/det|>
+seok- woo.lee@uconn.edu  
+
+<|ref|>text<|/ref|><|det|>[[50, 303, 635, 321]]<|/det|>
+University of Connecticut https://orcid.org/0000- 0001- 6752- 5694  
+
+<|ref|>text<|/ref|><|det|>[[44, 327, 275, 366]]<|/det|>
+Zhongyuan Li University of Connecticut  
+
+<|ref|>text<|/ref|><|det|>[[44, 373, 384, 413]]<|/det|>
+Ayush Bhardwaj University of Massachusetts Amherst  
+
+<|ref|>text<|/ref|><|det|>[[44, 419, 344, 459]]<|/det|>
+Jinlong He University of Wisconsin Madison  
+
+<|ref|>text<|/ref|><|det|>[[44, 466, 707, 506]]<|/det|>
+Wenxin Zhang California Institute of Technology https://orcid.org/0000- 0002- 6318- 0622  
+
+<|ref|>text<|/ref|><|det|>[[44, 512, 345, 552]]<|/det|>
+Thomas Tran California Institute of Technology  
+
+<|ref|>text<|/ref|><|det|>[[44, 559, 344, 599]]<|/det|>
+Ying Li University of Wisconsin Madison  
+
+<|ref|>text<|/ref|><|det|>[[44, 605, 384, 644]]<|/det|>
+Andrew McClung University of Massachusetts Amherst  
+
+<|ref|>text<|/ref|><|det|>[[44, 651, 384, 690]]<|/det|>
+Sravya Nuguri University of Massachusetts Amherst  
+
+<|ref|>text<|/ref|><|det|>[[44, 697, 384, 737]]<|/det|>
+James Watkins University of Massachusetts Amherst  
+
+<|ref|>text<|/ref|><|det|>[[44, 781, 103, 798]]<|/det|>
+Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 819, 135, 836]]<|/det|>
+Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 857, 348, 875]]<|/det|>
+Posted Date: December 19th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[44, 895, 474, 912]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 3699209/v1
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 44, 914, 88]]<|/det|>
+License: © ① This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 105, 535, 125]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 160, 920, 204]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on September 17th, 2024. See the published version at https://doi.org/10.1038/s41467-024-52359-6.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[118, 90, 880, 154]]<|/det|>
+# Nanoporous amorphous carbon nanopillars with lightweight, near-theoretical strength, large fracture strain, and high damping capability  
+
+<|ref|>text<|/ref|><|det|>[[161, 219, 840, 240]]<|/det|>
+Zhongyuan Li \(^{1,*}\) , Ayush Bhardwaj \(^{2,*}\) , Jinlong He \(^{3}\) , Wenxin Zhang \(^{4}\) , Thomas T. Tran \(^{4}\) ,  
+
+<|ref|>text<|/ref|><|det|>[[168, 255, 830, 276]]<|/det|>
+Ying Li \(^{3}\) , Andrew McClung \(^{6}\) , Sravya Nuguri \(^{2}\) , James J. Watkins \(^{28}\) , Seok- Woo Lee \(^{18}\)  
+
+<|ref|>text<|/ref|><|det|>[[120, 339, 884, 396]]<|/det|>
+1. Department of Materials Science and Engineering & Institute of Materials Science, University of Connecticut, 25 King Hill Road, Storrs CT 06269-3136, United States  
+
+<|ref|>text<|/ref|><|det|>[[120, 410, 884, 465]]<|/det|>
+2. Department of Polymer Science and Engineering, University of Massachusetts Amherst, 120 Governors Drive, Amherst, MA 01003, United States  
+
+<|ref|>text<|/ref|><|det|>[[120, 480, 884, 535]]<|/det|>
+3. Department of Mechanical Engineering, University of Wisconsin-Madison, 1513 University Ave, Madison, WI 53706, United States  
+
+<|ref|>text<|/ref|><|det|>[[120, 550, 884, 604]]<|/det|>
+4. Division of Engineering and Applied Sciences, California Institute of Technology, 1200 E. California Blvd, Pasadena, CA 91125, United States  
+
+<|ref|>text<|/ref|><|det|>[[120, 619, 884, 674]]<|/det|>
+5. Department of Electrical and Computer Engineering, University of Massachusetts Amherst, 100 Natural Resources Rd, Amherst, MA 01003, United States  
+
+<|ref|>text<|/ref|><|det|>[[113, 774, 694, 860]]<|/det|>
+\*Equal Contribution: Zhongyuan Li, Ayush Bhardwaj \(^{8}\) Corresponding Author: James J. Watkins (watkins@polysci.umass.edu), Seok- Woo Lee (seok- woo.lee@uconn.edu)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 85, 888, 636]]<|/det|>
+Abstract: Simultaneous achievement of lightweight, high strength, large fracture strain, and high damping capability has been challenging because some of these mechanical properties are mutually exclusive. Here, we have utilized self- assembled polymeric carbon precursor materials in combination with scalable nanoimprinting lithography to produce nanoporous carbon nanopillars. The produced amorphous carbon nanopillars exhibited ultrahigh strength similar to or even higher than one tenth of their Young's modulus \((E)\) , the most widely used approximation of a fundamental upper limit of material breaking strength. Remarkably, nanoporosity induced via sacrificial template significantly reduced the mass density of amorphous carbon to \(0.66 \sim 0.82g / cm^3\) while the strength of \(E / 10\) is still maintained. Moreover, these nanopillars displayed both elastic and plastic behavior with large fracture strain. A reversible part of the \(\mathrm{sp}^2\) - to- \(\mathrm{sp}^3\) transition produces large elastic strain and a high loss factor (up to 0.033) comparable to Ni- Ti shape memory alloys. The irreversible part of the \(\mathrm{sp}^2\) - to- \(\mathrm{sp}^3\) transition enables plastic deformation, leading to a large fracture strain of up to \(35\%\) . These findings have been substantiated using simulation studies. None of the existing structural materials exhibit a comparable combination of density, strength, deformability, and damping capability. Hence, the results of this study illustrate the potential of both dense and nanoporous amorphous carbon materials as superior structural nanomaterials.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 92, 268, 114]]<|/det|>
+## Introduction:  
+
+<|ref|>text<|/ref|><|det|>[[113, 142, 886, 409]]<|/det|>
+Nanoporous materials are structurally classified as materials having pore sizes typically less than \(100~\mathrm{nm}\) , whose properties are governed by matrix material as well as shape, size, and distribution of pores'. Their high surface area and the tunability of pore topologies facilitate unique physical and chemical properties, making them suitable for various applications such as ion exchange, separation, sensors, \(\mathrm{CO_2}\) capture and storage, water purification, catalysis, renewable energy, drug delivery, and tissue engineering?. In addition, their excellent mechanical properties with the combination of lightweight and high strength render them a potential candidate for future structural applications in aerospace, automobile, and military engineering?.  
+
+<|ref|>text<|/ref|><|det|>[[113, 428, 886, 905]]<|/det|>
+It is difficult to achieve lightweight and high strength simultaneously using classic microstructural control of bulk monolithic materials because the strength of materials is generally proportional to their mass density'. For instance, it is not easy to create a lightweight polymer material that possesses the strength of heavy steel. Interestingly, the creation of nanoporous structures could enable us to reduce the mass density within the same class of materials without the significant sacrifice of strength because nanoscale ligaments can impart ultrahigh strength due to the mechanical size effects, so- called 'Smaller is Stronger' behavior5- 8. Recent nanomechanics studies have demonstrated that a material exhibits a significant fraction of theoretical strength, the upper limit of material strength, if its dimension approaches the nanoscale. One tenth of Young's modulus (E/10) has been considered the most widely accepted approximation of the maximum theoretical breaking strength'. Nanopillars, nanowires, and nanoparticles often exhibit the strength of E/50- E/30, which is 20- 100 times higher strength than their bulk counterparts because the absence of flaws and defects in such a small volume can improve strength substantially'. It is expected that the nanoscale ligaments in a nanoporous structure could also exhibit ultrahigh
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 884, 144]]<|/det|>
+strength through the mechanical size effect, leading to the high effective strength of nanoporous materials.  
+
+<|ref|>text<|/ref|><|det|>[[113, 163, 886, 852]]<|/det|>
+Nanoporous materials have been successfully fabricated by a variety of methods, including dealloying, soft/hard templating, microwave irradiation, additive manufacturing, ion- beam processing, and laser processing<sup>2</sup>. Specifically, nanoporous metals and ceramics are typically fabricated using selective chemical removal of one phase from two- phase alloy mixtures<sup>10</sup> and hard templating with silica colloidal crystals<sup>11</sup>, respectively. Nanoporous metals, ceramics, and their mixtures exhibit reasonably high strength, \(10 \sim 200 \text{MPa}\) , with their high strength- to- weight ratio and a mass density of \(1 \sim 10 \text{g/cm}^3\)<sup>12</sup>. Nevertheless, several factors still inhibit the desired improvement in mechanical properties of nanoporous materials: First, conventional fabrication methods of nanoporous metals or ceramics do not allow the fabrication of nanoporous structures with extremely thin ligament dimensions. Dealloying and templating methods usually limit the ligament thickness down to \(50 \sim 100 \text{nm}^2\) . To exploit the size affected strength more effectively, it is desirable to reduce the ligament thickness down below \(50 \text{nm}\) , which would allow us to maximize the mechanical size effect. Interestingly, the size reduction has also been known to enhance the ductility due to the flaw tolerance effect at the nanoscale. Brittle- to- ductile transition has been observed in metallic glasses<sup>13</sup>, semiconductors<sup>14</sup>, quasicrystals<sup>15</sup>, and nanolattices<sup>16</sup> when their structural length- scale lies in the regime of \(100 \text{nm}\) or smaller. Thus, the reduction of thickness of ligaments could enable the creation of more ductile nanoporous structures. Second, metals and ceramics exhibit relatively low specific strength due to their heavy mass density. To achieve lightweight and ultrahigh strength simultaneously, it is critical to create nanoporous structures from high specific strength materials, such as carbonaceous materials<sup>17,18</sup>.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 886, 390]]<|/det|>
+Recently, nanoporous amorphous carbon materials have been successfully fabricated in various ways, including sequential chemical dealloying, hard and soft templating<sup>2</sup>. These methods successfully produce ligaments with thicknesses much thinner than 50nm. Most of these materials have been investigated to study the capacitance properties<sup>19</sup>, electrochemical properties<sup>20</sup>, and biocompatibility<sup>21</sup>. Moreover, their mechanical properties are also expected to be excellent, specifically strength and ductility, considering the high breaking strength of C- C bond and the mechanical size effect. Therefore, it is essential to systematically study the mechanical properties of nanoporous amorphous carbon and to compare them with those of other advanced structural materials.  
+
+<|ref|>text<|/ref|><|det|>[[113, 409, 886, 886]]<|/det|>
+In this work, therefore, we have developed lightweight nanoporous amorphous carbon nanopillars exhibiting ultrahigh strength, high deformability, and high damping- capability. Nanoporous nanopillar structures were derived by carbonization of the self- assembled nanocomposite of phenol- formaldehyde (PF) resin and bottlebrush block copolymers (BBCP) in combination with low cost and scalable nanoimprinting lithography. The use of BBCP enables precise control over porosity ranging from \(0\%\) to \(59\%\) with remarkably low shrinkage during carbonization (less than \(20\%\) ). These produced specimens possess extremely fine and uniform porous structures with pore diameter \(\sim 50 \mathrm{nm}\) and the ligament thickness in the range of \(20 \mathrm{nm}\) or thinner. Our fully dense amorphous carbon nanopillars and nanoporous structures represent exceptional mechanical properties, including ultrahigh strength comparable to or even higher than \(E / 10\) , a large elastic strain (up to \(13\%\) ), a large fracture strain (up to \(35\%\) ), and high damping capability with the loss factor (up to \(0.033\) ). Due to the lightweight and near- theoretical strength, the specific strength and the modulus of resilience of our materials is higher than any other engineering materials within a similar range of mass density. Micromechanical tests, electron
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 886, 319]]<|/det|>
+energy loss spectroscopy (EELS), Raman spectroscopy, and atomistic simulation revealed that reversible and irreversible portions of the \(\mathrm{sp}^{2} - \mathrm{to} - \mathrm{sp}^{3}\) transition and the nanoscale dimension of ligaments led to large elastic and fracture strains as well as high damping capability. Remarkably, none of the known structural materials achieves this unique combination of mechanical properties in terms of weight, elasticity, plasticity, and damping. Thus, this work demonstrates the preparation of next generation superior structural nanomaterials for aerospace, military, energy, biomedical, filtration, and catalyst applications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 385, 172, 402]]<|/det|>
+## Result  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 426, 572, 446]]<|/det|>
+## Fabrication of fully dense and nanoporous nanopillars  
+
+<|ref|>text<|/ref|><|det|>[[113, 468, 886, 908]]<|/det|>
+Nanopillars of fully dense amorphous carbon and their nanoporous structures were fabricated by spin- coating the film ( \(\sim 5 \mu \mathrm{m}\) ) of self- assembled BBCP and PF thin films followed by nanoimprinting and thermal annealing (Figures 1(a) and 1(b)) (See also Methods). Nanopillars ( \(\sim 300 \mathrm{nm}\) diameter and \(\sim 1 \mu \mathrm{m}\) height) with the five different porosities (0%, 40%, 46%, 51%, and 59%) were prepared to obtain samples with different mass densities ranging from 0.66 to \(1.6 \mathrm{g} / \mathrm{cm}^{3}\) . The details of the porosity and density measurement are available in Supplementary Materials. Scanning electron microscope (SEM) images (Figures 1(c)- 1(e)) show that fully dense and uniform nanoporous structures in nanopillars were created successfully. High- resolution transmission electron microscope (HRTEM) images of both fully dense and nanoporous carbon show the disordered atomic arrangement (Figure 1(f), 1(g) and 1(h)), confirming the formation of amorphous phase. The two halo rings in the selected area diffraction pattern (SADP) (marked by the white arrows in the inset of Figure 1(f) and 1(g)) correspond to the interplanar spacing of \(\sim 1.15 \mathrm{\AA}\) and \(\sim 1.95 \mathrm{\AA}\) , which are typically observed from fully amorphous carbon structures<sup>22</sup>. Moreover,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 886, 354]]<|/det|>
+there is no noticeable difference in atomic arrangements of fully dense and nanoporous structure, and hence the mechanical properties of these nanopillars are expected to be solely governed by the difference in their porosity/porous structure. Note that the atomic arrangement of our specimens is different from the other commonly observed structure of crumbled graphite/graphene networks, which possess folded graphene layers in HRTEM images and several halo diffraction rings due to the higher degree of atomic ordering \(^{23 - 25}\) . The different atomic arrangement of our nanopillars may result from the random orientation of the molecular chains of the PF carbon precursor and the carbonization temperature employed, which impedes the easy crystallization of carbon atoms \(^{19}\) .  
+
+<|ref|>text<|/ref|><|det|>[[113, 375, 886, 780]]<|/det|>
+It is worthwhile to note that nanoimprinting has a significant advantage over commonly used focused- ion- beam (FIB) milling. FIB milling requires a substantial amount of time to create a single nanopillar \(^{26,27}\) whereas a large number of nanopillars produced using nano imprinting lithography (NIL) quickly and efficiently allows us to test several nanopillars and obtain statistically reliable mechanical properties. Additionally, self- assembled BBCP and PF resin composite as an ink for NIL enabled the formation of nanopillar with homogenous porous structure after carbonization (see Method section). This can be attributed to the relatively very small domain size ( \(\sim 50 \mathrm{nm}\) ) of the nanocomposite as compared to smallest dimension of pillar ( \(\sim 300 \mathrm{nm}\) in diameter), permitting the formation of uniform porous structure in nanopillar. Remarkably, the NIL process and developed ink were also utilized for imprinting various three- dimensional architecture with maintained porous structure demonstrating the versatility of this fabrication technique (See Supplementary Materials).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 180, 875, 656]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[114, 682, 884, 805]]<|/det|>
+<center>Figure 1. Synthesis and microstructure of amorphous carbon nanopillar and its nanoporous structures. (a) Schematic diagram of nanopillar fabrication; (b) SEM image of nanopillar array; SEM images of nanopillar and surface structure (c) of fully dense amorphous carbon nanopillar, (d) 46% porosity nanopillar, and (e) 59% porosity nanopillar; HRTEM image of (f) the fully dense amorphous carbon and (g) the 59% porosity nanoporous carbon. The inset shows the selected area diffraction pattern, which includes two halo rings (marked by the white arrows); (h) HRTEM image of the 59% porosity nanoporous carbon. Pore surface is atomically smooth. </center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 462, 110]]<|/det|>
+## In situ micromechanical characterization  
+
+<|ref|>text<|/ref|><|det|>[[113, 125, 886, 830]]<|/det|>
+In situ uniaxial compression tests (Figures 2(a)- 2(d)) revealed a significantly large fracture strain from all nanopillars, comprised of substantial elastic recovery and plastic deformation (See also the Supplementary Video). A permanent height change was also observed when the test was stopped right before the fracture (Figure 2(b)). Interestingly, we could not find any signature of local fracture in the nanoporous structures even after a substantial amount of plastic deformation, implying that the plastic strain is produced by mechanisms that cannot be easily observed by SEM. The mechanical loading was also stopped right after a significant displacement jump, which can be regarded as the initiation of fracture, and the SEM image shows that the fracture occurred via shear banding (Figure 2(c)). Nanoindentation was also performed on the thin film area adjacent to nanopillars to measure Young's moduli precisely. The obtained values matched well to that of data derived from micropillar experiments (Figures 2(d) and 2(e)). We also found that there is negligible depth dependence in our Young's modulus data, indicating that nanoindentation data are not influenced by sink- in or pile- up effects. Nanopillar compression and nanoindentation data of all specimens are available in Supplementary Materials. In general, the yield strength \((\sigma_{y})\) , fracture strength \((\sigma_{f})\) , and Young's modulus \((E)\) decreased as the porosity increased (Figures 2(e) and 2(f)). The power law scaling relations \(^{28}E \sim \rho^{- m}\) (the inset of Figure 2(e)) and \(\sigma_{f} \sim \rho^{- n}\) (the inset of Figure 2(f)) with the relative density \((\rho)\) show \(m = 1.42\) and \(n = 2.03\) , respectively. These results imply that the elastic deformation is more bending- dominant \((m< 1.5)\) , but the fracture is more stretching- dominant \((n > 1.5)^{28}\) . This can be attributed to the complex nanoporous structures which may have resulted in different mechanical responses to elasticity and fracture.  
+
+<|ref|>text<|/ref|><|det|>[[115, 845, 884, 901]]<|/det|>
+Surprisingly, our mechanical data shows that the fracture and yield strengths of all specimens reached and even surpassed \(E / 10\) (marked by the blue dash line in Figure 2(f)), which
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 80, 886, 744]]<|/det|>
+is the most widely used value to approximate a fundamental upper limit of material breaking strength<sup>9</sup>. Even with the presence of nanoporous structure, the fracture and yield strengths do not fall below \(E / 10\). This result indicates that the creation of additional free surface, which sometimes acts as a source of plastic deformation or fracture, does not play a role while the creation of thin ligaments seem to preserve ultrahigh strength via the nanoscale size effect. It has been reported that nanoscale diamonds with \(\sim 100nm\) in thickness show ultrahigh strength and significantly large fracture strain due to the absence of internal defects and smooth surface in their small volume<sup>8,29</sup>. The ligaments in our nanoporous amorphous carbon are around 20\~50nm, which is thinner than these nanoscale diamonds. Thus, there is an even smaller chance of containing any critical flaws in ligaments of our nanoporous structures. Also, the HRTEM images depicts the formation of the atomically smooth pore surfaces, confirming that the produced nanoporous carbon is free from any critical surface flaws. Therefore, the robust C-C covalent bonding and the absence of defects within nanoscale ligaments and on the pore surfaces should be the main reasons for the ultrahigh strength of our nanopillars. We also fabricated micropillars with \(2\mu m\) in diameter and \(6\mu m\) in height (\~260 times larger volume) using FIB milling. We confirmed that all FIB-ed micropillars show nearly the same mechanical properties as nanoimprinted nanopillars (See also Supplementary Materials), affirming that the mechanical properties remain unchanged up to the micrometer scale. Thus, the sample dimension of our nanopillars is small enough not to contain any detrimental defects, such as abnormally large pores or cracks.  
+
+<|ref|>text<|/ref|><|det|>[[114, 758, 886, 884]]<|/det|>
+We also found the fracture strain exhibits an interesting transition with the porosity change (marked by black arrows in Figure 2(g)). In general, the fracture strain of our samples decreases as the porosity increases. However, nanopillar with the highest porosity (59%) shows a higher fracture strain of \(25 \pm 0.11\%\) than that of \(18 \pm 0.45\%\) of the nanopillar with the porosity of 51%.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 85, 886, 636]]<|/det|>
+These results were repeatable as indicated by the small error in data. This remarkable increase is primarily attributed to the increase in plastic strain limit. In other words, the ductility of the material increases as the porosity of the sample rises from \(51\%\) to \(59\%\) . We also observed a similar transition when the pore size \((21.83 \pm 3.32 \text{nm (small)}, 39.41 \pm 3.95 \text{nm (medium)},\) and \(79.42 \pm 10.00 \text{nm (large)})\) is varied while the porosity \((51\%)\) remains unchanged (Figure 2(g)). As the pore size decreases, the ligament thickness also decreases (Figures 1(d) and 1(e)). This implies that the transition of fracture strain is specifically being governed by the ligament thickness. Enhanced ductility (or fracture strain) has been frequently observed at the nanoscale. Metallic glasses \(^{13}\) , semiconductors \(^{14}\) , and quasicrystals \(^{15}\) are extremely brittle at bulk scale but showed significant improvement in ductility due to the flaw tolerance effect at the nanoscale when their dimensions become smaller than \(100nm\) . Also, some ceramic nanolattices, which could be considered as the highly ordered nanoporous structure, exhibited the improved ductility when the thickness of strut becomes smaller than \(50 \sim 100nm^{16,30,31}\) . The ligament thickness of our highest porosity nanopillar is only around \(20nm\) (Figure 1(e)), which is much smaller than \(100nm\) . Hence, based on the flaw tolerance at the nanoscale, our highest porosity nanopillar (or smaller pore size nanopillar) could also exhibit ductility.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 90, 876, 627]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[114, 647, 884, 858]]<|/det|>
+<center>Figure 2. Mechanical properties of amorphous carbon nanopillar and its nanoporous structures. (a) SEM snapshots of microcompression test of \(59\%\) porosity nanoporous carbon nanopillar (See also Supplementary Video); SEM images of \(59\%\) porosity nanoporous carbon nanopillar (b) before and after plastic deformation, and (c) after shear fracture; (d) Representative engineering stress-strain curves; (e) Young's modulus as a function of porosity. The inset shows the power law relation between Young's modulus and relative density; (f) Fracture strength and yield strength as a function of porosity. The sky-blue color bar indicates the magnitude of E/10. The inset shows the power law relation between strength and relative density; (g) (left) fracture, elastic, and plastic strains as a function of porosity. (right) fracture, elastic, and plastic strains as a function of pore size. These specimens have the \(51\%\) porosity, but different pore sizes: \(21.83 \pm 3.32\) nm (small), \(39.41 \pm 3.95\) nm (medium), and \(79.42 \pm 10.00\) nm (large). The black arrows in both figures show the presence of transition in fracture strain. </center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 525, 110]]<|/det|>
+## Mechanisms of elasticity, plasticity, and fracture  
+
+<|ref|>text<|/ref|><|det|>[[113, 131, 886, 397]]<|/det|>
+Mechanisms of elasticity, plasticity, and fractureTo thoroughly investigate the deformation mechanism, we studied deformation- induced changes in atomic arrangement/configuration of amorphous carbon. Here, we utilized fully dense carbon considering the fact that all nanoporous structures irrespective of their porosity are composed of same amorphous carbon as discussed in the previous section. The atomic scale structural change \((\mathrm{sp}^{2} - \mathrm{to} - \mathrm{sp}^{3}\) transition) of a fully dense amorphous carbon nanopillar was analyzed using electron energy loss spectroscopy (EELS), Raman spectroscopy, cyclic loading test, and atomistic simulations. Then, its implication for mechanical properties of nanoporous structures will be discussed in this section.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 419, 602, 438]]<|/det|>
+## Electron Energy Loss Spectroscopy and Raman Spectroscopy  
+
+<|ref|>text<|/ref|><|det|>[[113, 460, 886, 655]]<|/det|>
+EELS analysis was conducted on the pillar at two different locations. The first location was the region compressed to approximately \(35\%\) strain, while the second location was the pillar base, which underwent a negligible amount of plasticity (Figure 3(a)). Two- windows method \(^{32}\) was used to determine the fraction of \(\mathrm{sp}^{2}\) bond by integrating the intensity of \(1\mathrm{s} - \pi^{*}\) and \(1\mathrm{s} - \sigma^{*}\) from 282- \(286eV\) window and 288- \(298eV\) window, respectively. The result shows that the \(\mathrm{sp}^{2}\) fraction \((\sim 94\%)\) of the compressed nanopillar is about \(4\%\) lower to the pillar base \((\sim 98\%)\) .  
+
+<|ref|>text<|/ref|><|det|>[[113, 677, 886, 907]]<|/det|>
+Raman spectroscopy analysis was conducted before and after compressing identical nanopillars by \(\sim 35\%\) in strain (Figure 3(b)) to further explore the carbon microstructure changes. In the Raman spectra range of \(800cm^{- 1}\) to \(2000cm^{- 1}\) , two characteristic peaks are typically observed for amorphous carbon materials. The D peak, centered at \(\sim 1380cm^{- 1}\) , corresponds the \(\mathrm{A_{1g}}\) breathing mode only in aromatic rings, while the G peak, centered at \(\sim 1560cm^{- 1}\) , corresponds to the \(\mathrm{E_{2g}}\) stretching mode of \(\mathrm{sp}^{2}\) atoms in both aromatic rings and olefinic chains \(^{33}\) . The Raman spectra shows the increase in D peak intensity and the positive shift of G peak after the compression. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 87, 886, 530]]<|/det|>
+positive shift of G peak is related to the increase in \(\mathrm{sp}^{3}\) bonds in \(\mathrm{sp}^{2}\) - dominant amorphous carbon \(^{33}\) . The increase in D peak intensity occurs when the crystal symmetry of graphite or graphene is disrupted by local disorder, resulting from the localized formation of defects or \(\mathrm{sp}^{3}\) bonds \(^{34 - 36}\) . In addition, recent experimental studies on carbon nanotubes showed that the lateral compression of carbon nanotube could increase the D peak \(^{37}\) . We expect that uniaxial compression could induce a similar topological change locally. If a slightly curved hexagonal planar structure is present in a nanopillar, uniaxial compression could fold such structure and increase the local curvature, enhancing the D peak intensity. Although it is difficult to experimentally confirm the local curvature change, our molecular dynamic (MD) simulation captured a permanent folding of hexagonal planar structure after uniaxial compression (The detailed discussion is also available in Supplementary Materials; See also Supplementary Video). Thus, the creation of \(\mathrm{sp}^{3}\) bonds and the local topological change in internal structures could be the reasons for the increase in D peak intensity in our Raman spectra data.  
+
+<|ref|>text<|/ref|><|det|>[[113, 550, 886, 885]]<|/det|>
+The combination of EELS and Raman data suggest strongly that a local \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition occurs during compression. In fact, the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition has been widely observed phenomenon in amorphous carbon under hydrostatic \(^{38,39}\) , biaxial \(^{40}\) , and uniaxial compression \(^{41}\) . Hydrostatic compression tests often apply extremely high pressure (10- 40GPa) to induce the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. This raises a question of if the uniaxial fracture strength of our fully dense nanopillars, \(\sim 4\) GPa, may be too low for such transition. However, it is important to consider that the strain is the more critical factor than the stress (or pressure) because the strain is directly related to the atomic displacement, which influences the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. Note that hydrostatic pressure of 20- 40GPa changes the linear dimension of amorphous carbon by less than 10%. But in our case, the fracture strain is around 35%, which is 3 times larger than the hydrostatic pressure case.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 886, 321]]<|/det|>
+Therefore, it is feasible to induce a significant atomic displacement capable of triggering the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition within our nanopillar. Also, the stress required for the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is strongly dependent on the initial carbon arrangement. The recent density functional theory calculation claimed that over \(10\%\) increase in \(\mathrm{sp}^{3}\) content is possible through the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition in amorphous carbon with \(1.8g / cm^{3}\) in density (similar to our \(1.6g / cm^{3}\) ) only under around 5GPa of hydrostatic pressure<sup>38</sup>. Based on all these results, our amorphous carbon could have an initial carbon arrangement that can be adjusted easily even at relatively low stress level.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 343, 266, 361]]<|/det|>
+## Cyclic loading test  
+
+<|ref|>text<|/ref|><|det|>[[113, 382, 886, 578]]<|/det|>
+One typical feature of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is the mechanical hysteresis<sup>42</sup>. This hysteresis is related to the different energy barriers between the forward \((\mathrm{sp}^{2} \rightarrow \mathrm{sp}^{3})\) and the backward \((\mathrm{sp}^{3} \rightarrow \mathrm{sp}^{2})\) transition (Figure 3(c)). Due to the different energy barrier, the phase transformation during loading \((\mathrm{sp}^{2} \rightarrow \mathrm{sp}^{3})\) and unloading \((\mathrm{sp}^{3} \rightarrow \mathrm{sp}^{2})\) produces different critical stresses for transition under the load- controlled condition, which is our case. As a result, the stress- strain path is different between loading and unloading, i.e., mechanical hysteresis<sup>43</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[113, 598, 886, 880]]<|/det|>
+To confirm that the closed hysteresis loops indeed exist in stress- strain data, a cyclic loading test was conducted on the fully dense carbon pillar. The loss factor \((\eta)\) can be calculated with \(\eta = \frac{\Delta W}{\pi W_{max}}\) for the micropillar compression cyclic test<sup>44</sup>, where \(\Delta W\) is the dissipated energy per stress- release cycle, and \(W_{max}\) is the maximum stored energy per unit volume over the cycle. Our fully dense pillar showed the mechanical hysteresis loop (the magenta colored area in Figure 3(d)) with \(\eta = 0.033\) , which corresponds to the high damping materials \((\eta > 0.015)^{45}\) and is close to the loss factor of commercial Nitinol (Ni- Ti) shape memory alloys \((\eta = 0.028 \sim 0.041)\) (Cyclic stress- strain curves and loss factor data of all nanoporous nanopillars are available in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 885, 249]]<|/det|>
+Supplementary Materials). Based on EELS, Raman spectra, and cyclic loading test, a local \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is likely to be the main mechanism of both elastic and plastic deformation. A reversible part of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition produces large elastic strain and a high loss factor. The irreversible part of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition produces plastic deformation, leading to a large fracture strain of up to \(35\%\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 273, 288, 290]]<|/det|>
+## Atomistic simulations  
+
+<|ref|>text<|/ref|><|det|>[[113, 308, 886, 896]]<|/det|>
+Atomistic simulations were used to probe how atomic arrangement is changed under uniaxial loading and unloading. The amorphous carbon structure with the similar mass density \((1.6g / cm^3)\) with the high fraction of \(\mathrm{sp}^{2}\) bonds \((\sim 90\%)\) was constructed, and the fractions of \(\mathrm{sp}^{2}\) and \(\mathrm{sp}^{3}\) bonds were monitored during uniaxial compression up to \(40\%\) strain, which is similar to the experimental fracture strain \((\sim 35\%)\) , and during unloading. The simulation results show that the fraction of \(\mathrm{sp}^{3}\) increases by \(\sim 13\%\) at \(40\%\) strain, but a considerable amount of newly formed \(\mathrm{sp}^{3}\) bonds is transformed back to \(\mathrm{sp}^{2}\) bonds during unloading (Figure 3(e)). This result is similar to the hydrostatic pressure studies showing that most newly formed \(\mathrm{sp}^{3}\) bonds are transformed back to \(\mathrm{sp}^{2}\) bonds during unloading<sup>42</sup>. When unloading is completed, only \(3.7\%\) of \(\mathrm{sp}^{2}\) bonds were permanently reduced by forming the similar amount of \(\mathrm{sp}^{3}\) bonds. This simulation result is similar to our EELS data that showed about \(4\%\) reduction of the \(\mathrm{sp}^{2}\) bonds after unloading. Thus, our simulation data agrees well with our experimental findings and confirms the plastic strain resulted from the irreversible part of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. This atomic scale transformation also explains why the signature of plasticity could not be observed from the high- resolution SEM images even after the large amount of deformation (Figure 2(b)). We also found that the slope of the simulated unloading curve changes nonlinearly when the new \(\mathrm{sp}^{3}\) bonds are transformed back to the \(\mathrm{sp}^{2}\) bonds (two broken red lines in Figure 3(f)). This nonlinear change resembles the shape
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 886, 389]]<|/det|>
+of the experimental unloading curve (two broken red lines in Figure 3(d)), which could be the main reason for mechanical hysteresis. Note that the high stress level of atomistic simulations is typical and unavoidable due to the extremely high strain rate<sup>46,47</sup>. We also studied the strain rate effect (See also Supplementary Information) and confirmed that the high stress level in our simulation data is directly associated with the strain rate. As discussed before, the strain is the primary factor to control the atomic displacement, which is directly related to the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. Thus, our simulation results should be considered valid for studying the fundamental mechanisms of elastic and plastic deformation reasonably in terms of large strain (40%) similar to experimental value (35%).  
+
+<|ref|>text<|/ref|><|det|>[[113, 409, 886, 885]]<|/det|>
+Interestingly, plastic strain (Figure 2(g)) and loss factor (See also Supplementary Materials) of nanoporous nanopillars are generally lower than those of fully dense ones. These results can also be understood in terms of the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition. The \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition is driven by compressive stress because the \(\mathrm{sp}^{3}\) bond is more energetically preferred under compression. Thus, both plastic strain and loss factor must be affected by the distribution and magnitude of compressive stress. Because of complex nanoporous structures, ligaments in nanoporous nanopillars should undergo bending stress, which includes both compressive and tensile stresses. The local regions under tensile stress do not undergo the \(\mathrm{sp}^{2}\) - to- \(\mathrm{sp}^{3}\) transition, leading to the lower plastic strain and loss factor of nanoporous nanopillars. However, the plastic strain and loss factor could be enhanced if the size of ligaments become extremely thin. As discussed in the previous section, the flaw tolerance effect at the nanoscale could suppress the fracture of ligaments in regions under tension and enable further deformation. This could be the reason for nanopillar with the highest porosity (59%) to exhibit higher fracture strain and higher loss factor than the nanopillar with the porosity of 51%.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[130, 140, 876, 620]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[114, 648, 884, 825]]<|/det|>
+<center>Figure 3. Deformation mechanisms of amorphous carbon. (a) EELS measurements of the fraction of \(\mathrm{sp}^2\) bonded carbon atoms in micropillar (left graph) and in base (right graph); (b) Raman spectroscopy data before and after compression; (c) Schematic diagram of energy barriers for elastic loading and unloading; (d) Engineering stress-strain curve of cyclic compression test. Black curves show the recoverable hysteresis deformation. Blue curves show the non-linear recovery after plastic deformation. Red broken lines show how the slope of unloading curve changes; (e) Simulated \(\mathrm{sp}^2 /\mathrm{sp}^3\) fraction of fully dense amorphous carbon. (g) Simulated stress-strain curve of fully dense amorphous carbon and the snapshots of microstructural evolution during loading and unloading (See also Supplementary Video). Red broken lines in the graph show how the slope of unloading curve changes. </center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 92, 207, 109]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[113, 130, 886, 677]]<|/det|>
+Mechanical characterizations revealed that all nanopillars in this study exhibit remarkably high yield and fracture strengths, which are similar to or even higher than the theoretical breaking strength, \(E / 10\) (Figure 3(f)). In addition, our fully dense specimen has a low mass density \((1.6g / cm^3)\) , and the introduction of nanoscale pores reduces the mass density even further \((0.66 \sim 0.82g / cm^3)\) . As a result, the plot of the fracture strength versus mass density shows that our nanopillars are among the strongest materials for their mass density (Figure 4(a)). These superior mechanical properties could result primarily from the ligament size effects and the sample size effects. The nanoscale thickness of ligaments is too small to contain any significant flaws that would detrimentally affect the strength, and the overall volume of nanopillars may not be large enough to contain flaws. As shown by the FIB- milled micropillar compression tests, the sample size effect seems to be relatively weak because the self- assembly of BBCPs produces uniform and ordered nanoporous structures, which preserves the consistent mechanical properties up to the micrometer scale. The sample size effect would be more observable if the sample dimension goes beyond millimeters and centimeters because the weaker spots will be statistically probable to be present in such large volume. It would be important to study the sample size effect in a larger length scale as a future work.  
+
+<|ref|>text<|/ref|><|det|>[[114, 696, 886, 891]]<|/det|>
+In addition, our nanopillars exhibit large fracture strain (Figure 3(g)) and high damping capability (Figure 4(d)). Recent study on the state- of- the- art amorphous carbon nanolattices, which could be considered as a highly ordered porous structure, showed the near theoretical breaking strength \((E / 10)\) , too \(^9\) , but they exhibited extreme brittleness with no measurable plastic strain and no damping capability. In contrast, our nanoporous amorphous carbon nanopillars are much more deformable than that of the 3D architected amorphous carbon nanolattice structures with the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 884, 180]]<|/det|>
+fracture strain of \(\sim 10\%\) only. The latter requires a more time- consuming and costly fabrication process, not to mention the huge volume shrinkage during pyrolysis process which hampers precise control over the final feature dimensions.  
+
+<|ref|>text<|/ref|><|det|>[[113, 201, 886, 641]]<|/det|>
+Due to the ultrahigh strength and low mass density, our materials exhibit the extraordinary combination of specific yield strength and specific Young's modulus. They occupy the white space in the plot of specific yield strength vs. specific Young's modulus (Figure 4(b)). This indicates that our materials exhibit the unprecedentedly high specific modulus of resilience \((\sim 5,000\mathrm{MJ / m^3 / (kg / m^3)})\), which corresponds to the maximum possible elastic energy absorption and release per unit volume and per unit density. The straight contour lines show the magnitude of specific modulus of resilience. The specific modulus of resilience of our materials is certainly the highest within their range of specific yield strength and is comparable with that of elastomers, which show the highest modulus of resilience due to their extremely large elastic deformability. Also, for a given fracture strain, our materials show the highest specific strength among all materials, implying that they are highly deformable even with ultrahigh strength. Some high strength composites show relatively similar specific fracture strength, but their ductility is less than \(1\%\) , which is more than a magnitude much smaller than that of our nanopillars \((17 \sim 35\%)\) .  
+
+<|ref|>text<|/ref|><|det|>[[113, 661, 886, 891]]<|/det|>
+As mentioned in the previous section, due to the reversible \(\mathrm{sp}^{2} - \mathrm{to} - \mathrm{sp}^{3}\) transition, our materials can also be classified as highly damping capable materials, which have the loss factor of \(0.015 \sim 0.033\) . Thus, our amorphous carbon nanopillar and its nanoporous structures provide unusual combinations of lightweight, high strength, large fracture strain, and high damping capability. Although there have been similar studies on micromechanical testing of fully dense amorphous carbon<sup>17,48,49</sup>, these systems used different initial conditions with either the crumbled graphite networks or the higher density of \(\mathrm{sp}^{3}\) bonds, and their mass density was not measured
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 80, 888, 742]]<|/det|>
+experimentally. More importantly, our work is unique in aspect that the mass density is tunable by introducing nanoporous structure without sacrificing \(E / 10\) relation of yield strength, a result that no other approach has achieved. In addition, it is also noteworthy to emphasize that our synthesis method is both rapid and scalable. It is possible to create both fully dense and nanoporous structures with even centimeter- scale dimensions in width and hundreds of micrometers in thickness (See also Supplementary Materials). This efficient and scalable fabrication method makes our materials more suitable for potential engineering applications, compared to nanolattices that have limited scalability beyond \(200\mu m\) in dimension. Moreover, there is a large degree of freedom in materials design. The pore distribution can be varied by adding both linear block copolymer (LBCP) and BBCP or by varying the molecular weight of BBCP, so the non- uniform pore distribution, for instance, bimodal distribution, can be easily created<sup>50,51</sup>. Metallic or ceramic nanoparticles with the diameter less than 5nm can be embedded into ligaments of nanoporous structures<sup>52</sup>. A nanoscale ceramic layer can be coated on the pore surfaces using the atomic layer deposition<sup>53</sup>. All these nanoporous carbon and their composite materials will provide a wide range of structure- property- processing controllability. In summary, lightweight, superior mechanical properties, scalable fabrication method, and tunable microstructures of nanoporous amorphous carbon promote the fabrication of not only advanced structural materials for military and aerospace applications but also mechanically robust nanoporous structures for energy, biomedical, filtration, and catalyst applications.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[116, 171, 880, 707]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[114, 730, 884, 819]]<|/det|>
+<center>Figure 4. Mechanical properties of amorphous carbon nanopillar and its nanoporous structures. (a) Compressive strength vs. density. This plot includes all published data of nanoporous materials and nanolattices. Their citation is available in Supplementary Materials.; (b) Specific yield strength vs. fracture strain; (c) Specific Young's modulus vs. specific yield strength. The broken line contour shows the magnitude of specific modulus of resilience. </center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 207, 108]]<|/det|>
+## Reference:  
+
+<|ref|>text<|/ref|><|det|>[[111, 125, 888, 870]]<|/det|>
+1. Lu, G. Q. M. & Zhao, X. S. Nanoporous materials: science and engineering. vol. 4 (World Scientific, 2004).  
+2. Hadden, M., Martinez-Martin, D., Yong, K.-T., Ramaswamy, Y. & Singh, G. Recent Advancements in the Fabrication of Functional Nanoporous Materials and Their Biomedical Applications. Materials 15, 2111 (2022).  
+3. Gibson, L. J. & Ashby, M. F. Cellular Solids: Structure and Properties. Cambridge Solid State Science Series (Cambridge University Press, 1997). doi:DOI: 10.1017/CBO9781139878326.  
+4. Ashby, M. F. & CEBON, D. Materials selection in mechanical design. Le Journal de Physique IV 3, C7-1 (1993).  
+5. Zhang, H. et al. Approaching the ideal elastic strain limit in silicon nanowires. Sci Adv 2, e1501382 (2023).  
+6. Greer, J. R. & De Hosson, J. Th. M. Plasticity in small-sized metallic systems: Intrinsic versus extrinsic size effect. Prog Mater Sci 56, 654-724 (2011).  
+7. Zhu, T. & Li, J. Ultra-strength materials. Prog Mater Sci 55, 710-757 (2010).  
+8. Dang, C. et al. Achieving large uniform tensile elasticity in microfabricated diamond. Science (1979) 371, 76-78 (2021).  
+9. Bauer, J., Schroer, A., Schwaiger, R. & Kraft, O. Approaching theoretical strength in glassy carbon nanolattices. Nat Mater 15, 438-443 (2016).  
+10. Biener, J. et al. Size Effects on the Mechanical Behavior of Nanoporous Au. Nano Lett 6, 2379-2382 (2006).  
+11. Saito, Y. et al. Preparation of ordered nanoporous indium tin oxides with large crystallites and individual control over their thermal and electrical conductivities. ACS Appl Mater Interfaces 13, 15373-15382 (2021).  
+12. Yang, W. et al. Light, strong, and stable nanoporous aluminum with native oxide shell. Sci Adv 7, eabb9471 (2023).  
+13. Jang, D. & Greer, J. R. Transition from a strong-yet-brittle to a stronger-and-ductile state by size reduction of metallic glasses. Nat Mater 9, 215-219 (2010).  
+14. Östlund, F. et al. Brittle-to-Ductile Transition in Uniaxial Compression of Silicon Pillars at Room Temperature. Adv Funct Mater 19, 2439-2444 (2009).  
+15. Zou, Y. et al. Superior room-temperature ductility of typically brittle quasicrystals at small sizes. Nat Commun 7, 12261 (2016).  
+16. Lee, S.-W., Jafary-Zadeh, M., Chen, D. Z., Zhang, Y.-W. & Greer, J. R. Size Effect Suppresses Brittle Failure in Hollow Cu60Zr40 Metallic Glass Nanolattices Deformed at Cryogenic Temperatures. Nano Lett 15, 5673-5681 (2015).  
+17. Zhang, X. et al. Theoretical strength and rubber-like behaviour in micro-sized pyrolytic carbon. Nat Nanotechnol 14, 762-769 (2019).  
+18. Hu, M. et al. Compressed glassy carbon: An ultrastrong and elastic interpenetrating graphene network. Sci Adv 3, e1603213 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 886, 128]]<|/det|>
+19. Bhardwaj, A. et al. Large-Pore Ordered Mesoporous Turbostratic Carbon Films Prepared Using Rapid Thermal Annealing for High-Performance Micro-pseudocapacitors. ACS Appl Mater Interfaces 13, 61027-61038 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[112, 127, 884, 201]]<|/det|>
+20. Yun, Y. S. & Jin, H.-J. Electrochemical performance of heteroatom-enriched amorphous carbon with hierarchical porous structure as anode for lithium-ion batteries. Mater Lett 108, 311-315 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[112, 201, 884, 240]]<|/det|>
+21. Jain, S., Sharma, A. & Basu, B. Vertical electric field stimulated neural cell functionality on porous amorphous carbon electrodes. Biomaterials 34, 9252-9263 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[112, 240, 884, 297]]<|/det|>
+22. Neidhardt, J., Hultman, L. & Czigány, Zs. Correlated high resolution transmission electron microscopy and X-ray photoelectron spectroscopy studies of structured CNx (0<x<0.25) thin solid films. Carbon N Y 42, 2729-2734 (2004).  
+
+<|ref|>text<|/ref|><|det|>[[112, 297, 884, 335]]<|/det|>
+23. Shiell, T. B. et al. Graphitization of glassy carbon after compression at room temperature. Phys Rev Lett 120, 215701 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[112, 335, 884, 373]]<|/det|>
+24. Shiell, T. B. et al. Nanocrystalline hexagonal diamond formed from glassy carbon. Sci Rep 6, 37232 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[112, 373, 886, 411]]<|/det|>
+25. Wu, Y. et al. Strong amorphous carbon prepared by spark-plasma sintering C60. Journal of the American Ceramic Society 104, 1655-1660 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[112, 411, 884, 467]]<|/det|>
+26. Yang, X. et al. Integration of nanoimprint lithography with block copolymer directed self-assembly for fabrication of a sub-20 nm template for bit-patterned media. Nanotechnology 25, 395301 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[112, 467, 884, 505]]<|/det|>
+27. Lee, S. et al. FIB-induced dislocations in Al submicron pillars: Annihilation by thermal annealing and effects on deformation behavior. Acta Mater 110, 283-294 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[112, 505, 884, 542]]<|/det|>
+28. Bauer, J. et al. Nanolattices: An Emerging Class of Mechanical Metamaterials. Advanced Materials 29, 1701850 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[112, 542, 884, 579]]<|/det|>
+29. Banerjee, A. et al. Ultralarge elastic deformation of nanoscale diamond. Science (1979) 360, 300-302 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[112, 580, 884, 617]]<|/det|>
+30. Meza, L. R., Das, S. & Greer, J. R. Strong, lightweight, and recoverable three-dimensional ceramic nanolattices. Science (1979) 345, 1322-1326 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[112, 617, 884, 655]]<|/det|>
+31. Michelson, A., Flanagan, T. J., Lee, S.-W. & Gang, O. High-strength, lightweight nano-architected silica. Cell Rep Phys Sci 4, (2023).  
+
+<|ref|>text<|/ref|><|det|>[[112, 655, 884, 712]]<|/det|>
+32. Bruley, J., Williams, D. B., Cuomo, J. J. & Pappas, D. P. Quantitative near-edge structure analysis of diamond-like carbon in the electron microscope using a two-window method. J Microsc 180, 22-32 (1995).  
+
+<|ref|>text<|/ref|><|det|>[[112, 712, 884, 750]]<|/det|>
+33. Ferrari, A. C. & Robertson, J. Interpretation of Raman spectra of disordered and amorphous carbon. Phys Rev B 61, 14095 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[112, 750, 884, 788]]<|/det|>
+34. Lucchese, M. M. et al. Quantifying ion-induced defects and Raman relaxation length in graphene. Carbon N Y 48, 1592-1597 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[112, 788, 884, 825]]<|/det|>
+35. Eckmann, A. et al. Probing the nature of defects in graphene by Raman spectroscopy. Nano Lett 12, 3925-3930 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[112, 826, 884, 863]]<|/det|>
+36. Cancado, L. G. et al. Disentangling contributions of point and line defects in the Raman spectra of graphene-related materials. 2d Mater 4, 025039 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[112, 863, 884, 901]]<|/det|>
+37. Picheau, E. et al. Intense Raman D Band without Disorder in Flattened Carbon Nanotubes. ACS Nano 15, 596-603 (2021).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 886, 128]]<|/det|>
+38. Jiang, X., Ārhammar, C., Liu, P., Zhao, J. & Ahuja, R. The R3-carbon allotrope: a pathway towards glassy carbon under high pressure. Sci Rep 3, 1877 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[112, 128, 884, 166]]<|/det|>
+39. Shiell, T. B. et al. In situ analysis of the structural transformation of glassy carbon under compression at room temperature. Phys Rev B 99, 24114 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[112, 166, 884, 204]]<|/det|>
+40. McCulloch, D. G., McKenzie, D. R. & Goringe, C. M. Ab initio simulations of the structure of amorphous carbon. Phys Rev B 61, 2349 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[112, 204, 884, 260]]<|/det|>
+41. Zhao, Y., Qian, C., Gladkikh, V. & Ding, F. Simulated Pressure-temperature Carbon Structure Map obtained through uniaxial compression of Bulk C60. Carbon N Y 202, 554-560 (2023).  
+
+<|ref|>text<|/ref|><|det|>[[112, 260, 884, 297]]<|/det|>
+42. Zeng, Z. et al. Structural transition in cold-compressed glassy carbon. Phys Rev Mater 3, 033608 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[112, 297, 884, 335]]<|/det|>
+43. Sypek, J. T. et al. Superelasticity and cryogenic linear shape memory effects of CaFe2As2. Nat Commun 8, 1083 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[112, 335, 884, 373]]<|/det|>
+44. Juan, J. S., Nó, M. L. & Schuh, C. A. Nanoscale shape-memory alloys for ultrahigh mechanical damping. Nat Nanotechnol 4, 415-419 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[112, 373, 884, 410]]<|/det|>
+45. Zhang, J., Perez, R. J. & Lavernia, E. J. Documentation of damping capacity of metallic, ceramic and metal-matrix composite materials. J Mater Sci 28, 2395-2404 (1993).  
+
+<|ref|>text<|/ref|><|det|>[[112, 410, 886, 466]]<|/det|>
+46. Park, H. et al. Toward the constitutive modeling of epoxy matrix: Temperature-accelerated quasi-static molecular simulations consistent with the experimental test. Compos B Eng 142, 131-141 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[112, 466, 884, 522]]<|/det|>
+47. Fan, J., Anastassiou, A., Macosko, C. W. & Tadmor, E. B. Molecular dynamics predictions of thermomechanical properties of an epoxy thermosetting polymer. Polymer (Guildf) 196, 122477 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[112, 522, 884, 560]]<|/det|>
+48. Yoon, J. et al. In situ tensile and fracture behavior of monolithic ultra-thin amorphous carbon in TEM. Carbon N Y 196, 236-242 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[112, 560, 884, 598]]<|/det|>
+49. Liu, C., Lin, Y., Zhou, Z. & Li, K.-Y. Dual phase amorphous carbon ceramic achieves theoretical strength limit and large plasticity. Carbon N Y 122, 276-280 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[112, 598, 884, 636]]<|/det|>
+50. Fei, H.-F. et al. Ordered nanoporous carbons with broadly tunable pore size using bottlebrush block copolymer templates. J Am Chem Soc 141, 17006-17014 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[112, 636, 884, 692]]<|/det|>
+51. Fei, H.-F. et al. One-step synthesis of hierarchical, bimodal nanoporous carbons via co-templating with bottlebrush and linear block copolymers. Chemistry of Materials 32, 6055-6061 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[112, 693, 884, 748]]<|/det|>
+52. Song, D.-P. et al. Millisecond photothermal carbonization for in-situ fabrication of mesoporous graphitic carbon nanocomposite electrode films. Carbon N Y 174, 439-444 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[112, 749, 884, 805]]<|/det|>
+53. Bhardwaj, A. et al. Large-Pore Ordered Mesoporous Turbostratic Carbon Films Prepared Using Rapid Thermal Annealing for High-Performance Micro-pseudocapacitors. ACS Appl Mater Interfaces 13, 61027-61038 (2021).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 208, 108]]<|/det|>
+## METHOD  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 133, 681, 152]]<|/det|>
+## Synthesis of brush block copolymer and phenol formaldehyde resin  
+
+<|ref|>text<|/ref|><|det|>[[114, 167, 886, 360]]<|/det|>
+Polydimethylsiloxane- b- poly(ethylene oxide) (PDMS- b- PEO) brush block copolymer was synthesized following ring opening metathesis polymerization (ROMP) using Grubbs generation III catalyst following our previous report<sup>19,50,51</sup>. The molecular weight of brush block copolymer was varied by changing the overall degree of polymerization (DP) which in turn was controlled by changing the feed ratio of monomer and catalyst as described previously<sup>50</sup>. The obtained molecular weight at different DP was calculated to be 250 KDa/mol, 500 KDa/mol and 800 KDa/mol.  
+
+<|ref|>text<|/ref|><|det|>[[113, 375, 886, 640]]<|/det|>
+Phenol formaldehyde (PF) resin was synthesized in basic polymerization medium<sup>50</sup>. Phenol is melted at \(42^{\circ}\mathrm{C}\) and 20 wt.\% of NaOH (sodium hydroxide) was added dropwise to it. Formaldehyde was added in the molar ratio of 2 as compared to that of phenol and the entire mixture solution is stirred at \(70^{\circ}\mathrm{C}\) for 1 hr. After cooling down the mixture, the solution pH was neutralized using 0.5 M HCl solution. Water was removed in presence of nitrogen flow overnight, following that ethanol was added to the solution to remove the formed sodium chloride (NaCl). Finally, ethanol was removed by nitrogen and redissolved in the THF to get the desired concentration.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 690, 468, 709]]<|/det|>
+## Fabrication of nanoimprinted nanopillars  
+
+<|ref|>text<|/ref|><|det|>[[114, 724, 886, 884]]<|/det|>
+The carbon precursor film is comprised of PDMS- b- PEO brush block copolymer (BBCP) of different molecular weight as the soft sacrificial template (porogen) and PF resin as the carbon source. The BBCP and PF resin were dissolved in THF separately to achieve a concentration of 40 and \(50\mathrm{mg / mL}\) , respectively. The BBCP and PF resin were added in the different weight ratio namely 1:1.5, 1:2, 1:3 and 1:4.0 and the solvent was evaporated with a nitrogen flow to increase
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 84, 888, 811]]<|/det|>
+the concentration to about \(300\mathrm{mg / ml}\) . The prepared solution was spin- coated at \(1000\mathrm{rpm}\) for 30s on Si wafer substrate. The substrate was ultrasonically cleaned with isopropanol and deionized water for \(10\mathrm{min}\) each, followed by UV- ozone treatment for \(15\mathrm{min}\) prior to coating. Carbon precursor film was patterned via nano imprinting lithography (NIL) using h- pdms stamp. H- pdms stamp was placed on spin coated carbon precursor film followed by heating it at \(170^{\circ}\mathrm{C}\) for 10 minutes under constant pressure of \(180\mathrm{psi}\) . Applied pressure and temperature ensured pattern transfer from the h- pdms stamp to the carbon precursor film with simultaneous crosslinking of the PF resin. Preparation of h- pdms stamp from the Si imprint master was carried out as reported previously<sup>54</sup>. Stamp was removed from the carbon precursor film after cooling down the system to the room temperature. Finally, imprinted carbon precursor film was carbonized in tube furnace at \(750^{\circ}\mathrm{C}\) in nitrogen atmosphere for \(1\mathrm{hr}\) . with the heating rate of \(10^{\circ}\mathrm{C / min}\) . This resulted in the formation of nanostructured mesoporous carbon with film thickness of \(\sim 5\mathrm{um}\) . During the carbonization process BBCP degrades completely and PF resin gets converted into carbon resulting in the formation of porous carbon with the nanostructure transferred using NIL. The residual carbon film from the surface of nanostructured porous carbon was removed using reactive ion etching in \(\mathrm{O_2}\) plasma (HF power - \(10\mathrm{W}\) , pressure - \(12\mathrm{mTorr}\) , \(\mathrm{O_2} - 40\mathrm{scm}\) ) for 90s. Fully dense patterned carbon film was fabricated using the same procedure but without any BBCP in it. We also prepared nanoporous carbon thin film following the procedure described above for porosity and density measurement without performing NIL. Moreover, to demonstrate the versatility of this approach we produced nanoporous carbon with different architecture specially motheye, pyramid and pillar by using their respective desired h- pdms stamp keeping other processes same.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 91, 370, 110]]<|/det|>
+## Nanomechanical experiments:  
+
+<|ref|>text<|/ref|><|det|>[[114, 131, 885, 293]]<|/det|>
+Nanodimentation was performed on the thin film region of the carbon sample at room temperature by using an iNano™ system (Nanomechanics, TN, USA). The indentation was performed utilizing the standard Berkovich tip nanoindentation technique, with an indentation strain rate of \(0.2 \mathrm{s}^{- 1}\) . The indentation depths are set as 1/10 of the thickness of the carbon samples to make sure that the indentation data are not affected by the silicon substrate.  
+
+<|ref|>text<|/ref|><|det|>[[114, 313, 886, 509]]<|/det|>
+In situ uniaxial compression tests were performed by the nanoindenter (Nanoflip™, Nanomechanics Inc., TN, USA) with a flat diamond tip, which was installed in a field- emission gun scanning electron microscope (SEM) (JSM- 6335F, JEOL, Japan). The nanopillars were compressed under a constant displacement rate of \(10 \mathrm{nm / s}\) . The stress- strain curves were obtained through the corresponding load- displacement data and were corrected by the Sneddon punch method<sup>55</sup>. The entire deformation process was also recorded to avoid strain measurement errors.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 574, 395, 592]]<|/det|>
+## Microstructure characterization:  
+
+<|ref|>text<|/ref|><|det|>[[114, 614, 886, 882]]<|/det|>
+The microstructure of porous and fully dense carbon pillars was characterized by Titan Themis AC- STEM (ThermoFisher) at the accelerating voltage of \(300\mathrm{kV}\) . The cross- sectional TEM samples were prepared using the lift- out technique with a focused ion beam (FIB) instrument (FEI Helios, ThermaFisher). A protective carbon layer was first deposited on the carbon thin film and the micropillar. Milled graduated trenches were introduced into the bottom Si substrate utilizing \(\mathrm{Ga + }\) ion beams. With the samples mounted on the micromanipulator, careful cutting was carried out on the side and bottom sections. Subsequently, the samples were lifted out and positioned onto a copper grid. Finally, a thorough cleaning and thinning process was performed until achieving the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 884, 180]]<|/det|>
+desired final thin TEM sample. EELS analysis was also performed in Titan Themis AC- STEM with the accelerating voltage of \(250\mathrm{kV}\) and the EELS curves were analyzed by Two- window method \(^{32}\) to calculate the relative fractions of \(\mathrm{sp}^{2}\) and \(\mathrm{sp}^{3}\) phases.  
+
+<|ref|>text<|/ref|><|det|>[[115, 201, 884, 293]]<|/det|>
+The phase transformation upon compression was characterized using Raman spectroscopy with a \(633\mathrm{nm}\) laser (Renishaw Ramascope 2000). The Raman spectra were fitted with Gaussian peaks to quantitatively evaluate the D and G band changes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 357, 401, 376]]<|/det|>
+## Molecular Dynamics Simulations:  
+
+<|ref|>text<|/ref|><|det|>[[114, 398, 885, 627]]<|/det|>
+The amorphous carbon structures were generated employing two key methods: the liquid- quench method \(^{56,57}\) . The overall process of generating the model carbon structures, based on an experimental density of \(1.6\mathrm{g / cm}^3\) , involves seven steps (more details are described in Supplementary Materials). Based on the generated amorphous configuration, the uniaxial mechanical test was conducted using the Large- scale Atomic/Molecular Massively Parallel Simulator (LAMMPS) package \(^{58}\) . Interaction potentials involving carbon particles are characterized by Adaptive Intermolecular Reactive Empirical Bond Order (AIREBO) Potential \(^{59}\) .  
+
+<|ref|>text<|/ref|><|det|>[[113, 649, 885, 882]]<|/det|>
+During the loading process, a relaxation time of \(500\mathrm{ps}\) is first performed under the NPT ensemble at a constant temperature of \(300\mathrm{K}\) and pressure of \(0.0\mathrm{MPa}\) . Following the relaxation period, non- equilibrium Molecular Dynamics (NEMD) simulations are conducted to explore the system's response to loading- unloading with the strain rate, \(10^{9}\mathrm{s}^{- 1}\) , at \(300\mathrm{K}\) and a timestep of \(0.2\mathrm{fs}\) under the \(N\sigma_{ij}\epsilon_{ij}T\) ensemble. During the loading and unloading processes, carbon hybridization (sp, \(\mathrm{sp}^{2}\) and \(\mathrm{sp}^{3}\) ) is also computed based on the number of nearest neighbor carbon atoms within a cutoff radius of \(2.0\mathrm{\AA}\) . We define the bond type of a carbon atom based on a coordination criterion
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 884, 178]]<|/det|>
+as follows: a carbon atom is labeled as sp if it is bonded to 2 other carbon atoms, \(\mathrm{sp}^{2}\) if it forms bonds with 3 other carbon atoms, and \(\mathrm{sp}^{3}\) if it is bonded to 4 other carbon atoms, all within a 2.0 Å cutoff radius.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 245, 256, 263]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[115, 287, 884, 341]]<|/det|>
+The data presented in the main text and the Supplementary Information are available from the corresponding authors upon reasonable request.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 407, 207, 424]]<|/det|>
+## Reference:  
+
+<|ref|>text<|/ref|><|det|>[[60, 448, 886, 675]]<|/det|>
+54. Einck, V. J. et al. Scalable nanoimprint lithography process for manufacturing visible metasurfaces composed of high aspect ratio TiO2 meta-atoms. ACS Photonics 8, 2400–2409 (2021).  
+55. Lee, S.-W., Han, S. M. & Nix, W. D. Uniaxial compression of fcc Au nanopillars on an MgO substrate: The effects of prestraining and annealing. Acta Mater 57, 4404–4415 (2009).  
+56. Galli, G., Martin, R. M., Car, R. & Parrinello, M. Structural and electronic properties of amorphous carbon. Phys Rev Lett 62, 555 (1989).  
+57. Rosen, J., Warschkow, O., McKenzie, D. R. & Bilek, M. M. M. Amorphous and crystalline phases in thermal quench simulations of alumina. J Chem Phys 126, (2007).  
+58. Plimpton, S. Fast parallel algorithms for short-range molecular dynamics. J Comput Phys 117, 1–19 (1995).  
+59. Stuart, S. J., Tutein, A. B. & Harrison, J. A. A reactive potential for hydrocarbons with intermolecular interactions. J Chem Phys 112, 6472–6486 (2000).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 91, 273, 110]]<|/det|>
+## Acknowledgement  
+
+<|ref|>text<|/ref|><|det|>[[114, 131, 886, 258]]<|/det|>
+A.B., S.N., and J.J.W. acknowledge support from the Office of Naval Research (N00014- 23- 9- 0008) through the American Lightweight Materials Manufacturing Innovation Institute. Z.L. and S.- W.L. acknowledge support from the UConn/Thermo Fischer Scientific Center for Advanced Microscopy and Materials Analysis (CAMMA) for the FIB milling, TEM, and EELS experiment.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 315, 297, 333]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[113, 355, 886, 620]]<|/det|>
+Z.L., A.B., J.J.W. and S.- W.L conceived and designed the experiments and analysis. Nanomechanical testing, TEM, EELS, and Raman Spectroscopy experiments were carried out and analyzed by Z.L. and S.- W.L. Synthesis of BBCP and PF, self- assembly, and nanoimprinting experiments were carried out and analyzed by A.B. and J.J.W. W.Z. and T.T.T. helped Z.L. to obtain the in- situ deformation video. Molecular dynamic simulations were carried out by and analyzed by J.H. and Y.L. A.M. and S.N. conducted the porosity measurement. Z.L., A.B., J.J.W. and S.- W.L wrote the paper together. J.J.W. and S.- W.L. supervised and provided support through the paper. All authors have commented and edited the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 686, 287, 704]]<|/det|>
+## Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[116, 727, 460, 746]]<|/det|>
+The authors declare no competing interests.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[42, 42, 312, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[42, 92, 768, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[59, 130, 550, 230]]<|/det|>
+SupplementaryMaterials.pdf  SuppleMovie59percentporosityloadingunloading.mp4  SuppleMovieMDLocalStructuralFolding.mp4  SuppleMovieMDLoadingUnloading.mp4
+
+<--- Page Split --->

preprint/preprint__b345a63a0d2e91059489bde27ab056b68d5ce9330496d1ab7976d39d7786a904/preprint__b345a63a0d2e91059489bde27ab056b68d5ce9330496d1ab7976d39d7786a904.mmd

@@ -0,0 +1,496 @@
+
+# Differences in RNA polymerase II complexes and their interactions with surrounding chromatin on human and cytomegalovirus genomes  
+
+Benjamin Spector Carver College of Medicine University of Iowa  
+
+Mrutyunjaya Parida Carver College of Medicine University of Iowa  
+
+Christopher Ball Carver College of Medicine University of Iowa  
+
+Ming Li Carver College of Medicine University of Iowa https://orcid.org/0000- 0003- 0396- 4078  
+
+Jeffrey Meier Carver College of Medicine University of Iowa  
+
+Donal S. Luse Lerner Research Institute  
+
+David Price (David- price@uiowa.edu) Carver College of Medicine University of Iowa  
+
+## Genetics Article  
+
+Keywords: RNA polymerase II, preinitiation complexes, chromatin, human cytomegalovirus, TBP, UL87, DFF ChIP-Seq  
+
+Posted Date: October 8th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 912323/v1  
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Version of Record: A version of this preprint was published at Nature Communications on April 14th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29739-x.
+
+<--- Page Split --->
+
+# Differences in RNA polymerase II complexes and their interactions with surrounding chromatin on human and cytomegalovirus genomes  
+
+Benjamin M. Spector<sup>a</sup>, Mrutyunjaya Parida<sup>a</sup>, Ming Li<sup>a,b,c,d</sup>, Christopher B. Ball<sup>a</sup>, Jeffrey L. Meier<sup>b,c,d</sup>, Donal S. Luse<sup>a</sup>, and David H. Price<sup>a</sup>  
+
+<sup>a</sup>Department of Biochemistry, The University of Iowa, Iowa City, IA 52242, USA  
+
+<sup>b</sup>Department of Internal Medicine, The University of Iowa, Iowa City, IA 52242, USA  
+
+<sup>c</sup>Department of Epidemiology, The University of Iowa, Iowa City, IA 52242, USA  
+
+<sup>d</sup>Veterans Affairs Health Care System, Iowa City, IA 52242, USA  
+
+<sup>e</sup>Department of Cardiovascular and Metabolic Sciences, Lerner Research Institute, Cleveland Clinic, Cleveland, OH, 44195, USA.  
+
+Keywords: RNA polymerase II, preinitiation complexes, chromatin, human cytomegalovirus, TBP, UL87, DFF ChIP- Seq  
+
+## Summary  
+
+Interactions of the RNA polymerase II (Pol II) preinitiation complex (PIC) and paused early elongation complexes with the first downstream \((+1)\) nucleosome are thought to be functionally important. However, current methods are limited for investigating these relationships, both for cellular chromatin and the human cytomegalovirus (HCMV) genome. Digestion with human DNA fragmentation factor (DFF) before immunoprecipitation (DFF- ChIP) precisely revealed both similarities and major differences in PICs driven by TBP on the host genome in comparison with PICs driven by TBP or the viral- specific, late initiation factor UL87 on the viral genome. Host PICs and paused Pol II complexes are frequently found in contact with the \(+1\) nucleosome and paused Pol II can also be found in a complex involved in the initial invasion of the \(+1\) nucleosome. In contrast, viral transcription complexes have very limited nucleosomal interactions, reflecting a relative lack of chromatinization of transcriptionally active regions of HCMV genomes.  
+
+## Highlights  
+
+DFF- ChIP and fragMaps allow visualization of large transcription complexes in cells HCMV promoters are not surrounded by H3K4me3 marked nucleosomes PICs are major features found on host and viral genomes PICs driven by TBP and the HCMV late transcription factor UL87 are highly dissimilar
+
+<--- Page Split --->
+
+## Introduction  
+
+Regulation of human gene expression is accomplished by a highly orchestrated interplay between the transcription machinery and its chromatinized genomic template. The required general Pol II initiation factors are instructed by Mediator and a host of more specific transcription factors to utilize selected promoters as sites of initiation'. However, the default state of the genome is repressive because of global nucleosome deposition, which must be relieved by chromatin remodelers. Assembly of the preinitiation complex (PIC) occurs over sequences surrounding the TSS and an upstream region depleted of nucleosomes (NDR)2. A strongly positioned \(+1\) nucleosome has a boundary around 50 bp downstream from the transcription start site (TSS)3. This nucleosome and several downstream nucleosomes are marked by tri- methylation of histone H3 on lysine 4 (H3K4me3)4,5. Both PIC assembly and pausing by newly- initiated Pol II have been linked to interactions with the \(+1\) nucleosome6- 10, but the nature and functional significance of these interactions remain incompletely understood. Existing global methods to visualize and quantify sites of transcription and the locations of the transcriptional machinery within the local chromatin landscape have had limited success in addressing these questions. Occupancy by the \(+1\) nucleosome is near \(100\%\) while Pol II occupancy is generally much less than \(10\%\) , resulting in a potentially misleading correlation between two disparate signals. Recently there has been a number of highly informative structural studies of PICs11- 14, but the abundance of PICs on cellular chromatin has not been adequately determined and their positioning over promoter elements has been primarily inferred from in vitro studies.  
+
+The interface of transcription and chromatin on the HCMV genome is even more poorly defined. Lyti infection can propagate only in non- dividing cells. The process begins with delivery of a nucleosome- free viral genome to the nucleus, where the standard host Pol II machinery drives transcription from the major immediate early promoter. Expression of viral immediate early proteins then allow expression of early genes15,16. The early genes encode the machinery for viral DNA replication and for transcription of a group of late genes, which requires a special set of viral- specific Pol II initiation factors17- 20 that likely replace some of the host initiation factors. While ChIP- PCR on individual loci and genome- wide studies have revealed changes in chromatinization throughout the viral lifecycle21- 24, it is not clear what regions on each of the hundreds of viral genomes present late in infection are occupied by nucleosomes. Nothing is known about how Pol II transcription complexes interface with any chromatin that may be present. Our recent work demonstrated promiscuous transcription initiation from thousands of promoters across the \(\sim 240,000\) bp dsDNA genome, consistent with the idea that chromatinization of the HCMV genomes during lytic infection is incomplete25. Increasing our knowledge of how HCMV transcription is regulated is important for identification of potential therapeutic targets since the virus infects about \(60\%\) of the population. HCMV is a significant cause of death in immunocompromised individuals26 and a leading viral cause of birth defects27.  
+
+Our group recently described a nuclear run- off method to directly observe engaged Pol II interacting with the \(+1\) nucleosome. We digested nuclei with the double- stranded endonuclease human DNA fragmentation factor (DFF) and then chased nascent transcripts to the resulting run- off sites. We found that many, but not all, paused polymerases were abutted to the \(+1\) nucleosome. Because DFF digestion preserved the viability of transcription complexes and their relation to chromatin, we decided to investigate whether combining it with chromatin immunoprecipitation would allow improved insight into localization and positioning of transcription complexes. We found that we were able to quantitatively visualize PICs as well as interactions between PICs and paused transcription complexes with the downstream chromatin. Most of our experiments were done on primary human foreskin fibroblasts (HFFs) productively infected with HCMV, allowing a direct comparison of the chromatin neighborhoods of human genomic Pol II promoters with their more poorly characterized counterparts on the viral genome.
+
+<--- Page Split --->
+
+## Results  
+
+Our study began with a characterization of our initial DFF- Seq dataset (GSE139237) generated from HeLa cells6. Fragments from DFF- Seq were about 160 bp in length and primarily derived from nucleosomes covering inactive regions of the genome, but a very small percentage of fragments were sub- nucleosomal in size and many of these localized within promoter regions of actively transcribed genes6. To investigate the regions around promoters, active genes in HeLa cells were identified by truQuant28 which finds the most highly utilized TSS (MaxTSS) for each expressed gene from HeLa NasCap25 data. Fragments generated by DFF that were present in a 2000 bp region centered on the MaxTSS of each of the 12,201 genes were collected and the distribution of fragment lengths was quantified and compared to fragment lengths from the total DFF- Seq dataset (Supplementary Fig. 1A). In the total dataset, peaks of fragment sizes corresponding to mono- , di- , and tri- nucleosome were 163, 326, and 512 bp respectively, while those values were 161, 298 and 452 in the truQuant subset. This suggests that many nucleosomes around promoters are essentially close packed. DFF is a homodimer that mostly cuts DNA to form blunt ends29. To determine any sequence requirements for DFF cleavage, sequences surrounding 520 million cut sites were examined. A slight sequence preference was seen (Supplementary Fig. 1B) with \(6.8\%\) of the sites having AAANT(cut) directly on one of the two sides of the cut. There was no preference to having this sequence on both sides.  
+
+## DFF-ChIP reveals differing promoter architecture on human and HCMV genomes  
+
+Because DFF can generate primarily nucleosome- sized fragments without significant internal cutting, we performed two initial experiments (Exp1 and Exp2) to explore its use as a front end for H3K4me3 and Pol II ChIP. Nuclei from non- crosslinked HeLa or MRC5 cells expressing a GFP- tagged Pol II30 were digested with DFF for 1 hour to generate primarily mononucleosomes. The resulting chromatin was immunoprecipitated with antibodies to H3K4me3 modified nucleosomes or to GFP and the associated DNA was prepared for sequencing (Fig. 1A).  
+
+The DFF- ChIP results were compared to NasCap PRO- Seq data6 from HeLa cells. Paused Pol II is evident from the tall peaks in the PRO- Seq data and over a 500,000 bp region of the human genome peaks of Pol II and H3K4me3 occupancy in the DFF- ChIP tracks exhibited strong visual correlation (Fig. 1B). Greater detail can be observed across a 30,000 bp region (Fig. 1C). Each promoter exhibits clear nucleosome phasing in the H3K4me3 dataset surrounding a NDR that supports Pol II initiation in both cell types. Additionally, Pol II DFF- ChIP signal overlaps with the PRO- Seq signal but also extends further downstream. This downstream signal likely results from Pol II that is abutted to the \(+1\) nucleosome such that DFF cannot cleave between polymerase and the nucleosome6. The complex H3K4me3 and Pol II patterning on the majority of promoters is well replicated even across cell types and growth conditions between Exp1 and Exp2 indicating the robustness of the DFF- ChIP method. To further validate this method's reproducibility, reads found in a 10 kb window centered on each of 12,201 truQuant MaxTSSs from the different experiments were directly compared and strongly correlated between the datasets (Fig. 1D).  
+
+Given the success of these initial experiments, DFF- ChIP was then applied to contact inhibited HFFs infected with HCMV (TB40/E) for 48 hours (Exp3). DFF- ChIP results were compared to PRO- Seq data from similarly infected HFFs for broad regions across the host (Fig. 2A) and viral genomes (Fig. 2B). As expected on the host genome, Pol II and H3K4me3 correspond with paused Pol II evident from the PRO- Seq data. The viral genome was pervasively transcribed as previously demonstrated25 and Pol II DFF- ChIP correlated with the PRO- Seq signal when transcription of both strands is taken into account. Unlike the host genome where H3K4me3 is found only around promoter regions, the entirety of the HCMV genome is covered with H3K4me3, at levels ranging from relatively low to more enriched irrespective of Pol II occupancy.
+
+<--- Page Split --->
+
+Additional insight was obtained when the results from individual promoters on the host and viral genomes were examined. The host GAPDH promoter features nucleosome phasing around a NDR that supports Pol II initiation (Fig. 2C and 2D) as it did in Exp1 and Exp2. In contrast, the viral early gene promoter for UL4 and late gene promoter for UL76 show no obvious phasing of nucleosomes (Fig. 2E and 2F). Critically, the two viral promoters are not found in NDRs. Because initiation cannot take place when the promoter is occluded by a nucleosome \(^{31,32}\) , the results suggest that some of the nucleosomes detected are present only on regions of viral genomes that are not transcribed. As in Exp1 and Exp2, the Pol II DFF- ChIP signal on GAPDH overlaps with the PRO- Seq signal but also extends farther upstream and downstream of the MaxTSS (Fig. 2D). Signal from the free paused Pol II predominates for the HCMV early promoter (Fig. 2E), but is not the main signal for the late promoter (Fig. 2F). Neither viral promoter shows evidence of nucleosome- abutted Pol II. The location of upstream protection seen in all three promoters suggests the presence of Pol II in preinitiation complexes and surprisingly, those complexes predominate for the viral late promoter. A fourth experiment was performed on HCMV infected HFFs using a larger set of antibodies for ChIP (Exp4). Comparison of DFF- ChIP for H3K4me3 and Pol II across specific regions on the host and viral genomes for Exp3 and Exp4 clearly demonstrates high reproducibility of the patterns of occupancy (Supplementary Fig. 2A and 2B). In addition, a number of strong correlations between Exp3 and Exp4 were found when signals around host and viral promoters were compared (Supplementary Fig. 2C and 2D).  
+
+## Transcription complexes and their interactions with nearby nucleosomes can be visualized using fragMaps  
+
+Because the patterns of DFF protection around promoters varied, fragment length analysis was explored as a means to identify specific transcription complexes. The same set of MaxTSSs for all active host genes (12,229) were utilized in this analysis. The distribution of DFF- ChIP fragment sizes within a thousand bp of these TSSs from the Pol II and H3K4me3 Exp4 datasets revealed several distinctive groups of fragment sizes (Fig. 3A). A majority of H3K4me3 fragments in this window center on 158 bp in length and a smaller population were about 294 bp in length, corresponding to mono- and di- nucleosomes. In the Pol II dataset, two abundant fragment sizes at approximately 50 bp and 180 bp were the most prevalent, with a smaller population of fragment sizes around 75 bp. Fragment ranges corresponding to each of the three most common Pol II populations in the 2000 bp windows were then chosen and aligned relative to TSSs (Fig. 3B). The total amount of reads in each of these ranges was normalized to emphasize the protected footprints of the less abundant \(\sim 75\) bp fragments. Both the \(\sim 50\) bp and \(\sim 180\) bp fragments align slightly downstream of the TSS in the pause region while the \(\sim 75\) bp fragments span the TSS, as would be expected for PICs. The positioning of the \(\sim 50\) bp and the \(\sim 180\) bp fragments are consistent with free paused Pol II and Pol II abutted to the \(+1\) nucleosome. The same analysis was performed utilizing Exp3 data, which gave similar results (Supplementary Fig. 3A and 3B)  
+
+Since metaplots limit how many fragment lengths can be shown together in a meaningful and accurate way, we created a method that more holistically depicts the distribution of all fragments around TSSs. The output of this visualization method simultaneously captures the size, amount, and position of each fragment across all promoters in an easily viewable fragMap. For each fragment length, the coverage at each position \(+ / - 1000\) bp around all TSSs was calculated and averaged. These averages for each fragment length at each position were then stacked with the shortest fragments on top and longest on the bottom. Fragment lengths included in this view span from 18 to 400 bp. More focused fragMaps were also created \(+ / - 100\) bp around the MaxTSS using 18 to 120 bp fragments. For most fragMaps, black values (overall darkness of the image) are set by the maximum average read value in the window. Because the black value in each fragMap is influenced by the recovery of the IP, absolute amounts of visible
+
+<--- Page Split --->
+
+complexes should only be compared within the same fragMap. Host HFF fragMaps were generated utilizing the 12,229 TSSs found with truQuant and TSSs on the HCMV genome were found utilizing transcription start regions (TSRs) identified using TSR- finder on a PRO- Cap dataset generated from HCMV infected HFFs that resulted in 1,461 non- overlapping 200 bp regions centered on a MaxTSS \(^{25}\) .  
+
+FragMaps for the Pol II and H3K4me3 Exp4 datasets generated a snapshot of transcription complexes and chromatin features on the host and HCMV genomes. Host H3K4me3 fragMaps show very well positioned nucleosomes relative to the TSS and a clear NDR (Fig. 3C). This patterning of nucleosomes closely resembles fragMaps generated using the DFF- Seq HeLa truQuant dataset, except that the signal becomes fainter as the distance from the TSS increases (Supplementary Fig. 3C). Pol II fragMaps of the host genome show free paused Pol II (\~50 bp fragments downstream from the TSS), Pol II positioned over the TSS (\~75 bp fragments), Pol II abutted to the +1 nucleosome (\~180 bp fragments downstream of the TSS), and Pol II associated with the first two nucleosomes (\~320 bp fragments downstream of the TSS) (Fig. 3C). The size of the fragments bearing Pol II associated with the first two nucleosomes supports the finding that nucleosomes around promoters are more closely spaced than in bulk chromatin (Supplementary Fig. 1A). A complex of \~100 bp, located downstream of the TSS, is also visible in the Pol II fragMaps. Possible origins for this unanticipated complex will be discussed later. Divergent transcription occurs at variable distances upstream of the sense TSS on the host genome, so the upstream region displays a similar, but less well- defined pattern compared to transcription in the sense direction. Pol II and H3K4me3 fragMaps from Exp 3 datasets displayed all the same features and were virtually indistinguishable from those using Exp4 datasets (Supplementary Fig. 3D) demonstrating that the method is robust and reproducible.  
+
+Major differences from the host patterns were found when analyzing the HCMV genome. The H3K4me3 signal was not positioned around the majority of promoters (Fig. 3D). HCMV Pol II fragMaps confirm the pervasiveness of Pol II transcription with Pol II visible across the fragMap (Fig. 3D). The TSRs utilized to generate HCMV fragMaps are 200 bp so individual TSRs may be represented multiple times in the 2000 bp window. This causes a light background of particularly sized fragments across the entire visualized region. The fragments near the TSS show free paused Pol II, Pol II positioned over the TSS in \~75 bp fragments, and an additional complex of \~50 bp positioned just upstream of the TSS that is not present on host fragMaps. This \~50 bp protection over the TSS is especially evident in Exp3 viral fragMaps (Supplementary Fig. 3E). Nucleosome- abutted Pol II, which is the most abundant feature seen on the host genome, is present but at a vastly lower level than free Pol II on the HCMV genome.  
+
+The apparent differences in Pol II association with the +1 nucleosome on the host and HCMV genomes prompted us to quantify how frequently Pol II encounters an immediately downstream nucleosome in these two cases. The number of reads in features corresponding to the free Pol II, abutted Pol II, and +1 nucleosome based on genomic position of fragment centers and fragment size were quantified from Pol II and H3K4me3 datasets (Supplemental Excel file). Analysis of free or abutted Pol II demonstrated that 36% of the total engaged Pol II signal arises from free Pol II on the host genome. In contrast to the host, 74% of the engaged Pol II was free on the HCMV genome. On a promoter by promoter basis, 85% have more free than abutted Pol II signal on the HCMV genome, whereas only 18% do on host (Fig. 3E). Examination of the fragment count of +1 nucleosome sized fragments from the H3K4me3 dataset shows that the host promoters with the absolute highest percentage of free Pol II have little +1 nucleosome, but most promoters have a similar amount of +1 nucleosome (Fig. 3E). Although the H3K4me3 signal over the viral genome has a similar value in terms of reads, there are about a hundred times more viral genomes and thus the H3K4me3 modified nucleosome occupancy over the viral genomes is about 1% of that in the host.
+
+<--- Page Split --->
+
+## Detection and characterization of TBP-driven PICs and UL87-driven viral PICs  
+
+Detection and characterization of TBP- driven PICs and UL87- driven viral PICsThe existence of Pol II- containing complexes that extended upstream of the MaxTSS strongly suggested that these features correspond to PICs, so Pol II DFF- ChIP was repeated under conditions that would alter PIC prevalence. Treatment of cells for an hour with 1 \(\mu \mathrm{M}\) triptolide, an inhibitor of transcription initiation, should increase the PIC relative to paused Pol II and that is exactly what was found for the \(\sim 75\) bp feature on the host and viral genomes (Fig. 4A). Triptolide treatment also more clearly reveals the \(\sim 50\) bp feature directly upstream of the TSS, which is unique to the viral genome (Fig. 4A). As an additional verification that \(\sim 75\) bp protections arose from uninitiated Pol II, Exp1 and Exp2 DFF- ChIP were reanalyzed since they utilized different wash conditions in a GFP- Pol II tagged MRC5 cell line<sup>30</sup> and GFP nanobody beads. The immunoprecipitations in these experiments were carried out with 150 mM or 1 M salt wash conditions. The \(\sim 75\) bp feature was preferentially lost during the high salt conditions as expected<sup>33</sup> for uninitiated Pol II (Supplementary Fig. 4). DFF- ChIP was then performed targeting the TATA- binding protein (TBP), a critical component of the host PIC. FragMaps generated from the TBP dataset show primarily the \(\sim 75\) bp feature (Fig. 3A), indicating that these complexes are indeed TBP- containing PICs. Additionally, TBP fragMaps reveal protections that share the relatively sharp upstream edge with the full TBP PIC but are much smaller in size, about 40 bp. These likely result from TBP- containing complexes prior to incorporation of Pol II and assembly of the complete PIC. DFF- ChIP was also performed utilizing a Pol II antibody that targets the serine 5 phosphorylation on the CTD of Pol II (Ser5P). Ser5P modification is carried out by CDK7, a component of the initiation factor TFIIH<sup>34</sup>. The resulting fragMaps demonstrate that Ser5P antibodies recognize modified Pol II in the TBP PIC on both the host and viral genome. However, the \(\sim 50\) bp PICs on the viral genome were not detected (Fig. 4B) suggesting that the Pol II in those complexes is not phosphorylated. There is a difference in the relative amounts of PIC and free paused Pol II detected with the F12 antibody and the Ser5P antibody. The epitope on the Pol II large subunit recognized by F12 is located deep within the PIC<sup>13</sup> and potentially masked by TFIIE and TFIIH, while the Ser5P epitope would be difficult to mask because of its repetitive occurrence and disordered structure. Therefore, we favor the idea that the Ser5P results are more representative of Pol II levels in all TBP- driven promoter- proximal complexes. Supporting this idea, when MRC5 cells containing a GFP- tagged Pol II were immunoprecipitated with GFP nanobodies, the PIC was a major species (Supplementary Fig. 4B).  
+
+Large TBP fragMaps provided further insight into the chromatin differences between the host and HCMV genomes (Fig. 4C). Some TBP PICs on the host genome are associated with the \(+1\) nucleosome, resulting in fragments of \(\sim 250\) bp. These interactions are mostly absent on the HCMV genome presumably because the viral genome is less chromatized. Significant amounts of the host PICs driving divergent transcription also connect with the adjacent nucleosome (the - 1 nucleosome), analogous to the PIC- nucleosome complexes in the sense direction. As with the free- standing PIC, the PIC/ \(+1\) nucleosome feature on the host genome was salt sensitive (Supplementary Fig. 4C). Previous analysis of downstream sequences relative to TSSs revealed periodic elements that likely serve to position the \(+1\) nucleosome<sup>6</sup>, suggesting a connection between the TSS and nucleosome positioning. Considering that both genic- oriented and divergent PICs are normally associated with well- positioned immediately- adjacent nucleosomes, it is further likely that the PIC, TSS, and \(+1\) nucleosome connection is important for specifying and/or facilitating transcription initiation.  
+
+Unlike the early HCMV transcriptional program that relies entirely on the host general Pol II transcription machinery, beta- and gamma herpesviruses have unique late promoters containing a TATT upstream element that recruits virally- encoded late transcription factors<sup>35,36</sup>. Since only one of the two distinct Pol II- containing PICs on the viral genome corresponds to the host complex driven by TBP, we posited that the \(\sim 50\) bp virus- specific PIC is based on UL87, one of the viral late transcription factors which associates with the TATT element<sup>17</sup>. DFF- ChIP was
+
+<--- Page Split --->
+
+performed with a Towne strain of HCMV expressing a HA- tagged UL87 (Exp4). FragMaps demonstrated that the \(\sim 50\) bp viral PICs were almost exclusively recovered with the HA antibody (Fig. 4D). Interestingly, unlike the 75 bp TBP- PICs the 50 bp UL87- PICs did not cover the TSS.  
+
+In an attempt to correlate upstream sequence elements with the extent of PIC assembly, UL87 PIC and TBP PIC features at each gene on the host or on each TSR of the HCMV genome were quantified (Supplementary Excel file). The size and position of these features were selected such that no overlap was allowed between them. Each region was then rank ordered by the amount of the UL87 PIC feature or the TBP PIC feature. Logos were generated by MEME analysis<sup>37</sup> of the - 38 to - 19 region for the PICs in the top decile of occupancy on the host and viral genomes (Fig. 4E). Such analysis recapitulated the expected TA- rich binding motifs with TBP preferring sequences containing TATA and UL87 preferring sequences containing TATT. UL87 evidently has a stricter requirement for TATT containing sequences with 128/146 matches to its HCMV Logo, while TBP had a lower percentage of matches to its host (623/1230) and viral (88/146) Logos. It is important to note that not all host promoters with high levels of PIC occupancy are AT- rich in the - 38 to - 19 region<sup>28</sup>.  
+
+## TBP and UL87 are functionally distinct but not mutually exclusive on HCMV promoters  
+
+To characterize the transcription of the HCMV genome throughout lytic infection and its relation to UL87 and TBP usage, PRO- Seq was carried out at multiple different time points of infection and compared to 48 hpi DFF- ChIP data. Time points include two early times (4 and 12 hpi), 24 hpi which is the beginning of replication, and two late times (48 and 72 hpi) in which high levels of viral replication have occurred and during which UL87 function is critical. Representative regions of the HCMV genome containing early and late genes are depicted in a 1,400 bp region and an 800 bp region with corresponding HCMV genomic fragMaps below showing fragments from the UL87, TBP, Pol II, and Ser5P datasets (Fig. 5A). The early promoter UL29 has only TBP, Pol II, and Ser5P fragments associated with it (Fig. 5A, blue TBP arrow), while the late intragenic promoter in UL49 only has UL87 and Pol II fragments (Fig. 5A, red UL87 arrow). TBP and UL87 driven promoters can occur very close to each other (Fig. 5A, red and blue Both arrows) or can even drive transcription from the exact same TSS (Fig. 5A, purple Both arrow). Overlapping PICs such as these would be expected to compete for occupancy on the HCMV genome. Quantification of 5' end reads found in an 11 bp window around the MaxTSS of these five promoters shows that TBP- driven TSSs are more active early in comparison to UL87 PICs that are active late (Supplementary Fig. 5A). Because some promoters are driven by both TBP and UL87 the distinction between early versus late gene transcription is more complex than a simple separation of TBP and UL87 driven promoters.  
+
+To quantify how TBP and UL87 usage is related to early or late expression, transcription from each promoter was quantified by counting 5' ends in an 11 bp window centered on the MaxTSS of each TSR at each time point. The fractional usage for each promoter at each time point was displayed in a heatmap after normalizing to the total number HCMV reads in each time point. 795 of the 1,461 promoters that each had a total of 100 reads across the time course were then sorted based on PFA sensitivity<sup>28</sup>, on a slope calculated from the 5 time points for each promoter and by UL87 dependency<sup>36</sup> (Fig. 5B). Each promoter was colorized individually based on the relative usage across the time course. The three sorts gave similar patterns of genes with early (red to green) and late (green to red) transcription kinetics. Each TSR's reliance on DNA replication (PFA sensitivity) and UL87 dependency was then plotted against slope to classify how TSRs with primarily TBP (blue) and UL87 (red) PICs behaved in relation to time of expression (Supplementary Fig. 5B). The relative preference for TBP or UL87 PICs was calculated as the ratio of TBP/UL87 PICs after normalization of total counts in each feature. These plots show that UL87 primarily functions on genes with late transcription kinetics and that many of these promoters are also stimulated by DNA replication. However, and significantly,
+
+<--- Page Split --->
+
+many promoters with late kinetics are primarily TBP driven showing that TBP is also essential in transcription of some late genes. The UL22A promoter has a TBP/UL87 ratio of 0.93 and displays late kinetics. There are two main TSSs at all time points that are separated by 3 bp with a shift in the relative usage at early and late time points that is reverted when viral replication is blocked by PFA treatment (Supplementary Fig. 5C). This promoter exemplifies a clear competition for formation of TBP and UL87 PICs.  
+
+Nearly all promoters show some level of UL87 and TBP usage and to determine to what extent promoters are shared, the TBP/UL87 PIC ratio was calculated for all 1,461 TSRs and plotted (Fig. 5C). The results indicate that the relative levels of UL87 and TBP PICs vary by many orders of magnitude across the 1461 TSRs. Therefore, in order to make comparisons between TBP and UL87 PICs only the top and bottom \(5\%\) of TSRs sorted by PIC ratio were utilized to prevent contaminating signal from the other class of PIC. To compare initiation efficiency, fragMaps were generated utilizing the TBP and Ser5P datasets for the top TBP TSRs and UL87 and Pol II datasets for the top UL87 TSRs (Fig. 5D). As noted above, the F12 antibody epitope in the large subunit of Pol II is significantly masked in TBP PICs \(^{13}\) , so the Ser5P Pol II signal was used to visualize Pol II in TBP driven TSRs. The strong correlation of feature counts from the Ser5P dataset in comparison to TBP and Pol II datasets also advocates for this approach (Supplementary Fig. 5D). These fragMaps reveal that UL87 and TBP driven TSRs yield similar amounts of engaged Pol II in relation to PIC amounts. This result was initially surprising given the prominence of UL87 PIC peaks in comparison to detected paused Pol II downstream of those PICs, which seemed to indicate poor UL87 initiation. However, it is possible that a significant fraction of \(\sim 50\) bp UL87 features may not have Pol II associated with them and this is supported by the relative lack of protection downstream of the TSS. Critically, on both promoter classes the PIC is as prominent on chromatin as paused Pol II indicating that the PIC is far more prevalent on the genome than expected.  
+
+## Nucleosomes on the HCMV genome are irregularly spaced  
+
+To directly assess if sites of initiation correlated with H3K4me3 modified nucleosomes, genome browser tracks and genomic fragMaps for TBP, UL87 and H3K4me3 were compared. There is little evidence for a correlation of sites of transcription initiation and H3K4me3 modification regardless of PIC type (Fig. 6A). This comparison does allow for a rough estimation of the number of nucleosomes that span the 21,000 bp region shown, since approximately 70- 80 nucleosomes are distinguishable. This indicates that on average a nucleosome is positioned every 250- 300 bp on the HCMV genome. This average spacing is consistent across the entire HCMV genome (Supplementary Fig. 6). It is unclear what drives nucleosome positioning on HCMV DNA but given the scarcity of transcription- coupled nucleosomes it is unlikely to be related to transcription. Metaplot analysis of all 1,461 viral TSRs shows that nucleosomes around promoters are spaced approximately 250- 300 bp in contrast a much more compact spacing around promoters on the host genome of about 150 bp (Fig. 6B). A PIC containing TBP is capable of associating with the +1 nucleosome on the host (Fig. 4C) and this prompted us to investigate if viral TSRs with strong TBP signal have a better positioned downstream nucleosome. Selecting the top \(10\%\) of TSRs with the highest level of TBP PICs showed that indeed these TSRs have a slightly stronger +1 nucleosome signal (Fig. 6C). The same analysis performed with the top UL87 TSRs failed to reveal any clear nucleosome patterning (Fig. 6C). These findings suggest that while TBP PICs may aid in the positioning of a +1 nucleosome, this is not a driving force for nucleosome positioning on the HCMV genome.  
+
+## Increasing the extent of DFF digestion reveals a general robustness of features and captures Pol II association with sub-nucleosomal fragments  
+
+To confirm reproducibility and determine the effects of more extensive DFF digestion, five datasets were generated in duplicate using HFFs infected with the Towne strain of HCMV
+
+<--- Page Split --->
+
+(Exp5). The datasets correlated highly within Exp5 and between Exp5 and Exp4 (Supplementary Fig. 7A). Total library fragment length distributions for Pol II, TBP, and H3K4me3 were compared between the initial and more extensively digested experiments (Fig. 7A and 7B). It was apparent that higher levels of digestion resulted in a division of some complexes into subgroups typically separated by a 10 bp periodicity. Over- digestion did not change the ratio of the free to abutted Pol II (Fig. 7C and 7D), but the group of Pol II- containing fragments of \(\sim 90\) to 110 bp became more apparent after increased digestion. Connections of the TBP PIC to downstream nucleosomes were mostly lost with over- digestion.  
+
+The H3K4me3 fragment distributions in Fig 7 clearly show that DFF can cleave within nucleosomes at higher digestion levels. Two classes of products shorter than the expected protection from the full nucleosome were observed: fragments which apparently resulted from removal of 10 or 20 bp from the nucleosome ends and another population of centered fragments less than 80 bp consistent with protection by the H3/H4 tetramer. The lack of fragments between the two populations suggests that complete loss of the flanking H2A/H2B dimers on the edges of the nucleosome occurs before the central H3/H4 is invaded by DFF. All sub- nucleosomal fragments were centered including those less than 80 bp, further demonstrating that they arose from the H3/H4 tetramer (Supplementary Fig. 7C).  
+
+To better understand the interaction of Pol II with the \(+1\) nucleosome we specifically considered well positioned nucleosomes downstream of the most focused promoters<sup>6</sup> (Supplementary Fig. 7B). At such promoters, the upstream edge of the \(+1\) nucleosome is positioned on average at \(+47\) , just downstream of the average paused Pol II position at \(+41\) . Pol II protects approximately 20 bp of DNA downstream of the active site<sup>38</sup>, suggesting that the leading edge of the abutted Pol II intrudes \(\sim 1.5\) DNA turns into the nucleosome. Pol II at this location would disrupt the H3 contact at nucleosome entry but could leave the H2A/H2B contacts with DNA at least partially intact. If invasion by Pol II distorts the nucleosome, this could in turn reveal a cutting site for DFF downstream of the H2A/H2B dimer. DFF cleavage in such a complex would give rise to the 87- 107 bp fragment set that we observe in Pol II IPs, more prominently upon over- digestion (Fig. 7, and Supplementary Fig. 7). Other possible origins for this \(\sim 100\) bp Pol II fragment set can be envisioned, which will be addressed in the Discussion.  
+
+## Discussion  
+
+Our results provide a deeper understanding of transcription complexes and their interactions with chromatin on both the host and HCMV genomes. Critically, analysis of the length and position of fragments recovered by DFF- ChIP using fragMaps not only provided detailed footprints of PICs and paused Pol II but also directly revealed the interactions of those complexes with the \(+1\) nucleosome. Using HCMV infected HFFs, we demonstrated that Pol II encounters a vastly different chromatin environment on the viral genome than it does on the host genome. Furthermore, our direct visualization of both TBP and UL87- driven PICs sheds light on sequence preferences, dimensions, and shared usage at the two promoter classes.  
+
+Our results provide a new perspective into the transcriptional processes during lytic infection in relation to the chromatin structure of the HCMV genome. Earlier work suggested that during productive infection, viral genomes form irregular nucleosome arrays<sup>21,39,40</sup>. However, a more recent genome wide analysis utilizing MNase suggested that the HCMV genome is largely packaged into nucleosomes throughout the viral cycle<sup>24</sup>. Our data confirm that nucleosomes are deposited across the viral genome at low level of occupancy, but we also show in multiple ways that paused Pol II rarely encounters a nucleosome on HCMV DNA. The TBP, Pol II, Ser5P, and UL87 DFF- ChIP signals corresponding to initiating Pol II or PICs often reside in the middle of apparently nucleosome rich regions on HCMV DNA despite the known ability of a nucleosome to block initiation<sup>31,32</sup>. Therefore, those regions of HCMV genomes with nucleosomes over TSSs cannot be transcriptionally active. Transcription complexes themselves report on local
+
+<--- Page Split --->
+
+nucleosomes by conferring a larger protection footprint when they are in close proximity to a nucleosome. HCMV chromatin very rarely confers these protection patterns. Although it is likely that the large majority of HCMV promoters do not feature modified nucleosomes immediately downstream, it is not possible to prove that this is the case at every individual viral promoter. Post- translationally modified nucleosomes may be positioned on individual loci at some stages of the viral life cycle<sup>22,23</sup>. Overall, we conclude that during lytic infection the HCMV genome is transcribed in a predominantly nonchromatinized state.  
+
+The unique ability to recover PICs with DFF- ChIP without prior crosslinking was likely allowed by the use of EDTA during very rapid nuclei isolation<sup>41</sup> prior to digestion and immunoprecipitation steps. EDTA was included to halt transcription but it may also increase PIC retention by eliminating the destabilization caused by ATP<sup>33</sup> in abortive initiation and XPB function. Substantial recovery of PICs in DFF- ChIP has allowed a better understanding of their properties in the nucleus including documenting that TBP- and UL87- driven PICs are major features on the host (TBP) and viral genomes (TBP and UL87), equaling or surpassing paused Pol II in amount. The formation of the TBP- containing PIC requires Pol II in a hypophosphorylated state, but after PIC assembly in vitro, phosphorylation of the CTD may occur even prior to formation of the first phosphodiester bond<sup>42,43</sup>. Our data demonstrate that this actually occurs in cells, showing that Ser5P is present on TBP PICs (Fig. 3B). Additionally, we show that a substantial fraction of PICs directly interacts with the +1 nucleosome.  
+
+DFF- ChIP also uncovered differences between TBP and UL87 PICs and their interplay on the HCMV genome. TBP PIC footprints correspond to the known in vitro footprint of TFIID, from roughly 40 bp upstream of the TSS to 35 bp downstream<sup>6</sup>. Published human TFIID and PIC structures indicate that TFIID and XPB both contact DNA well downstream of the TSS<sup>2,44</sup>. In contrast, UL87 PIC footprints are only located upstream of the TSS, suggesting that UL87 PICs lack the subunits of TFIID that contact downstream DNA and crucially, at least the XPB subunit of TFIID. Since UL87 PICs also lack Ser5P modification, it seems likely that the UL87 PICs we detect lack TFIID. However, initiation at UL87 PICs is sensitive to inhibition of XPB by tripotide. We therefore propose that while TFIID is required for UL87 initiation, it is not a stable component of the UL87 PIC that we detect. A recent study regarding ORF24, a UL87 homolog in Kaposi's sarcoma associated virus<sup>20</sup>, showed association with the Pol II CTD only in the hypophosphorylated state, which is in agreement with our data demonstrating that the Pol II in UL87 PIC is predominantly unphosphorylated<sup>45</sup>. Regardless of these functional differences, both PICs function on a large shared subset of HCMV promoters.  
+
+We expected from our earlier study that paused Pol II complexes would be located upstream of the +1 nucleosome or abutted to that nucleosome<sup>6</sup>. Those complexes are evident from DFF- ChIP (Fig. 2C) but high levels of DFF digestion in particular revealed unanticipated Pol II- containing fragments of 87- 107 bp that extend downstream into the region that is expected to be protected by the proximal H2A/H2B dimer of the +1 nucleosome (Fig. S4D). We suggested above that polymerase invasion of the nucleosome could reveal a site for DFF cleavage downstream of the dimer. However, other recent work suggests a plausible alternative explanation. It was reported that the Chd1 chromatin remodeller associates with +1 nucleosomes, specifically on the promoter- proximal face<sup>46</sup>. Subsequent structural studies showed that in a Chd1- nucleosome complex, a Chd1 domain displaces DNA from the nucleosome surface normally occupied by H2A/H2B<sup>47</sup>. Thus, the 87- 107 bp Pol II- containing fragments we detected could have arisen from Pol II paused at the entry of a +1 nucleosome already occupied by Chd1. Presumably the interface of Chd1 and the nucleosome is more easily accessible at high levels of DFF digestion. In the earlier studies it was speculated that displacement of the Chd1 domain at nucleosome entry by the advancing polymerase would activate the remainder of Chd1 to drive displacement of the nucleosome and thus facilitate traversal by Pol II. Thus, this model predicts that once Pol II has displaced the proximal Chd1 domain, full traversal of the +1 nucleosome should be efficient<sup>47</sup>. This is consistent with the fact
+
+<--- Page Split --->
+
+that we did not detect Pol II- containing complexes by DFF- ChIP corresponding to pausing just upstream of H3/H4 tetramer of the \(+1\) nucleosome, as might have been predicted from earlier in vitro studies \(^{48}\) . Other results based on micrococcal nuclease digestion patterns indicated that in Drosophila \(+1\) nucleosomes frequently lack the proximal H2A/H2B dimer \(^{49}\) . However, as just noted, we do not have evidence from our experiments for a Pol II barrier at the H3/H4 tetramer of the \(+1\) nucleosome.  
+
+Future usage of DFF including DFF- ChIP holds great potential in uncovering intricacies of chromatin architecture. Targeted immunoprecipitation of various modified histones or histone replacements are but a few of the possibilities for these investigations. For direct inquiries into transcription, further DFF- ChIP experiments that target the general transcription factors involved in initiation, pausing, and productive elongation are of great interest and will aid in uncovering the ways that transcription complexes interact with nucleosomes. In addition, we expect DFF- ChIP to be applicable to the targeted investigation of many more specific chromatin associated factors.  
+
+## Methods  
+
+## Viruses  
+
+HCMV TB40/E BAC4 and Towne UL87HA were used in this study. The construction and use of the Towne UL87- HA recombinant virus was described previously \(^{28}\) .  
+
+## Infections and treatments  
+
+Primary human foreskin fibroblasts were maintained in Minimum Essential Medium (Gibco, 11095080) supplemented with \(5\%\) fetal bovine serum (Gibco, 26140079) and \(1\%\) penicillin- streptomycin (Gibco, 15140122). Confluent (contact inhibited) HFF monolayers in T- 150 cm \(^2\) flasks were used for these studies. The culture medium was refreshed \(24 \text{h}\) prior to infection. On the day of infection, all but \(12 \text{mL}\) of the conditioned medium was removed and set aside. The remaining \(12 \text{mL}\) of medium was inoculated with HCMV at a multiplicity of infection of 3 infectious units per cell (MOI of 3). Viral adsorption was carried out for \(90 \text{min}\) . The medium containing viral inoculum was then replaced with \(12 \text{mL}\) of the conditioned medium. For experiments involving treatment with flavopiridol (Flavo; final concentration, \(1 \mu \text{M}\) ), or triptolide (final concentration, \(1 \mu \text{M}\) ), \(6 \text{mL}\) of conditioned medium was temporarily removed \(1 \text{h}\) before cells were harvested. This medium was treated with \(6 \mu \text{L}\) of \(2 \text{mM}\) Flavo (NIH AIDS Reagent Program 9925z) in DMSO, \(6 \mu \text{L}\) of \(2 \text{mM}\) triptolide (Sigma, T3652) in DMSO or \(6 \mu \text{L}\) of DMSO alone. Once inoculated with drug, the \(6 \text{mL}\) of medium was immediately returned to the flask for a final \(12 \text{mL}\) of culture medium. At \(48 \text{h}\) post- infection, cells were lysed and cell nuclei were isolated and held in frozen storage until use, as described previously \(^{28}\) .  
+
+## DFF-ChIP Seq  
+
+DFF was purified as previously described \(^{6}\) . For Exp4 approximately 6 million nuclei from human foreskin fibroblasts infected for 48 hours with human cytomegalovirus strains TB40/E or a mutant Towne virus expressing UL87- HA were digested with approximately \(15 \mu \text{g}\) of DFF in \(20 \text{mM}\) HEPES (pH 7.6), \(5 \text{mM}\) magnesium acetate, \(100 \text{mM}\) K(Ac), \(5 \text{mM}\) DTT, for 1 hour at \(37^{\circ}\text{C}\) . Digestions were carried out in batch where possible for all nuclei of the same treatments conserving the ratio of nuclei to DFF. Digestion was halted with the addition EDTA to a concentration four times that of magnesium and nuclei were subsequently split for individual IPs. Nuclei were lightly sonicated for \(20 \text{s}\) at \(40\%\) amplitude using Qsonica Q800R3 Sonicator and the supernatant was collected and brought up to \(1 \text{mL}\) with solution containing \(10 \text{mM}\) Tris (pH 7.5), \(100 \text{mM}\) NaCl, \(1 \text{mM}\) EDTA, and TritonX- 100 such that the final concentration was \(0.1\%\) . The supernatants were precleared for 20 minutes over Protein A (Sigma P9424) or G Sepharose (Sigma P3296) beads. Afterwards, the supernatants were removed from beads and immunoprecipitated with approximately \(2.5 \mu \text{g}\) of antibodies for Pol II (Santa Cruz, sc- 55492),
+
+<--- Page Split --->
+
+TBP (Abcam, ab51841), Ser5P (Millipore, 3E8), HA- tag (Cell Signaling Technology, C29F4), and H3K4me3 (Abcam, ab8580) overnight at \(4^{\circ}\mathrm{C}\) with rotation. Next, samples were incubated with Protein A beads (Protein G beads for Ser5P IPs) for 2 hours at \(4^{\circ}\mathrm{C}\) with rotation. The beads were than washed five times with \(10\mathrm{mM}\) Tris (pH 7.5), \(150\mathrm{mM}\) NaCl, \(1\mathrm{mM}\) EDTA, and \(0.1\%\) TritonX for five minutes per wash. Bound material was than eluted twice with \(50\mu \mathrm{L}\) of \(10\mathrm{mM}\) Tris (pH 7.5), \(1\%\) SDS, and \(1\mathrm{mM}\) EDTA incubated at \(65^{\circ}\mathrm{C}\) for 5 minutes. Eluted material was subsequently treated with \(20\mu \mathrm{g}\) RNase A for 30 minutes at \(37^{\circ}\mathrm{C}\) and then \(40\mu \mathrm{g}\) of Proteinase K for 2 hours at \(65^{\circ}\mathrm{C}\) . The same protocol was used for the all DFF- ChIP experiments with slight modifications. For MRC- 5 expressing GFP- Pol II and HeLa cell lines in Exp1 and Exp2, 12 million nuclei were digested with \(30\mu \mathrm{g}\) . After splitting and preclearing on Protein A sepharose beads, GFP- Pol II samples were incubated with Chromotek GFP- Trap beads for 4 hours, washed five times with either the same buffer as above or one containing \(1\mathrm{M}\) NaCl for \(\sim 1\) min each, and eluted as described above. Samples from HeLa cells were immunoprecipitated, eluted, and treated the same as Exp4 except with \(\sim 1\) min washes. Exp3 was performed exactly as Exp4 was done, but again with \(\sim 1\) min washes. For Exp5 utilizing HFFs infected with Towne, 3 million HFF cells infected with Towne UL87- HA were incubated with approximately \(25\mu \mathrm{g}\) DFF. All libraries were cleaned up utilizing Invitrogen PureLink PCR purification kit. All libraries were prepared using the KAPA Hyper Prep Kit according to their protocol (Roche 7962312001). For libraries in experiments 1 and 2 we utilized Illumina TruSeq adapters. For all subsequent libraries, we created custom adapters utilizing the TruSeq sequences that contained an 8 bp UMI immediately downstream of the index. Test amplifications were performed on each library to determine the number of cycles necessary to obtain enough library material for sequencing. Full- scale amplification was then performed, fragments were quantified using an Agilent Bioanalyzer 2100, and then libraries were pooled and size selected from 135- 1000 bp using a BluePippin. Prior to submission, proper size selection was confirmed with reanalysis using the Agilent Bioanalyzer. Libraries were sequenced on either an Illumina HiSeq 4000 or a NovaSeq 6000 by the Iowa Institute of Human Genetics.  
+
+## PRO-Seq and PRO-Cap  
+
+PRO- Seq and PRO- Cap libraries were prepared as previously described \(^{28}\) , with a few modifications. Frozen nuclei isolated from HFF infected with HCMV TB40/E (MOI 3) for 48 or 72 h were thawed on ice, gently pelleted, and resuspended in \(40\mu \mathrm{L}\) of a nuclear run- on buffer (20 mM HEPES (pH 7.6), \(5\mathrm{mM}\) magnesium chloride, \(100\mathrm{mM}\) potassium chloride, \(5\mathrm{mM}\) dithiothreitol (DTT), and \(0.6\mathrm{U} / \mu \mathrm{L}\) SUPERase- In (Invitrogen AM2696)). Approximately 100,000 moth SF21 nuclei were spiked into HFF nuclei prior to pelleting. Nuclei were warmed to \(37^{\circ}\mathrm{C}\) and then combined with \(20\mu \mathrm{L}\) of a 3X nuclear run- on mix (20 mM HEPES (pH 7.6), \(5\mathrm{mM}\) magnesium chloride, \(100\mathrm{mM}\) potassium chloride, \(5\mathrm{mM}\) DTT, \(1.5\%\) Sarkosyl, and \(60\mathrm{uM}\) biotinylated ATP, UTP, GTP, and CTP (Perkin Elmer NEL544, NEL543, NEL545, and NEL542, respectively)). Samples were pulse- vortexed after the addition of nuclear run- on mix and incubated at \(37^{\circ}\mathrm{C}\) for 10 minutes. Reactions were quenched with \(40\mu \mathrm{L}\) of \(50\mathrm{mM}\) ethylenediaminetetraacetic acid (EDTA) and \(300\mu \mathrm{L}\) of Trizol LS (Ambion 10296028), and total RNA was extracted according to manufacturer protocol. All subsequent steps in PRO- Seq library preparation were carried out as previously described \(^{28}\) . PRO- Cap library preparation followed a similar procedure, except excluded the RNA hydrolysis step and included RNA polyphosphatase and terminator exonuclease treatments to ensure the integrity of nascent RNA 5' end capture. A detailed description of the PRO- Cap protocol has been published \(^{25}\) . Reverse- transcribed libraries were PCR amplified for 11 (PRO- Seq) or 17 (PRO- Cap) cycles, subjected to analysis on an Agilent Bioanalyzer 2100, pooled in equimolar ratios, size- selected on a Sage Science Blue Pippen (BDF2010 cassette, 135- 600 bp fragments selected), and sequenced on an Illumina HiSeq 4000 with 150 bp paired- end reads at the University of Iowa Genomics
+
+<--- Page Split --->
+
+Division. Raw data were trimmed, mapped, deduplicated, and processed into tracks as previously described<sup>28</sup>.  
+
+## DFF-ChIP analysis  
+
+Initial workup of data was performed using DNAfastqtoBigWig (https://github.com/P- TEFb/DNAfastqtoBigWig). This is a linux based, multi- thread capable, Next Generation Sequencing (NGS) data analysis program with a command line interface. It performs the standard NGS data processing steps including, downloading sequencing data from a given web server, trimming adapter sequences from the sequencing data, aligning the trimmed data to a given list of genomes, generating mapping statistics for the aligned data, deduplication of aligned fragments using their Unique Molecular Identifiers (UMI), and finally, generating tracks for each sample in bigwig format. The program automatically accomplishes the following steps. Raw sequences in Fastq format were downloaded from the Iowa Institute of Human Genetics (IIHG) Genome Sequencing web server using wget command. Next, adapter sequences were trimmed from these sequences using trim_galore 0.6 (https://github.com/FelixKrueger/TrimGalore/releases/tag/0.6.6) while retaining only paired end trimmed sequences of at least 18 bp in size. These sequences were aligned with UCSC hg38, and Genbank TB40/E and Towne HCMV assemblies using bowtie v1.2.2 to generate alignments in sam format. UMLs reads were used to deduplicate the aligned reads which were then converted into bed files. Unstranded tracks were generated for each sample by first converting bed into bedGraph format using bedtools v2.26, and subsequently into bigwig format using the Kent UCSC utility program called bedGraphToBigWig. All datasets are described in the supplementary Excel file. Raw and processed sequencing data can be obtained from GEO (GSEXXXX). All tracks can be viewed using the following links: HOST: https://genome.ucsc.edu/s/David%20Price/Host Towne: https://genome.ucsc.edu/s/David%20Price/Town E TB40/E: https://genome.ucsc.edu/s/David%20Price/TB40_E  
+
+## Fragment distribution plots  
+
+Fragment size frequencies were calculated by counting the number of times a single fragment size or a range of fragment sizes are present in a given sample or overlap to a list of genomic intervals of a specific size. Bedtools v2.26 intersect program was used to generate overlap data between fragment and genomic intervals. Bash and awk scripts were used to generate counts for a single or a range of fragment sizes. Next, fragment sizes and their associated counts were sorted from short to long order. Additionally, fragment size counts were normalized to generate their relative amounts by dividing them with the total number of fragments present in a sample. MS Excel was used to plot relative and absolute counts of fragment sizes for each sample using scatter with straight lines option.  
+
+## Average base distribution plots  
+
+The number of times a nucleotide is present at each position within a specified genomic interval was calculated as follows. Bedtools v2.26 getfasta program was used to generate FASTA files while maintaining same strand orientation for \(\pm 100\) bp genomic intervals centered on fragment starts and ends. For these analyses all intervals centered on fragment starts were marked positive strand and intervals centered on fragment ends were marked negative strand. A custom linux based, multi- thread capable python script called ABD.py (https://github.com/P- TEFb/ABD) was run to generate absolute counts for each nucleotide across the given genomic intervals. Absolute nucleotide counts were converted to fractions by dividing them with the total number of sequences. MS Excel was used to plot base fractions across the genomic interval using scatter with straight lines option. Finally, the following color code was used for these plots: A was blue
+
+<--- Page Split --->
+
+(hex code #2222ff), T was orange (hex code #ff6600), G was gray (hex code #bbbbbb), and C was yellow (hex code #dddd00).  
+
+## MaxTSSs analysis  
+
+The truQuant program<sup>28</sup> with an updated list of black listed regions was run on the PRO- Cap data (GSE113394)<sup>25</sup> generated from uninfected HFF cells and DMSO bound NAS- Cap data (GSE139237)<sup>6</sup> generated from HeLa cells using published parameters to generate host specific MaxTSSs. 12,229 (PRO- Cap) and 12,201 (NasCap) MaxTSSs associated with known host genes were used for further analysis. TsrFinderM<sup>26</sup> was run on PRO- Cap datasets from TB40/E infected HFFs 72 hpi (GSEXXXX) and Towne infected HFFs 96 hpi (GSE113394) using published parameters to generate HMCV specific MaxTSSs. MaxTSSs associated with RNA 4.9 and without a \(\pm 1000\) bp genomic sequence were excluded from further analysis. 1,461 TB40/E and 1,456 Towne MaxTSSs were used for further analysis.  
+
+## Feature analysis  
+
+Feature analysis was performed by only counting fragments of a specific length whose center lay within a specific genomic interval using Bedtools v2.26 intersect program and simple awk commands for each MaxTSSs in the host and HMCV (TB40/E and Towne) genomes. The features consisted of fragments 30- 65 bp with a fragment center between \(+10\) and \(+45\) for free Pol II, 140- 205 bp with a fragment center between \(+65\) and \(+140\) for abutted Pol II, 120- 175 with a fragment center between \(+65\) and \(+175\) for \(+1\) nucleosomes, 64- 88 with a fragment center between - 18 and - 2 for TBP PICs, and 40- 63 with a fragment center between - 36 and - 18 for UL87 PICs. The engaged Pol II feature was found by summing the free Pol II and abutted Pol II counts. Quantification of percentage of Pol II that was either abutted or free was done by selecting promoters with at least one read in either the abutted or free features and that have a MaxTSS of at least 10. Percentage of free Pol II was found relative to engaged Pol II for the whole dataset as well as a gene by gene basis. Quantification of PIC amounts to engaged Pol II was done by first normalizing the TBP PIC fragments from the TBP dataset and UL87 PIC fragments from the UL87 dataset. The ratio between these normalized values was used to rank order promoters by UL87 or TBP dominance and then the ratio of PIC to engaged Pol II was calculated. The Ser5P dataset was utilized for analysis of TBP PICs.  
+
+## FragMaps  
+
+FragMapsHeatmaps displaying the average distribution and position of fragments from DFF- Seq and DFF- ChIP across individual genome intervals or collections of intervals were created using fragMap.py (https://github.com/P- TEFb/fragMap). In general, collections were centered on the MaxTSSs from truQuant annotations that provide the most highly utilized TSS for each active gene with proper strand orientation. In some cases, subsets of the truQuant annotations were used. The data was generated by counting the total number of fragments of each fragment size between 18 and 400 bp across the genomic interval. The aspect ratio, number of pixels, intensities assigned, and the shape of major and minor tick marks were controlled. The desired aspect ratio was implemented by choosing a discrete number of pixels for each base or fragment size. The values at each horizontal position of the fragment sizes were used to assign intensities using the gray.colors function in R with 0 being white and maximum being black. A linear relationship between relative read value and intensity was utilized. Black was set at the maximum read value for most frag maps. To correct for human inaccuracies in perception of dark and light patterns on heatmaps, a gamma correction of 0.5 was applied to all fragMaps.  
+
+## Statistics  
+
+StatisticsPearson correlation coefficient “r” was calculated to demonstrate the reproducibility of our datasets using the MS excel CORREL function. Pearson’s r computes the effect of change in
+
+<--- Page Split --->
+
+one variable compared to the change in another variable. MEME motif discovery tool was used to short sequence motifs in a set of longer DNA sequences. Motifs with an E- value less than 0.05 were considered as significantly enriched in our dataset. E- value is an estimation of seeing similar motifs of identical width and contributing sites in a similar size dataset of random sequences.  
+
+## Acknowledgments  
+
+We thank J. A. Marteijn for providing MRC5 cells expressing GFP- tagged Pol II. This research was supported by the National Institute of General Medical Sciences (grant GM126908 to D.H.P. and GM121428 to D.S.L.) and the Department of Veterans Affairs (merit award 1BX001107 to J.L.M.).  
+
+## Author Contributions  
+
+BMS developed the DFF- ChIP protocol, performed all DFF- ChIP experiments, analyzed the data and wrote the paper. MP performed all bioinformatics. ML and JLM generated the HCMV infected HFFs. CBB performed the HCMV time course PRO- Seq experiment. DSL provided intellectual input throughout the study and helped refine the manuscript. DHP directed the research and helped create the bioinformatics approaches.  
+
+## Competing Interests  
+
+No authors have competing interests.
+
+<--- Page Split --->
+
+## References  
+
+1. Compe, E. & Egly, J.M. The Long Road to Understanding RNAPII Transcription Initiation and Related Syndromes. Annu Rev Biochem 90, 193-219 (2021).  
+2. Patel, A.B., Greber, B.J. & Nogales, E. Recent insights into the structure of TFIIID, its assembly, and its binding to core promoter. Curr Opin Struct Biol 61, 17-24 (2020).  
+3. Weber, C.M., Ramachandran, S. & Henikoff, S. Nucleosomes are context-specific, H2A.Z-modulated barriers to RNA polymerase. Mol Cell 53, 819-30 (2014).  
+4. Santos-Rosa, H. et al. Active genes are tri-methylated at K4 of histone H3. Nature 419, 407-11 (2002).  
+5. Howe, F.S., Fischl, H., Murray, S.C. & Mellor, J. Is H3K4me3 instructive for transcription activation? Bioessays 39, 1-12 (2017).  
+6. Luse, D.S., Parida, M., Spector, B.M., Nilson, K.A. & Price, D.H. A unified view of the sequence and functional organization of the human RNA polymerase II promoter. Nucleic Acids Res 48, 7767-7785 (2020).  
+7. Vermeulen, M. et al. Selective anchoring of TFIIID to nucleosomes by trimethylation of histone H3 lysine 4. Cell 131, 58-69 (2007).  
+8. Lauberth, S.M. et al. H3K4me3 interactions with TAF3 regulate preinitiation complex assembly and selective gene activation. Cell 152, 1021-36 (2013).  
+9. Day, D.S. et al. Comprehensive analysis of promoter-proximal RNA polymerase II pausing across mammalian cell types. Genome Biol 17, 120 (2016).  
+10. Jimeno-Gonzalez, S., Ceballos-Chavez, M. & Reyes, J.C. A positioned +1 nucleosome enhances promoter-proximal pausing. Nucleic Acids Res 43, 3068-78 (2015).  
+11. Verger, A., Monte, D. & Villeret, V. Take Your PIC. Trends Biochem Sci (2021).  
+12. Chen, X. et al. Structural insights into preinitiation complex assembly on core promoters. Science 372(2021).  
+13. Abdella, R. et al. Structure of the human Mediator-bound transcription preinitiation complex. Science (2021).  
+14. Aibara, S., Schilbach, S. & Cramer, P. Structures of mammalian RNA polymerase II pre-initiation complexes. Nature 594, 124-128 (2021).  
+15. Meier, J.L. & Stinski, M.F. Major Immediate-Early Enhancer and Its Gene Products. Cytomegaloviruses: From Molecular Pathogenesis to Intervention, Vol I, 152-173 (2013).  
+16. Paulus, C. & Nevels, M. The human cytomegalovirus major immediate-early proteins as antagonists of intrinsic and innate antiviral host responses. Viruses 1, 760-79 (2009).  
+17. Wyrwicz, L.S. & Rychlewski, L. Identification of Herpes TATT-binding protein. Antiviral Res 75, 167-72 (2007).  
+18. Aubry, V. et al. Epstein-Barr virus late gene transcription depends on the assembly of a virus-specific preinitiation complex. J Virol 88, 12825-38 (2014).  
+19. Gruffat, H., Kadjouf, F., Mariame, B. & Manet, E. The Epstein-Barr virus BcRF1 gene product is a TBP-like protein with an essential role in late gene expression. J Virol 86, 6023-32 (2012).  
+20. Davis, Z.H. et al. Global mapping of herpesvirus-host protein complexes reveals a transcription strategy for late genes. Mol Cell 57, 349-60 (2015).  
+21. Nitzsche, A., Paulus, C. & Nevels, M. Temporal dynamics of cytomegalovirus chromatin assembly in productively infected human cells. J Virol 82, 11167-80 (2008).  
+22. Cuevas-Bennett, C. & Shenk, T. Dynamic histone H3 acetylation and methylation at human cytomegalovirus promoters during replication in fibroblasts. J Virol 82, 9525-36 (2008).  
+23. Groves, I.J., Reeves, M.B. & Sinclair, J.H. Lytic infection of permissive cells with human cytomegalovirus is regulated by an intrinsic 'pre-immediate-early' repression of viral gene expression mediated by histone post-translational modification. J Gen Virol 90, 2364-2374 (2009).
+
+<--- Page Split --->
+
+24. Zalckvar, E. et al. Nucleosome maps of the human cytomegalovirus genome reveal a temporal switch in chromatin organization linked to a major IE protein. Proc Natl Acad Sci U S A 110, 13126-31 (2013).  
+
+25. Parida, M. et al. Nucleotide Resolution Comparison of Transcription of Human Cytomegalovirus and Host Genomes Reveals Universal Use of RNA Polymerase II Elongation Control Driven by Dissimilar Core Promoter Elements. mBio 10(2019).  
+26. Griffiths, P. & Reeves, M. Pathogenesis of human cytomegalovirus in the immunocompromised host. Nat Rev Microbiol (2021).  
+
+27. Britt, W.J. Maternal Immunity and the Natural History of Congenital Human Cytomegalovirus Infection. Viruses 10(2018).  
+
+28. Li, M. et al. Human cytomegalovirus IE2 drives transcription initiation from a select subset of late infection viral promoters by host RNA polymerase II. PLoS Pathog 16, e1008402 (2020).  
+
+29. Liu, X., Zou, H., Slaughter, C. & Wang, X. DFF, a heterodimeric protein that functions downstream of caspase-3 to trigger DNA fragmentation during apoptosis. Cell 89, 175-84 (1997).  
+
+30. Steurer, B. et al. Live-cell analysis of endogenous GFP-RPB1 uncovers rapid turnover of initiating and promoter-paused RNA Polymerase II. Proc Natl Acad Sci U S A 115, E4368-E4376 (2018).  
+
+31. Knezetic, J.A. & Luse, D.S. The presence of nucleosomes on a DNA template prevents initiation by RNA polymerase II in vitro. Cell 45, 95-104 (1986).  
+
+32. Lorch, Y., LaPointe, J.W. & Kornberg, R.D. Nucleosomes inhibit the initiation of transcription but allow chain elongation with the displacement of histones. Cell 49, 203-10 (1987).  
+
+33. Cai, H. & Luse, D.S. Transcription initiation by RNA polymerase II in vitro. Properties of preinitiation, initiation, and elongation complexes. J Biol Chem 262, 298-304 (1987).  
+
+34. Egly, J.M. & Coin, F. A history of TFIIH: two decades of molecular biology on a pivotal transcription/repair factor. DNA Repair (Amst) 10, 714-21 (2011).  
+
+35. Gruffat, H., Marchione, R. & Manet, E. Herpesvirus Late Gene Expression: A Viral-Specific Pre-initiation Complex Is Key. Front Microbiol 7, 869 (2016).  
+
+36. Li, M. et al. Cytomegalovirus late transcription factor target sequence diversity orchestrates viral early to late transcription. PLoS Pathog 17, e1009796 (2021).  
+
+37. Bailey, T.L. et al. MEME SUITE: tools for motif discovery and searching. Nucleic Acids Res 37, W202-8 (2009).  
+
+38. Samkuraishvili, I. & Luse, D.S. Translocation and transcriptional arrest during transcript elongation by RNA polymerase II. J Biol Chem 271, 23495-505 (1996).  
+
+39. Nevels, M., Nitzsche, A. & Paulus, C. How to control an infectious bead string: nucleosome-based regulation and targeting of herpesvirus chromatin. Rev Med Virol 21, 154-80 (2011).  
+
+40. Kent, J.R. et al. During lytic infection herpes simplex virus type 1 is associated with histones bearing modifications that correlate with active transcription. J Virol 78, 10178-86 (2004).  
+
+41. Ball, C.B., Nilson, K.A. & Price, D.H. Use of the nuclear walk-on methodology to determine sites of RNA polymerase II initiation and pausing and quantify nascent RNAs in cells. Methods 159-160, 165-176 (2019).  
+
+42. Kang, M.E. & Dahmus, M.E. RNA polymerases IIA and IIO have distinct roles during transcription from the TATA-less murine dihydrofolate reductase promoter. J Biol Chem 268, 25033-40 (1993).  
+
+43. Laybourn, P.J. & Dahmus, M.E. Phosphorylation of RNA polymerase IIA occurs subsequent to interaction with the promoter and before the initiation of transcription. J Biol Chem 265, 13165-73 (1990).  
+
+44. Kang, M.E. & Dahmus, M.E. RNA polymerases IIA and IIO have distinct roles during transcription from the TATA-less murine dihydrofolate reductase promoter. J Biol Chem 268, 25033-40 (1993).  
+
+45. Ball, C.B., Nilson, K.A. & Price, D.H. Use of the nuclear walk-on methodology to determine sites of RNA polymerase II initiation and pausing and quantify nascent RNAs in cells. Methods 159-160, 165-176 (2019).  
+
+46. Kang, M.E. & Dahmus, M.E. RNA polymerases IIA and IIO have distinct roles during transcription from the TATA-less murine dihydrofolate reductase promoter. J Biol Chem 268, 25033-40 (1993).  
+
+47. Laybourn, P.J. & Dahmus, M.E. Phosphorylation of RNA polymerase IIA occurs subsequent to interaction with the promoter and before the initiation of transcription. J Biol Chem 265, 13165-73 (1990).
+
+<--- Page Split --->
+
+44. He, Y. et al. Near-atomic resolution visualization of human transcription promoter opening. Nature 533, 359-65 (2016).45. Castaneda, A.F. et al. The gammaherpesviral TATA-box-binding protein directly interacts with the CTD of host RNA Pol II to direct late gene transcription. PLoS Pathog 16, e1008843 (2020).46. Skene, P.J., Hernandez, A.E., Groudine, M. & Henikoff, S. The nucleosomal barrier to promoter escape by RNA polymerase II is overcome by the chromatin remodeler Chd1. Elife 3, e02042 (2014).47. Farnung, L., Vos, S.M., Wigge, C. & Cramer, P. Nucleosome-Chd1 structure and implications for chromatin remodelling. Nature 550, 539-542 (2017).48. Bondarenko, V.A. et al. Nucleosomes can form a polar barrier to transcript elongation by RNA polymerase II. Mol Cell 24, 469-79 (2006).49. Ramachandran, S., Ahmad, K. & Henikoff, S. Transcription and Remodeling Produce Asymmetrically Unwrapped Nucleosomal Intermediates. Mol Cell 68, 1038-1053 e4 (2017).
+
+<--- Page Split --->
+
+## Figure Legends  
+
+## Figure 1. Reproducibility of Pol II and H3K4me3 DFF-ChIP  
+
+Figure 1. Reproducibility of Pol II and H3K4me3 DFF- ChIPA Diagram of the DFF- ChIP method. Isolated nuclei are digested with DFF without crosslinking and lightly sonicated to release soluble DNA complexes. The soluble DNA is then immunoprecipitated, library prepped, and sequenced. B,C Genome browser tracks of Pol II DFF- ChIP (purple) and H3K4me3 DFF- ChIP (orange) from HeLa and MRC5 GFP- Pol II cells generated in two different experiments (Exp1 and Exp2). Browser tracks of Flavo NasCap PROSeq (black/grey) show transcription data. D Correlation plots of datasets from Exp1 and Exp2. Read counts in 10,000 bp windows around 12,229 HFF truQuant MaxTSSs were summed and plotted against sums from other experiments.  
+
+Figure 2. Genome browser tracks of H3K4me3 and Pol II DFF- ChIP from Infected HFFs A,B Genome browser tracks of PRO- Seq, Pol II DFF- ChIP, and H3K4me3 DFF- ChIP from HFFs 48 hpi on the hg38 and TB40/E genomes. C,D Genome browser tracks of PRO- Seq, Pol II DFF- ChIP, and H3K4me3 DFF- ChIP showing the GAPDH promoter in 5,000 and 1,000 bp windows. A dotted line denotes the TSS. E,F Genome browser tracks of PRO- Seq, Pol II DFF- ChIP, and H3K4me3 DFF- ChIP showing the an early (E) and late (F) promoter in a 1,000 bp windows. A dotted line denotes the TSS.  
+
+## Figure 3. Visualizing and quantifying transcription complexes and chromatin utilizing fragMaps  
+
+Figure 3. Visualizing and quantifying transcription complexes and chromatin utilizing fragMapsA Length distribution of fragments +/- 1000 bp of each of 12,229 truQuant genes for the Pol II and H3K4me3 DFF- ChIP datasets from Exp4. B Fragment count of aligned fragments from Exp4 of specified length +/- 1000 bp relative to the TSS of all 12,229 truQuant genes. Total fragment length counts were normalized. C fragMaps of H3K4me3 and Pol II for the 12,229 truQuant genes showing fragments from 18- 400 bp that are +/- 1000 bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are +/- 100 bp relative to the TSS is also shown (right). D fragMaps of H3K4me3 and Pol II for the 1,461 TSRs showing fragments from 18- 400 bp that are +/- 1000 bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are +/- 100 bp relative to the TSS is also shown (right. Resulting fragMaps were lightened 100% to aid in visualization. E Quantification of percentage of the free Pol II feature signal relative to total Pol II feature signal (free + abutted) on both the host and TB40/E genomes on a gene by gene basis sorted by highest to lowest (Left). Fragment count of the Nuc1 feature signal from the H3K4me3 dataset on host (middle) and TB40/E (right) genomes utilizing the same sort.  
+
+Figure 4. Detection and characterization of TBP- driven PICs and UL87- driven viral PICs A,B fragMaps of fragments positioned +/- 100 bp around the TSS that are 18- 120 bp sized fragments. Host fragMaps were generated from 12,229 truQuant HFF promoters and HCMV (TB40/E) fragMaps were generated from 1,461 TSRs from the Pol II, Pol II + triptolide (Trp), TBP, and Ser5P DFF- ChIP datasets. A dotted line denotes the TSS. C fragMaps of fragments positioned +/- 1000 bp around the MaxTSS that are 18- 400 bp in size. The host fragMap was generated from 12,229 truQuant HFF promoters and HCMV (TB40/E) fragMap was generated from 1,461 TSRs using the TBP datasets. D UL87 fragMaps were generated from 1,456 Towne TSRs. A dotted line denotes the TSS. E LOGOs generated with MEME Suite 5.3.1 from the top 10% genes/TSRs with the most fragments present in the TBP PIC or UL87 PIC feature as detected by DFF- ChIP. Parameters were: ZOOPS, search only given strand, 1 motif, 6 bp motif. Fractions represent the number of sequences matching the sequence motif out of the number of input sequences. E values for the three LOGOs were: TBP host, 1.5e- 398; TBP HCMV, 3.6e- 038; UL87 HCMV, 1.4e- 082.
+
+<--- Page Split --->
+
+Figure 5. Classification of TBP and UL87 usage on HCMV genes according to lateness A PRO- Seq tracks depicting \(5'\) ends of reads from HCMV infection time course including 4, 12, 24, 48 and 72 hpi datasets in 1,400 and 800 bp regions of the viral genome. Below are corresponding DFF- ChIP tracks and genomic fragMaps of the same region. B A set of 795 TSRs with greater than 100 MaxTSS \(5'\) ends \((+ / - 5bp)\) when all time points are summed were selected and sorted based on PFA sensitivity, slope, and UL87 dependency. Each TSR had each time point value normalized to library size and each TSR was colored independently. The time point with the highest relative transcription was colored green and lowest colored red. C Quantification of the relative usage of TBP and UL87. The amount of TBP PIC feature counted from the TBP dataset was normalized to the amount of UL87 PIC feature counted from the UL87 dataset for all 1,461 TSRs. The ratio of TBP PIC to UL87 PIC was used to sort the TSRs by TBP PIC dominance (High value) and then plotted. D FragMaps for the top \(5\%\) TBP and UL87 dominated (C) TSRs utilizing 18- 120 bp sized fragments positioned \(+ / - 100\) bp around the TSSs. Top TBP TSRs are depicted using TBP and Ser5P datasets whereas UL87 TSRs are depicted using UL87 and Pol II datasets.  
+
+## Figure 6. Analysis of TBP and UL87 PICs on the HCMV genome  
+
+A Genome browser tracks from H3K4me3, TBP, and UL87 datasets (Exp4) compared directly to genomic fragMaps of the same region of the HCMV genome. B Normalized metaplot of H3K4me3 signal around all 12,229 host truQuant promoters and 1,461 HCMV TSRs. The inner graph shows the HCMV metaplot with a different Y- axis. C Normalized metaplot of H3K4me3 signal around the top \(10\%\) of HCMV TSRs determined by amount of TBP PIC feature or UL87 PIC feature.  
+
+## Figure 7. FragMaps resulting from different extents of DFF digestion  
+
+A,B Total fragment distributions from Pol II, TBP, and H3K4me3 datasets from the control digestion condition (Exp4) and the excess digestion condition (Exp5). C,D Pol II, H3K4me3, and TBP fragMaps of \(+ / - 1000\) bp around the TSS containing 18- 400 bp fragments of 12,229 truQuant genes with either (C) control (Exp4) or (D) excess digestion (Exp5).  
+
+## Figure S1. Analysis of DFF-Seq and comparison of DFF-Seq to MNase-Seq  
+
+A Fragment length distribution for the total DFF- Seq library and for \(+ / - 1000\) bp from 12,201 truQuant HeLa TSSs. Data was normalized such that total fragment counts were the same. Mono- , di- , and tri- nucleosome are demarcated. B Base distribution of the surrounding 60 bp around the \(5'\) ends DFF- Seq fragments.  
+
+## Figure S2. Reproducibility of DFF-ChIP on infected HFFs  
+
+A Genome browser tracks DFF- ChIP H3K4me3 and Pol II tracks from Exp3 and Exp4 and 72 hpi PRO- Cap transcription data. Representative 8000 bp regions are shown on the host and HCMV genomes. B Correlations between Exp3 and Exp4 datasets. Reads from 10,000 bp windows centered on each host truQuant promoter or each viral 200 bp TSR were summed and compared across datasets.  
+
+## Figure S3. Visualizing transcription complexes and chromatin from DFF-Seq and Exp3 replicate data  
+
+A Length distribution of fragments \(+ / - 1000\) bp of each of 12,229 truQuant genes for the Pol II and H3K4me3 DFF- ChIP Exp3 datasets. B Fragment count of aligned fragments of specified length \(+ / - 1000\) bp relative to the TSS of all 12,229 truQuant genes using Exp3 data. Total fragment length counts were normalized. C fragMap representation of DFF- Seq data shown using two different black values. D FragMaps of Exp3 H3K4me3 and Exp3 Pol II for the 12,229
+
+<--- Page Split --->
+
+truQuant genes showing fragments from 18- 400 bp that are \(+ / - 1000\) bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are \(+ / - 100\) bp relative to the TSS is also shown (right). E FragMaps of Exp3 H3K4me3 and Exp3 Pol II for the 1,461 TSRs showing fragments from 18- 400 bp that are \(+ / - 1000\) bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are \(+ / - 100\) bp relative to the TSS is also shown (right).  
+
+## Figure S4. DFF-Seq fragMap and salt sensitivity of TBP PICs on the host genome  
+
+A Genome browser tracks showing uninfected HFF PRO- Cap data and MRC5 Flavo GFP- Pol II DFF- ChIP tracks that were either washed with high or low salt as described in Methods. The dotted line demarcates the TSS. B fragMaps of \(+ / - 1000\) bp around the TSS containing 18- 400 bp sized fragments and zoomed in fragMaps of \(+ / - 100\) bp around the TSS containing 18- 120 bp sized fragments from 12,201 HeLa truQuant genes from the MRC5 GFP- Pol II datasets. Regions of free Pol II and PIC are demarcated by brackets. For the wider fragMaps, the black value was reduced 3 fold to emphasize smaller fragments. For the zoomed in fragMaps, max black levels were set based on the shown region.  
+
+## Figure S5. TBP PICs are functionally different from UL87 PICs even though both may appear at late promoters  
+
+A Quantification of 5' end reads in an 11 bp window around the five TSSs specifically indicated in Fig. 5A. B A subset TSRs from Fig. 5B plotted with PFA sensitivity and UL87 dependency against slope. TSRs with a TBP/UL87 ratio greater than 2 were colored blue and TSRs with a ratio less than 0.5 were colored red. C Correlations of the amount of engaged Pol II and TBP PIC features counted utilizing the Ser5P dataset in comparison to the Pol II and TBP dataset, respectively, on the host and HCMV genomes. D Genome browser tracks depicting shifting 5' ends from PRO- Seq CMV infection time course data and DFF- ChIP tracks showing UL87, TBP, Pol II, and Ser5P datasets.  
+
+## Figure S6. Genomic fragMaps spanning the entire HCMV genome for H3K4me3, TBP, and UL87 fragments  
+
+A Genomic fragMaps depicting overlapping 21,000 bp windows of the entire HCMV genome. H3K4me3 depict fragments between 18- 400 bp whereas TBP and UL87 fragMaps depict fragments between 18- 150 bp.  
+
+## Figure S7. Characterization of fragments resulting from excess DFF digestion  
+
+A Correlation plots comparing number of reads in 10,000 bp windows around all 12,229 truQuant genes. Graphs compare Exp5 replicates and a single Exp5 replicate to Exp4 data. B H3K4me3 fragMaps from the top \(25\%\) and bottom \(25\%\) of the 12,229 truQuant genes sorted by TSS focus (standard deviation of TSSs in TSRs; Exp4). C fragMaps from the H3K4me3 and Pol II excess digestion experiments (Exp5) with schematic depicting likely nucleosome protections and DFF cleavage sites.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+
+<--- Page Split --->
+![](images/Figure_unknown_1.jpg)
+
+
+<--- Page Split --->
+![](images/Figure_unknown_2.jpg)
+
+
+<--- Page Split --->
+![](images/Figure_unknown_3.jpg)
+
+ 
+
+![](images/Figure_unknown_4.jpg)
+
+ 
+
+![](images/Figure_unknown_5.jpg)
+
+ 
+
+![](images/Figure_unknown_6.jpg)
+
+ 
+
+![](images/Figure_unknown_7.jpg)
+
+ 
+
+<center>Figure 5</center>
+
+<--- Page Split --->
+![](images/Figure_unknown_8.jpg)
+
+<center>C </center>  
+
+![](images/Figure_unknown_9.jpg)
+
+<center>Host H3K4me3 </center>  
+
+![PLACEHOLDER_27_2]
+
+<center>Host TBP </center>  
+
+![PLACEHOLDER_27_3]
+
+<center>D </center>  
+
+![PLACEHOLDER_27_4]
+
+<center>Host H3K4me3 </center>  
+
+![PLACEHOLDER_27_5]
+
+<center>Host Pol II excess digestion </center>  
+
+![PLACEHOLDER_27_6]
+
+<center>Host Pol II excess digestion </center>
+
+<--- Page Split --->
+![PLACEHOLDER_28_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_29_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_30_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_31_0]
+
+ 
+
+<center>MRC5 GFP-Pol II Low Salt</center> 
+
+![PLACEHOLDER_31_1]
+
+ 
+
+<center>MRC5 GFP-Pol II High Salt</center> 
+
+![PLACEHOLDER_31_2]
+
+ 
+
+<center>Figure S4</center>
+
+<--- Page Split --->
+![PLACEHOLDER_32_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_33_0]
+
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+DataAnalysis.xlsx
+
+<--- Page Split --->

preprint/preprint__b345a63a0d2e91059489bde27ab056b68d5ce9330496d1ab7976d39d7786a904/preprint__b345a63a0d2e91059489bde27ab056b68d5ce9330496d1ab7976d39d7786a904_det.mmd

@@ -0,0 +1,620 @@
+<|ref|>title<|/ref|><|det|>[[44, 107, 899, 209]]<|/det|>
+# Differences in RNA polymerase II complexes and their interactions with surrounding chromatin on human and cytomegalovirus genomes  
+
+<|ref|>text<|/ref|><|det|>[[44, 230, 450, 272]]<|/det|>
+Benjamin Spector Carver College of Medicine University of Iowa  
+
+<|ref|>text<|/ref|><|det|>[[44, 278, 450, 319]]<|/det|>
+Mrutyunjaya Parida Carver College of Medicine University of Iowa  
+
+<|ref|>text<|/ref|><|det|>[[44, 325, 450, 365]]<|/det|>
+Christopher Ball Carver College of Medicine University of Iowa  
+
+<|ref|>text<|/ref|><|det|>[[44, 371, 808, 412]]<|/det|>
+Ming Li Carver College of Medicine University of Iowa https://orcid.org/0000- 0003- 0396- 4078  
+
+<|ref|>text<|/ref|><|det|>[[44, 417, 450, 457]]<|/det|>
+Jeffrey Meier Carver College of Medicine University of Iowa  
+
+<|ref|>text<|/ref|><|det|>[[44, 464, 270, 503]]<|/det|>
+Donal S. Luse Lerner Research Institute  
+
+<|ref|>text<|/ref|><|det|>[[44, 509, 450, 550]]<|/det|>
+David Price (David- price@uiowa.edu) Carver College of Medicine University of Iowa  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 591, 181, 609]]<|/det|>
+## Genetics Article  
+
+<|ref|>text<|/ref|><|det|>[[42, 628, 926, 671]]<|/det|>
+Keywords: RNA polymerase II, preinitiation complexes, chromatin, human cytomegalovirus, TBP, UL87, DFF ChIP-Seq  
+
+<|ref|>text<|/ref|><|det|>[[44, 689, 313, 708]]<|/det|>
+Posted Date: October 8th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 727, 463, 746]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 912323/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 764, 910, 807]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 843, 910, 886]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on April 14th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 29739-x.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[137, 89, 860, 123]]<|/det|>
+# Differences in RNA polymerase II complexes and their interactions with surrounding chromatin on human and cytomegalovirus genomes  
+
+<|ref|>text<|/ref|><|det|>[[115, 136, 812, 171]]<|/det|>
+Benjamin M. Spector<sup>a</sup>, Mrutyunjaya Parida<sup>a</sup>, Ming Li<sup>a,b,c,d</sup>, Christopher B. Ball<sup>a</sup>, Jeffrey L. Meier<sup>b,c,d</sup>, Donal S. Luse<sup>a</sup>, and David H. Price<sup>a</sup>  
+
+<|ref|>text<|/ref|><|det|>[[115, 184, 747, 202]]<|/det|>
+<sup>a</sup>Department of Biochemistry, The University of Iowa, Iowa City, IA 52242, USA  
+
+<|ref|>text<|/ref|><|det|>[[115, 216, 780, 234]]<|/det|>
+<sup>b</sup>Department of Internal Medicine, The University of Iowa, Iowa City, IA 52242, USA  
+
+<|ref|>text<|/ref|><|det|>[[115, 248, 752, 266]]<|/det|>
+<sup>c</sup>Department of Epidemiology, The University of Iowa, Iowa City, IA 52242, USA  
+
+<|ref|>text<|/ref|><|det|>[[115, 280, 630, 298]]<|/det|>
+<sup>d</sup>Veterans Affairs Health Care System, Iowa City, IA 52242, USA  
+
+<|ref|>text<|/ref|><|det|>[[115, 312, 860, 346]]<|/det|>
+<sup>e</sup>Department of Cardiovascular and Metabolic Sciences, Lerner Research Institute, Cleveland Clinic, Cleveland, OH, 44195, USA.  
+
+<|ref|>text<|/ref|><|det|>[[115, 360, 844, 393]]<|/det|>
+Keywords: RNA polymerase II, preinitiation complexes, chromatin, human cytomegalovirus, TBP, UL87, DFF ChIP- Seq  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 409, 200, 425]]<|/det|>
+## Summary  
+
+<|ref|>text<|/ref|><|det|>[[115, 425, 881, 616]]<|/det|>
+Interactions of the RNA polymerase II (Pol II) preinitiation complex (PIC) and paused early elongation complexes with the first downstream \((+1)\) nucleosome are thought to be functionally important. However, current methods are limited for investigating these relationships, both for cellular chromatin and the human cytomegalovirus (HCMV) genome. Digestion with human DNA fragmentation factor (DFF) before immunoprecipitation (DFF- ChIP) precisely revealed both similarities and major differences in PICs driven by TBP on the host genome in comparison with PICs driven by TBP or the viral- specific, late initiation factor UL87 on the viral genome. Host PICs and paused Pol II complexes are frequently found in contact with the \(+1\) nucleosome and paused Pol II can also be found in a complex involved in the initial invasion of the \(+1\) nucleosome. In contrast, viral transcription complexes have very limited nucleosomal interactions, reflecting a relative lack of chromatinization of transcriptionally active regions of HCMV genomes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 632, 206, 648]]<|/det|>
+## Highlights  
+
+<|ref|>text<|/ref|><|det|>[[115, 648, 795, 713]]<|/det|>
+DFF- ChIP and fragMaps allow visualization of large transcription complexes in cells HCMV promoters are not surrounded by H3K4me3 marked nucleosomes PICs are major features found on host and viral genomes PICs driven by TBP and the HCMV late transcription factor UL87 are highly dissimilar
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 222, 106]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[114, 105, 879, 425]]<|/det|>
+Regulation of human gene expression is accomplished by a highly orchestrated interplay between the transcription machinery and its chromatinized genomic template. The required general Pol II initiation factors are instructed by Mediator and a host of more specific transcription factors to utilize selected promoters as sites of initiation'. However, the default state of the genome is repressive because of global nucleosome deposition, which must be relieved by chromatin remodelers. Assembly of the preinitiation complex (PIC) occurs over sequences surrounding the TSS and an upstream region depleted of nucleosomes (NDR)2. A strongly positioned \(+1\) nucleosome has a boundary around 50 bp downstream from the transcription start site (TSS)3. This nucleosome and several downstream nucleosomes are marked by tri- methylation of histone H3 on lysine 4 (H3K4me3)4,5. Both PIC assembly and pausing by newly- initiated Pol II have been linked to interactions with the \(+1\) nucleosome6- 10, but the nature and functional significance of these interactions remain incompletely understood. Existing global methods to visualize and quantify sites of transcription and the locations of the transcriptional machinery within the local chromatin landscape have had limited success in addressing these questions. Occupancy by the \(+1\) nucleosome is near \(100\%\) while Pol II occupancy is generally much less than \(10\%\) , resulting in a potentially misleading correlation between two disparate signals. Recently there has been a number of highly informative structural studies of PICs11- 14, but the abundance of PICs on cellular chromatin has not been adequately determined and their positioning over promoter elements has been primarily inferred from in vitro studies.  
+
+<|ref|>text<|/ref|><|det|>[[114, 424, 880, 696]]<|/det|>
+The interface of transcription and chromatin on the HCMV genome is even more poorly defined. Lyti infection can propagate only in non- dividing cells. The process begins with delivery of a nucleosome- free viral genome to the nucleus, where the standard host Pol II machinery drives transcription from the major immediate early promoter. Expression of viral immediate early proteins then allow expression of early genes15,16. The early genes encode the machinery for viral DNA replication and for transcription of a group of late genes, which requires a special set of viral- specific Pol II initiation factors17- 20 that likely replace some of the host initiation factors. While ChIP- PCR on individual loci and genome- wide studies have revealed changes in chromatinization throughout the viral lifecycle21- 24, it is not clear what regions on each of the hundreds of viral genomes present late in infection are occupied by nucleosomes. Nothing is known about how Pol II transcription complexes interface with any chromatin that may be present. Our recent work demonstrated promiscuous transcription initiation from thousands of promoters across the \(\sim 240,000\) bp dsDNA genome, consistent with the idea that chromatinization of the HCMV genomes during lytic infection is incomplete25. Increasing our knowledge of how HCMV transcription is regulated is important for identification of potential therapeutic targets since the virus infects about \(60\%\) of the population. HCMV is a significant cause of death in immunocompromised individuals26 and a leading viral cause of birth defects27.  
+
+<|ref|>text<|/ref|><|det|>[[114, 696, 881, 888]]<|/det|>
+Our group recently described a nuclear run- off method to directly observe engaged Pol II interacting with the \(+1\) nucleosome. We digested nuclei with the double- stranded endonuclease human DNA fragmentation factor (DFF) and then chased nascent transcripts to the resulting run- off sites. We found that many, but not all, paused polymerases were abutted to the \(+1\) nucleosome. Because DFF digestion preserved the viability of transcription complexes and their relation to chromatin, we decided to investigate whether combining it with chromatin immunoprecipitation would allow improved insight into localization and positioning of transcription complexes. We found that we were able to quantitatively visualize PICs as well as interactions between PICs and paused transcription complexes with the downstream chromatin. Most of our experiments were done on primary human foreskin fibroblasts (HFFs) productively infected with HCMV, allowing a direct comparison of the chromatin neighborhoods of human genomic Pol II promoters with their more poorly characterized counterparts on the viral genome.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 183, 105]]<|/det|>
+## Results  
+
+<|ref|>text<|/ref|><|det|>[[115, 105, 880, 378]]<|/det|>
+Our study began with a characterization of our initial DFF- Seq dataset (GSE139237) generated from HeLa cells6. Fragments from DFF- Seq were about 160 bp in length and primarily derived from nucleosomes covering inactive regions of the genome, but a very small percentage of fragments were sub- nucleosomal in size and many of these localized within promoter regions of actively transcribed genes6. To investigate the regions around promoters, active genes in HeLa cells were identified by truQuant28 which finds the most highly utilized TSS (MaxTSS) for each expressed gene from HeLa NasCap25 data. Fragments generated by DFF that were present in a 2000 bp region centered on the MaxTSS of each of the 12,201 genes were collected and the distribution of fragment lengths was quantified and compared to fragment lengths from the total DFF- Seq dataset (Supplementary Fig. 1A). In the total dataset, peaks of fragment sizes corresponding to mono- , di- , and tri- nucleosome were 163, 326, and 512 bp respectively, while those values were 161, 298 and 452 in the truQuant subset. This suggests that many nucleosomes around promoters are essentially close packed. DFF is a homodimer that mostly cuts DNA to form blunt ends29. To determine any sequence requirements for DFF cleavage, sequences surrounding 520 million cut sites were examined. A slight sequence preference was seen (Supplementary Fig. 1B) with \(6.8\%\) of the sites having AAANT(cut) directly on one of the two sides of the cut. There was no preference to having this sequence on both sides.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 392, 805, 409]]<|/det|>
+## DFF-ChIP reveals differing promoter architecture on human and HCMV genomes  
+
+<|ref|>text<|/ref|><|det|>[[115, 409, 870, 520]]<|/det|>
+Because DFF can generate primarily nucleosome- sized fragments without significant internal cutting, we performed two initial experiments (Exp1 and Exp2) to explore its use as a front end for H3K4me3 and Pol II ChIP. Nuclei from non- crosslinked HeLa or MRC5 cells expressing a GFP- tagged Pol II30 were digested with DFF for 1 hour to generate primarily mononucleosomes. The resulting chromatin was immunoprecipitated with antibodies to H3K4me3 modified nucleosomes or to GFP and the associated DNA was prepared for sequencing (Fig. 1A).  
+
+<|ref|>text<|/ref|><|det|>[[115, 520, 880, 744]]<|/det|>
+The DFF- ChIP results were compared to NasCap PRO- Seq data6 from HeLa cells. Paused Pol II is evident from the tall peaks in the PRO- Seq data and over a 500,000 bp region of the human genome peaks of Pol II and H3K4me3 occupancy in the DFF- ChIP tracks exhibited strong visual correlation (Fig. 1B). Greater detail can be observed across a 30,000 bp region (Fig. 1C). Each promoter exhibits clear nucleosome phasing in the H3K4me3 dataset surrounding a NDR that supports Pol II initiation in both cell types. Additionally, Pol II DFF- ChIP signal overlaps with the PRO- Seq signal but also extends further downstream. This downstream signal likely results from Pol II that is abutted to the \(+1\) nucleosome such that DFF cannot cleave between polymerase and the nucleosome6. The complex H3K4me3 and Pol II patterning on the majority of promoters is well replicated even across cell types and growth conditions between Exp1 and Exp2 indicating the robustness of the DFF- ChIP method. To further validate this method's reproducibility, reads found in a 10 kb window centered on each of 12,201 truQuant MaxTSSs from the different experiments were directly compared and strongly correlated between the datasets (Fig. 1D).  
+
+<|ref|>text<|/ref|><|det|>[[115, 744, 875, 904]]<|/det|>
+Given the success of these initial experiments, DFF- ChIP was then applied to contact inhibited HFFs infected with HCMV (TB40/E) for 48 hours (Exp3). DFF- ChIP results were compared to PRO- Seq data from similarly infected HFFs for broad regions across the host (Fig. 2A) and viral genomes (Fig. 2B). As expected on the host genome, Pol II and H3K4me3 correspond with paused Pol II evident from the PRO- Seq data. The viral genome was pervasively transcribed as previously demonstrated25 and Pol II DFF- ChIP correlated with the PRO- Seq signal when transcription of both strands is taken into account. Unlike the host genome where H3K4me3 is found only around promoter regions, the entirety of the HCMV genome is covered with H3K4me3, at levels ranging from relatively low to more enriched irrespective of Pol II occupancy.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 410]]<|/det|>
+Additional insight was obtained when the results from individual promoters on the host and viral genomes were examined. The host GAPDH promoter features nucleosome phasing around a NDR that supports Pol II initiation (Fig. 2C and 2D) as it did in Exp1 and Exp2. In contrast, the viral early gene promoter for UL4 and late gene promoter for UL76 show no obvious phasing of nucleosomes (Fig. 2E and 2F). Critically, the two viral promoters are not found in NDRs. Because initiation cannot take place when the promoter is occluded by a nucleosome \(^{31,32}\) , the results suggest that some of the nucleosomes detected are present only on regions of viral genomes that are not transcribed. As in Exp1 and Exp2, the Pol II DFF- ChIP signal on GAPDH overlaps with the PRO- Seq signal but also extends farther upstream and downstream of the MaxTSS (Fig. 2D). Signal from the free paused Pol II predominates for the HCMV early promoter (Fig. 2E), but is not the main signal for the late promoter (Fig. 2F). Neither viral promoter shows evidence of nucleosome- abutted Pol II. The location of upstream protection seen in all three promoters suggests the presence of Pol II in preinitiation complexes and surprisingly, those complexes predominate for the viral late promoter. A fourth experiment was performed on HCMV infected HFFs using a larger set of antibodies for ChIP (Exp4). Comparison of DFF- ChIP for H3K4me3 and Pol II across specific regions on the host and viral genomes for Exp3 and Exp4 clearly demonstrates high reproducibility of the patterns of occupancy (Supplementary Fig. 2A and 2B). In addition, a number of strong correlations between Exp3 and Exp4 were found when signals around host and viral promoters were compared (Supplementary Fig. 2C and 2D).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 424, 805, 457]]<|/det|>
+## Transcription complexes and their interactions with nearby nucleosomes can be visualized using fragMaps  
+
+<|ref|>text<|/ref|><|det|>[[114, 456, 881, 730]]<|/det|>
+Because the patterns of DFF protection around promoters varied, fragment length analysis was explored as a means to identify specific transcription complexes. The same set of MaxTSSs for all active host genes (12,229) were utilized in this analysis. The distribution of DFF- ChIP fragment sizes within a thousand bp of these TSSs from the Pol II and H3K4me3 Exp4 datasets revealed several distinctive groups of fragment sizes (Fig. 3A). A majority of H3K4me3 fragments in this window center on 158 bp in length and a smaller population were about 294 bp in length, corresponding to mono- and di- nucleosomes. In the Pol II dataset, two abundant fragment sizes at approximately 50 bp and 180 bp were the most prevalent, with a smaller population of fragment sizes around 75 bp. Fragment ranges corresponding to each of the three most common Pol II populations in the 2000 bp windows were then chosen and aligned relative to TSSs (Fig. 3B). The total amount of reads in each of these ranges was normalized to emphasize the protected footprints of the less abundant \(\sim 75\) bp fragments. Both the \(\sim 50\) bp and \(\sim 180\) bp fragments align slightly downstream of the TSS in the pause region while the \(\sim 75\) bp fragments span the TSS, as would be expected for PICs. The positioning of the \(\sim 50\) bp and the \(\sim 180\) bp fragments are consistent with free paused Pol II and Pol II abutted to the \(+1\) nucleosome. The same analysis was performed utilizing Exp3 data, which gave similar results (Supplementary Fig. 3A and 3B)  
+
+<|ref|>text<|/ref|><|det|>[[114, 728, 880, 904]]<|/det|>
+Since metaplots limit how many fragment lengths can be shown together in a meaningful and accurate way, we created a method that more holistically depicts the distribution of all fragments around TSSs. The output of this visualization method simultaneously captures the size, amount, and position of each fragment across all promoters in an easily viewable fragMap. For each fragment length, the coverage at each position \(+ / - 1000\) bp around all TSSs was calculated and averaged. These averages for each fragment length at each position were then stacked with the shortest fragments on top and longest on the bottom. Fragment lengths included in this view span from 18 to 400 bp. More focused fragMaps were also created \(+ / - 100\) bp around the MaxTSS using 18 to 120 bp fragments. For most fragMaps, black values (overall darkness of the image) are set by the maximum average read value in the window. Because the black value in each fragMap is influenced by the recovery of the IP, absolute amounts of visible
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 876, 170]]<|/det|>
+complexes should only be compared within the same fragMap. Host HFF fragMaps were generated utilizing the 12,229 TSSs found with truQuant and TSSs on the HCMV genome were found utilizing transcription start regions (TSRs) identified using TSR- finder on a PRO- Cap dataset generated from HCMV infected HFFs that resulted in 1,461 non- overlapping 200 bp regions centered on a MaxTSS \(^{25}\) .  
+
+<|ref|>text<|/ref|><|det|>[[114, 170, 880, 473]]<|/det|>
+FragMaps for the Pol II and H3K4me3 Exp4 datasets generated a snapshot of transcription complexes and chromatin features on the host and HCMV genomes. Host H3K4me3 fragMaps show very well positioned nucleosomes relative to the TSS and a clear NDR (Fig. 3C). This patterning of nucleosomes closely resembles fragMaps generated using the DFF- Seq HeLa truQuant dataset, except that the signal becomes fainter as the distance from the TSS increases (Supplementary Fig. 3C). Pol II fragMaps of the host genome show free paused Pol II (\~50 bp fragments downstream from the TSS), Pol II positioned over the TSS (\~75 bp fragments), Pol II abutted to the +1 nucleosome (\~180 bp fragments downstream of the TSS), and Pol II associated with the first two nucleosomes (\~320 bp fragments downstream of the TSS) (Fig. 3C). The size of the fragments bearing Pol II associated with the first two nucleosomes supports the finding that nucleosomes around promoters are more closely spaced than in bulk chromatin (Supplementary Fig. 1A). A complex of \~100 bp, located downstream of the TSS, is also visible in the Pol II fragMaps. Possible origins for this unanticipated complex will be discussed later. Divergent transcription occurs at variable distances upstream of the sense TSS on the host genome, so the upstream region displays a similar, but less well- defined pattern compared to transcription in the sense direction. Pol II and H3K4me3 fragMaps from Exp 3 datasets displayed all the same features and were virtually indistinguishable from those using Exp4 datasets (Supplementary Fig. 3D) demonstrating that the method is robust and reproducible.  
+
+<|ref|>text<|/ref|><|det|>[[114, 473, 877, 650]]<|/det|>
+Major differences from the host patterns were found when analyzing the HCMV genome. The H3K4me3 signal was not positioned around the majority of promoters (Fig. 3D). HCMV Pol II fragMaps confirm the pervasiveness of Pol II transcription with Pol II visible across the fragMap (Fig. 3D). The TSRs utilized to generate HCMV fragMaps are 200 bp so individual TSRs may be represented multiple times in the 2000 bp window. This causes a light background of particularly sized fragments across the entire visualized region. The fragments near the TSS show free paused Pol II, Pol II positioned over the TSS in \~75 bp fragments, and an additional complex of \~50 bp positioned just upstream of the TSS that is not present on host fragMaps. This \~50 bp protection over the TSS is especially evident in Exp3 viral fragMaps (Supplementary Fig. 3E). Nucleosome- abutted Pol II, which is the most abundant feature seen on the host genome, is present but at a vastly lower level than free Pol II on the HCMV genome.  
+
+<|ref|>text<|/ref|><|det|>[[113, 650, 877, 890]]<|/det|>
+The apparent differences in Pol II association with the +1 nucleosome on the host and HCMV genomes prompted us to quantify how frequently Pol II encounters an immediately downstream nucleosome in these two cases. The number of reads in features corresponding to the free Pol II, abutted Pol II, and +1 nucleosome based on genomic position of fragment centers and fragment size were quantified from Pol II and H3K4me3 datasets (Supplemental Excel file). Analysis of free or abutted Pol II demonstrated that 36% of the total engaged Pol II signal arises from free Pol II on the host genome. In contrast to the host, 74% of the engaged Pol II was free on the HCMV genome. On a promoter by promoter basis, 85% have more free than abutted Pol II signal on the HCMV genome, whereas only 18% do on host (Fig. 3E). Examination of the fragment count of +1 nucleosome sized fragments from the H3K4me3 dataset shows that the host promoters with the absolute highest percentage of free Pol II have little +1 nucleosome, but most promoters have a similar amount of +1 nucleosome (Fig. 3E). Although the H3K4me3 signal over the viral genome has a similar value in terms of reads, there are about a hundred times more viral genomes and thus the H3K4me3 modified nucleosome occupancy over the viral genomes is about 1% of that in the host.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 780, 107]]<|/det|>
+## Detection and characterization of TBP-driven PICs and UL87-driven viral PICs  
+
+<|ref|>text<|/ref|><|det|>[[113, 105, 880, 601]]<|/det|>
+Detection and characterization of TBP- driven PICs and UL87- driven viral PICsThe existence of Pol II- containing complexes that extended upstream of the MaxTSS strongly suggested that these features correspond to PICs, so Pol II DFF- ChIP was repeated under conditions that would alter PIC prevalence. Treatment of cells for an hour with 1 \(\mu \mathrm{M}\) triptolide, an inhibitor of transcription initiation, should increase the PIC relative to paused Pol II and that is exactly what was found for the \(\sim 75\) bp feature on the host and viral genomes (Fig. 4A). Triptolide treatment also more clearly reveals the \(\sim 50\) bp feature directly upstream of the TSS, which is unique to the viral genome (Fig. 4A). As an additional verification that \(\sim 75\) bp protections arose from uninitiated Pol II, Exp1 and Exp2 DFF- ChIP were reanalyzed since they utilized different wash conditions in a GFP- Pol II tagged MRC5 cell line<sup>30</sup> and GFP nanobody beads. The immunoprecipitations in these experiments were carried out with 150 mM or 1 M salt wash conditions. The \(\sim 75\) bp feature was preferentially lost during the high salt conditions as expected<sup>33</sup> for uninitiated Pol II (Supplementary Fig. 4). DFF- ChIP was then performed targeting the TATA- binding protein (TBP), a critical component of the host PIC. FragMaps generated from the TBP dataset show primarily the \(\sim 75\) bp feature (Fig. 3A), indicating that these complexes are indeed TBP- containing PICs. Additionally, TBP fragMaps reveal protections that share the relatively sharp upstream edge with the full TBP PIC but are much smaller in size, about 40 bp. These likely result from TBP- containing complexes prior to incorporation of Pol II and assembly of the complete PIC. DFF- ChIP was also performed utilizing a Pol II antibody that targets the serine 5 phosphorylation on the CTD of Pol II (Ser5P). Ser5P modification is carried out by CDK7, a component of the initiation factor TFIIH<sup>34</sup>. The resulting fragMaps demonstrate that Ser5P antibodies recognize modified Pol II in the TBP PIC on both the host and viral genome. However, the \(\sim 50\) bp PICs on the viral genome were not detected (Fig. 4B) suggesting that the Pol II in those complexes is not phosphorylated. There is a difference in the relative amounts of PIC and free paused Pol II detected with the F12 antibody and the Ser5P antibody. The epitope on the Pol II large subunit recognized by F12 is located deep within the PIC<sup>13</sup> and potentially masked by TFIIE and TFIIH, while the Ser5P epitope would be difficult to mask because of its repetitive occurrence and disordered structure. Therefore, we favor the idea that the Ser5P results are more representative of Pol II levels in all TBP- driven promoter- proximal complexes. Supporting this idea, when MRC5 cells containing a GFP- tagged Pol II were immunoprecipitated with GFP nanobodies, the PIC was a major species (Supplementary Fig. 4B).  
+
+<|ref|>text<|/ref|><|det|>[[115, 600, 876, 809]]<|/det|>
+Large TBP fragMaps provided further insight into the chromatin differences between the host and HCMV genomes (Fig. 4C). Some TBP PICs on the host genome are associated with the \(+1\) nucleosome, resulting in fragments of \(\sim 250\) bp. These interactions are mostly absent on the HCMV genome presumably because the viral genome is less chromatized. Significant amounts of the host PICs driving divergent transcription also connect with the adjacent nucleosome (the - 1 nucleosome), analogous to the PIC- nucleosome complexes in the sense direction. As with the free- standing PIC, the PIC/ \(+1\) nucleosome feature on the host genome was salt sensitive (Supplementary Fig. 4C). Previous analysis of downstream sequences relative to TSSs revealed periodic elements that likely serve to position the \(+1\) nucleosome<sup>6</sup>, suggesting a connection between the TSS and nucleosome positioning. Considering that both genic- oriented and divergent PICs are normally associated with well- positioned immediately- adjacent nucleosomes, it is further likely that the PIC, TSS, and \(+1\) nucleosome connection is important for specifying and/or facilitating transcription initiation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 808, 879, 905]]<|/det|>
+Unlike the early HCMV transcriptional program that relies entirely on the host general Pol II transcription machinery, beta- and gamma herpesviruses have unique late promoters containing a TATT upstream element that recruits virally- encoded late transcription factors<sup>35,36</sup>. Since only one of the two distinct Pol II- containing PICs on the viral genome corresponds to the host complex driven by TBP, we posited that the \(\sim 50\) bp virus- specific PIC is based on UL87, one of the viral late transcription factors which associates with the TATT element<sup>17</sup>. DFF- ChIP was
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 870, 154]]<|/det|>
+performed with a Towne strain of HCMV expressing a HA- tagged UL87 (Exp4). FragMaps demonstrated that the \(\sim 50\) bp viral PICs were almost exclusively recovered with the HA antibody (Fig. 4D). Interestingly, unlike the 75 bp TBP- PICs the 50 bp UL87- PICs did not cover the TSS.  
+
+<|ref|>text<|/ref|><|det|>[[115, 154, 877, 346]]<|/det|>
+In an attempt to correlate upstream sequence elements with the extent of PIC assembly, UL87 PIC and TBP PIC features at each gene on the host or on each TSR of the HCMV genome were quantified (Supplementary Excel file). The size and position of these features were selected such that no overlap was allowed between them. Each region was then rank ordered by the amount of the UL87 PIC feature or the TBP PIC feature. Logos were generated by MEME analysis<sup>37</sup> of the - 38 to - 19 region for the PICs in the top decile of occupancy on the host and viral genomes (Fig. 4E). Such analysis recapitulated the expected TA- rich binding motifs with TBP preferring sequences containing TATA and UL87 preferring sequences containing TATT. UL87 evidently has a stricter requirement for TATT containing sequences with 128/146 matches to its HCMV Logo, while TBP had a lower percentage of matches to its host (623/1230) and viral (88/146) Logos. It is important to note that not all host promoters with high levels of PIC occupancy are AT- rich in the - 38 to - 19 region<sup>28</sup>.  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 360, 857, 377]]<|/det|>
+## TBP and UL87 are functionally distinct but not mutually exclusive on HCMV promoters  
+
+<|ref|>text<|/ref|><|det|>[[115, 377, 876, 666]]<|/det|>
+To characterize the transcription of the HCMV genome throughout lytic infection and its relation to UL87 and TBP usage, PRO- Seq was carried out at multiple different time points of infection and compared to 48 hpi DFF- ChIP data. Time points include two early times (4 and 12 hpi), 24 hpi which is the beginning of replication, and two late times (48 and 72 hpi) in which high levels of viral replication have occurred and during which UL87 function is critical. Representative regions of the HCMV genome containing early and late genes are depicted in a 1,400 bp region and an 800 bp region with corresponding HCMV genomic fragMaps below showing fragments from the UL87, TBP, Pol II, and Ser5P datasets (Fig. 5A). The early promoter UL29 has only TBP, Pol II, and Ser5P fragments associated with it (Fig. 5A, blue TBP arrow), while the late intragenic promoter in UL49 only has UL87 and Pol II fragments (Fig. 5A, red UL87 arrow). TBP and UL87 driven promoters can occur very close to each other (Fig. 5A, red and blue Both arrows) or can even drive transcription from the exact same TSS (Fig. 5A, purple Both arrow). Overlapping PICs such as these would be expected to compete for occupancy on the HCMV genome. Quantification of 5' end reads found in an 11 bp window around the MaxTSS of these five promoters shows that TBP- driven TSSs are more active early in comparison to UL87 PICs that are active late (Supplementary Fig. 5A). Because some promoters are driven by both TBP and UL87 the distinction between early versus late gene transcription is more complex than a simple separation of TBP and UL87 driven promoters.  
+
+<|ref|>text<|/ref|><|det|>[[115, 665, 880, 904]]<|/det|>
+To quantify how TBP and UL87 usage is related to early or late expression, transcription from each promoter was quantified by counting 5' ends in an 11 bp window centered on the MaxTSS of each TSR at each time point. The fractional usage for each promoter at each time point was displayed in a heatmap after normalizing to the total number HCMV reads in each time point. 795 of the 1,461 promoters that each had a total of 100 reads across the time course were then sorted based on PFA sensitivity<sup>28</sup>, on a slope calculated from the 5 time points for each promoter and by UL87 dependency<sup>36</sup> (Fig. 5B). Each promoter was colorized individually based on the relative usage across the time course. The three sorts gave similar patterns of genes with early (red to green) and late (green to red) transcription kinetics. Each TSR's reliance on DNA replication (PFA sensitivity) and UL87 dependency was then plotted against slope to classify how TSRs with primarily TBP (blue) and UL87 (red) PICs behaved in relation to time of expression (Supplementary Fig. 5B). The relative preference for TBP or UL87 PICs was calculated as the ratio of TBP/UL87 PICs after normalization of total counts in each feature. These plots show that UL87 primarily functions on genes with late transcription kinetics and that many of these promoters are also stimulated by DNA replication. However, and significantly,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 881, 186]]<|/det|>
+many promoters with late kinetics are primarily TBP driven showing that TBP is also essential in transcription of some late genes. The UL22A promoter has a TBP/UL87 ratio of 0.93 and displays late kinetics. There are two main TSSs at all time points that are separated by 3 bp with a shift in the relative usage at early and late time points that is reverted when viral replication is blocked by PFA treatment (Supplementary Fig. 5C). This promoter exemplifies a clear competition for formation of TBP and UL87 PICs.  
+
+<|ref|>text<|/ref|><|det|>[[114, 186, 881, 490]]<|/det|>
+Nearly all promoters show some level of UL87 and TBP usage and to determine to what extent promoters are shared, the TBP/UL87 PIC ratio was calculated for all 1,461 TSRs and plotted (Fig. 5C). The results indicate that the relative levels of UL87 and TBP PICs vary by many orders of magnitude across the 1461 TSRs. Therefore, in order to make comparisons between TBP and UL87 PICs only the top and bottom \(5\%\) of TSRs sorted by PIC ratio were utilized to prevent contaminating signal from the other class of PIC. To compare initiation efficiency, fragMaps were generated utilizing the TBP and Ser5P datasets for the top TBP TSRs and UL87 and Pol II datasets for the top UL87 TSRs (Fig. 5D). As noted above, the F12 antibody epitope in the large subunit of Pol II is significantly masked in TBP PICs \(^{13}\) , so the Ser5P Pol II signal was used to visualize Pol II in TBP driven TSRs. The strong correlation of feature counts from the Ser5P dataset in comparison to TBP and Pol II datasets also advocates for this approach (Supplementary Fig. 5D). These fragMaps reveal that UL87 and TBP driven TSRs yield similar amounts of engaged Pol II in relation to PIC amounts. This result was initially surprising given the prominence of UL87 PIC peaks in comparison to detected paused Pol II downstream of those PICs, which seemed to indicate poor UL87 initiation. However, it is possible that a significant fraction of \(\sim 50\) bp UL87 features may not have Pol II associated with them and this is supported by the relative lack of protection downstream of the TSS. Critically, on both promoter classes the PIC is as prominent on chromatin as paused Pol II indicating that the PIC is far more prevalent on the genome than expected.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 504, 617, 521]]<|/det|>
+## Nucleosomes on the HCMV genome are irregularly spaced  
+
+<|ref|>text<|/ref|><|det|>[[114, 521, 880, 825]]<|/det|>
+To directly assess if sites of initiation correlated with H3K4me3 modified nucleosomes, genome browser tracks and genomic fragMaps for TBP, UL87 and H3K4me3 were compared. There is little evidence for a correlation of sites of transcription initiation and H3K4me3 modification regardless of PIC type (Fig. 6A). This comparison does allow for a rough estimation of the number of nucleosomes that span the 21,000 bp region shown, since approximately 70- 80 nucleosomes are distinguishable. This indicates that on average a nucleosome is positioned every 250- 300 bp on the HCMV genome. This average spacing is consistent across the entire HCMV genome (Supplementary Fig. 6). It is unclear what drives nucleosome positioning on HCMV DNA but given the scarcity of transcription- coupled nucleosomes it is unlikely to be related to transcription. Metaplot analysis of all 1,461 viral TSRs shows that nucleosomes around promoters are spaced approximately 250- 300 bp in contrast a much more compact spacing around promoters on the host genome of about 150 bp (Fig. 6B). A PIC containing TBP is capable of associating with the +1 nucleosome on the host (Fig. 4C) and this prompted us to investigate if viral TSRs with strong TBP signal have a better positioned downstream nucleosome. Selecting the top \(10\%\) of TSRs with the highest level of TBP PICs showed that indeed these TSRs have a slightly stronger +1 nucleosome signal (Fig. 6C). The same analysis performed with the top UL87 TSRs failed to reveal any clear nucleosome patterning (Fig. 6C). These findings suggest that while TBP PICs may aid in the positioning of a +1 nucleosome, this is not a driving force for nucleosome positioning on the HCMV genome.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 840, 830, 873]]<|/det|>
+## Increasing the extent of DFF digestion reveals a general robustness of features and captures Pol II association with sub-nucleosomal fragments  
+
+<|ref|>text<|/ref|><|det|>[[115, 872, 830, 905]]<|/det|>
+To confirm reproducibility and determine the effects of more extensive DFF digestion, five datasets were generated in duplicate using HFFs infected with the Towne strain of HCMV
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 870, 219]]<|/det|>
+(Exp5). The datasets correlated highly within Exp5 and between Exp5 and Exp4 (Supplementary Fig. 7A). Total library fragment length distributions for Pol II, TBP, and H3K4me3 were compared between the initial and more extensively digested experiments (Fig. 7A and 7B). It was apparent that higher levels of digestion resulted in a division of some complexes into subgroups typically separated by a 10 bp periodicity. Over- digestion did not change the ratio of the free to abutted Pol II (Fig. 7C and 7D), but the group of Pol II- containing fragments of \(\sim 90\) to 110 bp became more apparent after increased digestion. Connections of the TBP PIC to downstream nucleosomes were mostly lost with over- digestion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 217, 874, 362]]<|/det|>
+The H3K4me3 fragment distributions in Fig 7 clearly show that DFF can cleave within nucleosomes at higher digestion levels. Two classes of products shorter than the expected protection from the full nucleosome were observed: fragments which apparently resulted from removal of 10 or 20 bp from the nucleosome ends and another population of centered fragments less than 80 bp consistent with protection by the H3/H4 tetramer. The lack of fragments between the two populations suggests that complete loss of the flanking H2A/H2B dimers on the edges of the nucleosome occurs before the central H3/H4 is invaded by DFF. All sub- nucleosomal fragments were centered including those less than 80 bp, further demonstrating that they arose from the H3/H4 tetramer (Supplementary Fig. 7C).  
+
+<|ref|>text<|/ref|><|det|>[[115, 361, 880, 570]]<|/det|>
+To better understand the interaction of Pol II with the \(+1\) nucleosome we specifically considered well positioned nucleosomes downstream of the most focused promoters<sup>6</sup> (Supplementary Fig. 7B). At such promoters, the upstream edge of the \(+1\) nucleosome is positioned on average at \(+47\) , just downstream of the average paused Pol II position at \(+41\) . Pol II protects approximately 20 bp of DNA downstream of the active site<sup>38</sup>, suggesting that the leading edge of the abutted Pol II intrudes \(\sim 1.5\) DNA turns into the nucleosome. Pol II at this location would disrupt the H3 contact at nucleosome entry but could leave the H2A/H2B contacts with DNA at least partially intact. If invasion by Pol II distorts the nucleosome, this could in turn reveal a cutting site for DFF downstream of the H2A/H2B dimer. DFF cleavage in such a complex would give rise to the 87- 107 bp fragment set that we observe in Pol II IPs, more prominently upon over- digestion (Fig. 7, and Supplementary Fig. 7). Other possible origins for this \(\sim 100\) bp Pol II fragment set can be envisioned, which will be addressed in the Discussion.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 586, 213, 600]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[115, 601, 854, 730]]<|/det|>
+Our results provide a deeper understanding of transcription complexes and their interactions with chromatin on both the host and HCMV genomes. Critically, analysis of the length and position of fragments recovered by DFF- ChIP using fragMaps not only provided detailed footprints of PICs and paused Pol II but also directly revealed the interactions of those complexes with the \(+1\) nucleosome. Using HCMV infected HFFs, we demonstrated that Pol II encounters a vastly different chromatin environment on the viral genome than it does on the host genome. Furthermore, our direct visualization of both TBP and UL87- driven PICs sheds light on sequence preferences, dimensions, and shared usage at the two promoter classes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 880, 905]]<|/det|>
+Our results provide a new perspective into the transcriptional processes during lytic infection in relation to the chromatin structure of the HCMV genome. Earlier work suggested that during productive infection, viral genomes form irregular nucleosome arrays<sup>21,39,40</sup>. However, a more recent genome wide analysis utilizing MNase suggested that the HCMV genome is largely packaged into nucleosomes throughout the viral cycle<sup>24</sup>. Our data confirm that nucleosomes are deposited across the viral genome at low level of occupancy, but we also show in multiple ways that paused Pol II rarely encounters a nucleosome on HCMV DNA. The TBP, Pol II, Ser5P, and UL87 DFF- ChIP signals corresponding to initiating Pol II or PICs often reside in the middle of apparently nucleosome rich regions on HCMV DNA despite the known ability of a nucleosome to block initiation<sup>31,32</sup>. Therefore, those regions of HCMV genomes with nucleosomes over TSSs cannot be transcriptionally active. Transcription complexes themselves report on local
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 875, 202]]<|/det|>
+nucleosomes by conferring a larger protection footprint when they are in close proximity to a nucleosome. HCMV chromatin very rarely confers these protection patterns. Although it is likely that the large majority of HCMV promoters do not feature modified nucleosomes immediately downstream, it is not possible to prove that this is the case at every individual viral promoter. Post- translationally modified nucleosomes may be positioned on individual loci at some stages of the viral life cycle<sup>22,23</sup>. Overall, we conclude that during lytic infection the HCMV genome is transcribed in a predominantly nonchromatinized state.  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 876, 392]]<|/det|>
+The unique ability to recover PICs with DFF- ChIP without prior crosslinking was likely allowed by the use of EDTA during very rapid nuclei isolation<sup>41</sup> prior to digestion and immunoprecipitation steps. EDTA was included to halt transcription but it may also increase PIC retention by eliminating the destabilization caused by ATP<sup>33</sup> in abortive initiation and XPB function. Substantial recovery of PICs in DFF- ChIP has allowed a better understanding of their properties in the nucleus including documenting that TBP- and UL87- driven PICs are major features on the host (TBP) and viral genomes (TBP and UL87), equaling or surpassing paused Pol II in amount. The formation of the TBP- containing PIC requires Pol II in a hypophosphorylated state, but after PIC assembly in vitro, phosphorylation of the CTD may occur even prior to formation of the first phosphodiester bond<sup>42,43</sup>. Our data demonstrate that this actually occurs in cells, showing that Ser5P is present on TBP PICs (Fig. 3B). Additionally, we show that a substantial fraction of PICs directly interacts with the +1 nucleosome.  
+
+<|ref|>text<|/ref|><|det|>[[115, 392, 879, 617]]<|/det|>
+DFF- ChIP also uncovered differences between TBP and UL87 PICs and their interplay on the HCMV genome. TBP PIC footprints correspond to the known in vitro footprint of TFIID, from roughly 40 bp upstream of the TSS to 35 bp downstream<sup>6</sup>. Published human TFIID and PIC structures indicate that TFIID and XPB both contact DNA well downstream of the TSS<sup>2,44</sup>. In contrast, UL87 PIC footprints are only located upstream of the TSS, suggesting that UL87 PICs lack the subunits of TFIID that contact downstream DNA and crucially, at least the XPB subunit of TFIID. Since UL87 PICs also lack Ser5P modification, it seems likely that the UL87 PICs we detect lack TFIID. However, initiation at UL87 PICs is sensitive to inhibition of XPB by tripotide. We therefore propose that while TFIID is required for UL87 initiation, it is not a stable component of the UL87 PIC that we detect. A recent study regarding ORF24, a UL87 homolog in Kaposi's sarcoma associated virus<sup>20</sup>, showed association with the Pol II CTD only in the hypophosphorylated state, which is in agreement with our data demonstrating that the Pol II in UL87 PIC is predominantly unphosphorylated<sup>45</sup>. Regardless of these functional differences, both PICs function on a large shared subset of HCMV promoters.  
+
+<|ref|>text<|/ref|><|det|>[[115, 616, 879, 905]]<|/det|>
+We expected from our earlier study that paused Pol II complexes would be located upstream of the +1 nucleosome or abutted to that nucleosome<sup>6</sup>. Those complexes are evident from DFF- ChIP (Fig. 2C) but high levels of DFF digestion in particular revealed unanticipated Pol II- containing fragments of 87- 107 bp that extend downstream into the region that is expected to be protected by the proximal H2A/H2B dimer of the +1 nucleosome (Fig. S4D). We suggested above that polymerase invasion of the nucleosome could reveal a site for DFF cleavage downstream of the dimer. However, other recent work suggests a plausible alternative explanation. It was reported that the Chd1 chromatin remodeller associates with +1 nucleosomes, specifically on the promoter- proximal face<sup>46</sup>. Subsequent structural studies showed that in a Chd1- nucleosome complex, a Chd1 domain displaces DNA from the nucleosome surface normally occupied by H2A/H2B<sup>47</sup>. Thus, the 87- 107 bp Pol II- containing fragments we detected could have arisen from Pol II paused at the entry of a +1 nucleosome already occupied by Chd1. Presumably the interface of Chd1 and the nucleosome is more easily accessible at high levels of DFF digestion. In the earlier studies it was speculated that displacement of the Chd1 domain at nucleosome entry by the advancing polymerase would activate the remainder of Chd1 to drive displacement of the nucleosome and thus facilitate traversal by Pol II. Thus, this model predicts that once Pol II has displaced the proximal Chd1 domain, full traversal of the +1 nucleosome should be efficient<sup>47</sup>. This is consistent with the fact
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 878, 186]]<|/det|>
+that we did not detect Pol II- containing complexes by DFF- ChIP corresponding to pausing just upstream of H3/H4 tetramer of the \(+1\) nucleosome, as might have been predicted from earlier in vitro studies \(^{48}\) . Other results based on micrococcal nuclease digestion patterns indicated that in Drosophila \(+1\) nucleosomes frequently lack the proximal H2A/H2B dimer \(^{49}\) . However, as just noted, we do not have evidence from our experiments for a Pol II barrier at the H3/H4 tetramer of the \(+1\) nucleosome.  
+
+<|ref|>text<|/ref|><|det|>[[115, 186, 880, 315]]<|/det|>
+Future usage of DFF including DFF- ChIP holds great potential in uncovering intricacies of chromatin architecture. Targeted immunoprecipitation of various modified histones or histone replacements are but a few of the possibilities for these investigations. For direct inquiries into transcription, further DFF- ChIP experiments that target the general transcription factors involved in initiation, pausing, and productive elongation are of great interest and will aid in uncovering the ways that transcription complexes interact with nucleosomes. In addition, we expect DFF- ChIP to be applicable to the targeted investigation of many more specific chromatin associated factors.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 330, 191, 345]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 347, 180, 360]]<|/det|>
+## Viruses  
+
+<|ref|>text<|/ref|><|det|>[[115, 361, 872, 394]]<|/det|>
+HCMV TB40/E BAC4 and Towne UL87HA were used in this study. The construction and use of the Towne UL87- HA recombinant virus was described previously \(^{28}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 410, 336, 425]]<|/det|>
+## Infections and treatments  
+
+<|ref|>text<|/ref|><|det|>[[115, 425, 880, 666]]<|/det|>
+Primary human foreskin fibroblasts were maintained in Minimum Essential Medium (Gibco, 11095080) supplemented with \(5\%\) fetal bovine serum (Gibco, 26140079) and \(1\%\) penicillin- streptomycin (Gibco, 15140122). Confluent (contact inhibited) HFF monolayers in T- 150 cm \(^2\) flasks were used for these studies. The culture medium was refreshed \(24 \text{h}\) prior to infection. On the day of infection, all but \(12 \text{mL}\) of the conditioned medium was removed and set aside. The remaining \(12 \text{mL}\) of medium was inoculated with HCMV at a multiplicity of infection of 3 infectious units per cell (MOI of 3). Viral adsorption was carried out for \(90 \text{min}\) . The medium containing viral inoculum was then replaced with \(12 \text{mL}\) of the conditioned medium. For experiments involving treatment with flavopiridol (Flavo; final concentration, \(1 \mu \text{M}\) ), or triptolide (final concentration, \(1 \mu \text{M}\) ), \(6 \text{mL}\) of conditioned medium was temporarily removed \(1 \text{h}\) before cells were harvested. This medium was treated with \(6 \mu \text{L}\) of \(2 \text{mM}\) Flavo (NIH AIDS Reagent Program 9925z) in DMSO, \(6 \mu \text{L}\) of \(2 \text{mM}\) triptolide (Sigma, T3652) in DMSO or \(6 \mu \text{L}\) of DMSO alone. Once inoculated with drug, the \(6 \text{mL}\) of medium was immediately returned to the flask for a final \(12 \text{mL}\) of culture medium. At \(48 \text{h}\) post- infection, cells were lysed and cell nuclei were isolated and held in frozen storage until use, as described previously \(^{28}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 681, 238, 697]]<|/det|>
+## DFF-ChIP Seq  
+
+<|ref|>text<|/ref|><|det|>[[115, 697, 876, 906]]<|/det|>
+DFF was purified as previously described \(^{6}\) . For Exp4 approximately 6 million nuclei from human foreskin fibroblasts infected for 48 hours with human cytomegalovirus strains TB40/E or a mutant Towne virus expressing UL87- HA were digested with approximately \(15 \mu \text{g}\) of DFF in \(20 \text{mM}\) HEPES (pH 7.6), \(5 \text{mM}\) magnesium acetate, \(100 \text{mM}\) K(Ac), \(5 \text{mM}\) DTT, for 1 hour at \(37^{\circ}\text{C}\) . Digestions were carried out in batch where possible for all nuclei of the same treatments conserving the ratio of nuclei to DFF. Digestion was halted with the addition EDTA to a concentration four times that of magnesium and nuclei were subsequently split for individual IPs. Nuclei were lightly sonicated for \(20 \text{s}\) at \(40\%\) amplitude using Qsonica Q800R3 Sonicator and the supernatant was collected and brought up to \(1 \text{mL}\) with solution containing \(10 \text{mM}\) Tris (pH 7.5), \(100 \text{mM}\) NaCl, \(1 \text{mM}\) EDTA, and TritonX- 100 such that the final concentration was \(0.1\%\) . The supernatants were precleared for 20 minutes over Protein A (Sigma P9424) or G Sepharose (Sigma P3296) beads. Afterwards, the supernatants were removed from beads and immunoprecipitated with approximately \(2.5 \mu \text{g}\) of antibodies for Pol II (Santa Cruz, sc- 55492),
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 878, 540]]<|/det|>
+TBP (Abcam, ab51841), Ser5P (Millipore, 3E8), HA- tag (Cell Signaling Technology, C29F4), and H3K4me3 (Abcam, ab8580) overnight at \(4^{\circ}\mathrm{C}\) with rotation. Next, samples were incubated with Protein A beads (Protein G beads for Ser5P IPs) for 2 hours at \(4^{\circ}\mathrm{C}\) with rotation. The beads were than washed five times with \(10\mathrm{mM}\) Tris (pH 7.5), \(150\mathrm{mM}\) NaCl, \(1\mathrm{mM}\) EDTA, and \(0.1\%\) TritonX for five minutes per wash. Bound material was than eluted twice with \(50\mu \mathrm{L}\) of \(10\mathrm{mM}\) Tris (pH 7.5), \(1\%\) SDS, and \(1\mathrm{mM}\) EDTA incubated at \(65^{\circ}\mathrm{C}\) for 5 minutes. Eluted material was subsequently treated with \(20\mu \mathrm{g}\) RNase A for 30 minutes at \(37^{\circ}\mathrm{C}\) and then \(40\mu \mathrm{g}\) of Proteinase K for 2 hours at \(65^{\circ}\mathrm{C}\) . The same protocol was used for the all DFF- ChIP experiments with slight modifications. For MRC- 5 expressing GFP- Pol II and HeLa cell lines in Exp1 and Exp2, 12 million nuclei were digested with \(30\mu \mathrm{g}\) . After splitting and preclearing on Protein A sepharose beads, GFP- Pol II samples were incubated with Chromotek GFP- Trap beads for 4 hours, washed five times with either the same buffer as above or one containing \(1\mathrm{M}\) NaCl for \(\sim 1\) min each, and eluted as described above. Samples from HeLa cells were immunoprecipitated, eluted, and treated the same as Exp4 except with \(\sim 1\) min washes. Exp3 was performed exactly as Exp4 was done, but again with \(\sim 1\) min washes. For Exp5 utilizing HFFs infected with Towne, 3 million HFF cells infected with Towne UL87- HA were incubated with approximately \(25\mu \mathrm{g}\) DFF. All libraries were cleaned up utilizing Invitrogen PureLink PCR purification kit. All libraries were prepared using the KAPA Hyper Prep Kit according to their protocol (Roche 7962312001). For libraries in experiments 1 and 2 we utilized Illumina TruSeq adapters. For all subsequent libraries, we created custom adapters utilizing the TruSeq sequences that contained an 8 bp UMI immediately downstream of the index. Test amplifications were performed on each library to determine the number of cycles necessary to obtain enough library material for sequencing. Full- scale amplification was then performed, fragments were quantified using an Agilent Bioanalyzer 2100, and then libraries were pooled and size selected from 135- 1000 bp using a BluePippin. Prior to submission, proper size selection was confirmed with reanalysis using the Agilent Bioanalyzer. Libraries were sequenced on either an Illumina HiSeq 4000 or a NovaSeq 6000 by the Iowa Institute of Human Genetics.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 553, 315, 569]]<|/det|>
+## PRO-Seq and PRO-Cap  
+
+<|ref|>text<|/ref|><|det|>[[113, 567, 878, 906]]<|/det|>
+PRO- Seq and PRO- Cap libraries were prepared as previously described \(^{28}\) , with a few modifications. Frozen nuclei isolated from HFF infected with HCMV TB40/E (MOI 3) for 48 or 72 h were thawed on ice, gently pelleted, and resuspended in \(40\mu \mathrm{L}\) of a nuclear run- on buffer (20 mM HEPES (pH 7.6), \(5\mathrm{mM}\) magnesium chloride, \(100\mathrm{mM}\) potassium chloride, \(5\mathrm{mM}\) dithiothreitol (DTT), and \(0.6\mathrm{U} / \mu \mathrm{L}\) SUPERase- In (Invitrogen AM2696)). Approximately 100,000 moth SF21 nuclei were spiked into HFF nuclei prior to pelleting. Nuclei were warmed to \(37^{\circ}\mathrm{C}\) and then combined with \(20\mu \mathrm{L}\) of a 3X nuclear run- on mix (20 mM HEPES (pH 7.6), \(5\mathrm{mM}\) magnesium chloride, \(100\mathrm{mM}\) potassium chloride, \(5\mathrm{mM}\) DTT, \(1.5\%\) Sarkosyl, and \(60\mathrm{uM}\) biotinylated ATP, UTP, GTP, and CTP (Perkin Elmer NEL544, NEL543, NEL545, and NEL542, respectively)). Samples were pulse- vortexed after the addition of nuclear run- on mix and incubated at \(37^{\circ}\mathrm{C}\) for 10 minutes. Reactions were quenched with \(40\mu \mathrm{L}\) of \(50\mathrm{mM}\) ethylenediaminetetraacetic acid (EDTA) and \(300\mu \mathrm{L}\) of Trizol LS (Ambion 10296028), and total RNA was extracted according to manufacturer protocol. All subsequent steps in PRO- Seq library preparation were carried out as previously described \(^{28}\) . PRO- Cap library preparation followed a similar procedure, except excluded the RNA hydrolysis step and included RNA polyphosphatase and terminator exonuclease treatments to ensure the integrity of nascent RNA 5' end capture. A detailed description of the PRO- Cap protocol has been published \(^{25}\) . Reverse- transcribed libraries were PCR amplified for 11 (PRO- Seq) or 17 (PRO- Cap) cycles, subjected to analysis on an Agilent Bioanalyzer 2100, pooled in equimolar ratios, size- selected on a Sage Science Blue Pippen (BDF2010 cassette, 135- 600 bp fragments selected), and sequenced on an Illumina HiSeq 4000 with 150 bp paired- end reads at the University of Iowa Genomics
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 805, 123]]<|/det|>
+Division. Raw data were trimmed, mapped, deduplicated, and processed into tracks as previously described<sup>28</sup>.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 138, 275, 154]]<|/det|>
+## DFF-ChIP analysis  
+
+<|ref|>text<|/ref|><|det|>[[113, 155, 879, 520]]<|/det|>
+Initial workup of data was performed using DNAfastqtoBigWig (https://github.com/P- TEFb/DNAfastqtoBigWig). This is a linux based, multi- thread capable, Next Generation Sequencing (NGS) data analysis program with a command line interface. It performs the standard NGS data processing steps including, downloading sequencing data from a given web server, trimming adapter sequences from the sequencing data, aligning the trimmed data to a given list of genomes, generating mapping statistics for the aligned data, deduplication of aligned fragments using their Unique Molecular Identifiers (UMI), and finally, generating tracks for each sample in bigwig format. The program automatically accomplishes the following steps. Raw sequences in Fastq format were downloaded from the Iowa Institute of Human Genetics (IIHG) Genome Sequencing web server using wget command. Next, adapter sequences were trimmed from these sequences using trim_galore 0.6 (https://github.com/FelixKrueger/TrimGalore/releases/tag/0.6.6) while retaining only paired end trimmed sequences of at least 18 bp in size. These sequences were aligned with UCSC hg38, and Genbank TB40/E and Towne HCMV assemblies using bowtie v1.2.2 to generate alignments in sam format. UMLs reads were used to deduplicate the aligned reads which were then converted into bed files. Unstranded tracks were generated for each sample by first converting bed into bedGraph format using bedtools v2.26, and subsequently into bigwig format using the Kent UCSC utility program called bedGraphToBigWig. All datasets are described in the supplementary Excel file. Raw and processed sequencing data can be obtained from GEO (GSEXXXX). All tracks can be viewed using the following links: HOST: https://genome.ucsc.edu/s/David%20Price/Host Towne: https://genome.ucsc.edu/s/David%20Price/Town E TB40/E: https://genome.ucsc.edu/s/David%20Price/TB40_E  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 537, 351, 553]]<|/det|>
+## Fragment distribution plots  
+
+<|ref|>text<|/ref|><|det|>[[115, 553, 880, 698]]<|/det|>
+Fragment size frequencies were calculated by counting the number of times a single fragment size or a range of fragment sizes are present in a given sample or overlap to a list of genomic intervals of a specific size. Bedtools v2.26 intersect program was used to generate overlap data between fragment and genomic intervals. Bash and awk scripts were used to generate counts for a single or a range of fragment sizes. Next, fragment sizes and their associated counts were sorted from short to long order. Additionally, fragment size counts were normalized to generate their relative amounts by dividing them with the total number of fragments present in a sample. MS Excel was used to plot relative and absolute counts of fragment sizes for each sample using scatter with straight lines option.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 713, 386, 729]]<|/det|>
+## Average base distribution plots  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 881, 887]]<|/det|>
+The number of times a nucleotide is present at each position within a specified genomic interval was calculated as follows. Bedtools v2.26 getfasta program was used to generate FASTA files while maintaining same strand orientation for \(\pm 100\) bp genomic intervals centered on fragment starts and ends. For these analyses all intervals centered on fragment starts were marked positive strand and intervals centered on fragment ends were marked negative strand. A custom linux based, multi- thread capable python script called ABD.py (https://github.com/P- TEFb/ABD) was run to generate absolute counts for each nucleotide across the given genomic intervals. Absolute nucleotide counts were converted to fractions by dividing them with the total number of sequences. MS Excel was used to plot base fractions across the genomic interval using scatter with straight lines option. Finally, the following color code was used for these plots: A was blue
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 866, 123]]<|/det|>
+(hex code #2222ff), T was orange (hex code #ff6600), G was gray (hex code #bbbbbb), and C was yellow (hex code #dddd00).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 138, 273, 153]]<|/det|>
+## MaxTSSs analysis  
+
+<|ref|>text<|/ref|><|det|>[[115, 154, 881, 298]]<|/det|>
+The truQuant program<sup>28</sup> with an updated list of black listed regions was run on the PRO- Cap data (GSE113394)<sup>25</sup> generated from uninfected HFF cells and DMSO bound NAS- Cap data (GSE139237)<sup>6</sup> generated from HeLa cells using published parameters to generate host specific MaxTSSs. 12,229 (PRO- Cap) and 12,201 (NasCap) MaxTSSs associated with known host genes were used for further analysis. TsrFinderM<sup>26</sup> was run on PRO- Cap datasets from TB40/E infected HFFs 72 hpi (GSEXXXX) and Towne infected HFFs 96 hpi (GSE113394) using published parameters to generate HMCV specific MaxTSSs. MaxTSSs associated with RNA 4.9 and without a \(\pm 1000\) bp genomic sequence were excluded from further analysis. 1,461 TB40/E and 1,456 Towne MaxTSSs were used for further analysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 315, 258, 330]]<|/det|>
+## Feature analysis  
+
+<|ref|>text<|/ref|><|det|>[[114, 330, 880, 585]]<|/det|>
+Feature analysis was performed by only counting fragments of a specific length whose center lay within a specific genomic interval using Bedtools v2.26 intersect program and simple awk commands for each MaxTSSs in the host and HMCV (TB40/E and Towne) genomes. The features consisted of fragments 30- 65 bp with a fragment center between \(+10\) and \(+45\) for free Pol II, 140- 205 bp with a fragment center between \(+65\) and \(+140\) for abutted Pol II, 120- 175 with a fragment center between \(+65\) and \(+175\) for \(+1\) nucleosomes, 64- 88 with a fragment center between - 18 and - 2 for TBP PICs, and 40- 63 with a fragment center between - 36 and - 18 for UL87 PICs. The engaged Pol II feature was found by summing the free Pol II and abutted Pol II counts. Quantification of percentage of Pol II that was either abutted or free was done by selecting promoters with at least one read in either the abutted or free features and that have a MaxTSS of at least 10. Percentage of free Pol II was found relative to engaged Pol II for the whole dataset as well as a gene by gene basis. Quantification of PIC amounts to engaged Pol II was done by first normalizing the TBP PIC fragments from the TBP dataset and UL87 PIC fragments from the UL87 dataset. The ratio between these normalized values was used to rank order promoters by UL87 or TBP dominance and then the ratio of PIC to engaged Pol II was calculated. The Ser5P dataset was utilized for analysis of TBP PICs.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 600, 202, 616]]<|/det|>
+## FragMaps  
+
+<|ref|>text<|/ref|><|det|>[[114, 616, 875, 840]]<|/det|>
+FragMapsHeatmaps displaying the average distribution and position of fragments from DFF- Seq and DFF- ChIP across individual genome intervals or collections of intervals were created using fragMap.py (https://github.com/P- TEFb/fragMap). In general, collections were centered on the MaxTSSs from truQuant annotations that provide the most highly utilized TSS for each active gene with proper strand orientation. In some cases, subsets of the truQuant annotations were used. The data was generated by counting the total number of fragments of each fragment size between 18 and 400 bp across the genomic interval. The aspect ratio, number of pixels, intensities assigned, and the shape of major and minor tick marks were controlled. The desired aspect ratio was implemented by choosing a discrete number of pixels for each base or fragment size. The values at each horizontal position of the fragment sizes were used to assign intensities using the gray.colors function in R with 0 being white and maximum being black. A linear relationship between relative read value and intensity was utilized. Black was set at the maximum read value for most frag maps. To correct for human inaccuracies in perception of dark and light patterns on heatmaps, a gamma correction of 0.5 was applied to all fragMaps.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 856, 198, 870]]<|/det|>
+## Statistics  
+
+<|ref|>text<|/ref|><|det|>[[115, 871, 857, 904]]<|/det|>
+StatisticsPearson correlation coefficient “r” was calculated to demonstrate the reproducibility of our datasets using the MS excel CORREL function. Pearson’s r computes the effect of change in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 877, 170]]<|/det|>
+one variable compared to the change in another variable. MEME motif discovery tool was used to short sequence motifs in a set of longer DNA sequences. Motifs with an E- value less than 0.05 were considered as significantly enriched in our dataset. E- value is an estimation of seeing similar motifs of identical width and contributing sites in a similar size dataset of random sequences.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 186, 275, 201]]<|/det|>
+## Acknowledgments  
+
+<|ref|>text<|/ref|><|det|>[[116, 202, 864, 266]]<|/det|>
+We thank J. A. Marteijn for providing MRC5 cells expressing GFP- tagged Pol II. This research was supported by the National Institute of General Medical Sciences (grant GM126908 to D.H.P. and GM121428 to D.S.L.) and the Department of Veterans Affairs (merit award 1BX001107 to J.L.M.).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 282, 299, 297]]<|/det|>
+## Author Contributions  
+
+<|ref|>text<|/ref|><|det|>[[115, 297, 861, 378]]<|/det|>
+BMS developed the DFF- ChIP protocol, performed all DFF- ChIP experiments, analyzed the data and wrote the paper. MP performed all bioinformatics. ML and JLM generated the HCMV infected HFFs. CBB performed the HCMV time course PRO- Seq experiment. DSL provided intellectual input throughout the study and helped refine the manuscript. DHP directed the research and helped create the bioinformatics approaches.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 394, 290, 409]]<|/det|>
+## Competing Interests  
+
+<|ref|>text<|/ref|><|det|>[[116, 410, 414, 425]]<|/det|>
+No authors have competing interests.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 215, 105]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[110, 103, 880, 905]]<|/det|>
+1. Compe, E. & Egly, J.M. The Long Road to Understanding RNAPII Transcription Initiation and Related Syndromes. Annu Rev Biochem 90, 193-219 (2021).  
+2. Patel, A.B., Greber, B.J. & Nogales, E. Recent insights into the structure of TFIIID, its assembly, and its binding to core promoter. Curr Opin Struct Biol 61, 17-24 (2020).  
+3. Weber, C.M., Ramachandran, S. & Henikoff, S. Nucleosomes are context-specific, H2A.Z-modulated barriers to RNA polymerase. Mol Cell 53, 819-30 (2014).  
+4. Santos-Rosa, H. et al. Active genes are tri-methylated at K4 of histone H3. Nature 419, 407-11 (2002).  
+5. Howe, F.S., Fischl, H., Murray, S.C. & Mellor, J. Is H3K4me3 instructive for transcription activation? Bioessays 39, 1-12 (2017).  
+6. Luse, D.S., Parida, M., Spector, B.M., Nilson, K.A. & Price, D.H. A unified view of the sequence and functional organization of the human RNA polymerase II promoter. Nucleic Acids Res 48, 7767-7785 (2020).  
+7. Vermeulen, M. et al. Selective anchoring of TFIIID to nucleosomes by trimethylation of histone H3 lysine 4. Cell 131, 58-69 (2007).  
+8. Lauberth, S.M. et al. H3K4me3 interactions with TAF3 regulate preinitiation complex assembly and selective gene activation. Cell 152, 1021-36 (2013).  
+9. Day, D.S. et al. Comprehensive analysis of promoter-proximal RNA polymerase II pausing across mammalian cell types. Genome Biol 17, 120 (2016).  
+10. Jimeno-Gonzalez, S., Ceballos-Chavez, M. & Reyes, J.C. A positioned +1 nucleosome enhances promoter-proximal pausing. Nucleic Acids Res 43, 3068-78 (2015).  
+11. Verger, A., Monte, D. & Villeret, V. Take Your PIC. Trends Biochem Sci (2021).  
+12. Chen, X. et al. Structural insights into preinitiation complex assembly on core promoters. Science 372(2021).  
+13. Abdella, R. et al. Structure of the human Mediator-bound transcription preinitiation complex. Science (2021).  
+14. Aibara, S., Schilbach, S. & Cramer, P. Structures of mammalian RNA polymerase II pre-initiation complexes. Nature 594, 124-128 (2021).  
+15. Meier, J.L. & Stinski, M.F. Major Immediate-Early Enhancer and Its Gene Products. Cytomegaloviruses: From Molecular Pathogenesis to Intervention, Vol I, 152-173 (2013).  
+16. Paulus, C. & Nevels, M. The human cytomegalovirus major immediate-early proteins as antagonists of intrinsic and innate antiviral host responses. Viruses 1, 760-79 (2009).  
+17. Wyrwicz, L.S. & Rychlewski, L. Identification of Herpes TATT-binding protein. Antiviral Res 75, 167-72 (2007).  
+18. Aubry, V. et al. Epstein-Barr virus late gene transcription depends on the assembly of a virus-specific preinitiation complex. J Virol 88, 12825-38 (2014).  
+19. Gruffat, H., Kadjouf, F., Mariame, B. & Manet, E. The Epstein-Barr virus BcRF1 gene product is a TBP-like protein with an essential role in late gene expression. J Virol 86, 6023-32 (2012).  
+20. Davis, Z.H. et al. Global mapping of herpesvirus-host protein complexes reveals a transcription strategy for late genes. Mol Cell 57, 349-60 (2015).  
+21. Nitzsche, A., Paulus, C. & Nevels, M. Temporal dynamics of cytomegalovirus chromatin assembly in productively infected human cells. J Virol 82, 11167-80 (2008).  
+22. Cuevas-Bennett, C. & Shenk, T. Dynamic histone H3 acetylation and methylation at human cytomegalovirus promoters during replication in fibroblasts. J Virol 82, 9525-36 (2008).  
+23. Groves, I.J., Reeves, M.B. & Sinclair, J.H. Lytic infection of permissive cells with human cytomegalovirus is regulated by an intrinsic 'pre-immediate-early' repression of viral gene expression mediated by histone post-translational modification. J Gen Virol 90, 2364-2374 (2009).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 90, 870, 134]]<|/det|>
+24. Zalckvar, E. et al. Nucleosome maps of the human cytomegalovirus genome reveal a temporal switch in chromatin organization linked to a major IE protein. Proc Natl Acad Sci U S A 110, 13126-31 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[112, 133, 840, 201]]<|/det|>
+25. Parida, M. et al. Nucleotide Resolution Comparison of Transcription of Human Cytomegalovirus and Host Genomes Reveals Universal Use of RNA Polymerase II Elongation Control Driven by Dissimilar Core Promoter Elements. mBio 10(2019).  
+26. Griffiths, P. & Reeves, M. Pathogenesis of human cytomegalovirus in the immunocompromised host. Nat Rev Microbiol (2021).  
+
+<|ref|>text<|/ref|><|det|>[[112, 200, 770, 214]]<|/det|>
+27. Britt, W.J. Maternal Immunity and the Natural History of Congenital Human Cytomegalovirus Infection. Viruses 10(2018).  
+
+<|ref|>text<|/ref|><|det|>[[112, 215, 770, 248]]<|/det|>
+28. Li, M. et al. Human cytomegalovirus IE2 drives transcription initiation from a select subset of late infection viral promoters by host RNA polymerase II. PLoS Pathog 16, e1008402 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[112, 248, 844, 297]]<|/det|>
+29. Liu, X., Zou, H., Slaughter, C. & Wang, X. DFF, a heterodimeric protein that functions downstream of caspase-3 to trigger DNA fragmentation during apoptosis. Cell 89, 175-84 (1997).  
+
+<|ref|>text<|/ref|><|det|>[[112, 296, 868, 346]]<|/det|>
+30. Steurer, B. et al. Live-cell analysis of endogenous GFP-RPB1 uncovers rapid turnover of initiating and promoter-paused RNA Polymerase II. Proc Natl Acad Sci U S A 115, E4368-E4376 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[112, 345, 870, 394]]<|/det|>
+31. Knezetic, J.A. & Luse, D.S. The presence of nucleosomes on a DNA template prevents initiation by RNA polymerase II in vitro. Cell 45, 95-104 (1986).  
+
+<|ref|>text<|/ref|><|det|>[[112, 393, 870, 427]]<|/det|>
+32. Lorch, Y., LaPointe, J.W. & Kornberg, R.D. Nucleosomes inhibit the initiation of transcription but allow chain elongation with the displacement of histones. Cell 49, 203-10 (1987).  
+
+<|ref|>text<|/ref|><|det|>[[112, 426, 870, 475]]<|/det|>
+33. Cai, H. & Luse, D.S. Transcription initiation by RNA polymerase II in vitro. Properties of preinitiation, initiation, and elongation complexes. J Biol Chem 262, 298-304 (1987).  
+
+<|ref|>text<|/ref|><|det|>[[112, 474, 866, 508]]<|/det|>
+34. Egly, J.M. & Coin, F. A history of TFIIH: two decades of molecular biology on a pivotal transcription/repair factor. DNA Repair (Amst) 10, 714-21 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[112, 507, 833, 523]]<|/det|>
+35. Gruffat, H., Marchione, R. & Manet, E. Herpesvirus Late Gene Expression: A Viral-Specific Pre-initiation Complex Is Key. Front Microbiol 7, 869 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[112, 523, 833, 538]]<|/det|>
+36. Li, M. et al. Cytomegalovirus late transcription factor target sequence diversity orchestrates viral early to late transcription. PLoS Pathog 17, e1009796 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[112, 538, 833, 553]]<|/det|>
+37. Bailey, T.L. et al. MEME SUITE: tools for motif discovery and searching. Nucleic Acids Res 37, W202-8 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[112, 553, 833, 568]]<|/det|>
+38. Samkuraishvili, I. & Luse, D.S. Translocation and transcriptional arrest during transcript elongation by RNA polymerase II. J Biol Chem 271, 23495-505 (1996).  
+
+<|ref|>text<|/ref|><|det|>[[112, 568, 800, 584]]<|/det|>
+39. Nevels, M., Nitzsche, A. & Paulus, C. How to control an infectious bead string: nucleosome-based regulation and targeting of herpesvirus chromatin. Rev Med Virol 21, 154-80 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[112, 584, 803, 599]]<|/det|>
+40. Kent, J.R. et al. During lytic infection herpes simplex virus type 1 is associated with histones bearing modifications that correlate with active transcription. J Virol 78, 10178-86 (2004).  
+
+<|ref|>text<|/ref|><|det|>[[112, 599, 870, 632]]<|/det|>
+41. Ball, C.B., Nilson, K.A. & Price, D.H. Use of the nuclear walk-on methodology to determine sites of RNA polymerase II initiation and pausing and quantify nascent RNAs in cells. Methods 159-160, 165-176 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[112, 632, 870, 666]]<|/det|>
+42. Kang, M.E. & Dahmus, M.E. RNA polymerases IIA and IIO have distinct roles during transcription from the TATA-less murine dihydrofolate reductase promoter. J Biol Chem 268, 25033-40 (1993).  
+
+<|ref|>text<|/ref|><|det|>[[112, 666, 875, 714]]<|/det|>
+43. Laybourn, P.J. & Dahmus, M.E. Phosphorylation of RNA polymerase IIA occurs subsequent to interaction with the promoter and before the initiation of transcription. J Biol Chem 265, 13165-73 (1990).  
+
+<|ref|>text<|/ref|><|det|>[[112, 714, 870, 763]]<|/det|>
+44. Kang, M.E. & Dahmus, M.E. RNA polymerases IIA and IIO have distinct roles during transcription from the TATA-less murine dihydrofolate reductase promoter. J Biol Chem 268, 25033-40 (1993).  
+
+<|ref|>text<|/ref|><|det|>[[112, 763, 870, 811]]<|/det|>
+45. Ball, C.B., Nilson, K.A. & Price, D.H. Use of the nuclear walk-on methodology to determine sites of RNA polymerase II initiation and pausing and quantify nascent RNAs in cells. Methods 159-160, 165-176 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[112, 811, 870, 860]]<|/det|>
+46. Kang, M.E. & Dahmus, M.E. RNA polymerases IIA and IIO have distinct roles during transcription from the TATA-less murine dihydrofolate reductase promoter. J Biol Chem 268, 25033-40 (1993).  
+
+<|ref|>text<|/ref|><|det|>[[112, 860, 857, 907]]<|/det|>
+47. Laybourn, P.J. & Dahmus, M.E. Phosphorylation of RNA polymerase IIA occurs subsequent to interaction with the promoter and before the initiation of transcription. J Biol Chem 265, 13165-73 (1990).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 89, 875, 330]]<|/det|>
+44. He, Y. et al. Near-atomic resolution visualization of human transcription promoter opening. Nature 533, 359-65 (2016).45. Castaneda, A.F. et al. The gammaherpesviral TATA-box-binding protein directly interacts with the CTD of host RNA Pol II to direct late gene transcription. PLoS Pathog 16, e1008843 (2020).46. Skene, P.J., Hernandez, A.E., Groudine, M. & Henikoff, S. The nucleosomal barrier to promoter escape by RNA polymerase II is overcome by the chromatin remodeler Chd1. Elife 3, e02042 (2014).47. Farnung, L., Vos, S.M., Wigge, C. & Cramer, P. Nucleosome-Chd1 structure and implications for chromatin remodelling. Nature 550, 539-542 (2017).48. Bondarenko, V.A. et al. Nucleosomes can form a polar barrier to transcript elongation by RNA polymerase II. Mol Cell 24, 469-79 (2006).49. Ramachandran, S., Ahmad, K. & Henikoff, S. Transcription and Remodeling Produce Asymmetrically Unwrapped Nucleosomal Intermediates. Mol Cell 68, 1038-1053 e4 (2017).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 251, 107]]<|/det|>
+## Figure Legends  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 113, 609, 130]]<|/det|>
+## Figure 1. Reproducibility of Pol II and H3K4me3 DFF-ChIP  
+
+<|ref|>text<|/ref|><|det|>[[115, 130, 878, 259]]<|/det|>
+Figure 1. Reproducibility of Pol II and H3K4me3 DFF- ChIPA Diagram of the DFF- ChIP method. Isolated nuclei are digested with DFF without crosslinking and lightly sonicated to release soluble DNA complexes. The soluble DNA is then immunoprecipitated, library prepped, and sequenced. B,C Genome browser tracks of Pol II DFF- ChIP (purple) and H3K4me3 DFF- ChIP (orange) from HeLa and MRC5 GFP- Pol II cells generated in two different experiments (Exp1 and Exp2). Browser tracks of Flavo NasCap PROSeq (black/grey) show transcription data. D Correlation plots of datasets from Exp1 and Exp2. Read counts in 10,000 bp windows around 12,229 HFF truQuant MaxTSSs were summed and plotted against sums from other experiments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 272, 870, 386]]<|/det|>
+Figure 2. Genome browser tracks of H3K4me3 and Pol II DFF- ChIP from Infected HFFs A,B Genome browser tracks of PRO- Seq, Pol II DFF- ChIP, and H3K4me3 DFF- ChIP from HFFs 48 hpi on the hg38 and TB40/E genomes. C,D Genome browser tracks of PRO- Seq, Pol II DFF- ChIP, and H3K4me3 DFF- ChIP showing the GAPDH promoter in 5,000 and 1,000 bp windows. A dotted line denotes the TSS. E,F Genome browser tracks of PRO- Seq, Pol II DFF- ChIP, and H3K4me3 DFF- ChIP showing the an early (E) and late (F) promoter in a 1,000 bp windows. A dotted line denotes the TSS.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 400, 844, 434]]<|/det|>
+## Figure 3. Visualizing and quantifying transcription complexes and chromatin utilizing fragMaps  
+
+<|ref|>text<|/ref|><|det|>[[114, 433, 872, 658]]<|/det|>
+Figure 3. Visualizing and quantifying transcription complexes and chromatin utilizing fragMapsA Length distribution of fragments +/- 1000 bp of each of 12,229 truQuant genes for the Pol II and H3K4me3 DFF- ChIP datasets from Exp4. B Fragment count of aligned fragments from Exp4 of specified length +/- 1000 bp relative to the TSS of all 12,229 truQuant genes. Total fragment length counts were normalized. C fragMaps of H3K4me3 and Pol II for the 12,229 truQuant genes showing fragments from 18- 400 bp that are +/- 1000 bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are +/- 100 bp relative to the TSS is also shown (right). D fragMaps of H3K4me3 and Pol II for the 1,461 TSRs showing fragments from 18- 400 bp that are +/- 1000 bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are +/- 100 bp relative to the TSS is also shown (right. Resulting fragMaps were lightened 100% to aid in visualization. E Quantification of percentage of the free Pol II feature signal relative to total Pol II feature signal (free + abutted) on both the host and TB40/E genomes on a gene by gene basis sorted by highest to lowest (Left). Fragment count of the Nuc1 feature signal from the H3K4me3 dataset on host (middle) and TB40/E (right) genomes utilizing the same sort.  
+
+<|ref|>text<|/ref|><|det|>[[114, 671, 880, 897]]<|/det|>
+Figure 4. Detection and characterization of TBP- driven PICs and UL87- driven viral PICs A,B fragMaps of fragments positioned +/- 100 bp around the TSS that are 18- 120 bp sized fragments. Host fragMaps were generated from 12,229 truQuant HFF promoters and HCMV (TB40/E) fragMaps were generated from 1,461 TSRs from the Pol II, Pol II + triptolide (Trp), TBP, and Ser5P DFF- ChIP datasets. A dotted line denotes the TSS. C fragMaps of fragments positioned +/- 1000 bp around the MaxTSS that are 18- 400 bp in size. The host fragMap was generated from 12,229 truQuant HFF promoters and HCMV (TB40/E) fragMap was generated from 1,461 TSRs using the TBP datasets. D UL87 fragMaps were generated from 1,456 Towne TSRs. A dotted line denotes the TSS. E LOGOs generated with MEME Suite 5.3.1 from the top 10% genes/TSRs with the most fragments present in the TBP PIC or UL87 PIC feature as detected by DFF- ChIP. Parameters were: ZOOPS, search only given strand, 1 motif, 6 bp motif. Fractions represent the number of sequences matching the sequence motif out of the number of input sequences. E values for the three LOGOs were: TBP host, 1.5e- 398; TBP HCMV, 3.6e- 038; UL87 HCMV, 1.4e- 082.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 105, 875, 345]]<|/det|>
+Figure 5. Classification of TBP and UL87 usage on HCMV genes according to lateness A PRO- Seq tracks depicting \(5'\) ends of reads from HCMV infection time course including 4, 12, 24, 48 and 72 hpi datasets in 1,400 and 800 bp regions of the viral genome. Below are corresponding DFF- ChIP tracks and genomic fragMaps of the same region. B A set of 795 TSRs with greater than 100 MaxTSS \(5'\) ends \((+ / - 5bp)\) when all time points are summed were selected and sorted based on PFA sensitivity, slope, and UL87 dependency. Each TSR had each time point value normalized to library size and each TSR was colored independently. The time point with the highest relative transcription was colored green and lowest colored red. C Quantification of the relative usage of TBP and UL87. The amount of TBP PIC feature counted from the TBP dataset was normalized to the amount of UL87 PIC feature counted from the UL87 dataset for all 1,461 TSRs. The ratio of TBP PIC to UL87 PIC was used to sort the TSRs by TBP PIC dominance (High value) and then plotted. D FragMaps for the top \(5\%\) TBP and UL87 dominated (C) TSRs utilizing 18- 120 bp sized fragments positioned \(+ / - 100\) bp around the TSSs. Top TBP TSRs are depicted using TBP and Ser5P datasets whereas UL87 TSRs are depicted using UL87 and Pol II datasets.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 361, 656, 378]]<|/det|>
+## Figure 6. Analysis of TBP and UL87 PICs on the HCMV genome  
+
+<|ref|>text<|/ref|><|det|>[[115, 378, 874, 473]]<|/det|>
+A Genome browser tracks from H3K4me3, TBP, and UL87 datasets (Exp4) compared directly to genomic fragMaps of the same region of the HCMV genome. B Normalized metaplot of H3K4me3 signal around all 12,229 host truQuant promoters and 1,461 HCMV TSRs. The inner graph shows the HCMV metaplot with a different Y- axis. C Normalized metaplot of H3K4me3 signal around the top \(10\%\) of HCMV TSRs determined by amount of TBP PIC feature or UL87 PIC feature.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 489, 700, 506]]<|/det|>
+## Figure 7. FragMaps resulting from different extents of DFF digestion  
+
+<|ref|>text<|/ref|><|det|>[[115, 505, 879, 568]]<|/det|>
+A,B Total fragment distributions from Pol II, TBP, and H3K4me3 datasets from the control digestion condition (Exp4) and the excess digestion condition (Exp5). C,D Pol II, H3K4me3, and TBP fragMaps of \(+ / - 1000\) bp around the TSS containing 18- 400 bp fragments of 12,229 truQuant genes with either (C) control (Exp4) or (D) excess digestion (Exp5).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 584, 753, 600]]<|/det|>
+## Figure S1. Analysis of DFF-Seq and comparison of DFF-Seq to MNase-Seq  
+
+<|ref|>text<|/ref|><|det|>[[115, 600, 855, 664]]<|/det|>
+A Fragment length distribution for the total DFF- Seq library and for \(+ / - 1000\) bp from 12,201 truQuant HeLa TSSs. Data was normalized such that total fragment counts were the same. Mono- , di- , and tri- nucleosome are demarcated. B Base distribution of the surrounding 60 bp around the \(5'\) ends DFF- Seq fragments.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 680, 600, 697]]<|/det|>
+## Figure S2. Reproducibility of DFF-ChIP on infected HFFs  
+
+<|ref|>text<|/ref|><|det|>[[115, 697, 878, 776]]<|/det|>
+A Genome browser tracks DFF- ChIP H3K4me3 and Pol II tracks from Exp3 and Exp4 and 72 hpi PRO- Cap transcription data. Representative 8000 bp regions are shown on the host and HCMV genomes. B Correlations between Exp3 and Exp4 datasets. Reads from 10,000 bp windows centered on each host truQuant promoter or each viral 200 bp TSR were summed and compared across datasets.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 792, 855, 824]]<|/det|>
+## Figure S3. Visualizing transcription complexes and chromatin from DFF-Seq and Exp3 replicate data  
+
+<|ref|>text<|/ref|><|det|>[[115, 825, 870, 904]]<|/det|>
+A Length distribution of fragments \(+ / - 1000\) bp of each of 12,229 truQuant genes for the Pol II and H3K4me3 DFF- ChIP Exp3 datasets. B Fragment count of aligned fragments of specified length \(+ / - 1000\) bp relative to the TSS of all 12,229 truQuant genes using Exp3 data. Total fragment length counts were normalized. C fragMap representation of DFF- Seq data shown using two different black values. D FragMaps of Exp3 H3K4me3 and Exp3 Pol II for the 12,229
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 872, 170]]<|/det|>
+truQuant genes showing fragments from 18- 400 bp that are \(+ / - 1000\) bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are \(+ / - 100\) bp relative to the TSS is also shown (right). E FragMaps of Exp3 H3K4me3 and Exp3 Pol II for the 1,461 TSRs showing fragments from 18- 400 bp that are \(+ / - 1000\) bp relative to the MaxTSS. A zoomed fragMap showing 18- 120 bp fragments that are \(+ / - 100\) bp relative to the TSS is also shown (right).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 184, 808, 201]]<|/det|>
+## Figure S4. DFF-Seq fragMap and salt sensitivity of TBP PICs on the host genome  
+
+<|ref|>text<|/ref|><|det|>[[115, 201, 875, 330]]<|/det|>
+A Genome browser tracks showing uninfected HFF PRO- Cap data and MRC5 Flavo GFP- Pol II DFF- ChIP tracks that were either washed with high or low salt as described in Methods. The dotted line demarcates the TSS. B fragMaps of \(+ / - 1000\) bp around the TSS containing 18- 400 bp sized fragments and zoomed in fragMaps of \(+ / - 100\) bp around the TSS containing 18- 120 bp sized fragments from 12,201 HeLa truQuant genes from the MRC5 GFP- Pol II datasets. Regions of free Pol II and PIC are demarcated by brackets. For the wider fragMaps, the black value was reduced 3 fold to emphasize smaller fragments. For the zoomed in fragMaps, max black levels were set based on the shown region.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 344, 840, 377]]<|/det|>
+## Figure S5. TBP PICs are functionally different from UL87 PICs even though both may appear at late promoters  
+
+<|ref|>text<|/ref|><|det|>[[115, 377, 876, 504]]<|/det|>
+A Quantification of 5' end reads in an 11 bp window around the five TSSs specifically indicated in Fig. 5A. B A subset TSRs from Fig. 5B plotted with PFA sensitivity and UL87 dependency against slope. TSRs with a TBP/UL87 ratio greater than 2 were colored blue and TSRs with a ratio less than 0.5 were colored red. C Correlations of the amount of engaged Pol II and TBP PIC features counted utilizing the Ser5P dataset in comparison to the Pol II and TBP dataset, respectively, on the host and HCMV genomes. D Genome browser tracks depicting shifting 5' ends from PRO- Seq CMV infection time course data and DFF- ChIP tracks showing UL87, TBP, Pol II, and Ser5P datasets.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 520, 875, 553]]<|/det|>
+## Figure S6. Genomic fragMaps spanning the entire HCMV genome for H3K4me3, TBP, and UL87 fragments  
+
+<|ref|>text<|/ref|><|det|>[[115, 553, 855, 601]]<|/det|>
+A Genomic fragMaps depicting overlapping 21,000 bp windows of the entire HCMV genome. H3K4me3 depict fragments between 18- 400 bp whereas TBP and UL87 fragMaps depict fragments between 18- 150 bp.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 616, 780, 633]]<|/det|>
+## Figure S7. Characterization of fragments resulting from excess DFF digestion  
+
+<|ref|>text<|/ref|><|det|>[[115, 633, 876, 729]]<|/det|>
+A Correlation plots comparing number of reads in 10,000 bp windows around all 12,229 truQuant genes. Graphs compare Exp5 replicates and a single Exp5 replicate to Exp4 data. B H3K4me3 fragMaps from the top \(25\%\) and bottom \(25\%\) of the 12,229 truQuant genes sorted by TSS focus (standard deviation of TSSs in TSRs; Exp4). C fragMaps from the H3K4me3 and Pol II excess digestion experiments (Exp5) with schematic depicting likely nucleosome protections and DFF cleavage sites.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[37, 0, 936, 1005]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[26, 0, 500, 952]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[515, 7, 960, 280]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[515, 280, 960, 495]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[515, 495, 960, 710]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[515, 710, 960, 925]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[520, 971, 620, 992]]<|/det|>
+<center>Figure 5</center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[110, 25, 480, 220]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[110, 212, 130, 234]]<|/det|>
+<center>C </center>  
+
+<|ref|>image<|/ref|><|det|>[[110, 250, 485, 700]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[240, 240, 360, 254]]<|/det|>
+<center>Host H3K4me3 </center>  
+
+<|ref|>image<|/ref|><|det|>[[110, 710, 485, 924]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[240, 695, 328, 708]]<|/det|>
+<center>Host TBP </center>  
+
+<|ref|>image<|/ref|><|det|>[[510, 25, 880, 220]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[510, 220, 530, 234]]<|/det|>
+<center>D </center>  
+
+<|ref|>image<|/ref|><|det|>[[510, 250, 880, 460]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[510, 240, 550, 254]]<|/det|>
+<center>Host H3K4me3 </center>  
+
+<|ref|>image<|/ref|><|det|>[[510, 470, 880, 700]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[510, 460, 550, 473]]<|/det|>
+<center>Host Pol II excess digestion </center>  
+
+<|ref|>image<|/ref|><|det|>[[510, 710, 880, 924]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[510, 462, 550, 475]]<|/det|>
+<center>Host Pol II excess digestion </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 25, 925, 260]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[392, 277, 680, 300]]<|/det|>
+<center>MRC5 GFP-Pol II Low Salt</center> 
+
+<|ref|>image<|/ref|><|det|>[[67, 300, 928, 600]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[392, 612, 680, 635]]<|/det|>
+<center>MRC5 GFP-Pol II High Salt</center> 
+
+<|ref|>image<|/ref|><|det|>[[67, 635, 928, 930]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[444, 980, 552, 998]]<|/det|>
+<center>Figure S4</center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[61, 131, 240, 150]]<|/det|>
+DataAnalysis.xlsx
+
+<--- Page Split --->

preprint/preprint__b348061eddea6b751dcd9ad3ed3d9ea9fede2c7bf3b6611740dffa36b794578d/images_list.json

@@ -0,0 +1,10 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Thrombosis in vitro and in vivo",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 13
+  }
+]

preprint/preprint__b348061eddea6b751dcd9ad3ed3d9ea9fede2c7bf3b6611740dffa36b794578d/preprint__b348061eddea6b751dcd9ad3ed3d9ea9fede2c7bf3b6611740dffa36b794578d.mmd

@@ -0,0 +1,304 @@
+
+# NETosis and thrombosis in vaccine-induced immune thrombotic thrombocytopenia  
+
+Beng Chong ( \(\boxed{\infty}\) beng.chong@unsw.edu.au)  
+
+University of New South Wales  
+
+Halina Leung University of New South Wales  
+
+Jose Perdomo University of New South Wales  
+
+Zohra Ahmadi University of New South Wales  
+
+Fairooj Rashid University of Sydney  
+
+Anoop Enjeti Calvary Mater Hospital  
+
+Stephen Ting Monash University https://orcid.org/0000- 0001- 7755- 8326  
+
+James Chong Westmead Institute for Medical Research https://orcid.org/0000- 0002- 5201- 4856  
+
+## Article  
+
+Keywords: adenoviral vector vaccines, COVID- 19, neutrophil extracellular traps  
+
+Posted Date: September 30th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 753825/v1  
+
+License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Version of Record: A version of this preprint was published at Nature Communications on September 5th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 32946- 1.
+
+<--- Page Split --->
+
+1 Abstract2 Vaccine- induced immune thrombotic thrombocytopenia (VITT) is a rare yet serious3 adverse effect of adenoviral vector vaccines (AstraZeneca and Johnson & Johnson)4 against COVID- 19<sup>1</sup>. Anti- platelet factor 4 (PF4) antibodies are present in VITT5 patients<sup>2,3</sup>. Although the current view suggests that platelet activation by anti- PF4 6 antibodies is the cause of thrombosis there is as yet no direct evidence that the 7 antibodies induce clot formation and thrombocytopenia (reduction in platelet counts) 8 in VITT and the mechanisms involved remain unknown<sup>4</sup>. Here we show that VITT 9 antibodies induce thrombosis and thrombocytopenia, and that thrombus formation is 10 mediated by neutrophil extracellular traps (NETs). We found markers of NETosis, 11 abundance of neutrophil/platelet aggregates and presence of neutrophils undergoing 12 NETosis in patients with active VITT. VITT antibodies directly stimulate neutrophils to 13 release NETs and induce thrombus formation containing abundant platelets, 14 neutrophils, fibrin, extracellular DNA and citrullinated histone H3 using an in vitro blood 15 flow microfluidic system. In transgenic mice expressing human PF4 and FcγRIIa, VITT 16 antibodies lead to thrombosis, thrombocytopenia and formation of low density 17 granulocytes. Pharmacological and genetic inhibition of NETosis prevents VITT- 18 induced thrombosis in mice but not thrombocytopenia. In contrast, in vivo blockage of 19 FcγRIIa abrogates both thrombosis and thrombocytopenia suggesting they are distinct 20 processes. Our findings indicate that VITT antibodies activate cells via FcγRIIa and 21 are responsible for thrombosis and thrombocytopenia. This study identifies NETosis 22 as a pathogenic mechanism for thrombus formation in VITT. We anticipate our findings 23 will motivate future development of NETosis and FcγRIIa inhibitors as potential specific 24 therapies for VITT and consequently better patient outcomes.
+
+<--- Page Split --->
+
+## Main  
+
+Vaccine- induced immune thrombotic thrombocytopenia (VITT), also known as thrombosis with thrombocytopenia syndrome (TTS), is an uncommon but serious adverse effect of adenoviral vector- based SARS- CoV- 2 (COVID- 19) vaccines, specifically ChAdOx1 nCoV- 19 (Vaxzevria, AstraZeneca) and Ad26. COV2. S (Johnson & Johnson)1,5. VITT resembles heparin- induced thrombocytopenia (HIT) which is an immune reaction to a commonly used anticoagulant drug, heparin6. Like patients with HIT, patients with VITT present with thrombocytopenia (low platelets) and thrombosis (blood clots, often at unusual sites) and have an anti- platelet factor 4 (PF4) antibody which induces platelet activation4. The high mortality of VITT (fatality rate estimated at \(23\% 3\) to \(40\% 4\) ) has caused serious concerns among physicians, public health officials and the public, leading to vaccine hesitancy undermining vaccine roll- out in many countries. This is exacerbated by the lack of knowledge of its underlying disease mechanism.  
+
+It is generally believed that platelet activation by the anti- PF4 antibody causes thrombosis in VITT despite the lack of scientific evidence of this antibody inducing clot formation either in vitro or in vivo. Besides, this antibody is also present in individuals without thrombosis7. Experts have suggested that there is a need to show in vivo thrombus formation by the anti- PF4 antibody in a VITT animal model4 and also to understand the mechanism that causes thrombosis.  
+
+In HIT, we have previously shown that thrombosis is driven by NETosis8,9. Upon activation by pathogens, immune complexes and other stimuli, neutrophils release their granules and decondensed chromatin in the form of a DNA network, termed Neutrophil Extracellular Traps (NETs). NETs have two characteristic components, myeloperoxidase and citrullinated histone H3 (Cith3) which are often used as markers of NETs formation. NETosis is the process by which NETs are formed. NETs serve as a framework for thrombus formation and are highly thrombogenic - they activate platelets and other immune cells, damage endothelial cells10 and activate blood coagulation pathways11. NETosis is known to promote venous and arterial thrombosis12,13.  
+
+In this report, using a microfluidics blood flow assay we showed that the anti- PF4 antibody (purified IgG from patients with VITT) when added to circulating normal whole blood induced blood clot formation in vitro and when administered into FcγRIIa+/hPF4+ transgenic mice (a VITT animal model) induced thrombosis in vivo. We also demonstrated that the antibody- induced thrombosis was mediated by platelet and neutrophil activation and NETosis.  
+
+## Results  
+
+## VITT patients  
+
+VITT patients (n=7) from five hospitals in Australia participated in the study. Mean age was 62 years (range: 44 - 82 years), 3 were female. Their clinical features and laboratory test results are consistent with those of previously reported cases of VITT2,3,5. All received their first dose of COVID- 19 vaccine (Vaxzevria, AstraZeneca) 12 - 32 days (mean: 19 days) before their admission to the hospitals, blood samples were collected soon after. All had thrombocytopenia (mean platelet count at admission: \(58 \times 10^{9} / \mathrm{L}\) , range: \(8 - 128 \times 10^{9} / \mathrm{L}\) ) and thrombosis (cerebral venous sinus thrombosis, CVST: 2 patients, splanchnic vein thrombosis: 2, bilateral pulmonary
+
+<--- Page Split --->
+
+1 thromboembolism and deep vein thrombosis: 3). All had elevated D- dimer levels, 2 reduced or normal plasma fibrinogen, anti- PF4 antibodies detected by enzyme- linked 3 immunosorbent (Fig. 1a) and positive for platelet activation functional assays (Fig. 1b, 4 c).  
+
+## NETosis in VITT  
+
+We next investigated the presence of markers of NETosis in VITT patients' plasma and whole blood from patients with active VITT. We assessed both the presence of citrullinated histone H3 (CitH3)14 and the concentration of cell free DNA (cfDNA) in plasma. The levels of CitH3 and cfDNA were significantly increased relative to healthy controls (Fig. 1d, e), which is consistent with the presence of NETosis9,15. Moreover, analysis of fresh blood from patients with active VITT showed the presence of abundant activated neutrophils (low density granulocytes or LDG) (Fig. 1f), neutrophil- platelet aggregates (NPA) (Fig. 1g, h) and neutrophils undergoing NETosis (Fig. 1i, j). Overall, this suggests that NETosis is present in patients with active VITT.  
+
+## VITT IgG induces NETosis in vitro  
+
+Pathogenic anti- PF4 antibodies bind to endogenous PF4 and form immune complexes16. These complexes interact with FcγRIIa to activate platelets and neutrophils9,16. To determine the effect of VITT antibodies in thrombosis, we first isolated total IgG from VITT patients' plasma and assessed its impact on whole blood from healthy donors. Compared to buffer (PBS) and normal IgG (Ctrl), incubation with VITT IgG led to a pronounced increase in the formation of LDG (Fig. 2a) and induction of neutrophils to undergo NETosis (CD15+CitH3+MPO+ cells) (Fig. 2b). By comparison, a HIT antibody, known to induce NETosis9, stimulated neutrophils to a comparable level (Fig. 2a, b). NETs induction in the absence of other cells was corroborated by treatment of purified neutrophils in the presence of PF4 with VITT or normal IgG and assessment of DNA release with the cell impermeant dye Sytox green. Significantly increased DNA release was triggered by VITT IgG relative to normal IgG (Fig 2c, d), indicating that purified IgG from VITT patients strongly initiated NETs formation in healthy donors' whole blood and purified neutrophils in vitro.  
+
+## VITT IgG induces thrombosis in vitro  
+
+To examine the capacity of VITT antibodies to induce thrombosis, fresh whole blood from healthy donors were treated with VITT IgG or normal IgG and flowed through von Willebrand factor (vWf)- coated microchannels in a microfluidics system. The presence of VITT IgG led to thrombus formation. Confocal microscopy imaging of thrombi formed following treatment with VITT IgG showed that the thrombi were formed by platelets, neutrophils and extracellular DNA, while no clots were formed in blood treated with control IgG (Extended data Fig. 1a, b). Further analysis of VITT IgG- induced thrombi showed an abundance of fibrin (Fig. 2e) and CitH3 (Fig. 2f) confirming the strong thrombogenic activity of VITT antibodies and their ability to induce NETosis in vitro. To confirm the role of FcγRIIa and NETosis in VITT- induced thrombosis, blood was pre- treated with anti- FcγRIIa monoclonal antibody, IV.3 or DNase I prior to incubation with VITT IgG. The presence of IV.3 strongly inhibited deposition of platelets (Fig. 2g, i) and neutrophils (Fig. 2g, j). Furthermore, there was no induction of NETosis as indicated by the absence of DNA release in the presence of IV.3 (Fig. 2g, h). Similarly, the presence of DNase I resulted in inhibition of thrombus formation (Fig. 2g- j). These data suggest that direct blocking of FcγRIIa inhibits NETosis and
+
+<--- Page Split --->
+
+1 thrombosis, and digestion of extracellular DNA also inhibits thrombus formation in vitro.  
+
+## VITT IgG induces thrombosis in vivo  
+
+To assess whether anti- PF4 antibodies are responsible for the clinical features of thrombocytopenia and thrombosis in VITT patients, we used a FcγRIIa+/hPF4+ double transgenic mouse model. These mice are necessary to assess the activity of VITT IgG in vivo, since they express two essential components, human PF4 and FcγRIIa on platelets and neutrophils. VITT IgG was administered into the VITT mouse model and lungs extracted to examine the levels of thrombosis. Examination of extracted lungs (Fig. 3a, Extended Data Fig. 1c) from VITT IgG- treated mice showed extensive thrombi deposition in this organ. Thrombi were absent in control IgG treated animals (Fig 3a, Extended Data Fig. 1c). These clots contain abundant platelets, neutrophils (Fig. 3) and fibrin (Fig. 3c). These data suggest that VITT IgG is responsible for thrombosis in vivo.  
+
+## Role of FcγRIIa and NETosis in thrombosis in vivo  
+
+To investigate the role of FcγRIIa and neutrophil activation and NETosis in VITT IgG- induced thrombosis, inhibitors of FcγRIIa (aglycosylated IV.3<sup>9</sup>, agIV.3) and NETosis (GSK484)<sup>9,17</sup> were administered in vivo. In support of our in vitro findings (Fig. 2) blocking either FcγRIIa or NETosis was effective in preventing the formation of clots in vivo as shown by the lack of clots (platelet and neutrophil accumulation) in lung sections of mice treated with agIV.3 (Fig. 3c). The dramatic reduction in thrombus deposition is also confirmed using whole organ imaging (Fig. 4a, b) and quantitative analysis of platelet accumulation in mice treated with VITT IgG plus agIV.3 or GSK484 compared to mice treated with VITT IgG without either inhibitor (Fig. 4a, b). Moreover, inhibitor- treated mice were not only protected from thrombosis but also had significantly less low density granulocytes present in peripheral blood compared to control mice (Extended Data Fig. 1d).  
+
+The contribution of NETosis to thrombosis in VITT was further assessed using the VITT mouse model deficient in PAD4 (FcγRIIa+/hPF4+/PAD4<sup>- /-</sup>). PAD4 is the enzyme responsible for the citrullination of histones necessary for induction of NETosis<sup>18</sup>. Consistent with findings in animals treated with VITT IgG plus GSK484, mice lacking PAD4 treated with VITT IgG had a dramatic reduction in clot formation compared to control VITT mice (which are wild type for PAD4) (Fig. 3c, Fig. 4a, b). There were also significantly fewer circulating low density granulocytes in PAD4 deficient mice compared to control (Extended Data Fig. 1d). Collectively, our data indicate that inhibition of platelet and neutrophil activation by blocking FcγRIIa or inhibition of NETosis can efficiently abolish VITT IgG- induced thrombosis in vivo.  
+
+## VITT IgG induces thrombocytopenia in VITT mouse model  
+
+Unlike mice treated with normal IgG, mice treated with VITT IgG experienced thrombocytopenia (Fig. 4c, d) and systemic reactions such as hypothermia (Extended Data Fig. 2a). AgIV.3 was effective in preventing both thrombocytopenia (Fig. 4d) and thrombosis (Fig. 3c, Fig. 4a). In contrast, NETosis inhibitor GSK484 and PAD4 knockout had no effect on the development of thrombocytopenia (Fig. 4e) although they strongly inhibited thrombosis (Fig. 4a, b). Altogether, these results indicate that VITT IgG- induced thrombosis and thrombocytopenia are distinct processes.
+
+<--- Page Split --->
+
+## Discussion  
+
+Although vaccines against COVID- 19 infection have been very effective, there have been some serious side- effects. One of these is a rare clotting disorder termed VITT or TTS that has caused significant morbidity and mortality. It has generated much public concern globally resulting in vaccine hesitancy and undermining of vaccine roll- out in many jurisdictions.  
+
+Despite numerous recent publications \(^{1,19 - 21}\) , there are still significant knowledge gaps in VITT, in particular regarding its underlying disease mechanism(s) \(^{4}\) . There is yet no direct evidence that VITT antibodies cause thrombosis and thrombocytopenia in vivo. Better knowledge of VITT will improve public confidence which might contribute to increased vaccine uptake. The conventional concept is that the platelet activating anti- PF4 antibody causes clot formation in VITT despite no direct evidence. The presence of anti- PF4 antibodies in individuals without thrombosis \(^{7}\) has created doubts about this concept.  
+
+## Evidence that anti-PF4 antibody induces clot formation  
+
+Here we provide evidence that VITT antibodies directly induce thrombus formation in vitro and in vivo, not by platelet activation alone but also through neutrophil activation and NETosis.  
+
+In this study, we demonstrated that VITT IgG stimulated platelet activation via serotonin release and platelet aggregation assays as had other investigators previously \(^{22}\) . More importantly, we further showed that VITT IgG when added in vitro to circulating whole blood induced clot formation in the microchannels using a microfluidics system. In contrast, normal IgG failed to induce thrombosis. Similarly, administration of VITT IgG but not normal IgG led to development of multiple thrombi in the lungs of the VITT mouse model (FcγRIIa+/hPF4+ double transgenic mice). These data provide direct evidence that VITT IgG (or more specifically immune complexes formed by VITT IgG and PF4) induced clot formation in vitro and in vivo, filling a crucial knowledge gap in VITT pathogenesis. Thrombosis was blocked by anti- FcγRIIa monoclonal antibody, IV.3 suggesting that it was mediated by FcγRIIa receptors on platelets and neutrophils.  
+
+## NETosis is a critical driver of thrombosis in VITT  
+
+Our study and the study by Holm et al \(^{21}\) both showed the presence of NETosis in patients with VITT. In the latter study, NETosis markers occurred together with numerous markers of inflammation, activated innate immune pathways, activated blood cells and endothelium, and damaged tissues in VITT patients. These markers were present in the blood, in a thrombus and in the immune precipitates extracted by a goat anti- human PF4 antibody from plasma of VITT patients. These findings were not unexpected as the VITT patients had robust immune responses, intense inflammation and severe thromboses. However, there was no data implicating NETosis as the cause of thrombosis in the VITT patients. Even the presence of neutrophils and NETosis markers in the thrombus does not necessary indicate that it is the cause of thrombosis as neutrophils and NETs are frequently observed in thrombi in various conditions including stroke, acute myocardial infarction \(^{23,24}\) and deep vein thrombosis \(^{25}\) . The authors, in fact, speculated in their report whether the adenovirus in the vaccine or even the spike protein could have triggered the pronounced inflammatory processes including NETosis.
+
+<--- Page Split --->
+
+1 In contrast, our study not only provides evidence of the presence of NETosis in VITT, 2 but we also show that NETosis directly drives thrombosis in VITT in vivo in the VITT 3 animal model. Administration of VITT IgG but not normal IgG induced development of 4 multiple thrombi in the lungs of the mice. Thrombosis could be prevented or 5 substantially suppressed by administration of NETosis inhibitor, GSK484 or by using 6 PAD4 knock- out mice (which blocks NETosis).  
+
+## VITT IgG induces thrombocytopenia  
+
+We demonstrated here that VITT antibodies induced thrombocytopenia (platelet count decrease) in the VITT mouse model by binding to platelet FcγRIIa. Thrombocytopenia was substantially prevented by anti- FcγRIIa monoclonal antibody, IV.3. In contrast, NETosis inhibitor GSK484 and absence of PAD4 (FcγRIIa+/hPF4+/PAD4+/- mice) which significantly blocked thrombosis in VITT had no effect on thrombocytopenia, suggesting that thrombosis and thrombocytopenia in VITT are two distinct processes as we have previously shown in heparin- induced thrombocytopenia<sup>9</sup>.  
+
+In summary, our findings showed that anti- PF4 antibodies are the pathogenic or disease- causing antibodies in VITT. They induce platelet and neutrophil activation leading to development of NETosis which is the major driver of thrombosis in VITT (Extended Data Fig. 2b). FcγRIIA blockage prevented both thrombocytopenia and thrombosis but NETosis inhibition which effectively suppressed thrombosis, had no effect on thrombocytopenia. Thrombosis and thrombocytopenia appeared to be mediated by two distinct mechanisms.  
+
+Our results have contributed to a better understanding of pathogenesis in VITT and may also lead to development of disease biomarkers and improved diagnosis and new more efficacious therapies for VITT and consequently better clinical outcomes for the patients.
+
+<--- Page Split --->
+
+## Methods  
+
+## Human samples  
+
+VITT samples were collected from patients in Australia from the following hospitals: St George Hospital, Kogarah, Sydney, New South Wales; Calvary Mater Hospital, Wallsend, New South Wales; Box Hill Hospital, Box Hill, Victoria; University Hospital Geelong, Geelong, Victoria and Townsville University Hospital, Townsville, Queensland. Blood was collected from patients clinically diagnosed with HIT and VITT and positive for laboratory tests (ELISA and serotonin release assay) \(^{5,26}\) . Blood from healthy donors was used in control experiments. This study was approved by the Human Research Ethics Committee of South Eastern Sydney Local Health District (17/211 LNR/17/POWH/501). Informed consent was obtained from all study participants. Sera and plasma samples were stored in aliquots at \(- 80^{\circ}C\) until required for analysis.  
+
+## Diagnostic assays  
+
+The abundance of anti- PF4 or anti- PF4/heparin antibodies in patient sera was measured using a solid phase PF4 or PF4/heparin ELISA performed in microwell plates. Sera from patients or healthy donors were added to each well and incubated for 60 min at room temperature and then washed. Conjugated anti- human IgG was added, incubated for 60 min at room temperature and washed. Chromogenic substrate reaction was stopped with 1 M \(\mathsf{H}_2\mathsf{SO}_4\) . Optical density was measured using an automatic plate reader (Tecan Infinite Pro).  
+
+\(^{14}\mathrm{C}\) serotonin- release assay ( \(^{14}\mathrm{C}\) - SRA) was performed as previously described \(^{27}\) . Briefly, washed donor platelets were incubated with radiolabelled \(^{14}\mathrm{C}\) and heat inactivated patient's sera, in the presence and absence of PF4 (10 \(\mu \mathrm{g / mL}\) ), 0.1 U/mL heparin, IV.3 antibody (50 \(\mu \mathrm{g / mL}\) ) or 100 U/mL heparin, for 60 min at room temperature while stirring. Reaction was stopped using PBS- EDTA buffer and centrifuged. Radioactivity (counts per minute) of the supernatant was measured using a beta- counter. Levels greater than \(20\%\) were considered positive.  
+
+## Antibodies  
+
+Purification of immunoglobulin G antibodies from patients' or healthy donor's sera was performed using Protein G Agarose (Roche Mannheim, Germany). The AKTA purifier chromatography system (GE Healthcare) was used for purification. Eluted peak fractions were pooled and concentrated using ultracentrifugal units. Purity of IgG was \(>95\%\) as determined by SDS PAGE Gel analysis. Functional activity of purified IgG was determined by platelet aggregation and serotonin release assays. Hybridoma cells producing IV.3 were obtained from ATCC (clone HB- 217). Cells were cultured in DMEM medium containing \(10\%\) FBS at \(37^{\circ}\mathrm{C}\) , \(5\%\) CO2. Cells were cultured in serum- free DMEM 24h prior to collection of antibody- containing supernatant. Protein G Sepharose affinity chromatography was used to purify IV.3.  
+
+## Platelet aggregation  
+
+Light transmission platelet aggregometry was used to determine antibody activity and role of FcγRIIa in VITT- induced platelet aggregation. Platelet- rich plasma (PRP) was prepared from citrate- anticoagulated healthy donor blood by centrifugation at room temperature at \(150\mathrm{g}\) for 10 min. \(50~\mu \mathrm{L}\) of VITT or normal sera was added to a cuvette with \(300~\mu \mathrm{L}\) of PRP with or without FcγRIIa- inhibitor, IV.3 (20 \(\mu \mathrm{g / mL}\) ), whilst stirring at \(37^{\circ}\mathrm{C}\) for 15 min. Platelet poor plasma was used as blank.
+
+<--- Page Split --->
+
+## Quantification of NETosis markers  
+
+Quantification of NETosis markersCell- free DNA was measured in plasma of VITT and healthy donor samples using Quant- iT™ PicoGreen™ dsDNA assay kit (P11496, Invitrogen), as described by the manufacturer. Plasma levels of citrullinated histone H3 were determined using the H3R8Cit ELISA Capture and Detection kit (R&D143002, EpiCypher)14 following the manufacturer's instructions.  
+
+## Cell isolation  
+
+Cell isolationNeutrophils were purified using EDTA- anticoagulated blood and the EasySep Direct Human Neutrophil Isolation kit (19666, StemCell Technologies) following the manufacturer's instructions. Purified neutrophils are free of platelets and other blood cells as assessed by flow cytometry. Washed platelets were prepared from citrate- anticoagulated blood. For low density granulocytes, whole blood was diluted with PBS and Lymphoprep (07851, StemCell Technologies) was gently underlayed to avoid mixing with the diluted blood. Sample was then centrifuged at \(800 \times \mathrm{g}\) for 20 min at room temperature. Peripheral blood mononuclear cell layer was harvested.  
+
+## Flow cytometry  
+
+Flow cytometryFresh citrate- anticoagulated blood from VITT patients or healthy donors was diluted with PBS. Platelet- neutrophil aggregates were analysed using anti- CD15 (Alexa Fluor 647, BD 562369) and anti- CD41a (PE, BD 555467), NETs were identified using anti- citrullinated histone H3 (ab5103), anti- MPO (PE, BD 341642) and goat anti- rabbit IgG (BV421, BD 565014). Monocytes and low density granulocytes were identified using anti- CD14 (V500, BD 561391) and anti- CD15 (Alexa Fluor 647, BD 562369) or anti- Ly6G (V450, BD560603) and anti- CD11b (PE, BD 557397). Platelet counts in mouse blood were determined by number of events acquired in 60s relative to time 0. Flow cytometry data were analysed using FlowJo software (LCC, USA.  
+
+## Timelapse  
+
+TimelapsePurified neutrophils were stained with Hoechst 33342 (14533, Sigma) and seeded into eight- well Nunc Lab- Tek II chambers. Purified VITT IgG (5 mg/mL) or normal IgG (5 mg/mL) with PF4 (12 \(\mu \mathrm{g} / \mathrm{mL}\) ) were added to each reaction. Release of extracellular DNA was measured using Sytox Green (S7020, Invitrogen). Wells were imaged using a confocal laser- scanning microscope (Leica TCS SP8). Sytox green fluorescence relative to Hoechst 33342 fluorescence was calculated with ImageJ software (version 2.1.0/1.53c, NIH).  
+
+## Microfluidics  
+
+MicrofluidicsCitrate- anticoagulated blood was diluted 1:2 with PBS, supplemented with purified IgG (VITT IgG 3 mg/mL, normal IgG 3 mg/mL) and incubated at \(37^{\circ}\mathrm{C}\) for 90 min. In selected experiments, blood was pre- incubated with IV.3 (20 \(\mu \mathrm{g} / \mathrm{mL}\) ) or DNase I (160 U/mL). Blood was stained with combinations of Hoechst 33342 (3 \(\mu \mathrm{g} / \mathrm{mL}\) ), Sytox green (0.3 \(\mu \mathrm{M}\) ), anti- CD41 Alexa 647 (15 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- CD41- FITC (15 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- CD15 Alexa 647 (15 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- fibrin Alexa 594 (30 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- CitH3 Alexa 594 (30 \(\mu \mathrm{g} / \mathrm{mL}\) ) prior to perfusion through Vena8 Fluoro+™ biochip microchannels coated with vWf (Haematologic Technologies United BioResearch Products Pty Ltd). Biochips were mounted on a fluorescent microscope (Zeiss Axio Observer.A1) and fluorescence images from different microscopic fields were captured in real time with a Q- Imaging EXi Blue™ camera (Q- Imaging, Surry, BC, Canada) with the fluid shear
+
+<--- Page Split --->
+
+stress set at 67 dyne/cm² (shear rate 1500/s) for 30 min. Selected samples were fixed with 2% paraformaldehyde and imaged by confocal microscopy.  
+
+## Mouse model  
+
+Mouse modelMice expressing the \(\mathrm{R^{131}}\) isoform of human FcyRlla and human PF4 were generated in C57BL/6 background. Double transgenic (FcyRlla+/hPF4+) and FcyRlla+/hPF4+/PAD4- have been characterised previously<sup>9,28</sup>. VITT was recreated in these mice by intravenous injection of purified VITT IgG (250 \(\mu \mathrm{g / g}\) ). Inhibitors of NETosis (GSK484, Cayman chemicals) or anti FcyRlla (aglycosylated IV.3, 1 \(\mu \mathrm{g / g}\) ) were injected at time 0. Anti-CD42c Dylight- 649 antibody (Emfret, Germany) and Alexa Fluor 594- fibrinogen were used to label mouse platelets and fibrin in vivo, respectively. Following euthanasia, lungs were perfused with PBS followed by formalin, extracted and imaged using the IVIS Spectrum (Perkin Elmer). Fluorescence was calculated in radiant efficiency using living Image 4.5.5 software (Perkin Elmer). All animal experiments were approved by the University of New South Wales Animal Care and Ethics Committee.  
+
+## Histology  
+
+HistologyFormalin- fixed lungs were embedded in paraffin, sectioned at 4 microns and mounted onto slides. Slides were deparaffinised, rehydrated, and underwent heat- induced antigen retrieval. Slides were probed with anti- Ly6G (Alexa Fluor 488, 127626 Biolegend). Vectashield antifade mounting medium with DAPI (H- 1200, Vector Laboratories) was used to mount glass coverslips onto the slides. Slides were imaged by confocal microscopy. Slides were also stained with H&E and imaged with a Zeiss Axioskop microscope.  
+
+## Statistical analyses  
+
+Statistical tests were performed using GraphPad Prism version 8 (GraphPad Software, USA). The following statistical tests were used in this study as described in the figure legends: (1) Shapiro- Wilk normality test. (2) Student's t test was performed when comparing between two groups. (3) Multiple comparisons were analysed by one- way ANOVA with post- test correction for multiple comparisons. Each individual healthy donor for in vitro experiments and each mouse used for animal experiments was considered a biological replicate. P values \(< 0.05\) were considered statistically significant.  
+
+Acknowledgments: The authors wish to thank Steven McKenzie (Philadelphia, USA) for providing FcyRlla+/hPF4+ mice, Drs Feng Yan, Rose Wong and Kathryn Evans for valuable technical assistance, Drs Sumita Ratnasingam, John Cassey and Silvia Zheng for management of VITT patients and valuable clinical input, O Szeto, J Bennett, M Poxton, E Heyer and P Rojanski for assistance in obtaining human research ethics/governance approvals, and members of the THANZ VITT Advisory Group for helpful discussion of VITT patients. This work was supported by grants from National Health and Medical Research Council, Australia, Program Grant APP1052616 and New South Wales Capacity Program Senior Researcher Grant RG201677 to BHC; NSW Health Cardiovascular Disease Clinician Scientist Grant and National Health and Medical Research Council Australia, Investigator Grant to JJHC
+
+<--- Page Split --->
+
+1 Author contributions: BHC conceived the idea, designed and supervised the 2 research, analysed the data and wrote the manuscript, HL and JP designed and 3 carried out the experiments, collected and analysed the data and wrote the 4 manuscript, ZA performed platelet function assays and microfluidic studies, collected 5 and analysed the data, FR carried out histology and immunochemistry studies, 6 collected and analysed the data, JC provided conceptual input, designed experiments 7 and analysed data, ST and AA provide intellectual input, analysed clinical data and 8 managed VITT patients. All authors reviewed and edited the manuscript and approved 9 the final version of the manuscript. 10 11 Conflict of interest statement. The authors declare no conflicts of interest.
+
+<--- Page Split --->
+
+## References  
+
+1. Greinacher, A. et al. Thrombotic Thrombocytopenia after ChAdOx1 nCov-19 Vaccination. N Engl J Med 384, 2092-2101, (2021).  
+2. Scully, M. et al. Pathologic Antibodies to Platelet Factor 4 after ChAdOx1 nCov-19 Vaccination. N Engl J Med 384, 2202-2211, (2021).  
+3. Pavord, S. et al. Clinical Features of Vaccine-Induced Immune Thrombocytopenia and Thrombosis. N. Engl. J. Med., (2021).  
+4. Cines, D. B. & Bussel, J. B. SARS-CoV-2 Vaccine-Induced Immune Thrombotic Thrombocytopenia. N Engl J Med 384, 2254-2256, (2021).  
+5. Schultz, N. H. et al. Thrombosis and Thrombocytopenia after ChAdOx1 nCov-19 Vaccination. N Engl J Med 384, 2124-2130, (2021).  
+6. Chong, B. H. Evolving concepts of pathogenesis of heparin-induced thrombocytopenia: Diagnostic and therapeutic implications. Int. J. Lab. Hematol. 42, 25-32, (2020).  
+7. Thiele, T. et al. Frequency of positive anti-PF4/polyanion antibody tests after COVID-19 vaccination with ChAdOx1 nCov-19 and BNT162b2. Blood 138, 299-303, (2021).  
+8. Leung, H. et al. Inhibition of NADPH oxidase blocks NETosis and reduces thrombosis in heparin-induced thrombocytopenia. Blood Advances, (2021).  
+9. Perdomo, J. et al. Neutrophil activation and NETosis are the major drivers of thrombosis in heparin-induced thrombocytopenia. Nat. Commun. 10, 1322, (2019).  
+10. Folco, E. J. et al. Neutrophil Extracellular Traps Induce Endothelial Cell Activation and Tissue Factor Production Through Interleukin-1α and Cathepsin G. Arterioscler. Thromb. Vasc. Biol. 38, 1901-1912, (2018).  
+11. Thälin, C., Hisada, Y., Lundström, S., Mackman, N. & Wallén, H. Neutrophil Extracellular Traps. Arterioscler. Thromb. Vasc. Biol. 39, 1724-1738, (2019).  
+12. Fuchs, T. A. et al. Extracellular DNA traps promote thrombosis. Proc. Natl. Acad. Sci. U. S. A. 107, 15880, (2010).  
+13. Laridan, E., Martinod, K. & De Meyer, S. F. Neutrophil Extracellular Traps in Arterial and Venous Thrombosis. Semin. Thromb. Hemost. 45, 86-93, (2019).  
+14. Thälin, C. et al. Quantification of citrullinated histones: Development of an improved assay to reliably quantify nucleosomal H3Cit in human plasma. J. Thromb. Haemost. 18, 2732-2743, (2020).  
+15. Masuda, S. et al. NETosis markers: Quest for specific, objective, and quantitative markers. Clin. Chim. Acta 459, 89-93, (2016).  
+16. Nguyen, T.-H., Medvedev, N., Delcea, M. & Greinacher, A. Anti-platelet factor 4/polyanion antibodies mediate a new mechanism of autoimmunity. Nat. Commun. 8, 14945, (2017).  
+17. Lewis, H. D. et al. Inhibition of PAD4 activity is sufficient to disrupt mouse and human NET formation. Nat. Chem. Biol. 11, 189-191, (2015).  
+18. Thiam, H. R. et al. NETosis proceeds by cytoskeleton and endomembrane disassembly and PAD4-mediated chromatin decondensation and nuclear envelope rupture. Proc. Natl. Acad. Sci. U. S. A. 117, 7326, (2020).  
+19. Huynh, A., Kelton, J. G., Arnold, D. M., Daka, M. & Nazy, I. Antibody epitopes in vaccine-induced immune thrombotic thrombocytopenia. Nature, (2021).
+
+<--- Page Split --->
+
+20. Greinacher, A. et al. Anti-Platelet Factor 4 Antibodies Causing VITT do not Cross-React with SARS-CoV-2 Spike Protein. Blood, (2021).  
+21. Holm, S. et al. Immune complexes, innate immunity, and NETosis in ChAdOx1 vaccine-induced thrombocytopenia. Eur. Heart J., (2021).  
+22. Althaus, K. et al. Antibody-mediated procoagulant platelets in SARS-CoV-2-vaccination associated immune thrombotic thrombocytopenia. Haematologica 106, 2170-2179, (2021).  
+23. Novotny, J. et al. Thrombus NET content is associated with clinical outcome in stroke and myocardial infarction. Neurology 94, e2346, (2020).  
+24. Chilingaryan, Z. et al. Erythrocyte interaction with neutrophil extracellular traps in coronary artery thrombosis following myocardial infarction. Pathology, (2021).  
+25. Brill, A. et al. Neutrophil extracellular traps promote deep vein thrombosis in mice. J. Thromb. Haemost. 10, 136-144, (2012).  
+26. Chong, B. H. & Isaacs, A. Heparin-induced thrombocytopenia: What clinicians need to know. Thromb. Haemost. 101, 279-283, (2009).  
+27. Sheridan, D., Carter, C. & Kelton, J. G. A diagnostic test for heparin-induced thrombocytopenia. Blood 67, 27-30, (1986).  
+28. Reilly, M. P. et al. Heparin-induced thrombocytopenia/thrombosis in a transgenic mouse model requires human platelet factor 4 and platelet activation through FcgammaRIIA. Blood 98, 2442-2447, (2001).
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+ 
+
+![PLACEHOLDER_13_1]
+
+ 
+
+![PLACEHOLDER_13_2]
+
+ 
+
+![PLACEHOLDER_13_3]
+
+ 
+
+![PLACEHOLDER_13_4]
+
+
+<--- Page Split --->
+
+Figure 1. Platelet activation and NETosis in VITT. a PF4 and PF4- heparin ELISA experiment of VITT serum and normal controls. The cut- off, 0.50 OD units. b \(^{14}\mathrm{C}\) - serotonin release assay for VITT samples with buffer alone, PF4 (10 \(\mu \mathrm{g} / \mathrm{mL}\) ), 0.1 or 100 U/mL heparin or IV.3 antibody (50 \(\mu \mathrm{g} / \mathrm{mL}\) ). Each dot represents the mean of assays done in triplicate. The cut- off was set at \(20\%\) CPM. c Platelet aggregation responses. Purified IgG from VITT patients induced aggregation in platelet rich plasma (red and blue traces). Blockage of \(\mathrm{Fc\gamma / R} \mathrm{lla}\) with IV.3 inhibited aggregation (purple and green traces). d Nucleosomal CitH3 levels in VITT patients' plasma (n=7) relative to normal controls (n=8) was determined by H3R8Cit ELISA. e cfDNA in VITT patients' plasma (n=7) relative to normal controls (n=6) determined by PicoGreen fluorescence assay. f Representative side and forward scatter flow cytometry plot backgated for neutrophils (yellow) and monocytes (blue) from VITT patient's and normal blood. LDG are indicated. g Representative plot of NPA from VITT and normal blood. h Quantification of NPA in VITT. i Representative plot of NETs from VITT and normal blood. j quantification of NETs in VITT. MPO\*, CitH3\* double positive cells within the CD15+ population were defined as neutrophils undergoing NETosis. The percentage of gated events is indicated in each quadrant. Statistics, Mann- Whitney test. \*P < 0.05; \*\*P < 0.01; \*\*\*\*P < 0.0001. OD, optical density units; CPM, counts per minute; NPA, neutrophil-platelet aggregates; LDG, low density granulocytes; cfDNA, cell- free DNA; CitH3, citrullinated histone H3; Pt, patient.
+
+<--- Page Split --->
+![PLACEHOLDER_15_0]
+
+
+<--- Page Split --->
+
+Figure 2. Effect of VITT IgG on donor's blood. a Quantification of LDG and b NETs following treatment of healthy donor blood with VITT, normal controls and HIT IgG. c Purified neutrophils treated with VITT IgG or normal IgG plus PF4 were stained for extracellular DNA (green) and nuclei (blue). d DNA release calculated as fluorescence intensity ratio of extracellular DNA (Sytox staining)/total DNA (Hoechst staining) vs. time (n=3). e VITT IgG induces thrombosis. Healthy donors' blood treated with VITT IgG was stained for total DNA (blue), platelets (green), fibrin (red) and neutrophils (magenta). Thrombi were imaged with a confocal laser-scanning microscope (overlap of green and red shown as yellow). Scale bar: \(10 \mu \mathrm{m}\) . f Thrombi contain CthH3. Thrombi were generated and imaged as in (e), and stained for DNA (blue), platelets (green), CthH3 (yellow) and neutrophils (magenta). Overlap of yellow and green is shown as white. g IV.3 and DNase I prevent VITT IgG-induced thrombus formation in microfluidics system. Treated blood was stained for DNA (blue), platelets (green) and neutrophils (red). Scale bar: \(50 \mu \mathrm{m}\) . Graphs show area coverage percentage for h total DNA, i platelets and j neutrophils. n=3, mean \(\pm\) s.d. Statistics: (a, b) Kruskal-Wallis test with uncorrected Dunn's test, (d) One-way ANOVA followed by Dunn's test for multiple comparisons, (h, i, j) One-way ANOVA with Tukey's correction for multiple comparisons. \*P < 0.05; \*\*P < 0.01; \*\*\*P < 0.001, \*\*\*\*P < 0.0001. LDG, low density granulocytes; ext. DNA, extracellular DNA; Ctrl, control; Pt, patient.
+
+<--- Page Split --->
+![PLACEHOLDER_17_0]
+
+
+<--- Page Split --->
+
+Figure 3. VITT IgG induces thrombosis in FcyRIIa+/hPF4+ mice. a Representative H&E staining of lung sections of mice treated with nlgG or VITT IgG. Scale bar 50 μm. b Fluorescent images of lung sections of mice treated with VITT IgG. Platelets were labelled in vivo with anti- CD42c- Dylight 649 (magenta). Neutrophil were stained with anti- Ly6G (green). Neutrophil infiltration in the clot is shown. Cell nuclei were stained with DAPI (blue). Scale bars 50 μm. c Fluorescent images of lung sections of FcyRIIa+/hPF4+ mice treated with nlgG, VITT IgG or VITT IgG plus agIV.3 or FcyRIIa+/hPF4+/PAD4+/- mice treated with VITT IgG. Fibrin labelled with AF594 (red) resulted from injection of AF594- labelled fibrinogen at 1 μg/g. Platelets were labelled in vivo with anti- CD42c- Dylight 649 (magenta). Cell nuclei were stained with DAPI (blue). Scale bar 10 μm. nlgG, normal IgG; agIV.3, aglycosylated IV.3 antibody.
+
+<--- Page Split --->
+![PLACEHOLDER_19_0]
+
+ 
+
+![PLACEHOLDER_19_1]
+
+
+<--- Page Split --->
+
+Figure 4. Thrombosis and thrombocytopenia. a Representative images of lungs following treatment. The level of fluorescence indicates accumulation of platelets labelled with anti- CD42c- Dylight 649 in the lungs. b Graph of lung fluorescence for the VITT patients indicated. c Representative graph showing platelet counts following treatment of FcyRlla+/hPF4+ mice with normal IgG (nIgG) or VITT IgG or VITT IgG plus agIV.3 determined at 1h and 4h after treatment. d Quantification of platelet counts in FcyRlla+/hPF4+ mice following the treatments indicated in the figure. e Graph showing platelet counts following treatment of FcyRlla+/hPF4+ mice with VITT IgG with or without GSK or FcyRlla+/hPF4+/PAD4+/- mice plus VITT IgG determined at 1h and 4h after treatment. Statistics. b One- way ANOVA with Dunnet's test for multiple comparisons. Unpaired t test for comparison between Pt5 in FcyRlla+/hPF4+ and Pt 5 in FcyRlla+/hPF4+/PAD4+/- mice. d One- way ANOVA with Dunnet's test for multiple comparisons. nIgG, normal IgG; PAD4 KO, PAD4 knockout FcyRlla+/hPF4+ mice; agIV.3, aglycosylated IV.3 antibody; Pt, patient.
+
+<--- Page Split --->
+![PLACEHOLDER_21_0]
+
+ 
+
+<center>Thrombosis in vitro and in vivo</center> 
+
+<center>VITT IgG</center> 
+
+![PLACEHOLDER_21_1]
+
+
+<--- Page Split --->
+
+Extended Data Figure 1. a VITT IgG and thrombosis. Healthy donors' blood treated with VITT IgG was flowed in vWf- coated microchannels. Extracellular DNA was stained with Sytox green (green), platelets with anti- CD41 AF647 (magenta) and neutrophils with anti- CD15 AF594 (red). Thrombi were imaged with a confocal laser- scanning microscope (Leica TCS SP8 running Leica's LAS X software) with a 63x oil immersion objective. Scale bar 20 μm. b Healthy donors' blood treated with normal IgG was flowed in vWf- coated microchannels. Total DNA was stained with Hoechst 33342 (blue), platelets with anti- CD41- FITC (green) and neutrophils with anti- CD15 AF594 (red). Scale bar 50 μm. c Fluorescent images of lung lobes from mice treated with VITT IgG or control IgG. DAPI- stained nuclei (blue), platelet- rich thrombi (magenta). Scale bar 500 μm. d Level of low density granulocytes (LDG) in blood from mice following the treatments indicated in the Figure. Statistics: Unpaired t test. nlgG, normal IgG; PAD4 KO, PAD4 knockout FcγRIIa+/hPF4+ mice.
+
+<--- Page Split --->
+
+# Systemic response and proposed mechanism of thrombocytopenia and thrombosis in VITT  
+
+![PLACEHOLDER_23_0]
+  
+
+![PLACEHOLDER_23_1]
+
+
+<--- Page Split --->
+
+Extended Data Figure 2. a Changes in temperature following the treatments indicated in the figure. Dotted line represents the mean temperature of mice before treatment \((38.3^{\circ}\mathrm{C}, \mathrm{n} = 30)\) . b Model of mechanism of thrombosis and thrombocytopenia in VITT. Anti- PF4 antibodies from VITT patients form a complex with PF4 and interact with FcγRIIa. Interaction of the complex with platelets results in thrombocytopenia, which can be blocked with the monoclonal antibody IV.3. In the case of neutrophils, the interaction of the complex with FcγRIIa leads to NETs formation and subsequent thrombosis. Thrombosis can be blocked by neutralisation of FcγRIIa with IV.3 or by inhibition of NETosis using NETs inhibitor or in PAD4 knockout mice. In vitro, addition of DNase I disrupts thrombus formation.
+
+<--- Page Split --->

preprint/preprint__b348061eddea6b751dcd9ad3ed3d9ea9fede2c7bf3b6611740dffa36b794578d/preprint__b348061eddea6b751dcd9ad3ed3d9ea9fede2c7bf3b6611740dffa36b794578d_det.mmd

@@ -0,0 +1,365 @@
+<|ref|>title<|/ref|><|det|>[[45, 107, 814, 175]]<|/det|>
+# NETosis and thrombosis in vaccine-induced immune thrombotic thrombocytopenia  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 438, 216]]<|/det|>
+Beng Chong ( \(\boxed{\infty}\) beng.chong@unsw.edu.au)  
+
+<|ref|>text<|/ref|><|det|>[[52, 219, 325, 237]]<|/det|>
+University of New South Wales  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 325, 283]]<|/det|>
+Halina Leung University of New South Wales  
+
+<|ref|>text<|/ref|><|det|>[[44, 289, 325, 330]]<|/det|>
+Jose Perdomo University of New South Wales  
+
+<|ref|>text<|/ref|><|det|>[[44, 336, 325, 376]]<|/det|>
+Zohra Ahmadi University of New South Wales  
+
+<|ref|>text<|/ref|><|det|>[[44, 382, 235, 422]]<|/det|>
+Fairooj Rashid University of Sydney  
+
+<|ref|>text<|/ref|><|det|>[[44, 428, 255, 468]]<|/det|>
+Anoop Enjeti Calvary Mater Hospital  
+
+<|ref|>text<|/ref|><|det|>[[44, 474, 576, 515]]<|/det|>
+Stephen Ting Monash University https://orcid.org/0000- 0001- 7755- 8326  
+
+<|ref|>text<|/ref|><|det|>[[44, 521, 770, 562]]<|/det|>
+James Chong Westmead Institute for Medical Research https://orcid.org/0000- 0002- 5201- 4856  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 604, 102, 621]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 641, 717, 661]]<|/det|>
+Keywords: adenoviral vector vaccines, COVID- 19, neutrophil extracellular traps  
+
+<|ref|>text<|/ref|><|det|>[[44, 679, 350, 699]]<|/det|>
+Posted Date: September 30th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 717, 463, 737]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 753825/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 754, 910, 797]]<|/det|>
+License: © \(\circledast\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 832, 950, 876]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on September 5th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 32946- 1.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[67, 87, 884, 480]]<|/det|>
+1 Abstract2 Vaccine- induced immune thrombotic thrombocytopenia (VITT) is a rare yet serious3 adverse effect of adenoviral vector vaccines (AstraZeneca and Johnson & Johnson)4 against COVID- 19<sup>1</sup>. Anti- platelet factor 4 (PF4) antibodies are present in VITT5 patients<sup>2,3</sup>. Although the current view suggests that platelet activation by anti- PF4 6 antibodies is the cause of thrombosis there is as yet no direct evidence that the 7 antibodies induce clot formation and thrombocytopenia (reduction in platelet counts) 8 in VITT and the mechanisms involved remain unknown<sup>4</sup>. Here we show that VITT 9 antibodies induce thrombosis and thrombocytopenia, and that thrombus formation is 10 mediated by neutrophil extracellular traps (NETs). We found markers of NETosis, 11 abundance of neutrophil/platelet aggregates and presence of neutrophils undergoing 12 NETosis in patients with active VITT. VITT antibodies directly stimulate neutrophils to 13 release NETs and induce thrombus formation containing abundant platelets, 14 neutrophils, fibrin, extracellular DNA and citrullinated histone H3 using an in vitro blood 15 flow microfluidic system. In transgenic mice expressing human PF4 and FcγRIIa, VITT 16 antibodies lead to thrombosis, thrombocytopenia and formation of low density 17 granulocytes. Pharmacological and genetic inhibition of NETosis prevents VITT- 18 induced thrombosis in mice but not thrombocytopenia. In contrast, in vivo blockage of 19 FcγRIIa abrogates both thrombosis and thrombocytopenia suggesting they are distinct 20 processes. Our findings indicate that VITT antibodies activate cells via FcγRIIa and 21 are responsible for thrombosis and thrombocytopenia. This study identifies NETosis 22 as a pathogenic mechanism for thrombus formation in VITT. We anticipate our findings 23 will motivate future development of NETosis and FcγRIIa inhibitors as potential specific 24 therapies for VITT and consequently better patient outcomes.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 166, 100]]<|/det|>
+## Main  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 880, 330]]<|/det|>
+Vaccine- induced immune thrombotic thrombocytopenia (VITT), also known as thrombosis with thrombocytopenia syndrome (TTS), is an uncommon but serious adverse effect of adenoviral vector- based SARS- CoV- 2 (COVID- 19) vaccines, specifically ChAdOx1 nCoV- 19 (Vaxzevria, AstraZeneca) and Ad26. COV2. S (Johnson & Johnson)1,5. VITT resembles heparin- induced thrombocytopenia (HIT) which is an immune reaction to a commonly used anticoagulant drug, heparin6. Like patients with HIT, patients with VITT present with thrombocytopenia (low platelets) and thrombosis (blood clots, often at unusual sites) and have an anti- platelet factor 4 (PF4) antibody which induces platelet activation4. The high mortality of VITT (fatality rate estimated at \(23\% 3\) to \(40\% 4\) ) has caused serious concerns among physicians, public health officials and the public, leading to vaccine hesitancy undermining vaccine roll- out in many countries. This is exacerbated by the lack of knowledge of its underlying disease mechanism.  
+
+<|ref|>text<|/ref|><|det|>[[116, 330, 880, 428]]<|/det|>
+It is generally believed that platelet activation by the anti- PF4 antibody causes thrombosis in VITT despite the lack of scientific evidence of this antibody inducing clot formation either in vitro or in vivo. Besides, this antibody is also present in individuals without thrombosis7. Experts have suggested that there is a need to show in vivo thrombus formation by the anti- PF4 antibody in a VITT animal model4 and also to understand the mechanism that causes thrombosis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 880, 606]]<|/det|>
+In HIT, we have previously shown that thrombosis is driven by NETosis8,9. Upon activation by pathogens, immune complexes and other stimuli, neutrophils release their granules and decondensed chromatin in the form of a DNA network, termed Neutrophil Extracellular Traps (NETs). NETs have two characteristic components, myeloperoxidase and citrullinated histone H3 (Cith3) which are often used as markers of NETs formation. NETosis is the process by which NETs are formed. NETs serve as a framework for thrombus formation and are highly thrombogenic - they activate platelets and other immune cells, damage endothelial cells10 and activate blood coagulation pathways11. NETosis is known to promote venous and arterial thrombosis12,13.  
+
+<|ref|>text<|/ref|><|det|>[[115, 608, 880, 706]]<|/det|>
+In this report, using a microfluidics blood flow assay we showed that the anti- PF4 antibody (purified IgG from patients with VITT) when added to circulating normal whole blood induced blood clot formation in vitro and when administered into FcγRIIa+/hPF4+ transgenic mice (a VITT animal model) induced thrombosis in vivo. We also demonstrated that the antibody- induced thrombosis was mediated by platelet and neutrophil activation and NETosis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 722, 193, 737]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 756, 247, 772]]<|/det|>
+## VITT patients  
+
+<|ref|>text<|/ref|><|det|>[[115, 773, 880, 903]]<|/det|>
+VITT patients (n=7) from five hospitals in Australia participated in the study. Mean age was 62 years (range: 44 - 82 years), 3 were female. Their clinical features and laboratory test results are consistent with those of previously reported cases of VITT2,3,5. All received their first dose of COVID- 19 vaccine (Vaxzevria, AstraZeneca) 12 - 32 days (mean: 19 days) before their admission to the hospitals, blood samples were collected soon after. All had thrombocytopenia (mean platelet count at admission: \(58 \times 10^{9} / \mathrm{L}\) , range: \(8 - 128 \times 10^{9} / \mathrm{L}\) ) and thrombosis (cerebral venous sinus thrombosis, CVST: 2 patients, splanchnic vein thrombosis: 2, bilateral pulmonary
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 84, 880, 152]]<|/det|>
+1 thromboembolism and deep vein thrombosis: 3). All had elevated D- dimer levels, 2 reduced or normal plasma fibrinogen, anti- PF4 antibodies detected by enzyme- linked 3 immunosorbent (Fig. 1a) and positive for platelet activation functional assays (Fig. 1b, 4 c).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 167, 273, 182]]<|/det|>
+## NETosis in VITT  
+
+<|ref|>text<|/ref|><|det|>[[115, 183, 880, 331]]<|/det|>
+We next investigated the presence of markers of NETosis in VITT patients' plasma and whole blood from patients with active VITT. We assessed both the presence of citrullinated histone H3 (CitH3)14 and the concentration of cell free DNA (cfDNA) in plasma. The levels of CitH3 and cfDNA were significantly increased relative to healthy controls (Fig. 1d, e), which is consistent with the presence of NETosis9,15. Moreover, analysis of fresh blood from patients with active VITT showed the presence of abundant activated neutrophils (low density granulocytes or LDG) (Fig. 1f), neutrophil- platelet aggregates (NPA) (Fig. 1g, h) and neutrophils undergoing NETosis (Fig. 1i, j). Overall, this suggests that NETosis is present in patients with active VITT.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 346, 444, 362]]<|/det|>
+## VITT IgG induces NETosis in vitro  
+
+<|ref|>text<|/ref|><|det|>[[115, 363, 880, 592]]<|/det|>
+Pathogenic anti- PF4 antibodies bind to endogenous PF4 and form immune complexes16. These complexes interact with FcγRIIa to activate platelets and neutrophils9,16. To determine the effect of VITT antibodies in thrombosis, we first isolated total IgG from VITT patients' plasma and assessed its impact on whole blood from healthy donors. Compared to buffer (PBS) and normal IgG (Ctrl), incubation with VITT IgG led to a pronounced increase in the formation of LDG (Fig. 2a) and induction of neutrophils to undergo NETosis (CD15+CitH3+MPO+ cells) (Fig. 2b). By comparison, a HIT antibody, known to induce NETosis9, stimulated neutrophils to a comparable level (Fig. 2a, b). NETs induction in the absence of other cells was corroborated by treatment of purified neutrophils in the presence of PF4 with VITT or normal IgG and assessment of DNA release with the cell impermeant dye Sytox green. Significantly increased DNA release was triggered by VITT IgG relative to normal IgG (Fig 2c, d), indicating that purified IgG from VITT patients strongly initiated NETs formation in healthy donors' whole blood and purified neutrophils in vitro.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 608, 474, 624]]<|/det|>
+## VITT IgG induces thrombosis in vitro  
+
+<|ref|>text<|/ref|><|det|>[[115, 625, 880, 890]]<|/det|>
+To examine the capacity of VITT antibodies to induce thrombosis, fresh whole blood from healthy donors were treated with VITT IgG or normal IgG and flowed through von Willebrand factor (vWf)- coated microchannels in a microfluidics system. The presence of VITT IgG led to thrombus formation. Confocal microscopy imaging of thrombi formed following treatment with VITT IgG showed that the thrombi were formed by platelets, neutrophils and extracellular DNA, while no clots were formed in blood treated with control IgG (Extended data Fig. 1a, b). Further analysis of VITT IgG- induced thrombi showed an abundance of fibrin (Fig. 2e) and CitH3 (Fig. 2f) confirming the strong thrombogenic activity of VITT antibodies and their ability to induce NETosis in vitro. To confirm the role of FcγRIIa and NETosis in VITT- induced thrombosis, blood was pre- treated with anti- FcγRIIa monoclonal antibody, IV.3 or DNase I prior to incubation with VITT IgG. The presence of IV.3 strongly inhibited deposition of platelets (Fig. 2g, i) and neutrophils (Fig. 2g, j). Furthermore, there was no induction of NETosis as indicated by the absence of DNA release in the presence of IV.3 (Fig. 2g, h). Similarly, the presence of DNase I resulted in inhibition of thrombus formation (Fig. 2g- j). These data suggest that direct blocking of FcγRIIa inhibits NETosis and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 84, 880, 118]]<|/det|>
+1 thrombosis, and digestion of extracellular DNA also inhibits thrombus formation in vitro.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 134, 470, 150]]<|/det|>
+## VITT IgG induces thrombosis in vivo  
+
+<|ref|>text<|/ref|><|det|>[[115, 151, 880, 334]]<|/det|>
+To assess whether anti- PF4 antibodies are responsible for the clinical features of thrombocytopenia and thrombosis in VITT patients, we used a FcγRIIa+/hPF4+ double transgenic mouse model. These mice are necessary to assess the activity of VITT IgG in vivo, since they express two essential components, human PF4 and FcγRIIa on platelets and neutrophils. VITT IgG was administered into the VITT mouse model and lungs extracted to examine the levels of thrombosis. Examination of extracted lungs (Fig. 3a, Extended Data Fig. 1c) from VITT IgG- treated mice showed extensive thrombi deposition in this organ. Thrombi were absent in control IgG treated animals (Fig 3a, Extended Data Fig. 1c). These clots contain abundant platelets, neutrophils (Fig. 3) and fibrin (Fig. 3c). These data suggest that VITT IgG is responsible for thrombosis in vivo.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 348, 600, 365]]<|/det|>
+## Role of FcγRIIa and NETosis in thrombosis in vivo  
+
+<|ref|>text<|/ref|><|det|>[[115, 366, 880, 566]]<|/det|>
+To investigate the role of FcγRIIa and neutrophil activation and NETosis in VITT IgG- induced thrombosis, inhibitors of FcγRIIa (aglycosylated IV.3<sup>9</sup>, agIV.3) and NETosis (GSK484)<sup>9,17</sup> were administered in vivo. In support of our in vitro findings (Fig. 2) blocking either FcγRIIa or NETosis was effective in preventing the formation of clots in vivo as shown by the lack of clots (platelet and neutrophil accumulation) in lung sections of mice treated with agIV.3 (Fig. 3c). The dramatic reduction in thrombus deposition is also confirmed using whole organ imaging (Fig. 4a, b) and quantitative analysis of platelet accumulation in mice treated with VITT IgG plus agIV.3 or GSK484 compared to mice treated with VITT IgG without either inhibitor (Fig. 4a, b). Moreover, inhibitor- treated mice were not only protected from thrombosis but also had significantly less low density granulocytes present in peripheral blood compared to control mice (Extended Data Fig. 1d).  
+
+<|ref|>text<|/ref|><|det|>[[115, 581, 880, 748]]<|/det|>
+The contribution of NETosis to thrombosis in VITT was further assessed using the VITT mouse model deficient in PAD4 (FcγRIIa+/hPF4+/PAD4<sup>- /-</sup>). PAD4 is the enzyme responsible for the citrullination of histones necessary for induction of NETosis<sup>18</sup>. Consistent with findings in animals treated with VITT IgG plus GSK484, mice lacking PAD4 treated with VITT IgG had a dramatic reduction in clot formation compared to control VITT mice (which are wild type for PAD4) (Fig. 3c, Fig. 4a, b). There were also significantly fewer circulating low density granulocytes in PAD4 deficient mice compared to control (Extended Data Fig. 1d). Collectively, our data indicate that inhibition of platelet and neutrophil activation by blocking FcγRIIa or inhibition of NETosis can efficiently abolish VITT IgG- induced thrombosis in vivo.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 764, 675, 781]]<|/det|>
+## VITT IgG induces thrombocytopenia in VITT mouse model  
+
+<|ref|>text<|/ref|><|det|>[[115, 782, 880, 897]]<|/det|>
+Unlike mice treated with normal IgG, mice treated with VITT IgG experienced thrombocytopenia (Fig. 4c, d) and systemic reactions such as hypothermia (Extended Data Fig. 2a). AgIV.3 was effective in preventing both thrombocytopenia (Fig. 4d) and thrombosis (Fig. 3c, Fig. 4a). In contrast, NETosis inhibitor GSK484 and PAD4 knockout had no effect on the development of thrombocytopenia (Fig. 4e) although they strongly inhibited thrombosis (Fig. 4a, b). Altogether, these results indicate that VITT IgG- induced thrombosis and thrombocytopenia are distinct processes.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 228, 100]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[118, 101, 880, 183]]<|/det|>
+Although vaccines against COVID- 19 infection have been very effective, there have been some serious side- effects. One of these is a rare clotting disorder termed VITT or TTS that has caused significant morbidity and mortality. It has generated much public concern globally resulting in vaccine hesitancy and undermining of vaccine roll- out in many jurisdictions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 880, 331]]<|/det|>
+Despite numerous recent publications \(^{1,19 - 21}\) , there are still significant knowledge gaps in VITT, in particular regarding its underlying disease mechanism(s) \(^{4}\) . There is yet no direct evidence that VITT antibodies cause thrombosis and thrombocytopenia in vivo. Better knowledge of VITT will improve public confidence which might contribute to increased vaccine uptake. The conventional concept is that the platelet activating anti- PF4 antibody causes clot formation in VITT despite no direct evidence. The presence of anti- PF4 antibodies in individuals without thrombosis \(^{7}\) has created doubts about this concept.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 346, 649, 363]]<|/det|>
+## Evidence that anti-PF4 antibody induces clot formation  
+
+<|ref|>text<|/ref|><|det|>[[118, 364, 880, 412]]<|/det|>
+Here we provide evidence that VITT antibodies directly induce thrombus formation in vitro and in vivo, not by platelet activation alone but also through neutrophil activation and NETosis.  
+
+<|ref|>text<|/ref|><|det|>[[116, 428, 880, 627]]<|/det|>
+In this study, we demonstrated that VITT IgG stimulated platelet activation via serotonin release and platelet aggregation assays as had other investigators previously \(^{22}\) . More importantly, we further showed that VITT IgG when added in vitro to circulating whole blood induced clot formation in the microchannels using a microfluidics system. In contrast, normal IgG failed to induce thrombosis. Similarly, administration of VITT IgG but not normal IgG led to development of multiple thrombi in the lungs of the VITT mouse model (FcγRIIa+/hPF4+ double transgenic mice). These data provide direct evidence that VITT IgG (or more specifically immune complexes formed by VITT IgG and PF4) induced clot formation in vitro and in vivo, filling a crucial knowledge gap in VITT pathogenesis. Thrombosis was blocked by anti- FcγRIIa monoclonal antibody, IV.3 suggesting that it was mediated by FcγRIIa receptors on platelets and neutrophils.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 644, 585, 660]]<|/det|>
+## NETosis is a critical driver of thrombosis in VITT  
+
+<|ref|>text<|/ref|><|det|>[[116, 661, 880, 907]]<|/det|>
+Our study and the study by Holm et al \(^{21}\) both showed the presence of NETosis in patients with VITT. In the latter study, NETosis markers occurred together with numerous markers of inflammation, activated innate immune pathways, activated blood cells and endothelium, and damaged tissues in VITT patients. These markers were present in the blood, in a thrombus and in the immune precipitates extracted by a goat anti- human PF4 antibody from plasma of VITT patients. These findings were not unexpected as the VITT patients had robust immune responses, intense inflammation and severe thromboses. However, there was no data implicating NETosis as the cause of thrombosis in the VITT patients. Even the presence of neutrophils and NETosis markers in the thrombus does not necessary indicate that it is the cause of thrombosis as neutrophils and NETs are frequently observed in thrombi in various conditions including stroke, acute myocardial infarction \(^{23,24}\) and deep vein thrombosis \(^{25}\) . The authors, in fact, speculated in their report whether the adenovirus in the vaccine or even the spike protein could have triggered the pronounced inflammatory processes including NETosis.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 84, 881, 183]]<|/det|>
+1 In contrast, our study not only provides evidence of the presence of NETosis in VITT, 2 but we also show that NETosis directly drives thrombosis in VITT in vivo in the VITT 3 animal model. Administration of VITT IgG but not normal IgG induced development of 4 multiple thrombi in the lungs of the mice. Thrombosis could be prevented or 5 substantially suppressed by administration of NETosis inhibitor, GSK484 or by using 6 PAD4 knock- out mice (which blocks NETosis).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 199, 466, 216]]<|/det|>
+## VITT IgG induces thrombocytopenia  
+
+<|ref|>text<|/ref|><|det|>[[117, 216, 880, 334]]<|/det|>
+We demonstrated here that VITT antibodies induced thrombocytopenia (platelet count decrease) in the VITT mouse model by binding to platelet FcγRIIa. Thrombocytopenia was substantially prevented by anti- FcγRIIa monoclonal antibody, IV.3. In contrast, NETosis inhibitor GSK484 and absence of PAD4 (FcγRIIa+/hPF4+/PAD4+/- mice) which significantly blocked thrombosis in VITT had no effect on thrombocytopenia, suggesting that thrombosis and thrombocytopenia in VITT are two distinct processes as we have previously shown in heparin- induced thrombocytopenia<sup>9</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[117, 349, 880, 466]]<|/det|>
+In summary, our findings showed that anti- PF4 antibodies are the pathogenic or disease- causing antibodies in VITT. They induce platelet and neutrophil activation leading to development of NETosis which is the major driver of thrombosis in VITT (Extended Data Fig. 2b). FcγRIIA blockage prevented both thrombocytopenia and thrombosis but NETosis inhibition which effectively suppressed thrombosis, had no effect on thrombocytopenia. Thrombosis and thrombocytopenia appeared to be mediated by two distinct mechanisms.  
+
+<|ref|>text<|/ref|><|det|>[[117, 480, 880, 549]]<|/det|>
+Our results have contributed to a better understanding of pathogenesis in VITT and may also lead to development of disease biomarkers and improved diagnosis and new more efficacious therapies for VITT and consequently better clinical outcomes for the patients.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 203, 100]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 118, 275, 133]]<|/det|>
+## Human samples  
+
+<|ref|>text<|/ref|><|det|>[[115, 134, 880, 315]]<|/det|>
+VITT samples were collected from patients in Australia from the following hospitals: St George Hospital, Kogarah, Sydney, New South Wales; Calvary Mater Hospital, Wallsend, New South Wales; Box Hill Hospital, Box Hill, Victoria; University Hospital Geelong, Geelong, Victoria and Townsville University Hospital, Townsville, Queensland. Blood was collected from patients clinically diagnosed with HIT and VITT and positive for laboratory tests (ELISA and serotonin release assay) \(^{5,26}\) . Blood from healthy donors was used in control experiments. This study was approved by the Human Research Ethics Committee of South Eastern Sydney Local Health District (17/211 LNR/17/POWH/501). Informed consent was obtained from all study participants. Sera and plasma samples were stored in aliquots at \(- 80^{\circ}C\) until required for analysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 331, 296, 347]]<|/det|>
+## Diagnostic assays  
+
+<|ref|>text<|/ref|><|det|>[[115, 348, 880, 460]]<|/det|>
+The abundance of anti- PF4 or anti- PF4/heparin antibodies in patient sera was measured using a solid phase PF4 or PF4/heparin ELISA performed in microwell plates. Sera from patients or healthy donors were added to each well and incubated for 60 min at room temperature and then washed. Conjugated anti- human IgG was added, incubated for 60 min at room temperature and washed. Chromogenic substrate reaction was stopped with 1 M \(\mathsf{H}_2\mathsf{SO}_4\) . Optical density was measured using an automatic plate reader (Tecan Infinite Pro).  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 880, 578]]<|/det|>
+\(^{14}\mathrm{C}\) serotonin- release assay ( \(^{14}\mathrm{C}\) - SRA) was performed as previously described \(^{27}\) . Briefly, washed donor platelets were incubated with radiolabelled \(^{14}\mathrm{C}\) and heat inactivated patient's sera, in the presence and absence of PF4 (10 \(\mu \mathrm{g / mL}\) ), 0.1 U/mL heparin, IV.3 antibody (50 \(\mu \mathrm{g / mL}\) ) or 100 U/mL heparin, for 60 min at room temperature while stirring. Reaction was stopped using PBS- EDTA buffer and centrifuged. Radioactivity (counts per minute) of the supernatant was measured using a beta- counter. Levels greater than \(20\%\) were considered positive.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 595, 225, 610]]<|/det|>
+## Antibodies  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 880, 777]]<|/det|>
+Purification of immunoglobulin G antibodies from patients' or healthy donor's sera was performed using Protein G Agarose (Roche Mannheim, Germany). The AKTA purifier chromatography system (GE Healthcare) was used for purification. Eluted peak fractions were pooled and concentrated using ultracentrifugal units. Purity of IgG was \(>95\%\) as determined by SDS PAGE Gel analysis. Functional activity of purified IgG was determined by platelet aggregation and serotonin release assays. Hybridoma cells producing IV.3 were obtained from ATCC (clone HB- 217). Cells were cultured in DMEM medium containing \(10\%\) FBS at \(37^{\circ}\mathrm{C}\) , \(5\%\) CO2. Cells were cultured in serum- free DMEM 24h prior to collection of antibody- containing supernatant. Protein G Sepharose affinity chromatography was used to purify IV.3.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 792, 312, 808]]<|/det|>
+## Platelet aggregation  
+
+<|ref|>text<|/ref|><|det|>[[115, 809, 880, 908]]<|/det|>
+Light transmission platelet aggregometry was used to determine antibody activity and role of FcγRIIa in VITT- induced platelet aggregation. Platelet- rich plasma (PRP) was prepared from citrate- anticoagulated healthy donor blood by centrifugation at room temperature at \(150\mathrm{g}\) for 10 min. \(50~\mu \mathrm{L}\) of VITT or normal sera was added to a cuvette with \(300~\mu \mathrm{L}\) of PRP with or without FcγRIIa- inhibitor, IV.3 (20 \(\mu \mathrm{g / mL}\) ), whilst stirring at \(37^{\circ}\mathrm{C}\) for 15 min. Platelet poor plasma was used as blank.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[120, 85, 450, 101]]<|/det|>
+## Quantification of NETosis markers  
+
+<|ref|>text<|/ref|><|det|>[[118, 101, 880, 183]]<|/det|>
+Quantification of NETosis markersCell- free DNA was measured in plasma of VITT and healthy donor samples using Quant- iT™ PicoGreen™ dsDNA assay kit (P11496, Invitrogen), as described by the manufacturer. Plasma levels of citrullinated histone H3 were determined using the H3R8Cit ELISA Capture and Detection kit (R&D143002, EpiCypher)14 following the manufacturer's instructions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 200, 245, 215]]<|/det|>
+## Cell isolation  
+
+<|ref|>text<|/ref|><|det|>[[118, 216, 880, 348]]<|/det|>
+Cell isolationNeutrophils were purified using EDTA- anticoagulated blood and the EasySep Direct Human Neutrophil Isolation kit (19666, StemCell Technologies) following the manufacturer's instructions. Purified neutrophils are free of platelets and other blood cells as assessed by flow cytometry. Washed platelets were prepared from citrate- anticoagulated blood. For low density granulocytes, whole blood was diluted with PBS and Lymphoprep (07851, StemCell Technologies) was gently underlayed to avoid mixing with the diluted blood. Sample was then centrifuged at \(800 \times \mathrm{g}\) for 20 min at room temperature. Peripheral blood mononuclear cell layer was harvested.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 364, 268, 379]]<|/det|>
+## Flow cytometry  
+
+<|ref|>text<|/ref|><|det|>[[117, 380, 880, 528]]<|/det|>
+Flow cytometryFresh citrate- anticoagulated blood from VITT patients or healthy donors was diluted with PBS. Platelet- neutrophil aggregates were analysed using anti- CD15 (Alexa Fluor 647, BD 562369) and anti- CD41a (PE, BD 555467), NETs were identified using anti- citrullinated histone H3 (ab5103), anti- MPO (PE, BD 341642) and goat anti- rabbit IgG (BV421, BD 565014). Monocytes and low density granulocytes were identified using anti- CD14 (V500, BD 561391) and anti- CD15 (Alexa Fluor 647, BD 562369) or anti- Ly6G (V450, BD560603) and anti- CD11b (PE, BD 557397). Platelet counts in mouse blood were determined by number of events acquired in 60s relative to time 0. Flow cytometry data were analysed using FlowJo software (LCC, USA.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 544, 219, 559]]<|/det|>
+## Timelapse  
+
+<|ref|>text<|/ref|><|det|>[[117, 560, 880, 675]]<|/det|>
+TimelapsePurified neutrophils were stained with Hoechst 33342 (14533, Sigma) and seeded into eight- well Nunc Lab- Tek II chambers. Purified VITT IgG (5 mg/mL) or normal IgG (5 mg/mL) with PF4 (12 \(\mu \mathrm{g} / \mathrm{mL}\) ) were added to each reaction. Release of extracellular DNA was measured using Sytox Green (S7020, Invitrogen). Wells were imaged using a confocal laser- scanning microscope (Leica TCS SP8). Sytox green fluorescence relative to Hoechst 33342 fluorescence was calculated with ImageJ software (version 2.1.0/1.53c, NIH).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 692, 245, 707]]<|/det|>
+## Microfluidics  
+
+<|ref|>text<|/ref|><|det|>[[116, 708, 880, 890]]<|/det|>
+MicrofluidicsCitrate- anticoagulated blood was diluted 1:2 with PBS, supplemented with purified IgG (VITT IgG 3 mg/mL, normal IgG 3 mg/mL) and incubated at \(37^{\circ}\mathrm{C}\) for 90 min. In selected experiments, blood was pre- incubated with IV.3 (20 \(\mu \mathrm{g} / \mathrm{mL}\) ) or DNase I (160 U/mL). Blood was stained with combinations of Hoechst 33342 (3 \(\mu \mathrm{g} / \mathrm{mL}\) ), Sytox green (0.3 \(\mu \mathrm{M}\) ), anti- CD41 Alexa 647 (15 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- CD41- FITC (15 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- CD15 Alexa 647 (15 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- fibrin Alexa 594 (30 \(\mu \mathrm{g} / \mathrm{mL}\) ), anti- CitH3 Alexa 594 (30 \(\mu \mathrm{g} / \mathrm{mL}\) ) prior to perfusion through Vena8 Fluoro+™ biochip microchannels coated with vWf (Haematologic Technologies United BioResearch Products Pty Ltd). Biochips were mounted on a fluorescent microscope (Zeiss Axio Observer.A1) and fluorescence images from different microscopic fields were captured in real time with a Q- Imaging EXi Blue™ camera (Q- Imaging, Surry, BC, Canada) with the fluid shear
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 84, 880, 120]]<|/det|>
+stress set at 67 dyne/cm² (shear rate 1500/s) for 30 min. Selected samples were fixed with 2% paraformaldehyde and imaged by confocal microscopy.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 134, 250, 149]]<|/det|>
+## Mouse model  
+
+<|ref|>text<|/ref|><|det|>[[115, 150, 880, 346]]<|/det|>
+Mouse modelMice expressing the \(\mathrm{R^{131}}\) isoform of human FcyRlla and human PF4 were generated in C57BL/6 background. Double transgenic (FcyRlla+/hPF4+) and FcyRlla+/hPF4+/PAD4- have been characterised previously<sup>9,28</sup>. VITT was recreated in these mice by intravenous injection of purified VITT IgG (250 \(\mu \mathrm{g / g}\) ). Inhibitors of NETosis (GSK484, Cayman chemicals) or anti FcyRlla (aglycosylated IV.3, 1 \(\mu \mathrm{g / g}\) ) were injected at time 0. Anti-CD42c Dylight- 649 antibody (Emfret, Germany) and Alexa Fluor 594- fibrinogen were used to label mouse platelets and fibrin in vivo, respectively. Following euthanasia, lungs were perfused with PBS followed by formalin, extracted and imaged using the IVIS Spectrum (Perkin Elmer). Fluorescence was calculated in radiant efficiency using living Image 4.5.5 software (Perkin Elmer). All animal experiments were approved by the University of New South Wales Animal Care and Ethics Committee.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 362, 213, 378]]<|/det|>
+## Histology  
+
+<|ref|>text<|/ref|><|det|>[[117, 378, 880, 495]]<|/det|>
+HistologyFormalin- fixed lungs were embedded in paraffin, sectioned at 4 microns and mounted onto slides. Slides were deparaffinised, rehydrated, and underwent heat- induced antigen retrieval. Slides were probed with anti- Ly6G (Alexa Fluor 488, 127626 Biolegend). Vectashield antifade mounting medium with DAPI (H- 1200, Vector Laboratories) was used to mount glass coverslips onto the slides. Slides were imaged by confocal microscopy. Slides were also stained with H&E and imaged with a Zeiss Axioskop microscope.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 510, 306, 526]]<|/det|>
+## Statistical analyses  
+
+<|ref|>text<|/ref|><|det|>[[117, 526, 880, 660]]<|/det|>
+Statistical tests were performed using GraphPad Prism version 8 (GraphPad Software, USA). The following statistical tests were used in this study as described in the figure legends: (1) Shapiro- Wilk normality test. (2) Student's t test was performed when comparing between two groups. (3) Multiple comparisons were analysed by one- way ANOVA with post- test correction for multiple comparisons. Each individual healthy donor for in vitro experiments and each mouse used for animal experiments was considered a biological replicate. P values \(< 0.05\) were considered statistically significant.  
+
+<|ref|>text<|/ref|><|det|>[[115, 674, 880, 870]]<|/det|>
+Acknowledgments: The authors wish to thank Steven McKenzie (Philadelphia, USA) for providing FcyRlla+/hPF4+ mice, Drs Feng Yan, Rose Wong and Kathryn Evans for valuable technical assistance, Drs Sumita Ratnasingam, John Cassey and Silvia Zheng for management of VITT patients and valuable clinical input, O Szeto, J Bennett, M Poxton, E Heyer and P Rojanski for assistance in obtaining human research ethics/governance approvals, and members of the THANZ VITT Advisory Group for helpful discussion of VITT patients. This work was supported by grants from National Health and Medical Research Council, Australia, Program Grant APP1052616 and New South Wales Capacity Program Senior Researcher Grant RG201677 to BHC; NSW Health Cardiovascular Disease Clinician Scientist Grant and National Health and Medical Research Council Australia, Investigator Grant to JJHC
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[70, 83, 881, 234]]<|/det|>
+1 Author contributions: BHC conceived the idea, designed and supervised the 2 research, analysed the data and wrote the manuscript, HL and JP designed and 3 carried out the experiments, collected and analysed the data and wrote the 4 manuscript, ZA performed platelet function assays and microfluidic studies, collected 5 and analysed the data, FR carried out histology and immunochemistry studies, 6 collected and analysed the data, JC provided conceptual input, designed experiments 7 and analysed data, ST and AA provide intellectual input, analysed clinical data and 8 managed VITT patients. All authors reviewed and edited the manuscript and approved 9 the final version of the manuscript. 10 11 Conflict of interest statement. The authors declare no conflicts of interest.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 84, 229, 100]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[113, 128, 880, 884]]<|/det|>
+1. Greinacher, A. et al. Thrombotic Thrombocytopenia after ChAdOx1 nCov-19 Vaccination. N Engl J Med 384, 2092-2101, (2021).  
+2. Scully, M. et al. Pathologic Antibodies to Platelet Factor 4 after ChAdOx1 nCov-19 Vaccination. N Engl J Med 384, 2202-2211, (2021).  
+3. Pavord, S. et al. Clinical Features of Vaccine-Induced Immune Thrombocytopenia and Thrombosis. N. Engl. J. Med., (2021).  
+4. Cines, D. B. & Bussel, J. B. SARS-CoV-2 Vaccine-Induced Immune Thrombotic Thrombocytopenia. N Engl J Med 384, 2254-2256, (2021).  
+5. Schultz, N. H. et al. Thrombosis and Thrombocytopenia after ChAdOx1 nCov-19 Vaccination. N Engl J Med 384, 2124-2130, (2021).  
+6. Chong, B. H. Evolving concepts of pathogenesis of heparin-induced thrombocytopenia: Diagnostic and therapeutic implications. Int. J. Lab. Hematol. 42, 25-32, (2020).  
+7. Thiele, T. et al. Frequency of positive anti-PF4/polyanion antibody tests after COVID-19 vaccination with ChAdOx1 nCov-19 and BNT162b2. Blood 138, 299-303, (2021).  
+8. Leung, H. et al. Inhibition of NADPH oxidase blocks NETosis and reduces thrombosis in heparin-induced thrombocytopenia. Blood Advances, (2021).  
+9. Perdomo, J. et al. Neutrophil activation and NETosis are the major drivers of thrombosis in heparin-induced thrombocytopenia. Nat. Commun. 10, 1322, (2019).  
+10. Folco, E. J. et al. Neutrophil Extracellular Traps Induce Endothelial Cell Activation and Tissue Factor Production Through Interleukin-1α and Cathepsin G. Arterioscler. Thromb. Vasc. Biol. 38, 1901-1912, (2018).  
+11. Thälin, C., Hisada, Y., Lundström, S., Mackman, N. & Wallén, H. Neutrophil Extracellular Traps. Arterioscler. Thromb. Vasc. Biol. 39, 1724-1738, (2019).  
+12. Fuchs, T. A. et al. Extracellular DNA traps promote thrombosis. Proc. Natl. Acad. Sci. U. S. A. 107, 15880, (2010).  
+13. Laridan, E., Martinod, K. & De Meyer, S. F. Neutrophil Extracellular Traps in Arterial and Venous Thrombosis. Semin. Thromb. Hemost. 45, 86-93, (2019).  
+14. Thälin, C. et al. Quantification of citrullinated histones: Development of an improved assay to reliably quantify nucleosomal H3Cit in human plasma. J. Thromb. Haemost. 18, 2732-2743, (2020).  
+15. Masuda, S. et al. NETosis markers: Quest for specific, objective, and quantitative markers. Clin. Chim. Acta 459, 89-93, (2016).  
+16. Nguyen, T.-H., Medvedev, N., Delcea, M. & Greinacher, A. Anti-platelet factor 4/polyanion antibodies mediate a new mechanism of autoimmunity. Nat. Commun. 8, 14945, (2017).  
+17. Lewis, H. D. et al. Inhibition of PAD4 activity is sufficient to disrupt mouse and human NET formation. Nat. Chem. Biol. 11, 189-191, (2015).  
+18. Thiam, H. R. et al. NETosis proceeds by cytoskeleton and endomembrane disassembly and PAD4-mediated chromatin decondensation and nuclear envelope rupture. Proc. Natl. Acad. Sci. U. S. A. 117, 7326, (2020).  
+19. Huynh, A., Kelton, J. G., Arnold, D. M., Daka, M. & Nazy, I. Antibody epitopes in vaccine-induced immune thrombotic thrombocytopenia. Nature, (2021).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 84, 870, 432]]<|/det|>
+20. Greinacher, A. et al. Anti-Platelet Factor 4 Antibodies Causing VITT do not Cross-React with SARS-CoV-2 Spike Protein. Blood, (2021).  
+21. Holm, S. et al. Immune complexes, innate immunity, and NETosis in ChAdOx1 vaccine-induced thrombocytopenia. Eur. Heart J., (2021).  
+22. Althaus, K. et al. Antibody-mediated procoagulant platelets in SARS-CoV-2-vaccination associated immune thrombotic thrombocytopenia. Haematologica 106, 2170-2179, (2021).  
+23. Novotny, J. et al. Thrombus NET content is associated with clinical outcome in stroke and myocardial infarction. Neurology 94, e2346, (2020).  
+24. Chilingaryan, Z. et al. Erythrocyte interaction with neutrophil extracellular traps in coronary artery thrombosis following myocardial infarction. Pathology, (2021).  
+25. Brill, A. et al. Neutrophil extracellular traps promote deep vein thrombosis in mice. J. Thromb. Haemost. 10, 136-144, (2012).  
+26. Chong, B. H. & Isaacs, A. Heparin-induced thrombocytopenia: What clinicians need to know. Thromb. Haemost. 101, 279-283, (2009).  
+27. Sheridan, D., Carter, C. & Kelton, J. G. A diagnostic test for heparin-induced thrombocytopenia. Blood 67, 27-30, (1986).  
+28. Reilly, M. P. et al. Heparin-induced thrombocytopenia/thrombosis in a transgenic mouse model requires human platelet factor 4 and platelet activation through FcgammaRIIA. Blood 98, 2442-2447, (2001).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[23, 80, 916, 641]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[25, 670, 272, 900]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[297, 670, 483, 900]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[515, 670, 722, 900]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[770, 667, 956, 900]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 100, 880, 431]]<|/det|>
+Figure 1. Platelet activation and NETosis in VITT. a PF4 and PF4- heparin ELISA experiment of VITT serum and normal controls. The cut- off, 0.50 OD units. b \(^{14}\mathrm{C}\) - serotonin release assay for VITT samples with buffer alone, PF4 (10 \(\mu \mathrm{g} / \mathrm{mL}\) ), 0.1 or 100 U/mL heparin or IV.3 antibody (50 \(\mu \mathrm{g} / \mathrm{mL}\) ). Each dot represents the mean of assays done in triplicate. The cut- off was set at \(20\%\) CPM. c Platelet aggregation responses. Purified IgG from VITT patients induced aggregation in platelet rich plasma (red and blue traces). Blockage of \(\mathrm{Fc\gamma / R} \mathrm{lla}\) with IV.3 inhibited aggregation (purple and green traces). d Nucleosomal CitH3 levels in VITT patients' plasma (n=7) relative to normal controls (n=8) was determined by H3R8Cit ELISA. e cfDNA in VITT patients' plasma (n=7) relative to normal controls (n=6) determined by PicoGreen fluorescence assay. f Representative side and forward scatter flow cytometry plot backgated for neutrophils (yellow) and monocytes (blue) from VITT patient's and normal blood. LDG are indicated. g Representative plot of NPA from VITT and normal blood. h Quantification of NPA in VITT. i Representative plot of NETs from VITT and normal blood. j quantification of NETs in VITT. MPO\*, CitH3\* double positive cells within the CD15+ population were defined as neutrophils undergoing NETosis. The percentage of gated events is indicated in each quadrant. Statistics, Mann- Whitney test. \*P < 0.05; \*\*P < 0.01; \*\*\*\*P < 0.0001. OD, optical density units; CPM, counts per minute; NPA, neutrophil-platelet aggregates; LDG, low density granulocytes; cfDNA, cell- free DNA; CitH3, citrullinated histone H3; Pt, patient.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[25, 45, 971, 970]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 880, 397]]<|/det|>
+Figure 2. Effect of VITT IgG on donor's blood. a Quantification of LDG and b NETs following treatment of healthy donor blood with VITT, normal controls and HIT IgG. c Purified neutrophils treated with VITT IgG or normal IgG plus PF4 were stained for extracellular DNA (green) and nuclei (blue). d DNA release calculated as fluorescence intensity ratio of extracellular DNA (Sytox staining)/total DNA (Hoechst staining) vs. time (n=3). e VITT IgG induces thrombosis. Healthy donors' blood treated with VITT IgG was stained for total DNA (blue), platelets (green), fibrin (red) and neutrophils (magenta). Thrombi were imaged with a confocal laser-scanning microscope (overlap of green and red shown as yellow). Scale bar: \(10 \mu \mathrm{m}\) . f Thrombi contain CthH3. Thrombi were generated and imaged as in (e), and stained for DNA (blue), platelets (green), CthH3 (yellow) and neutrophils (magenta). Overlap of yellow and green is shown as white. g IV.3 and DNase I prevent VITT IgG-induced thrombus formation in microfluidics system. Treated blood was stained for DNA (blue), platelets (green) and neutrophils (red). Scale bar: \(50 \mu \mathrm{m}\) . Graphs show area coverage percentage for h total DNA, i platelets and j neutrophils. n=3, mean \(\pm\) s.d. Statistics: (a, b) Kruskal-Wallis test with uncorrected Dunn's test, (d) One-way ANOVA followed by Dunn's test for multiple comparisons, (h, i, j) One-way ANOVA with Tukey's correction for multiple comparisons. \*P < 0.05; \*\*P < 0.01; \*\*\*P < 0.001, \*\*\*\*P < 0.0001. LDG, low density granulocytes; ext. DNA, extracellular DNA; Ctrl, control; Pt, patient.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[30, 45, 927, 970]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 881, 266]]<|/det|>
+Figure 3. VITT IgG induces thrombosis in FcyRIIa+/hPF4+ mice. a Representative H&E staining of lung sections of mice treated with nlgG or VITT IgG. Scale bar 50 μm. b Fluorescent images of lung sections of mice treated with VITT IgG. Platelets were labelled in vivo with anti- CD42c- Dylight 649 (magenta). Neutrophil were stained with anti- Ly6G (green). Neutrophil infiltration in the clot is shown. Cell nuclei were stained with DAPI (blue). Scale bars 50 μm. c Fluorescent images of lung sections of FcyRIIa+/hPF4+ mice treated with nlgG, VITT IgG or VITT IgG plus agIV.3 or FcyRIIa+/hPF4+/PAD4+/- mice treated with VITT IgG. Fibrin labelled with AF594 (red) resulted from injection of AF594- labelled fibrinogen at 1 μg/g. Platelets were labelled in vivo with anti- CD42c- Dylight 649 (magenta). Cell nuclei were stained with DAPI (blue). Scale bar 10 μm. nlgG, normal IgG; agIV.3, aglycosylated IV.3 antibody.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[80, 66, 950, 560]]<|/det|>
+ 
+
+<|ref|>image<|/ref|><|det|>[[68, 596, 355, 768]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 314]]<|/det|>
+Figure 4. Thrombosis and thrombocytopenia. a Representative images of lungs following treatment. The level of fluorescence indicates accumulation of platelets labelled with anti- CD42c- Dylight 649 in the lungs. b Graph of lung fluorescence for the VITT patients indicated. c Representative graph showing platelet counts following treatment of FcyRlla+/hPF4+ mice with normal IgG (nIgG) or VITT IgG or VITT IgG plus agIV.3 determined at 1h and 4h after treatment. d Quantification of platelet counts in FcyRlla+/hPF4+ mice following the treatments indicated in the figure. e Graph showing platelet counts following treatment of FcyRlla+/hPF4+ mice with VITT IgG with or without GSK or FcyRlla+/hPF4+/PAD4+/- mice plus VITT IgG determined at 1h and 4h after treatment. Statistics. b One- way ANOVA with Dunnet's test for multiple comparisons. Unpaired t test for comparison between Pt5 in FcyRlla+/hPF4+ and Pt 5 in FcyRlla+/hPF4+/PAD4+/- mice. d One- way ANOVA with Dunnet's test for multiple comparisons. nIgG, normal IgG; PAD4 KO, PAD4 knockout FcyRlla+/hPF4+ mice; agIV.3, aglycosylated IV.3 antibody; Pt, patient.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[45, 40, 874, 800]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[339, 19, 678, 38]]<|/det|>
+<center>Thrombosis in vitro and in vivo</center> 
+
+<|ref|>image_caption<|/ref|><|det|>[[339, 944, 600, 959]]<|/det|>
+<center>VITT IgG</center> 
+
+<|ref|>image<|/ref|><|det|>[[339, 800, 610, 972]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 880, 298]]<|/det|>
+Extended Data Figure 1. a VITT IgG and thrombosis. Healthy donors' blood treated with VITT IgG was flowed in vWf- coated microchannels. Extracellular DNA was stained with Sytox green (green), platelets with anti- CD41 AF647 (magenta) and neutrophils with anti- CD15 AF594 (red). Thrombi were imaged with a confocal laser- scanning microscope (Leica TCS SP8 running Leica's LAS X software) with a 63x oil immersion objective. Scale bar 20 μm. b Healthy donors' blood treated with normal IgG was flowed in vWf- coated microchannels. Total DNA was stained with Hoechst 33342 (blue), platelets with anti- CD41- FITC (green) and neutrophils with anti- CD15 AF594 (red). Scale bar 50 μm. c Fluorescent images of lung lobes from mice treated with VITT IgG or control IgG. DAPI- stained nuclei (blue), platelet- rich thrombi (magenta). Scale bar 500 μm. d Level of low density granulocytes (LDG) in blood from mice following the treatments indicated in the Figure. Statistics: Unpaired t test. nlgG, normal IgG; PAD4 KO, PAD4 knockout FcγRIIa+/hPF4+ mice.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[80, 29, 919, 74]]<|/det|>
+# Systemic response and proposed mechanism of thrombocytopenia and thrombosis in VITT  
+
+<|ref|>image<|/ref|><|det|>[[50, 100, 888, 680]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[128, 730, 863, 880]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 880, 249]]<|/det|>
+Extended Data Figure 2. a Changes in temperature following the treatments indicated in the figure. Dotted line represents the mean temperature of mice before treatment \((38.3^{\circ}\mathrm{C}, \mathrm{n} = 30)\) . b Model of mechanism of thrombosis and thrombocytopenia in VITT. Anti- PF4 antibodies from VITT patients form a complex with PF4 and interact with FcγRIIa. Interaction of the complex with platelets results in thrombocytopenia, which can be blocked with the monoclonal antibody IV.3. In the case of neutrophils, the interaction of the complex with FcγRIIa leads to NETs formation and subsequent thrombosis. Thrombosis can be blocked by neutralisation of FcγRIIa with IV.3 or by inhibition of NETosis using NETs inhibitor or in PAD4 knockout mice. In vitro, addition of DNase I disrupts thrombus formation.
+
+<--- Page Split --->

preprint/preprint__b3507d9826a318cafe485c6b8d4bed844b52604a56d4587e009677fe91f86a69/images_list.json

@@ -0,0 +1,32 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Identification of cell-specific exons and SLED vector construction (a) Diagramatic sketch of SLED vector design strategy. SLED is compatible with any promoter. A frameshifting mutation is introduced into a cell-specific alternative exon",
+    "footnote": [],
+    "bbox": [
+      [
+        80,
+        50,
+        920,
+        840
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. SLED vectors maintain cell-specific expression when delivered using AAV. (a) Diagramatic sketch of pan-neuronal (SLED.NPL), photoreceptor-specific (SLED.RAB), and excitatory neuron-specific (SLED.ENS) vectors designed for AAV packaging. In all SLED vectors, EGFP is translated when cell-specific exons are spliced-in. (b) SLED.RAB, packaged in AAV2.7m8, was intravertically injected into P0 mouse retinas and processed at P30. EGFP is highly enriched in retinal photoreceptors while inner retinal neurons are strongly positive for dsRed. ONL = outer nuclear layer, INL = inner nuclear layer. (c-e) SLED.NPL, packaged in AAV9, was used to transduce primary rat hippocampal cultures (c), mouse cortex (d), and human iPSC-derived neurons (e). In all cases, EGFP is highly enriched in HuC/D+",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        40,
+        931,
+        710
+      ]
+    ],
+    "page_idx": 9
+  }
+]

preprint/preprint__b3507d9826a318cafe485c6b8d4bed844b52604a56d4587e009677fe91f86a69/preprint__b3507d9826a318cafe485c6b8d4bed844b52604a56d4587e009677fe91f86a69.mmd

@@ -0,0 +1,487 @@
+
+# Cell-specific regulation of gene expression using splicing-dependent frameshifting.  
+
+Seth Blackshaw ( \(\boxed{\bullet}\) sblack@jhm.edu) Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 1338- 8476  
+
+Jonathan Ling Johns Hopkins University School of Medicine https://orcid.org/0000- 0003- 1927- 9729  
+
+Alexei Bygrave Johns Hopkins University School of Medicine  
+
+Clayton Santiago Johns Hopkins University School of Medicine https://orcid.org/0000- 0001- 7191- 668X  
+
+Roger Carmen- Orozco Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 2460- 9270  
+
+Vickie Trinh Johns Hopkins University School of Medicine  
+
+Minzhong Yu Cleveland Clinic  
+
+Yini Li Johns Hopkins University School of Medicine https://orcid.org/0000- 0003- 2213- 0650  
+
+Jeong Han Johns Hopkins University School of Medicine https://orcid.org/0000- 0003- 3277- 7669  
+
+Kamil Taneja Johns Hopkins University School of Medicine  
+
+Ying Liu Johns Hopkins University School of Medicine  
+
+Rochinelle Dongmo Johns Hopkins University School of Medicine  
+
+Travis Babola Johns Hopkins University School of Medicine  
+
+Patrick Parker Johns Hopkins University School of Medicine  
+
+Lizhi Jiang Johns Hopkins University School of Medicine  
+
+Patrick Leavey Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 2822- 1118  
+
+Jennifer Smith
+
+<--- Page Split --->
+
+Johns Hopkins University School of Medicine  
+
+Rachel Vistein Johns Hopkins University School of Medicine  
+
+Megan Gimmel Johns Hopkins University School of Medicine  
+
+Benjamin Dubner Johns Hopkins University School of Medicine  
+
+Patric Teodorescu Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 4129- 8478  
+
+Theodoros K arantanos Johns Hopkins University School of Medicine  
+
+Gabriel Ghiaur Johns Hopkins University School of Medicine  
+
+Patrick Kanold Johns Hopkins University https://orcid.org/0000- 0002- 7529- 5435  
+
+Dwight Bergles Johns Hopkins University https://orcid.org/0000- 0002- 7133- 7378  
+
+Ben Langmead Johns Hopkins University https://orcid.org/0000- 0003- 2437- 1976  
+
+Shuying Sun Johns Hopkins University School of Medicine  
+
+Kristina Nielsen Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 9155- 2972  
+
+Neal Peachy Cleveland Clinic  
+
+Mandeep Singh Johns Hopkins University School of Medicine  
+
+William Dalton Johns Hopkins University School of Medicine  
+
+Fatemeh Rajaii Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 1012- 2293  
+
+Richard Huganir Johns Hopkins School of Medicine https://orcid.org/0000- 0001- 9783- 5183  
+
+Article  
+
+Keywords:  
+
+Posted Date: March 24th, 2022
+
+<--- Page Split --->
+
+DOI: https://doi.org/10.21203/rs.3.rs- 1416757/v1  
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Version of Record: A version of this preprint was published at Nature Communications on October 1st, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33523- 2.
+
+<--- Page Split --->
+
+Title: Cell- specific regulation of gene expression using splicing- dependent frameshifting.  
+
+Authors: Jonathan P. Ling \(^{1,10,*,\#}\) , Alexei M. Bygrave \(^{2,\#}\) , Clayton P. Santiago \(^{2,\#}\) , Rogger P. Carmen- Orozco \(^{2}\) , Vickie Trinh \(^{2}\) , Minzhong Yu \(^{11,12}\) , Yini Li \(^{1}\) , Jeong Han \(^{4}\) , Kamil Taneja \(^{4}\) , Ying Liu \(^{4}\) , Rocinelle Dongmo \(^{2}\) , Travis A. Babola \(^{2,3}\) , Patrick Parker \(^{2}\) , Lizhi Jiang \(^{2}\) , Patrick J. Leavey \(^{2}\) , Jennifer J. Smith \(^{2,6}\) , Rachel Vistein \(^{2,6}\) , Megan Y. Gimen \(^{2}\) , Benjamin Dubner \(^{7}\) , Eric Helmenstine \(^{7}\) , Patric Teodorescu \(^{7}\) , Theodore Karantanos \(^{7}\) , Gabriel Ghiaur \(^{7}\) , Patrick O. Kanold \(^{2,3,10}\) , Dwight Bergles \(^{2,10}\) , Ben Langmead \(^{9,10}\) , Shuying Sun \(^{1}\) , Kristina J. Nielsen \(^{2,6,10}\) , Neal Peachey \(^{11,12,13}\) , Mandeep S. Singh \(^{4}\) , W. Brian Dalton \(^{7}\) , Fatemeh Rajaii \(^{4}\) , Richard L. Huganir \(^{2,10}\) , and Seth Blackshaw \(^{2,5,8,10,*}\)  
+
+Affiliations: \(^{1}\) Departments of Pathology, \(^{2}\) Solomon H. Snyder Department of Neuroscience, \(^{3}\) Biomedical Engineering, \(^{4}\) Wilmer Eye Institute, \(^{5}\) Neurology, \(^{6}\) Zanvyl Krieger Mind/Brain Institute, \(^{7}\) Oncology, and \(^{8}\) Institute for Cell Engineering, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA; \(^{9}\) Department of Computer Science and \(^{10}\) Kavli Neuroscience Discovery Institute, Johns Hopkins University, Baltimore, MD 21218, USA; \(^{11}\) Department of Ophthalmic Research, Cole Eye Institute, Cleveland Clinic Foundation, Cleveland, OH 44195, USA; \(^{12}\) Department of Ophthalmology, Cleveland Clinic Lerner College of Medicine of Case Western Reserve University, Cleveland, OH 44195, USA; \(^{13}\) Research Service, Louis Stokes Cleveland VA Medical Center, Cleveland, OH 44106, USA.  
+
+\(^{\#}\) indicates equal contributions \(*\) indicates corresponding authors
+
+<--- Page Split --->
+
+## Abstract  
+
+Precise and reliable cell- specific gene delivery remains technically challenging. Here we report a splicing- based approach for controlling gene expression whereby separate translational reading frames are coupled to the inclusion or exclusion of cell- specific alternative exons. Candidate exons are identified by analyzing thousands of publicly available RNA sequencing datasets and filtering by cell specificity, sequence conservation, and local intron length. This method, which we denote splicing- linked expression design (SLED), can be combined in a Boolean manner with existing techniques such as minipromoters and viral capsids. SLED vectors can leverage the strong expression of constitutive promoters, without sacrificing precision, by decoupling the tradeoff between promoter strength and selectivity. We generated SLED vectors to selectively target all neurons, photoreceptors, or excitatory neurons, and demonstrated that specificity was retained in vivo when delivered using AAVs. We further demonstrated the utility of SLED by creating what would otherwise be unobtainable research tools, specifically a GluA2 flip/flop reporter and a dual excitatory/inhibitory neuronal calcium indicator. Finally, we show the translational potential of SLED by rescuing photoreceptor degeneration in \(P r p h 2^{r d s / r d s}\) mice and by developing an oncolytic vector that can selectively induce apoptosis in SF3B1 mutant cancer cells. The flexibility of SLED technology enables new avenues for basic and translational research.  
+
+## Introduction  
+
+Cell type- specific control of gene expression is essential for both basic and translational biological research. Though this is often achieved using transgenic animal models, these are costly, difficult to scale, and restricted to a limited number of model organisms. An alternative approach, which is directly applicable for therapeutic purposes, is to use exogenous viral or plasmid constructs to selectively express genes of interest in specific cell types \(^{1 - 3}\) . These methods rely on the use of minimal promoters and enhancers that place constructs under the regulation of cell type- specific transcription factors \(^{4 - 8}\) , unique capsid proteins or surface features to limit the range of cell types infected by viral constructs \(^{9,10}\) , or the inclusion of specific miRNA seed sequences to inhibit off- target expression \(^{11 - 13}\) .  
+
+Current approaches, however, have important limitations. Minipromoter and enhancer- based constructs are difficult to develop and test in a systematic manner. For example, when removed from their genomic context, or tested in other species, they often show unpredictable patterns of cell- specific expression, despite showing high sequence conservation and patterns of chromatin accessibility \(^{14}\) . Furthermore, while viral serotypes typically provide enriched cell- specificity, thus far they are not strictly cell type- specific, and are not relevant for viral- independent gene delivery strategies. Likewise, microRNA- based approaches can help reduce off- target delivery in certain
+
+<--- Page Split --->
+
+cells, but must be used in conjunction with other methods to achieve cell type- specific expression.  
+
+An orthogonal strategy that can be combined with the above approaches would be to harness alternative splicing of mRNA (RNA splicing) to direct cell- specific gene expression. RNA splicing is a highly regulated process that generates transcriptomic and proteomic diversity and many splicing patterns are correlated with unique cell types or cellular states. Fluorescent reporter vectors have been used to study the mechanistic regulation of alternative splicing events \(^{15 - 18}\) , but the large size of most intronic sequences precludes their inclusion in the most commonly used viral vectors. Adeno- associated virus (AAV) vectors are a leading platform for gene therapy due to their demonstrated safety and long- term efficacy across a variety of tissues \(^{19 - 22}\) , but these viruses are limited by a maximum packaging size of \(\sim 4.7kb^{23}\) . Since the average intron length in the human genome is \(\sim 5.4kb\) in length \(^{24}\) , it has been historically difficult to identify cell type- specific patterns of alternative splicing that are potentially compatible with AAV vectors \(^{25,26}\) . However, rapid adoption of full- length RNA sequencing (RNA- Seq) over the past decade has led to the public archival of datasets obtained from various cell types across multiple species. Furthermore, recent computational methods have been developed to comprehensively analyze patterns of alternative splicing across hundreds of thousands of publicly archived RNA- Seq datasets \(^{27 - 29}\) . We have used these databases to identify many cell type- specific alternative exons that are suitable for use in AAV vectors.  
+
+In this study, we have developed a suite of AAV- based tools that direct pan- neuronal, excitatory neuron, and photoreceptor- specific gene expression via a splicing- linked expression design (SLED) strategy. This method uses splicing- dependent frameshifting, in combination with both ubiquitous and cell type- specific promoters, to drive cell type- specific expression of fluorescent proteins and other genes of interest. We show that, due to their small size, SLED constructs can be packaged into AAV vectors and that cell specificity is maintained in vivo across multiple species. Furthermore, the SLED method can be used to create previously unobtainable research tools. We miniaturized the Gria2 flip/flop intron for AAV packaging to monitor this mutually exclusive splicing event at single- neuron resolution. We also demonstrated that dual calcium sensors can be simultaneously expressed in different cell types using a single expression vector, instead of using multiple viruses or transgenic animals. Finally, we demonstrated that SLED- based AAV constructs perform as efficiently as state- of- the- art minipromoter vectors for functional rescue of photoreceptor dystrophies, and also show that SLED can be used to selectively target SF3B1 mutant cancer cells for oncolytic therapy. These results demonstrate that SLED- based tools are compatible with existing methods for regulating cell type- specific gene expression, and that SLED is broadly useful for a range of basic and translational research applications.
+
+<--- Page Split --->
+
+## Results  
+
+## Identification of cell-specific exons and SLED vector construction  
+
+To test the ability of alternative splicing to mediate cell type- specific expression of reporter constructs, we modified an existing bichromatic reporter plasmid 15. In this construct, dsRed is expressed when the default splicing pathway is used. When a cell type- specific alternative exon is spliced in, however, this results in a reading frame shift that leads to the expression of EGFP. In cases where the sequence length of the cell type- specific exon is a multiple of 3 and lacks a stop codon in the initiating translational reading frame, point mutations were introduced to create a frameshifting cell type- specific exon. Importantly, the dsRed sequence is modified to remove stop codons that would otherwise occur in the EGFP reading frame (Fig. 1a). 2A self- cleaving peptide sequences 30 were also included in front of each fluorescent protein to allow expression of the fluorescent protein independent of leader sequences (Fig. S1).  
+
+We used three criteria to select alternative exons for analysis. First, alternative exons needed to show highly cell type- specific patterns of inclusion. Second, cell type- specific patterns of splicing needed to be conserved between mice and humans. Finally, the size of the intronic sequence used needed to be less than 2 kb. Using a computational resource that catalogs cell type- specific splicing patterns (ASCOT) 27, we identified \(\sim 1000\) neuronal- enriched alternative exons, of which \(\sim 200\) had intronic lengths of \(< 2\) kb (Fig. 1b,c). \(\sim 99\%\) of exons show high conservation (vertebrate phyloP score \(>1.5\) ) of neuron- enriched splicing between mouse and human (Fig. 1d,e). A neuronally- enriched exon in the gene encoding the ubiquitously- expressed actin- binding protein Plastin 3 (PLS3) was selected for characterization. A similar process was used to identify a photoreceptor- specific exon in the gene encoding the ubiquitously expressed subunit of the ATPase \(\mathrm{Na + / K + }\) Transporting Subunit Beta 2 (ATP1B2) (Fig. S2). For proof- of- concept, we transfected the pan- neuronal and photoreceptor- specific SLED vectors into HEK293, HepG2, and N2a neuroblastoma cell lines to determine specificity (Fig. 1f). While dsRed was expressed in all cells, EGFP was only observed with pan- neuronal SLED in N2a cells, which exhibit neuronal precursor- like characteristics 31. No expression of EGFP was observed in any cells when transfected with the photoreceptor- specific SLED construct, supporting the cell type specificity of the ATP1B2 alternative exon. Specificity was determined at the single cell level by evaluating the \(\log_2\) ratio of EGFP/dsRed fluorescence (Fig. 1f).  
+
+## SLED vectors maintain cell-specific expression when delivered using AAV  
+
+To test whether SLED vectors retain specificity when packaged into AAV, we cloned pan- neuronal SLED (SLED.NPL) and photoreceptor- specific SLED (SLED.RAB) into an AAV backbone (Fig. 2a). Furthermore, we sought to test whether cell type- specific minipromoters could be combined with SLED- based constructs in a Boolean manner to provide more selective cell type- specific expression using AAV vectors. To
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Identification of cell-specific exons and SLED vector construction (a) Diagramatic sketch of SLED vector design strategy. SLED is compatible with any promoter. A frameshifting mutation is introduced into a cell-specific alternative exon </center>
+
+<--- Page Split --->
+
+to create two potential translational reading frames from an upstream start codon. In most SLED vectors, exon skipping will produce a red fluorescent protein while exon inclusion will shift the reading frame to produce a green fluorescent protein. (b) Ranking all neuron- enriched exons by the percent spliced- in (PSI) difference between neurons and other cell types. Exons were identified from mouse RNA- Seq datasets analyzed with the ASCOT pipeline \(^{27}\) . Approximately 1000 neuron- enriched exons have a \(\Delta \mathrm{PSI}\) greater than 20. (c) Among these top 1000 exons, approximately 200 candidates reside in introns \(< 2\mathrm{kb}\) in length. (d, e) UCSC genome browser views of the neuron- specific exon in \(P / s3\) that is used in SLED.NPL. Exon incorporation is only observed in neuronal datasets (red arrows) from both mouse (d) and human (e). A similar strategy was used to identify the photoreceptor- specific exon in \(Atp1b2\) . These exons were used to generate SLED vectors that were then tested in HEK293, HepG2, and N2a cancer cell lines (f). As predicted, neither vector showed EGFP expression, indicating an absence of cell- specific exon incorporation, except when the neuron- specific SLED was transfected into N2a cells, reflecting the neuronal characteristics of N2a neuroblastoma cells. Scale bars \(= 50\mu \mathrm{m}\) .  
+
+do this, we combined the pan- neuronal hSyn minipromoter \(^{32}\) with an alternative exon of the gene that encodes the ubiquitously expressed clathrin complex interactor Synergin gamma (SYNRG), which in the brain is specific to excitatory neurons and glia (SLED.ENS, Fig. 2a, Fig. S2).  
+
+We first tested the expression of these constructs using plasmid transfection and electroporation. As expected, we observed selective expression of EGFP in neurons following transfection of SLED.NPL into primary rat hippocampal cultures, although expression of the default splicing- driven dsRed was observed in transfected neurons and glia (Fig. S3). Likewise, in neonatal mouse retinal explants electroporated with the SLED.RAB construct \(^{33}\) , we observed expression of dsRed in all postnatally- generated cell types, but EGFP reporter expression is restricted to photoreceptors (Fig. S3). Lastly, we observe that transfection of the SLED.ENS constructs into primary rat hippocampal cultures resulted in selective exclusion of hSyn- driven EGFP expression from somatostatin- expressing GABAergic interneurons (Fig. S3). To further validate the photoreceptor- specificity of the SLED.RAB construct in vivo, postnatal day 0 (P0) mouse retinas were transduced with photoreceptor- specific AAV2.7m8.SLED.RAB and processed 4 weeks later at P30. This revealed highly enriched expression of EGFP in retinal photoreceptors, with inner retinal neurons strongly positive for dsRed (Fig. 2b).  
+
+We next tested the neuronal- specificity of SLED.NPL packaged into AAV9 by transducing primary rat hippocampal cultures at 1 day in vitro (DIV). At DIV 15, cells were fixed and immunofluorescence conducted for the neuronal marker HuC/D.. Comparison of the ratio of EGFP to dsRed fluorescence revealed that EGFP was highly enriched in HuC/D- positive neurons (Fig. 2c). To test SLED.NPL in vivo, we performed
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2. SLED vectors maintain cell-specific expression when delivered using AAV. (a) Diagramatic sketch of pan-neuronal (SLED.NPL), photoreceptor-specific (SLED.RAB), and excitatory neuron-specific (SLED.ENS) vectors designed for AAV packaging. In all SLED vectors, EGFP is translated when cell-specific exons are spliced-in. (b) SLED.RAB, packaged in AAV2.7m8, was intravertically injected into P0 mouse retinas and processed at P30. EGFP is highly enriched in retinal photoreceptors while inner retinal neurons are strongly positive for dsRed. ONL = outer nuclear layer, INL = inner nuclear layer. (c-e) SLED.NPL, packaged in AAV9, was used to transduce primary rat hippocampal cultures (c), mouse cortex (d), and human iPSC-derived neurons (e). In all cases, EGFP is highly enriched in HuC/D+ </center>
+
+<--- Page Split --->
+
+(rat, mouse) or NeuN+ (human) neurons, while non- neuronal cells are strongly positive for dsRed. (f- h) SLED.ENS, packaged in AAV9, was used to transduce primary rat hippocampal cultures (f), mouse hippocampus (g), and ferret cortex (h). In all cases, EGFP is highly enriched in GAD67- excitatory neurons, while GAD67+ inhibitory neurons are strongly positive for dsRed. \*\*\* indicate \(p < 0.0001\) , two- tailed t- test. For ratio calculations in panels b to h, \(n = 205\) (b), \(n = 94\) (c), \(n = 78\) (d), \(n = 55\) (e), \(n = 121\) (f), \(n = 198\) (g), and \(n = 154\) (h). Scale bars \(= 50\mu \mathrm{m}\) .  
+
+stereotactic injection of AAV9. SLED.NPL into the mouse hippocampal region. Efficient and widespread infection was observed and the neuronal specificity of EGFP expression was maintained (Fig. 2d). AAV- based gene therapies are being explored as treatment options for neurological disorders and SLED vectors may improve the safety and efficacy of these methods. To determine whether AAV9. SLED.NPL maintains neuron- specific expression in human cells, we transduced human iPSC- derived mixed neuronal and glial cultures (Fig. 2e). Here too, we observed strong EGFP expression in NeuN- positive neurons but only dsRed expression in NeuN- negative glial cells.  
+
+Lastly, we tested the specificity of the excitatory neuron- specific AAV9. SLED.ENS construct. Primary rat hippocampal cultures show strong expression of EGFP in Gad67- negative excitatory neurons, but little or no expression of EGFP in Gad67- positive GABAergic interneurons (Fig. 2f). Stereotactic injection of AAV9. SLED.ENS into mouse hippocampus likewise resulted in strong and broad neuronal expression of EGFP, but exclusion of EGFP signal from dsRed- positive, Gad67- positive interneurons (Fig. 2g). Finally, a similar pattern of exclusion from Gad67- positive interneurons was observed following transduction of primary ferret visual cortex (Fig. 2h). Together, these findings demonstrate that SLED cell specificity is maintained in vitro and in vivo across multiple species, and that mutually exclusive splicing events can be simultaneously monitored using AAV- based SLED tools.  
+
+## Generation of unique splicing-based tools using SLED vectors  
+
+We next sought to test whether SLED- based AAV vectors could be adapted to study in vivo patterns of flip/flop splicing in the AMPA- type glutamate receptor subunit GluA2. Flip/flop alternative splicing occurs within the ligand binding domain, and influences AMPA receptors assembly and channel kinetics \(^{34 - 37}\) . The short lengths and high sequence similarity of the mutually exclusive flip/flop exons precludes the use of immunostaining or in situ hybridization to detect their localization in situ, which has effectively restricted previous efforts investigating flip/flop splicing in learning and plasticity to using qRT- PCR analysis \(^{38,39}\) . To address this, we generated an hSyn- driven AAV vector which expresses EGFP when the flop exon is incorporated, and dsRed when the flip exon is incorporated (SLED.GluA2, Fig. 3a). To validate that AAV9. SLED.GluA2 splicing reflected endogenous flip/flop splicing, we electroporated primary rat cortical neurons and FACS- isolated high EGFP- expressing cells. We
+
+<--- Page Split --->
+![PLACEHOLDER_11_0]
+
+
+<--- Page Split --->
+
+## Figure 3: Generation of unique splicing-based tools using SLED vectors  
+
+(a) Diagramatic sketch of GluA2 (Gria2) flip/flop SLED vector design (SLED.GluA2). 
+(b) To validate that SLED.GluA2 reflected endogenous GluA2 flip/flop splicing patterns, we designed endogenous mRNA-specific primers to PCR amplify the GluA2 flip/flop locus. Although the mutually exclusive flip and flop exons are identical in length and highly similar in sequence, Hpal will selectively digest the flop PCR product into two fragments. SLED.GluA2, packaged into AAV9, was used to electroporate primary rat neuronal cultures and EGFPhigh/mCherrylow cells were isolated using FACS (Fig. S4). RNA was extracted from EGFPhigh/mCherrylow cells and bulk rat neuronal cultures and primers (b) were used to amplify PCR products. 
+(c) Hpal incubation yielded digestion products in the EGFPhigh/mCherrylow cells, which was further confirmed using Sanger sequencing (d, Fig. S4, n=36, \*\*\* indicate p < 0.0001, two-tailed t-test.). (e) Primary rat neuronal cultures transduced with SLED.GluA2 show significantly different EGFP/mCherry ratios between excitatory (GAD67-) and inhibitory (GAD67+) neurons. \*\*\* indicate p < 0.0001, two-tailed t-test. 
+(f) Diagramatic sketch of bicistronic jRGECO1a (inhibitory neurons) and jGCaMP7b (excitatory neurons) SLED vector design (SLED.CaRPv1). (g) Transfection of primary rat neuronal cultures yielded divergent ratios in jGCaMP7b and RGECO1a intensities in excitatory (mDlx-Azurite-) and inhibitory (mDlx-Azurite+) neurons. Data presented represent jGCaMP7b (top row) and jRGECO1a (middle row) intensity values over a 60s time-lapse (4Hz). Bottom row represents a normalized representation of total Ca intensity scaled by the delta between jGCaMP7b and jRGECO1a pixel values. Individual Ca indicator traces are demonstrated in the bottom panels. For ratio calculations in panel e, n=266 (cortex) and n=316 (hippocampus). Scale bars = 50μm (panel e) and 20μm (panel g).  
+
+designed primers to selectively amplify spliced flip/flop exons in GluA2 mRNA (Fig. 3b). Digestion using Hpal, which selectively cleaves the flop exon into two smaller fragments, revealed expected enrichment of flop exon fragments in the EGFP- enriched fraction (Fig. 3c). This was further confirmed using Sanger sequencing of amplified products (Fig. 3d, Fig. S4).  
+
+Transduction of AAV9.SLED.GluA2 into both primary rat hippocampal and cortical cultures (Fig. 3e) revealed a variety of different cellular patterns of reporter expression, with EGFP- dominant, dsRed- dominant and mixed cells all present. However, we observe that Gad67- positive hippocampal neurons are enriched for EGFP- dominant expression, matching previous observations obtained using single- cell SMART- Seq analysis and bulk RNA- Seq analysis of RiboTRAP- expressing interneurons \(^{40 - 42}\) (Fig. S5).  
+
+The translational frameshifting used in SLED vectors also offers the potential to deliver multiple functional payloads, such as genetically encoded calcium sensors or optogenetic actuators, using a single viral vector. As proof of concept, we created a
+
+<--- Page Split --->
+
+bicistronic calcium indicator vector based on the excitatory neuron- specific SLED.ENS. We termed this calcium reporter plasmid version 1 (CaRPv1) (Fig. 3f). In CaRPv1, GCaMP7b is expressed in excitatory neurons, while RGECO1a is expressed in the default translational reading frame (inhibitory neurons) \(^{43,44}\) . Identification of excitatory vs inhibitory neurons was established using the \(\Delta\) pixel intensity of normalized fluorescence values, due to differences in dynamic range and baseline fluorescence at resting calcium concentrations for GCaMP7b and RGECO1a (see methods). Transfection of CaRPv1 into primary rat hippocampal cultures revealed the expected patterns of calcium transients (Fig. S6, Supplemental Videos 1 & 2). Furthermore, large and synchronous calcium transients were observed following the addition of bicuculline (a GABA receptor antagonist, used to induce disinhibition), indicating that CaRPv1 was reporting cellular activity as expected (Fig. 3g). In its current design, CaRPv1 is unable to be packaged into AAV due to the size of the ENS intron ( \(\sim 1600bp\) ). However, future deletion mutagenesis and sequence optimization should enable AAV packaging of CaRPv1.  
+
+## Adapting SLED vectors for translational studies  
+
+Mutations in the photoreceptor outer segment structural gene PRPH2 cause human retinal degeneration \(^{45}\) and null mutations in Prph2 lead to slow- onset photoreceptor degeneration in mice \(^{46 - 49}\) . AAV- based constructs driven by the photoreceptor- specific minipromoter mOps have been previously used to rescue Prph2 expression in rds/rdr Prph2- deficient mice (Prph2 \(^{rds/rdr}\) ), although only modest photoreceptor preservation, and no long- term recovery of visual function was observed due to weak promoter efficiency \(^{50}\) . We modified the photoreceptor- specific SLED.RAB to selectively express PRPH2 under control of the ubiquitous CBh promoter. In parallel, we generated mOps minipromoter- driven rescue vectors that were used in previous studies \(^{50 - 52}\) . CMV- driven EGFP vectors were also obtained as controls (Fig. 4a). These were all packaged into AAV2.7m8 capsids and injected subretinally at P28 into Prph2 \(^{rds/rdr}\) mice.  
+
+Mice were injected with AAV at one month of age, and three months later were then analyzed using optical coherence tomography (OCT) to measure the relative thickness of the retinal outer nuclear layer (ONL), where photoreceptors reside. We observed that ONL thickness was similar in both SLED and mOps- regulated AAV constructs, and significantly greater than mice injected with CMV.GFP control virus (Fig. 4b). The amplitude of the light- adapted, cone- mediated, full- field electroretinogram (ERG) was larger in SLED relative to mOps- based rescue constructs, with both showing significantly higher b- wave responses relative to CMV.GFP controls (Fig. 4c). Immunostaining for Prph2 in transduced retina showed no detectable expression in CMV.GFP controls (Fig. 4d), but Prph2 signal was detected in photoreceptor inner
+
+<--- Page Split --->
+![PLACEHOLDER_14_0]
+
+
+<--- Page Split --->
+
+## Figure 4: Adapting SLED vectors for translational studies  
+
+(a) Diagramatic sketch of GFP under the control of the constitutive CMV promoter (CMV.GFP), PRPH2 under the control of the photoreceptor specific mOps promoter (mOps.PRPH2), and PRPH2 regulated by the photoreceptor-specific SLED.RAB (SLED.PRPH2) vectors designed for AAV packaging. CMV.GFP, mOps.PRPH2, and SLED.PRPH2 were packaged into AAV2.7m8 for testing in Prph2<sup>rds</sup>/<sup>rds</sup> animals. For experimental design, n=6 for each AAV treatment. (b) Average ONL/INL ratios and (c) average light-adapted ERG b-wave amplitudes in three month post-injected Prph2<sup>rds</sup>/<sup>rds</sup> animals treated with CMV.GFP, SLED.PRPH2 or mOps.PRPH2 viral constructs (asterisks indicate p < 0.05, two-tailed t-test, comparison between mOps.PRPH2 and SLED.PRPH2). (d-f) Immunofluorescence staining of retinal sections from Prph2<sup>rds</sup>/<sup>rds</sup> eyes injected with CMV.GFP (d), mOps.PRPH2 (e), and SLED.PRPH2 (f). ONL = outer nuclear layer, INL = inner nuclear layer. EGFP is only detected in CMV.GFP treated controls, but Prph2 signal (yellow arrow) is detected in photoreceptor inner segments of retinas transduced with both mOps.PRPH2 and SLED.PRPH2. (g) UCSC genome browser view of a cryptic exon in UBA1 (green arrow) that is present in cancers with oncogenic SF3B1 mutations (TCGA 76) and absent in all normal human tissues sequenced by the GTEx consortium 77. (h) Diagramatic sketch of bichromatic fluorescent reporter based on the SF3B1<sup>mut</sup> associated exon (SLED.SFUv1) and a similar vector where an inducible iCaspase9 kill switch is coupled to incorporation of the SF3B1<sup>mut</sup>-associated exon. (i) As a proof of concept, SLED.SFUv1 was transfected into uveal melanoma cell lines with (Mel-202) and without (92-1) SF3B1 mutations. EGFP was highly enriched in only Mel-202 cells while dsRed was strongly expressed in 92-1 cells, which was validated using FACS (j). Isogenic cell lines derived from Mel-202 with the SF3B1<sup>R625G</sup> mutation genetically inactivated (PC76B6) and maintained (MR5) showed similarly concordance, with strong EGFP expression only present in the SF3B1<sup>R625G</sup> MR5 cell line. Likewise, strong EGFP expression was only present in SF3B1<sup>K700E</sup> K562 leukemia cells, as compared to wildtype K562 cells. **** indicate p < 0.0001, two-tailed t-test. For ratio calculations, n=245 (92-1), n=1158 (Mel-202), n=377 (PC76B6), n=526 (MR5), n=49 (K562<sup>WT</sup>), n=677 (K562<sup>MUT</sup>). (k) Transfection of wildtype K562 cells and mutant SF3B1<sup>K700E</sup> K562 cells with SLED.SFUv2 revealed strong apoptosis only in SF3B1<sup>K700E</sup> K562 cells treated with the iCaspase9 activating dimerizer (n=3 FACS replicates). Scale bars = 50μm.  
+
+segments in retinas transduced with both mOps (Fig. 4e) and SLED- based (Fig. 4f) Prph2 rescue constructs.  
+
+Finally, because oncolytic virotherapy is now an approved treatment modality in oncology 53, we sought to leverage tumor- specific RNA splicing patterns to generate SLED- based oncolytic vectors. Specifically, we identified a cryptic alternative exon in the constitutively expressed ubiquitin- like modifier activating enzyme 1 (UBA1) that was
+
+<--- Page Split --->
+
+observed exclusively in \(SF3B1\) mutant cancer cells (Fig. 4g). We designed two SLED- based vectors incorporating this alternative exon: one that express a bichromatic fluorescent reporter that selectively expresses EGFP in \(SF3B1\) mutant cells (SLED.SFUv1, where EGFP will only express in mutant \(SF3B1\) cells), and one that selectively expresses an oncolytic inducible Caspase 9 \(54,55\) in \(SF3B1\) mutant cells (SLED.SFUv2) (Fig. 4h). We first tested SFUv1 specificity by transfected 92- 1 and Mel- 202 uveal melanoma cell lines. We observed that EGFP expression was present in \(SF3B1^{R625G}\) Mel- 202 cell lines \(^{56}\) , but absent in the 92- 1 uveal melanoma cell line, which is wildtype for \(SF3B1\) (Fig. 4i) \(^{57}\) . We next quantified this using FACS analysis and analyzed four additional cell lines, two of which were isogenic to Mel- 202: PC76B6, in which AAV- based gene targeting was used to revert the mutant \(SF3B1\) status to wildtype through inactivation of the \(SF3B1^{R625G}\) allele, and MR5, a gene targeting control clone of Mel- 202 that retains the \(SF3B1^{R625G}\) mutation (Fig. S7). The other two cell lines analyzed by FACS were wildtype and \(SF3B1^{K700E}\) K562 leukemia cells. FACS analysis revealed that the EGFP/dsRed ratio is strongly dependent on the presence of either the \(SF3B1^{R625G}\) or \(SF3B1^{K700E}\) mutation (Fig. 4j). Finally, we tested the efficacy of SLED.SFUv2 by transfecting the wildtype and \(SF3B1^{K700E}\) K562 cells and observed efficient and selective induction of apoptosis in cells carrying the \(SF3B1^{K700E}\) following induction of Caspase9 dimerization (Fig. 4k).  
+
+## Discussion  
+
+In this study, we demonstrate that SLED- based vectors can produce cell- specific expression in a variety of constructs, and that SLED- based approaches are both orthogonal and complementary to existing methodology. SLED- based alternative splicing can be combined in a Boolean fashion with minipromoters to achieve higher levels of cell type specificity, as demonstrated by the integration of the pan- neuronal hSyn minipromoter and the excitatory neuron and glial- specific SYNRG exon to generate the excitatory neuron- specific SLED.ENS vector. SLED- based AAV vectors can also be used to study previously intractable problems without the use of complex transgenics, such as the in vivo dynamics of GluA2 flip/flop splicing. The use of splicing- related frameshifting allows efficient cell type- specific expression of multiple reporter or effector constructs in a single vector. SLED- based vectors also enable new strategies to improve gene therapies. For instance, SLED vectors can use any promoter, potentially allowing for stronger and more sustained levels of expression relative to conventional minipromoters and enabling more consistent patterns of cell- specific expression across multiple model organisms \(^{58,59}\) . This is critical, as photoreceptor minipromoters have encountered complex issues when tested in various mammalian species. For instance, the hRK1 minipromoter, which is widely used to drive expression in both rods and cones in rodents, is unable to drive efficient expression in cones of other model organisms such as dogs and pigs unless used at very high titers, and expresses at lower levels in
+
+<--- Page Split --->
+
+rods than the rod- specific mOps promoter \(^{59 - 61}\) . Finally, SLED vectors can also selectively target disease states associated with abnormal splicing that would not be accessible using minipromoters. The use of photoreceptor minipromoters in AAV vectors can lead to long- term toxicity, but this is avoided using constitutive promoters \(^{62}\) .  
+
+Identification of evolutionarily- conserved patterns of cell- specific alternative splicing is straightforward, provided that good quality full- length RNA- Seq data is available. As transcriptomes from more tissues and cell types are profiled and deposited in public archives, our ability to identify highly cell- specific patterns of alternative splicing will increase and these datasets will guide the design of the next generation of SLED vectors. While transcriptome analysis has increasingly shifted towards 3'- directed short read single cell RNA- Seq platforms in recent years, emerging techniques such as long- read nanopore sequencing \(^{63}\) and economical full- length scRNA- Seq techniques such as SMART- Seq v3 will continue to improve our knowledge of splicing patterns \(^{64,65}\) . Recent compendia of splice- junction and transcript- level expression have surveyed 100,000s to millions of datasets \(^{28,29,66}\) , making these patterns easier to discover computationally.  
+
+With detailed characterization of alternative splicing patterns in the tremendous diversity of cell types, particularly in the human central nervous system, an intersectional approach combining SLED and cell- specific minipromoters may generate vectors that can selectively target to date untargetable cell types. Indeed, alternative splicing generates another layer of transcriptional complexity to the nervous system \(^{25,67,68}\) .  
+
+SLED- based vectors are still intrinsically limited by the size of the genomic intronic sequences used to control alternative splicing, which are generally substantially larger than minipromoters. While this is a less severe obstacle for transfection- or nanoparticle- based gene delivery, it is still a substantial limitation for AAV- based delivery. While the effects of deletion mutagenesis on cell- specific splicing can be unpredictable, recently developed machine learning algorithms may help facilitate rational design of smaller SLED vectors \(^{69,70}\) . Drug- inducible approaches to regulate splicing \(^{71 - 73}\) and the inclusion of miRNA target sites \(^{74,75}\) may enable further control of SLED- based constructs.
+
+<--- Page Split --->
+
+## Materials and Methods:  
+
+## Molecular cloning and cancer cell line culture:  
+
+Materials and Methods:Molecular cloning and cancer cell line culture:To generate SLED plasmids, gene fragments were commercially synthesized using Twist Biosciences and ThermoFisher GeneArt and cloned into an AAV vector backbone (Addgene #105922) using restriction enzyme cloning. HEK293, HepG2, and N2a cells were cultured in Dulbecco's Modified Eagle's Medium (Corning, 10- 017- CV) supplemented with 1x GlutaMAX (ThermoFisher Scientific, 35050061), 10% FBS (Corning, 35- 010- CV). Human uveal melanoma cell lines 92- 1 (generously provided by Charles Eberhart, Johns Hopkins University), MP41 (ATCC), and Mel- 202 (Sigma) were cultured in RPMI medium with 10% fetal bovine serum (FBS), penicillin/streptomycin, and l- glutamine. SF3B1<sup>K700E</sup> and control K562 cells were obtained from Horizon Discovery and cultured in RPMI with 20% FBS. The isolation, early characterization and further genetic and molecular characterization of the cell lines have been described elsewhere<sup>78- 80</sup>. Transfection of SLED vectors in uveal melanoma cells was achieved using Lipofectamine 3000 (ThermoFisher Scientific, L3000008) and with the 4D- Nucleofector X (Lonza) for K562 cells. The SF3B1<sup>R702R</sup> AAV targeting vector as described<sup>81</sup> was applied to SF3B1<sup>R625G</sup> Mel202 cells. iCaspase9 dimerization was induced by 100nM AP21087 (Sigma- Aldrich).  
+
+## Antibodies  
+
+AntibodiesThe following antibodies were used for primary culture neurons (supplier, catalog number and working dilution are indicated): anti- Gad67 ms IgG2a (Millipore MAB5406, 1:500); anti- somatostatin rat (Millipore MAB354, 1:400); anti- HuCD IgG2B (Thermo 16A11, 1:200); anti- NeuN mouse IgG1 (Thermo MAB377, 1:500). The following antibodies were used for brain sections (supplier, catalog number and working dilution are indicated): anti- GFP chicken polyclonal (Abcam ab13970, 1:2000); anti- dsRED rabbit polyclonal (Tanaka LivingColors 632496, 1:1000) (this antibody also detects mCherry); anti- NeuN mouse monoclonal IgG1 (Thermo MAB377, 1:500); anti- Gad67 mouse monoclonal IgG2a (Millipore MAB5406, 1:200); anti- PV mouse monoclonal IgG1 (Swant PV235, 1:2000); anti- somatostatin rat monoclonal IgG2b (Millipore MAB354, 1:400).  
+
+## Preparation and treatment of rat primary cultures:  
+
+Preparation and treatment of rat primary cultures:Hippocampi and cortices were dissected from embryonic day 18 rats, incubated with papain (Worthington Biochemical) and gently triturated with polished glass pipettes. Hippocampal neurons were plated on 18mm glass coverslips precoated with poly- L- Lysine (1mg/ml) in NeuroBasal media (Gibco) supplemented with 2% B27 (Gibco), 50 U/ml penicillin, 50 mg/ml streptomycin, 2mM GlutaMax (Gibco) and 5% horse serum (Hyclone). Hippocampal cells were plated at a density of 150K/coverslip (in a 12- well
+
+<--- Page Split --->
+
+plate). At days in vitro 1 (DIV1) media was replaced with NMO consisting of the above plating media without the addition of serum. Cells were then fed every 7 days with NMO. Hippocampal cultures were transduced with viral vectors at DIV1 and fixed at DIV15 for immunofluorescence analysis. SLED constructs were also transfected into hippocampal neurons using Lipofectamine 2000 (Invitrogen) following the manufacturer's instruction. Cortical neurons were used to evaluate SLED.GluA2. Before plating, cortical cells (6M per reaction) were electroporated with SLED.GluA2 plasmid DNA (3- 5ug) using a Rat Neurofection kit (Amaxa) and split evenly between 3 wells of a 6- well plate. As a comparison group, neurons were plated at equivalent density without electroporation. Cortical cells were harvested at (DIV4- 6) for FACS and downstream evaluation of Gria2 flip/flop splicing.  
+
+## Quantification of SLED.ENS with IMARIS:  
+
+To calculate the green/red ratios of SLED.ENS in excitatory and inhibitory neurons, Z- stacks were imported into IMARIS (Bitplane version 9.7.0) and 3D surfaces created around the cell bodies of transduced cells. The average EGFR and mCherry signal within the 3D surface was extracted for each cell. The Gad67 immunohistochemical signal was used to classify the cells into Gad67 positive (inhibitory interneuron) or Gad67 negative (putative excitatory neurons). To determine the statistical specificity of SLED vectors, log base 2 transformed ratios of green/red fluorescence were compared between sample groups using unpaired t- tests (GraphPad).  
+
+## Quantification of SLED.ENS, SLED.NPL and SLED.GluA2 in Fiji:  
+
+Confocal Z- stack images acquired with a 40x objective were processed using Fiji \(^{82}\) . Maximum intensity projections were generated, and the EGFR or mCherry/dsRed channel used to draw circle/oval ROIs around the cell body of transduced neurons without looking at signal in the 405 or 647 channels that contained cell- specific immunofluorescent markers. The raw EGFR and mCherry/dsRED average ROI intensities were then extracted for each cell. Subsequently, the cell- specific identification of each cell was observed from immunofluorescence using the 405 and 647 channel (for SLED.NPL this was HuC/D expression, for SLED.ENS this was Gad67 expression and for SLED.GluA2 this was Gad67). This enables clustering of the individual cells for comparisons of the green/red ratios. To determine the statistical specificity of SLED vectors, log base 2 transformed ratios of green/red fluorescence were compared between sample groups using unpaired t- tests (GraphPad).  
+
+## Live-cell imaging of SLED.CaRPv1  
+
+Rat primary hippocampal neurons at DIV11 were transfected with \(1\mu \mathrm{g}\) SLED.CaRPv1 alongside \(1\mu \mathrm{g}\) of mDlx- Azurite (as an enhancer based marker of
+
+<--- Page Split --->
+
+interneurons \(^{4,83)}\) per coverslip in a 12- well plate. At DIV13 coverslips were imaged live in ACSF (NaCl 120mM, KCl 5mM, HEPES 10mM, D- Glucose 10mM, CaCl2.2H2O 2mM, MgCl2 1mM) at pH 7.4 using a Zeiss LSM 880 confocal microscope in a temperature \((37^{\circ}C)\) and humidity controlled chamber. Interneurons were identified by presence of the mDlx- Azurite signal, and mDlx- Azurite- negative cells were considered putative excitatory neurons. Time series were collected using 20x or 10x objectives for single cell or multiple cell imaging, respectively. Images were acquired at baseline, and also following addition of Bicuculline (20μM) to promote network activity through disinhibition. Files were processed and analyzed using Fiji \(^{82}\) . Fluorescence signals from jGCaMP7b and jRGECO1a were normalized to maximize variation between mDlx- Azurite positive and negative cells and the difference (Δintensity) between jGCaMP7b and jRGECO1a was calculated across each pixel and image frame (a gaussian blur (1px) was applied to each image before Δintensity to avoid pixelation artifacts). The sum of jGCaMP7b and jRGECO1a pixel values (sumGR) were calculated across each pixel and image frame. To generate the CaGreen- CaRed heatmap in Figure 3g and Supplemental Videos 1 and 2, sumGR was multiplied by Δintensity and colored using a custom lookup table.  
+
+## Immunofluorescence analysis of primary cultured neurons:  
+
+Unless otherwise stated, hippocampal neurons were fixed at DIV15. Media was aspirated and cells were washed with PBS at RT once before being incubated with \(4\%\) PFA (Electron Microscopy Sciences) made up in PBS with the addition of \(4\%\) sucrose at RT for 15 mins. Coverslips were washed 4 times with PBS then immunofluorescence commenced using a gelatin- based buffer (15 mM phosphate buffer (pH 7.4) containing \(0.1\%\) gelatin, \(0.3\%\) Triton X- 100, and \(0.25M\) NaCl) for combined blocking/antibody incubation. Primary antibodies (see below) were incubated with coverslips O/N at \(4^{\circ}C\) Secondary antibodies (Invitrogen for 488, 568 and 647 and Jackson labs and Abcam for 405; all at 1:500 dilution) were incubated for 1hr at RT. Between antibody incubations cells were washed with PBS, with some experiments including a brief DAPI incubation to label cell nuclei. Coverslips were mounted on slides with PermaFluor (Thermo Fisher Scientific). Samples were imaged on a Zeiss LSM 880 confocal microscope. Care was taken to ensure pixels in each channel were not over saturated. Images were analyzed using Fiji. In cultured neurons the SLED- driven fluorophores were not antibody boosted.  
+
+## AAV production:  
+
+SLED.ENS (1E+12 vg/ml, mouse and ferret cortex) and SLED.GluA2 (1E+12 vg/ml) were generated by the UNC Vector Core and SLED.NPL (2E+13 vg/ml) and SLED.RAB (2E+13 vg/ml) were generated by Virovek. SLED.ENS used for primary rat neuronal cultures was generated by Virovek (2E+13 vg/ml)
+
+<--- Page Split --->
+
+## Stereotaxic surgery and virus injections:  
+
+All animals were treated in accordance with the Johns Hopkins University Animal Care and Use Committee (IACUC) guidelines. Adult Blk6/J mice were used to evaluate SLED.AAV vectors. Animals were anesthetized with isofluorane (Baxter) using a SomnoSuite system (Kent Scientific) and secured in a stereotaxic frame (Kopf). The animal's temperature was controlled with a closed- loop system (RightTemp, Kent Scientific). The animal's scalp skin was cleaned with an ethanol wipe, and the hair removed. Animals were injected with \(0.5ml\) sterile saline (VetOne) to maintain hydration, buprenorphine (ZooPharm; \(1mg / ml\) ) and lidocaine (VetOne; \(2\%\) ) subcutaneously. The lidocaine was injected under the scalp as a local anesthetic. An incision was made to expose the skull surface, and to enable a small craniotomy to be made (see below for coordinates) exposing the brain surface. A glass pipette (Drummond Science Company; Wiretrol II) was pulled (Sutter Instruments) and sharpened to a \(30^{\circ}\) angle (Medical System Corp) and controlled by a pneumatic injector (Narishige) to enable controlled virus injection. Pipettes loaded with SLED viruses were slowly lowered to the desired stereotaxic coordinate, and after a delay of 2 minutes, virus was injected at a rate of \(100\mathrm{ml / min}\) . The pipette was left in position for 5 minutes after virus injection to reduce backflow up the injection tract. After pipette removal, the skin was sutured (Ethicon) and sealed with glue (Vetbond). Mice were closely monitored during the recovery phase, and placed in a clean cage on a warmed surface with access to a softened chow diet. Animals were given 2 weeks to recover, and for the virus to express, before being euthanized for perfusion/fixation.  
+
+Stereotaxic coordinates: Targeting of dorsal hippocampus (all with respect to Bregma): [AP: - 2 | ML: 1.5 | Z: - 1.5, - 1.3, - 1.1 (from pia, 300nl at each site)]. Note, for SLED.NPL deep cortical layers above the hippocampus were imaged (with overflow virus injection into the dorsal hippocampus).  
+
+## Perfusions and immunohistochemistry:  
+
+Mice were terminally anesthetized and transcardially perfused with PBS followed by \(4\%\) PFA (Electron Microscopy Sciences), both ice cold. Brains were then postfixed for 2 hours at \(4^{\circ}C\) and then washed with PBS. Brains were then either sliced on a vibratome (Leica; VT- 1000; 60 \(\mu \mathrm{m}\) thick) or incubated overnight in \(30\%\) sucrose and cut into \(40\mu \mathrm{m}\) sections using a cryostat (Leica Biosystems). For slices that required Gad67 staining the following IHC protocol was followed as previously described4. Sections were washed x3 in PBS (10 minutes each) and permeabilized with PBS containing \(0.1\%\) Triton- X (Sigma) for 30 minutes at RT. Slices were blocked with PBS containing \(3\%\) BSA and \(5\%\) normal goat serum (Vector Laboratories) for 1 hour at RT. Primary antibodies were made up in the same blocking buffer and incubated at RT for 24hrs. Slices were washed 4x with PBS and then incubated with fluorescently conjugated secondary antibodies (Invitrogen, all at 1:500) ON at \(4^{\circ}C\) . Slices were then washed x4
+
+<--- Page Split --->
+
+with PBS and mounted on slides with PermaFluor (Thermo Fisher Scientific). For other antibody combinations the same overall protocol was followed with the following differences. Slices were permeabilized with PBS containing \(0.3\%\) Triton- X for 20 minutes. Slices were blocked with PBS containing \(5\%\) normal goat serum with the addition of \(0.15\%\) Trixon- X. Primary antibodies were incubated in the same blocking buffer but at \(4^{\circ}C\) overnight. Secondary antibodies were made up in the same blocking buffer and incubated ON at \(4^{\circ}C\) . In some instances, slices were washed with PBS containing DAPI to label nuclei after the secondary antibody incubation. For SLED.MEv2 the fluorophores were not antibody boosted. For evaluation of SLED.NPL and SLED.ENS the EGFP and mCherry/dsRed was antibody boosted. Slides were imaged on a confocal microscope (as described above), or on an Apotome epifluorescence scope (Zeiss) and analyzed further in Fiji.  
+
+## Ferret Cortex AAV injections:  
+
+An adult female ferret (Mustela putoris furo, Marshall Farms) was used for the virus injection. Anesthesia was induced with ketamine \((40 \mathrm{mg / kg})\) and maintained with isoflurane \((1.5 - 3\%)\) . Atropine \((0.05 \mathrm{mg / kg})\) was given at the start of the surgery. Burenorphine \((0.01 - 0.03 \mathrm{mg / kg})\) was administered pre- and post- operatively for analgesia in combination with a subcutaneous injection of lidocaine during the surgery, and post- operative administration of meloxicam \((0.1 - 0.2 \mathrm{mg / kg})\) . Animals were maintained at normal body temperature during the surgery using a heating pad. Skin and muscle over primary visual cortex were reflexed, and a small craniotomy was made over the brain region of interest. Virus was then injected through a pulled glass pipette sharpened to a tip angle of about 60 deg. Approximately 1 uL of virus was then injected, distributed across multiple depths at a single site in the craniotomy. After the injection, muscle and skin were closed and the animal was recovered and returned to its home cage. The animal was perfused 3 months after the virus injection.  
+
+## Ocular AAV injections:  
+
+For subretinal injections, AAVs were injected into the subretinal space of 28 day old Prph2rds/rds mice (#001979 Jackson Laboratory, Bar Harbor, ME). Briefly, mice were anesthetized by intraperitoneal injection of ketamine \((100 \mathrm{mg / kg})\) and xylazine hydrochloride \((20 \mathrm{mg / kg})\) . The pupils were dilated with \(1\%\) tropicamide (Alcon, Ft. Worth, TX). The corneas were covered with Healon GV sodium hyaluronate solution (Abbott Medical Optics Inc., Santa Ana, CA) and cover glass to facilitate transpupillary visualization. 1uL of AAV \((10^{\wedge}13\) viral genomes/mL) were loaded into a 33G needle micro- syringe (Hamilton Company, Reno, NV), then tangentially injected into the subretinal space through the sclera of the mice. A successful injection was verified by direct visualization through the dilated pupil of the recipient under the surgical microscope (Leica, Wetzlar, Germany).
+
+<--- Page Split --->
+
+For intravitreal injections, AAVs were injected into the vitreous cavity of 0 day old CD1 pups (Charles River Laboratories, Wilmington, MA). The neonatal animals were anesthetized by placing them on a waterproof surface over crushed ice until the pup was no longer responsive to touch. The eyelids were surgically separated before injecting \(1 \mu \mathrm{L}\) of AAV ( \(10^{\wedge}13\) viral genomes/mL) into the vitreous space using a custom 33G sharp needle micro- syringe (Hamilton Company, Reno, NV). The needle was held in place for 10 seconds to avoid outflow before being gently removed.  
+
+## Spectral Domain-Optical Coherence Tomography:  
+
+For in vivo retinal imaging, Spectral Domain- OCT images were obtained and analyzed as previously described \(^{84}\) . Mice were first anesthetized with ketamine (100 mg/kg) and xylazine hydrochloride (20 mg/kg), followed by dilation with \(1\%\) tropicamide and \(2.5\%\) phenylephrine. The clarity of the cornea and lens was maintained using GenTeal lubricating eye gel (Novartis Pharmaceuticals, Basel, Switzerland). The mice were secured using a bite bar to a movable stage. The stage was adjusted manually to center the image of the retina at the optic nerve head. Cross- sectional images were generated using 1000 rectangular volume scans using the Envisu OCT system (Leica Microsystems, Wetzlar, Germany). Outer nuclear layer and inner nuclear layer thickness was measured using the linear caliper function in the software by a masked observer using a pre- established uniform grid.  
+
+## Electrotinography (ERG):  
+
+Full- field flash ERGs were performed as previously described \(^{85}\) . In brief, mice were dark adapted overnight and anesthetized with ketamine (80 mg/kg) and xylazine hydrochloride (16 mg/kg) prior to recording. The pupils were dilated with \(1\%\) tropicamide, \(1\%\) cyclopentolate and \(2.5\%\) phenylephrine and the corneal surface was anesthetized with \(0.5\%\) proparacaine HCl eye drops. For recording retinal electrical responses, stainless- steel wire electrodes were placed on the corneas as the active electrodes, contacting the center of the corneal surface through a thin layer of artificial tear. Needle electrodes were subcutaneously inserted into the cheek and the tail as reference and ground electrodes, respectively. To maintain body temperature during the procedure, the animals were placed on a temperature- controlled heating pad. Using the UTAS Bigshot ERG system (LKC Technologies, Gaithersburg, MD), ERG responses were differentially amplified (0.3- 300 Hz), digitized at 1,000 Hz, averaged and stored. The recording epoch was 512 ms, with a 20 ms pre- stimulation baseline. After 7 min of light adaptation, ERGs were obtained to strobe flashes (- 0.8 to 1.9 log cd.s/m \(^2\) ) superimposed upon a steady 30 cd.s/m \(^2\) white background. in response to a series of flashes ranging from - 0.8 to 1.9 log cd.s/m \(^2\) . The b- wave amplitude was measured from the a- wave trough to the peak of the b- wave.
+
+<--- Page Split --->
+
+## Disclosures:  
+
+Disclosures:SB receives research support from Genentech, is a co- founder and shareholder in CDI Labs LLC, and was a consultant to Third Rock Ventures. JPL receives research support from Takeda Pharmaceuticals. SB and JPL have filed a patent application covering SLED technology.  
+
+## Acknowledgements:  
+
+Acknowledgements:This work was supported by NIH grants RF1MH123237 (S.B.), R24EY027283 (S.B.), K08EY027093 (F.R.), and R01EY033103 (M.S.S.), a Stein Innovation Award from Research to Prevent Blindness to S.B., an unrestricted departmental grant to the Wilmer Eye Institute from Research to Prevent Blindness awarded to F.R., an NSF NeuroNex grant #1934288 awarded to K.J.N., a Visual Sciences Training grant 2T32EY007143 awarded to C.P.S., a Johns Hopkins Kavli NDI fellowship awarded to J.P.L., and a Johns Hopkins IDIES Seed Fund awarded to J.P.L. We thank the Ross Flow Cytometry Core (JHMI), the Wilmer Microscopy module supported by the EY001765 core grant for flow cytometry, the Single Cell & Transcriptomics Core (JHMI), and the Cleveland Clinic core grant (EY025585). We thank W. Yap, W. Xin, and R. Roth for comments on the manuscript.  
+
+## Author contributions:  
+
+Author contributions:S.B. and J.P.L. conceived and oversaw all aspects of the study. J.P.L., A.M.B., C.P.S., R.P.C.O., V.T., M.Y.G., R.D., L.Z., T.A.B., K.T., J.H., Y.Li and P.J.L. analyzed cellular specificity of SLED vectors. Y.Liu and M.S.S. assisted with subretinal injections. Y.Li and S.S. assisted with human iPSC neuronal culture. D.B., P.P., P.O.K., and T.A.B. assisted with development of calcium indicator SLED vectors. M.Y. and N.P. conducted ERG analysis. B.L. assisted with computational efforts. R.L.H. supervised analysis of SLED.ENS and SLED.NPL vectors. W.B.D., F.R., B.D., K.T. and J.H. carried out studies of oncolytic SLED vectors. K.J.N., J.J.S., and R.V. carried out all ferret studies. J.P.L., A.M.B, C.P.S. and S.B. drafted the manuscript. All authors approved the final manuscript.
+
+<--- Page Split --->
+
+## References  
+
+1. Liu, Y., Hegarty, S., Winter, C., Wang, F. & He, Z. Viral vectors for neuronal cell type-specific visualization and manipulations. Curr. Opin. Neurobiol. 63, 67–76 (2020).2. Weissleder, R., Kelly, K., Sun, E. Y., Shtatland, T. & Josephson, L. Cell-specific targeting of nanoparticles by multivalent attachment of small molecules. Nat. Biotechnol. 23, 1418–1423 (2005).3. Bedbrook, C. N., Deverman, B. E. & Gradinaru, V. Viral Strategies for Targeting the Central and Peripheral Nervous Systems. Annu. Rev. Neurosci. 41, 323–348 (2018).4. Dimidschstein, J. et al. A viral strategy for targeting and manipulating interneurons across vertebrate species. Nat. Neurosci. 19, 1743–1749 (2016).5. de Leeuw, C. N. et al. rAAV-compatible MiniPromoters for restricted expression in the brain and eye. Mol. Brain 9, 52 (2016).6. Mich, J. K. et al. Functional enhancer elements drive subclass-selective expression from mouse to primate neocortex. Cell Rep. 34, 108754 (2021).7. Nair, R. R., Blankvoort, S., Lagartos, M. J. & Kentros, C. Enhancer-Driven Gene Expression (EDGE) Enables the Generation of Viral Vectors Specific to Neuronal Subtypes. iScience 23, 100888 (2020).8. Doronin, K. et al. Tissue-specific, tumor-selective, replication-competent adenovirus vector for cancer gene therapy. J. Virol. 75, 3314–3324 (2001).9. Srivastava, A. In vivo tissue-tropism of adeno-associated viral vectors. Curr. Opin. Virol. 21, 75–80 (2016).10. Ortiz, J. L. S. Engineering of Novel Adeno-Associated Virus Vectors for Gene Therapy Applications. (2016).11. Dhungel, B., Ramlogan-Steel, C. A. & Steel, J. C. MicroRNA-Regulated Gene Delivery Systems for Research and Therapeutic Purposes. Molecules 23, (2018).12. Singh, H. M. et al. MicroRNA-sensitive oncolytic measles virus for chemovirotherapy of pancreatic cancer. Mol Ther Oncolytics 21, 340–355 (2021).
+
+<--- Page Split --->
+
+13. Keaveney, M. K. et al. A MicroRNA-Based Gene-Targeting Tool for Virally Labeling Interneurons in the Rodent Cortex. Cell Rep. 24, 294–303 (2018).  
+
+14. Buck, T. M. & Wijnholds, J. Recombinant Adeno-Associated Viral Vectors (rAAV)-Vector Elements in Ocular Gene Therapy Clinical Trials and Transgene Expression and Bioactivity Assays. Int. J. Mol. Sci. 21, (2020).  
+
+15. Orengo, J. P., Bundman, D. & Cooper, T. A. A bichromatic fluorescent reporter for cell-based screens of alternative splicing. Nucleic Acids Res. 34, e148 (2006).  
+
+16. Hu, N. et al. Non-invasive monitoring of alternative splicing outcomes to identify candidate therapies for myotonic dystrophy type 1. Nat. Commun. 9, 5227 (2018).  
+
+17. Star, E. et al. A drug-repositioning screen using splicing-sensitive fluorescent reporters identifies novel modulators of VEGF-A splicing with anti-angiogenic properties. Oncogenesis 10, 1–12 (2021).  
+
+18. Takeuchi, A., Hosokawa, M., Nojima, T. & Hagiwara, M. Splicing reporter mice revealed the evolutionally conserved switching mechanism of tissue-specific alternative exon selection. PLoS One 5, e10946 (2010).  
+
+19. Massaro, G. et al. Fetal gene therapy for neurodegenerative disease of infants. Nat. Med. 24, 1317–1323 (2018).  
+
+20. Tabebordbar, M. et al. In vivo gene editing in dystrophic mouse muscle and muscle stem cells. Science 351, 407–411 (2016).  
+
+21. Nathwani, A. C. et al. Long-term safety and efficacy of factor IX gene therapy in hemophilia B. N. Engl. J. Med. 371, 1994–2004 (2014).  
+
+22. Bennett, J. et al. Safety and durability of effect of contralateral-eye administration of AAV2 gene therapy in patients with childhood-onset blindness caused by RPE65 mutations: a follow-on phase 1 trial. Lancet 388, 661–672 (2016).  
+
+23. Wang, D., Tai, P. W. L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy delivery. Nat. Rev. Drug Discov. 18, 358–378 (2019).  
+
+24. Sakharkar, M. K., Chow, V. T. K. & Kanguene, P. Distributions of exons and introns in the human genome. In Silico Biol. 4, 387–393 (2004).  
+
+25. Raj, B. & Blencowe, B. J. Alternative Splicing in the Mammalian Nervous System: Recent Insights into Mechanisms and Functional Roles. Neuron 87, 14–27 (2015).  
+
+26. Hallegger, M., Llorian, M. & Smith, C. W. J. Alternative splicing: global insights.
+
+<--- Page Split --->
+
+FEBS J. 277, 856- 866 (2010).  
+
+27. Ling, J. P. et al. ASCOT identifies key regulators of neuronal subtype-specific splicing. Nat. Commun. 11, 137 (2020).  
+
+28. Wilks, C., Gaddipati, P., Nellore, A. & Langmead, B. Snaptron: querying splicing patterns across tens of thousands of RNA-seq samples. Bioinformatics 34, 114-116 (2018).  
+
+29. Wilks, C. et al. recount3: summaries and queries for large-scale RNA-seq expression and splicing. Genome Biol. 22, 323 (2021).  
+
+30. Kim, J. H. et al. High cleavage efficiency of a 2A peptide derived from porcine teschovirus-1 in human cell lines, zebrafish and mice. PLoS One 6, e18556 (2011).  
+
+31. Tremblay, R. G. et al. Differentiation of mouse Neuro 2A cells into dopamine neurons. J. Neurosci. Methods 186, 60-67 (2010).  
+
+32. Kugler, S., Kilic, E. & Bahr, M. Human synapsin 1 gene promoter confers highly neuron-specific long-term transgene expression from an adenoviral vector in the adult rat brain depending on the transduced area. Gene Ther. 10, 337-347 (2003).  
+
+33. de Melo, J. & Blackshaw, S. In Vivo Electroporation of Developing Mouse Retina. Methods Mol. Biol. 1715, 101-111 (2018).  
+
+34. Swanson, G. T., Kamboj, S. K. & Cull-Candy, S. G. Single-channel properties of recombinant AMPA receptors depend on RNA editing, splice variation, and subunit composition. J. Neurosci. 17, 58-69 (1997).  
+
+35. Mosbacher, J. et al. A molecular determinant for submillisecond desensitization in glutamate receptors. Science 266, 1059-1062 (1994).  
+
+36. Koike, M., Tsukada, S., Tsuzuki, K., Kijima, H. & Ozawa, S. Regulation of kinetic properties of GluR2 AMPA receptor channels by alternative splicing. J. Neurosci. 20, 2166-2174 (2000).  
+
+37. Sommer, B. et al. Flip and flop: a cell-specific functional switch in glutamate-operated channels of the CNS. Science 249, 1580-1585 (1990).  
+
+38. Penn, A. C., Balik, A., Wozny, C., Cais, O. & Greger, I. H. Activity-mediated AMPA receptor remodeling, driven by alternative splicing in the ligand-binding domain. Neuron 76, 503-510 (2012).  
+
+39. Wright, A. & Vissel, B. The essential role of AMPA receptor GluR2 subunit RNA
+
+<--- Page Split --->
+
+editing in the normal and diseased brain. Front. Mol. Neurosci. 5, 34 (2012).  
+
+40. Tasic, B. et al. Shared and distinct transcriptomic cell types across neocortical areas. Nature 563, 72–78 (2018).  
+
+41. Furlanis, E., Traunmüller, L., Fucile, G. & Scheiffele, P. Landscape of ribosome-engaged transcript isoforms reveals extensive neuronal-cell-class-specific alternative splicing programs. Nat. Neurosci. 22, 1709–1717 (2019).  
+
+42. Huntley, M. A. et al. Genome-Wide Analysis of Differential Gene Expression and Splicing in Excitatory Neurons and Interneuron Subtypes. J. Neurosci. 40, 958–973 (2020).  
+
+43. Dana, H. et al. High-performance calcium sensors for imaging activity in neuronal populations and microcompartments. Nat. Methods 16, 649–657 (2019).  
+
+44. Dana, H. et al. Sensitive red protein calcium indicators for imaging neural activity. Elife 5, (2016).  
+
+45. Boon, C. J. F. et al. The spectrum of retinal dystrophies caused by mutations in the peripherin/RDS gene. Prog. Retin. Eye Res. 27, 213–235 (2008).  
+
+46. Cheng, T. et al. The effect of peripherin/rds haploinsufficiency on rod and cone photoreceptors. J. Neurosci. 17, 8118–8128 (1997).  
+
+47. Jansen, H. G. & Sanyal, S. Development and degeneration of retina in rds mutant mice: electron microscopy. J. Comp. Neurol. 224, 71–84 (1984).  
+
+48. Sohocki, M. M. et al. Prevalence of mutations causing retinitis pigmentosa and other inherited retinopathies. Hum. Mutat. 17, 42–51 (2001).  
+
+49. Stuck, M. W., Conley, S. M. & Naash, M. I. PRPH2/RDS and ROM-1: Historical context, current views and future considerations. Prog. Retin. Eye Res. 52, 47–63 (2016).  
+
+50. Sarra, G. M. et al. Gene replacement therapy in the retinal degeneration slow (rds) mouse: the effect on retinal degeneration following partial transduction of the retina. Hum. Mol. Genet. 10, 2353–2361 (2001).  
+
+51. Flannery, J. G. et al. Efficient photoreceptor-targeted gene expression in vivo by recombinant adeno-associated virus. Proc. Natl. Acad. Sci. U. S. A. 94, 6916–6921 (1997).  
+
+52. Ali, R. R. et al. Restoration of photoreceptor ultrastructure and function in retinal
+
+<--- Page Split --->
+
+degeneration slow mice by gene therapy. Nat. Genet. 25, 306–310 (2000).  
+
+53. Andtbacka, R. H. I. et al. Talimogene Laherparepvec Improves Durable Response Rate in Patients With Advanced Melanoma. J. Clin. Oncol. 33, 2780–2788 (2015).  
+
+54. Straathof, K. C. et al. An inducible caspase 9 safety switch for T-cell therapy. Blood vol. 105 4247–4254 (2005).  
+
+55. Wu, C. et al. Development of an inducible caspase-9 safety switch for pluripotent stem cell–based therapies. Molecular Therapy - Methods & Clinical Development vol. 1 14053 (2014).  
+
+56. Inoue, D. et al. Spliceosomal disruption of the non-canonical BAF complex in cancer. Nature 574, 432–436 (2019).  
+
+57. Jager, M. J., Magner, J. A. B., Ksander, B. R. & Dubovy, S. R. Uveal Melanoma Cell Lines: Where do they come from? (An American Ophthalmological Society Thesis). Trans. Am. Ophthalmol. Soc. 114, T5 (2016).  
+
+58. Boye, S. E. et al. The human rhodopsin kinase promoter in an AAV5 vector confers rod- and cone-specific expression in the primate retina. Hum. Gene Ther. 23, 1101–1115 (2012).  
+
+59. Manfredi, A. et al. Combined rod and cone transduction by adeno-associated virus 2/8. Hum. Gene Ther. 24, 982–992 (2013).  
+
+60. Beltran, W. A. et al. rAAV2/5 gene-targeting to rods: dose-dependent efficiency and complications associated with different promoters. Gene Ther. 17, 1162–1174 (2010).  
+
+61. Khani, S. C. et al. AAV-mediated expression targeting of rod and cone photoreceptors with a human rhodopsin kinase promoter. Invest. Ophthalmol. Vis. Sci. 48, 3954–3961 (2007).  
+
+62. Xiong, W. et al. AAV -regulatory sequences are correlated with ocular toxicity. Proc. Natl. Acad. Sci. U. S. A. 116, 5785–5794 (2019).  
+
+63. Philpott, M. et al. Nanopore sequencing of single-cell transcriptomes with scCOLOR-seq. Nat. Biotechnol. 39, 1517–1520 (2021).  
+
+64. Hagemann-Jensen, M. et al. Single-cell RNA counting at allele and isoform resolution using Smart-seq3. Nat. Biotechnol. 38, 708–714 (2020).  
+
+65. Hagemann-Jensen, M., Ziegenhain, C. & Sandberg, R. Scalable full-transcript
+
+<--- Page Split --->
+
+coverage single cell RNA sequencing with Smart- seq3xpress. bioRxiv 2021.07.10.451889 (2021) doi:10.1101/2021.07.10.451889.  
+
+66. Lachmann, A. et al. Massive mining of publicly available RNA-seq data from human and mouse. Nat. Commun. 9, 1366 (2018).  
+
+67. Zheng, S. Alternative splicing programming of axon formation. Wiley Interdiscip. Rev. RNA 11, e1585 (2020).  
+
+68. Gomez, A. M., Traunmüller, L. & Scheiffele, P. Neurexins: molecular codes for shaping neuronal synapses. Nat. Rev. Neurosci. 22, 137–151 (2021).  
+
+69. Cheng, J., Çelik, M. H., Kundaje, A. & Gagneur, J. Publisher Correction: MTSplice predicts effects of genetic variants on tissue-specific splicing. Genome Biol. 22, 107 (2021).  
+
+70. Jaganathan, K. et al. Predicting Splicing from Primary Sequence with Deep Learning. Cell 176, 535–548.e24 (2019).  
+
+71. Monteys, A. M. et al. Regulated control of gene therapies by drug-induced splicing. Nature 596, 291–295 (2021).  
+
+72. Mathur, M. et al. Programmable mutually exclusive alternative splicing for generating RNA and protein diversity. Nat. Commun. 10, 2673 (2019).  
+
+73. Jiang, K. et al. Programmable eukaryotic protein expression with RNA sensors. bioRxiv 2022.01.26.477951 (2022) doi:10.1101/2022.01.26.477951.  
+
+74. Jain, R. et al. MicroRNAs Enable mRNA Therapeutics to Selectively Program Cancer Cells to Self-Destruct. Nucleic Acid Ther. 28, 285–296 (2018).  
+
+75. Sago, C. D. et al. Augmented lipid-nanoparticle-mediated in vivo genome editing in the lungs and spleen by disrupting Cas9 activity in the liver. Nat Biomed Eng 6, 157–167 (2022).  
+
+76. Weinstein, J. N. et al. The Cancer Genome Atlas Pan-Cancer analysis project. Nat. Genet. 45, 1113–1120 (2013).  
+
+77. GTEx Consortium. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science 369, 1318–1330 (2020).  
+
+78. De Waard-Siebinga, I. et al. Establishment and characterization of an uveal-melanoma cell line. Int. J. Cancer 62, 155–161 (1995).
+
+<--- Page Split --->
+
+79. Skander, B. R., Rubsamen, P. E., Olsen, K. R., Cousins, S. W. & Streilein, J. W. Studies of tumor-infiltrating lymphocytes from a human choroidal melanoma. Invest. Ophthalmol. Vis. Sci. 32, 3198–3208 (1991).80. Griewank, K. G. et al. Genetic and molecular characterization of uveal melanoma cell lines. Pigment Cell Melanoma Res. 25, 182–187 (2012).81. Dalton, W. B. et al. Hotspot SF3B1 mutations induce metabolic reprogramming and vulnerability to serine deprivation. J. Clin. Invest. 129, 4708–4723 (2019).82. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012).83. Bygrave, A. M. et al. Btbd11 is an inhibitory interneuron specific synaptic scaffolding protein that supports excitatory synapse structure and function. bioRxiv 2021.11.01.466782 (2021) doi:10.1101/2021.11.01.466782.84. Santiago, C. P. et al. A Drug-Tunable Gene Therapy for Broad-Spectrum Protection against Retinal Degeneration. Mol. Ther. 26, 2407–2417 (2018).85. Kinoshita, J. & Peachey, N. S. Noninvasive Electroretinographic Procedures for the Study of the Mouse Retina. Curr. Protoc. Mouse Biol. 8, 1–16 (2018).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- SLEDSupplementalFigures031422.pdf- SupplementalVideo1CaRPv110x.mp4- SupplementalVideo2CaRPv120x.mp4
+
+<--- Page Split --->

preprint/preprint__b3507d9826a318cafe485c6b8d4bed844b52604a56d4587e009677fe91f86a69/preprint__b3507d9826a318cafe485c6b8d4bed844b52604a56d4587e009677fe91f86a69_det.mmd

@@ -0,0 +1,660 @@
+<|ref|>title<|/ref|><|det|>[[44, 106, 896, 178]]<|/det|>
+# Cell-specific regulation of gene expression using splicing-dependent frameshifting.  
+
+<|ref|>text<|/ref|><|det|>[[44, 194, 810, 240]]<|/det|>
+Seth Blackshaw ( \(\boxed{\bullet}\) sblack@jhm.edu) Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 1338- 8476  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 808, 285]]<|/det|>
+Jonathan Ling Johns Hopkins University School of Medicine https://orcid.org/0000- 0003- 1927- 9729  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 450, 331]]<|/det|>
+Alexei Bygrave Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 336, 810, 378]]<|/det|>
+Clayton Santiago Johns Hopkins University School of Medicine https://orcid.org/0000- 0001- 7191- 668X  
+
+<|ref|>text<|/ref|><|det|>[[44, 382, 808, 425]]<|/det|>
+Roger Carmen- Orozco Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 2460- 9270  
+
+<|ref|>text<|/ref|><|det|>[[44, 428, 450, 470]]<|/det|>
+Vickie Trinh Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 475, 198, 515]]<|/det|>
+Minzhong Yu Cleveland Clinic  
+
+<|ref|>text<|/ref|><|det|>[[44, 521, 808, 563]]<|/det|>
+Yini Li Johns Hopkins University School of Medicine https://orcid.org/0000- 0003- 2213- 0650  
+
+<|ref|>text<|/ref|><|det|>[[44, 567, 808, 610]]<|/det|>
+Jeong Han Johns Hopkins University School of Medicine https://orcid.org/0000- 0003- 3277- 7669  
+
+<|ref|>text<|/ref|><|det|>[[44, 614, 450, 655]]<|/det|>
+Kamil Taneja Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 660, 450, 702]]<|/det|>
+Ying Liu Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 707, 450, 749]]<|/det|>
+Rochinelle Dongmo Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 754, 450, 795]]<|/det|>
+Travis Babola Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 800, 450, 841]]<|/det|>
+Patrick Parker Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 846, 450, 888]]<|/det|>
+Lizhi Jiang Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[44, 893, 808, 935]]<|/det|>
+Patrick Leavey Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 2822- 1118  
+
+<|ref|>text<|/ref|><|det|>[[44, 939, 175, 958]]<|/det|>
+Jennifer Smith
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 45, 451, 64]]<|/det|>
+Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 70, 450, 110]]<|/det|>
+Rachel Vistein Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 116, 450, 157]]<|/det|>
+Megan Gimmel Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 163, 450, 203]]<|/det|>
+Benjamin Dubner Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 210, 808, 250]]<|/det|>
+Patric Teodorescu Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 4129- 8478  
+
+<|ref|>text<|/ref|><|det|>[[45, 255, 450, 296]]<|/det|>
+Theodoros K arantanos Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 301, 450, 342]]<|/det|>
+Gabriel Ghiaur Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 348, 636, 389]]<|/det|>
+Patrick Kanold Johns Hopkins University https://orcid.org/0000- 0002- 7529- 5435  
+
+<|ref|>text<|/ref|><|det|>[[45, 394, 636, 435]]<|/det|>
+Dwight Bergles Johns Hopkins University https://orcid.org/0000- 0002- 7133- 7378  
+
+<|ref|>text<|/ref|><|det|>[[45, 440, 636, 481]]<|/det|>
+Ben Langmead Johns Hopkins University https://orcid.org/0000- 0003- 2437- 1976  
+
+<|ref|>text<|/ref|><|det|>[[45, 486, 450, 527]]<|/det|>
+Shuying Sun Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 533, 808, 574]]<|/det|>
+Kristina Nielsen Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 9155- 2972  
+
+<|ref|>text<|/ref|><|det|>[[45, 579, 198, 619]]<|/det|>
+Neal Peachy Cleveland Clinic  
+
+<|ref|>text<|/ref|><|det|>[[45, 625, 450, 666]]<|/det|>
+Mandeep Singh Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 671, 450, 712]]<|/det|>
+William Dalton Johns Hopkins University School of Medicine  
+
+<|ref|>text<|/ref|><|det|>[[45, 718, 808, 759]]<|/det|>
+Fatemeh Rajaii Johns Hopkins University School of Medicine https://orcid.org/0000- 0002- 1012- 2293  
+
+<|ref|>text<|/ref|><|det|>[[45, 764, 808, 806]]<|/det|>
+Richard Huganir Johns Hopkins School of Medicine https://orcid.org/0000- 0001- 9783- 5183  
+
+<|ref|>text<|/ref|><|det|>[[45, 848, 101, 866]]<|/det|>
+Article  
+
+<|ref|>text<|/ref|><|det|>[[45, 886, 137, 904]]<|/det|>
+Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[45, 923, 314, 942]]<|/det|>
+Posted Date: March 24th, 2022
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 475, 64]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 1416757/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 82, 911, 125]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 161, 925, 205]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on October 1st, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 33523- 2.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 821, 128]]<|/det|>
+Title: Cell- specific regulation of gene expression using splicing- dependent frameshifting.  
+
+<|ref|>text<|/ref|><|det|>[[113, 147, 881, 309]]<|/det|>
+Authors: Jonathan P. Ling \(^{1,10,*,\#}\) , Alexei M. Bygrave \(^{2,\#}\) , Clayton P. Santiago \(^{2,\#}\) , Rogger P. Carmen- Orozco \(^{2}\) , Vickie Trinh \(^{2}\) , Minzhong Yu \(^{11,12}\) , Yini Li \(^{1}\) , Jeong Han \(^{4}\) , Kamil Taneja \(^{4}\) , Ying Liu \(^{4}\) , Rocinelle Dongmo \(^{2}\) , Travis A. Babola \(^{2,3}\) , Patrick Parker \(^{2}\) , Lizhi Jiang \(^{2}\) , Patrick J. Leavey \(^{2}\) , Jennifer J. Smith \(^{2,6}\) , Rachel Vistein \(^{2,6}\) , Megan Y. Gimen \(^{2}\) , Benjamin Dubner \(^{7}\) , Eric Helmenstine \(^{7}\) , Patric Teodorescu \(^{7}\) , Theodore Karantanos \(^{7}\) , Gabriel Ghiaur \(^{7}\) , Patrick O. Kanold \(^{2,3,10}\) , Dwight Bergles \(^{2,10}\) , Ben Langmead \(^{9,10}\) , Shuying Sun \(^{1}\) , Kristina J. Nielsen \(^{2,6,10}\) , Neal Peachey \(^{11,12,13}\) , Mandeep S. Singh \(^{4}\) , W. Brian Dalton \(^{7}\) , Fatemeh Rajaii \(^{4}\) , Richard L. Huganir \(^{2,10}\) , and Seth Blackshaw \(^{2,5,8,10,*}\)  
+
+<|ref|>text<|/ref|><|det|>[[114, 329, 883, 528]]<|/det|>
+Affiliations: \(^{1}\) Departments of Pathology, \(^{2}\) Solomon H. Snyder Department of Neuroscience, \(^{3}\) Biomedical Engineering, \(^{4}\) Wilmer Eye Institute, \(^{5}\) Neurology, \(^{6}\) Zanvyl Krieger Mind/Brain Institute, \(^{7}\) Oncology, and \(^{8}\) Institute for Cell Engineering, Johns Hopkins University School of Medicine, Baltimore, MD 21205, USA; \(^{9}\) Department of Computer Science and \(^{10}\) Kavli Neuroscience Discovery Institute, Johns Hopkins University, Baltimore, MD 21218, USA; \(^{11}\) Department of Ophthalmic Research, Cole Eye Institute, Cleveland Clinic Foundation, Cleveland, OH 44195, USA; \(^{12}\) Department of Ophthalmology, Cleveland Clinic Lerner College of Medicine of Case Western Reserve University, Cleveland, OH 44195, USA; \(^{13}\) Research Service, Louis Stokes Cleveland VA Medical Center, Cleveland, OH 44106, USA.  
+
+<|ref|>text<|/ref|><|det|>[[114, 569, 410, 608]]<|/det|>
+\(^{\#}\) indicates equal contributions \(*\) indicates corresponding authors
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[174, 91, 257, 108]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[113, 109, 877, 469]]<|/det|>
+Precise and reliable cell- specific gene delivery remains technically challenging. Here we report a splicing- based approach for controlling gene expression whereby separate translational reading frames are coupled to the inclusion or exclusion of cell- specific alternative exons. Candidate exons are identified by analyzing thousands of publicly available RNA sequencing datasets and filtering by cell specificity, sequence conservation, and local intron length. This method, which we denote splicing- linked expression design (SLED), can be combined in a Boolean manner with existing techniques such as minipromoters and viral capsids. SLED vectors can leverage the strong expression of constitutive promoters, without sacrificing precision, by decoupling the tradeoff between promoter strength and selectivity. We generated SLED vectors to selectively target all neurons, photoreceptors, or excitatory neurons, and demonstrated that specificity was retained in vivo when delivered using AAVs. We further demonstrated the utility of SLED by creating what would otherwise be unobtainable research tools, specifically a GluA2 flip/flop reporter and a dual excitatory/inhibitory neuronal calcium indicator. Finally, we show the translational potential of SLED by rescuing photoreceptor degeneration in \(P r p h 2^{r d s / r d s}\) mice and by developing an oncolytic vector that can selectively induce apoptosis in SF3B1 mutant cancer cells. The flexibility of SLED technology enables new avenues for basic and translational research.  
+
+<|ref|>sub_title<|/ref|><|det|>[[174, 491, 291, 508]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[114, 510, 880, 708]]<|/det|>
+Cell type- specific control of gene expression is essential for both basic and translational biological research. Though this is often achieved using transgenic animal models, these are costly, difficult to scale, and restricted to a limited number of model organisms. An alternative approach, which is directly applicable for therapeutic purposes, is to use exogenous viral or plasmid constructs to selectively express genes of interest in specific cell types \(^{1 - 3}\) . These methods rely on the use of minimal promoters and enhancers that place constructs under the regulation of cell type- specific transcription factors \(^{4 - 8}\) , unique capsid proteins or surface features to limit the range of cell types infected by viral constructs \(^{9,10}\) , or the inclusion of specific miRNA seed sequences to inhibit off- target expression \(^{11 - 13}\) .  
+
+<|ref|>text<|/ref|><|det|>[[114, 710, 880, 870]]<|/det|>
+Current approaches, however, have important limitations. Minipromoter and enhancer- based constructs are difficult to develop and test in a systematic manner. For example, when removed from their genomic context, or tested in other species, they often show unpredictable patterns of cell- specific expression, despite showing high sequence conservation and patterns of chromatin accessibility \(^{14}\) . Furthermore, while viral serotypes typically provide enriched cell- specificity, thus far they are not strictly cell type- specific, and are not relevant for viral- independent gene delivery strategies. Likewise, microRNA- based approaches can help reduce off- target delivery in certain
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 861, 128]]<|/det|>
+cells, but must be used in conjunction with other methods to achieve cell type- specific expression.  
+
+<|ref|>text<|/ref|><|det|>[[113, 130, 881, 509]]<|/det|>
+An orthogonal strategy that can be combined with the above approaches would be to harness alternative splicing of mRNA (RNA splicing) to direct cell- specific gene expression. RNA splicing is a highly regulated process that generates transcriptomic and proteomic diversity and many splicing patterns are correlated with unique cell types or cellular states. Fluorescent reporter vectors have been used to study the mechanistic regulation of alternative splicing events \(^{15 - 18}\) , but the large size of most intronic sequences precludes their inclusion in the most commonly used viral vectors. Adeno- associated virus (AAV) vectors are a leading platform for gene therapy due to their demonstrated safety and long- term efficacy across a variety of tissues \(^{19 - 22}\) , but these viruses are limited by a maximum packaging size of \(\sim 4.7kb^{23}\) . Since the average intron length in the human genome is \(\sim 5.4kb\) in length \(^{24}\) , it has been historically difficult to identify cell type- specific patterns of alternative splicing that are potentially compatible with AAV vectors \(^{25,26}\) . However, rapid adoption of full- length RNA sequencing (RNA- Seq) over the past decade has led to the public archival of datasets obtained from various cell types across multiple species. Furthermore, recent computational methods have been developed to comprehensively analyze patterns of alternative splicing across hundreds of thousands of publicly archived RNA- Seq datasets \(^{27 - 29}\) . We have used these databases to identify many cell type- specific alternative exons that are suitable for use in AAV vectors.  
+
+<|ref|>text<|/ref|><|det|>[[112, 510, 881, 870]]<|/det|>
+In this study, we have developed a suite of AAV- based tools that direct pan- neuronal, excitatory neuron, and photoreceptor- specific gene expression via a splicing- linked expression design (SLED) strategy. This method uses splicing- dependent frameshifting, in combination with both ubiquitous and cell type- specific promoters, to drive cell type- specific expression of fluorescent proteins and other genes of interest. We show that, due to their small size, SLED constructs can be packaged into AAV vectors and that cell specificity is maintained in vivo across multiple species. Furthermore, the SLED method can be used to create previously unobtainable research tools. We miniaturized the Gria2 flip/flop intron for AAV packaging to monitor this mutually exclusive splicing event at single- neuron resolution. We also demonstrated that dual calcium sensors can be simultaneously expressed in different cell types using a single expression vector, instead of using multiple viruses or transgenic animals. Finally, we demonstrated that SLED- based AAV constructs perform as efficiently as state- of- the- art minipromoter vectors for functional rescue of photoreceptor dystrophies, and also show that SLED can be used to selectively target SF3B1 mutant cancer cells for oncolytic therapy. These results demonstrate that SLED- based tools are compatible with existing methods for regulating cell type- specific gene expression, and that SLED is broadly useful for a range of basic and translational research applications.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[173, 91, 247, 108]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[172, 109, 789, 128]]<|/det|>
+## Identification of cell-specific exons and SLED vector construction  
+
+<|ref|>text<|/ref|><|det|>[[114, 130, 879, 348]]<|/det|>
+To test the ability of alternative splicing to mediate cell type- specific expression of reporter constructs, we modified an existing bichromatic reporter plasmid 15. In this construct, dsRed is expressed when the default splicing pathway is used. When a cell type- specific alternative exon is spliced in, however, this results in a reading frame shift that leads to the expression of EGFP. In cases where the sequence length of the cell type- specific exon is a multiple of 3 and lacks a stop codon in the initiating translational reading frame, point mutations were introduced to create a frameshifting cell type- specific exon. Importantly, the dsRed sequence is modified to remove stop codons that would otherwise occur in the EGFP reading frame (Fig. 1a). 2A self- cleaving peptide sequences 30 were also included in front of each fluorescent protein to allow expression of the fluorescent protein independent of leader sequences (Fig. S1).  
+
+<|ref|>text<|/ref|><|det|>[[113, 350, 881, 750]]<|/det|>
+We used three criteria to select alternative exons for analysis. First, alternative exons needed to show highly cell type- specific patterns of inclusion. Second, cell type- specific patterns of splicing needed to be conserved between mice and humans. Finally, the size of the intronic sequence used needed to be less than 2 kb. Using a computational resource that catalogs cell type- specific splicing patterns (ASCOT) 27, we identified \(\sim 1000\) neuronal- enriched alternative exons, of which \(\sim 200\) had intronic lengths of \(< 2\) kb (Fig. 1b,c). \(\sim 99\%\) of exons show high conservation (vertebrate phyloP score \(>1.5\) ) of neuron- enriched splicing between mouse and human (Fig. 1d,e). A neuronally- enriched exon in the gene encoding the ubiquitously- expressed actin- binding protein Plastin 3 (PLS3) was selected for characterization. A similar process was used to identify a photoreceptor- specific exon in the gene encoding the ubiquitously expressed subunit of the ATPase \(\mathrm{Na + / K + }\) Transporting Subunit Beta 2 (ATP1B2) (Fig. S2). For proof- of- concept, we transfected the pan- neuronal and photoreceptor- specific SLED vectors into HEK293, HepG2, and N2a neuroblastoma cell lines to determine specificity (Fig. 1f). While dsRed was expressed in all cells, EGFP was only observed with pan- neuronal SLED in N2a cells, which exhibit neuronal precursor- like characteristics 31. No expression of EGFP was observed in any cells when transfected with the photoreceptor- specific SLED construct, supporting the cell type specificity of the ATP1B2 alternative exon. Specificity was determined at the single cell level by evaluating the \(\log_2\) ratio of EGFP/dsRed fluorescence (Fig. 1f).  
+
+<|ref|>sub_title<|/ref|><|det|>[[170, 770, 868, 789]]<|/det|>
+## SLED vectors maintain cell-specific expression when delivered using AAV  
+
+<|ref|>text<|/ref|><|det|>[[115, 790, 876, 889]]<|/det|>
+To test whether SLED vectors retain specificity when packaged into AAV, we cloned pan- neuronal SLED (SLED.NPL) and photoreceptor- specific SLED (SLED.RAB) into an AAV backbone (Fig. 2a). Furthermore, we sought to test whether cell type- specific minipromoters could be combined with SLED- based constructs in a Boolean manner to provide more selective cell type- specific expression using AAV vectors. To
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[80, 50, 920, 840]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 848, 861, 909]]<|/det|>
+<center>Figure 1. Identification of cell-specific exons and SLED vector construction (a) Diagramatic sketch of SLED vector design strategy. SLED is compatible with any promoter. A frameshifting mutation is introduced into a cell-specific alternative exon </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[123, 93, 870, 416]]<|/det|>
+to create two potential translational reading frames from an upstream start codon. In most SLED vectors, exon skipping will produce a red fluorescent protein while exon inclusion will shift the reading frame to produce a green fluorescent protein. (b) Ranking all neuron- enriched exons by the percent spliced- in (PSI) difference between neurons and other cell types. Exons were identified from mouse RNA- Seq datasets analyzed with the ASCOT pipeline \(^{27}\) . Approximately 1000 neuron- enriched exons have a \(\Delta \mathrm{PSI}\) greater than 20. (c) Among these top 1000 exons, approximately 200 candidates reside in introns \(< 2\mathrm{kb}\) in length. (d, e) UCSC genome browser views of the neuron- specific exon in \(P / s3\) that is used in SLED.NPL. Exon incorporation is only observed in neuronal datasets (red arrows) from both mouse (d) and human (e). A similar strategy was used to identify the photoreceptor- specific exon in \(Atp1b2\) . These exons were used to generate SLED vectors that were then tested in HEK293, HepG2, and N2a cancer cell lines (f). As predicted, neither vector showed EGFP expression, indicating an absence of cell- specific exon incorporation, except when the neuron- specific SLED was transfected into N2a cells, reflecting the neuronal characteristics of N2a neuroblastoma cells. Scale bars \(= 50\mu \mathrm{m}\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 428, 880, 508]]<|/det|>
+do this, we combined the pan- neuronal hSyn minipromoter \(^{32}\) with an alternative exon of the gene that encodes the ubiquitously expressed clathrin complex interactor Synergin gamma (SYNRG), which in the brain is specific to excitatory neurons and glia (SLED.ENS, Fig. 2a, Fig. S2).  
+
+<|ref|>text<|/ref|><|det|>[[113, 509, 877, 790]]<|/det|>
+We first tested the expression of these constructs using plasmid transfection and electroporation. As expected, we observed selective expression of EGFP in neurons following transfection of SLED.NPL into primary rat hippocampal cultures, although expression of the default splicing- driven dsRed was observed in transfected neurons and glia (Fig. S3). Likewise, in neonatal mouse retinal explants electroporated with the SLED.RAB construct \(^{33}\) , we observed expression of dsRed in all postnatally- generated cell types, but EGFP reporter expression is restricted to photoreceptors (Fig. S3). Lastly, we observe that transfection of the SLED.ENS constructs into primary rat hippocampal cultures resulted in selective exclusion of hSyn- driven EGFP expression from somatostatin- expressing GABAergic interneurons (Fig. S3). To further validate the photoreceptor- specificity of the SLED.RAB construct in vivo, postnatal day 0 (P0) mouse retinas were transduced with photoreceptor- specific AAV2.7m8.SLED.RAB and processed 4 weeks later at P30. This revealed highly enriched expression of EGFP in retinal photoreceptors, with inner retinal neurons strongly positive for dsRed (Fig. 2b).  
+
+<|ref|>text<|/ref|><|det|>[[114, 790, 878, 890]]<|/det|>
+We next tested the neuronal- specificity of SLED.NPL packaged into AAV9 by transducing primary rat hippocampal cultures at 1 day in vitro (DIV). At DIV 15, cells were fixed and immunofluorescence conducted for the neuronal marker HuC/D.. Comparison of the ratio of EGFP to dsRed fluorescence revealed that EGFP was highly enriched in HuC/D- positive neurons (Fig. 2c). To test SLED.NPL in vivo, we performed
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[55, 40, 931, 710]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[123, 714, 863, 913]]<|/det|>
+<center>Figure 2. SLED vectors maintain cell-specific expression when delivered using AAV. (a) Diagramatic sketch of pan-neuronal (SLED.NPL), photoreceptor-specific (SLED.RAB), and excitatory neuron-specific (SLED.ENS) vectors designed for AAV packaging. In all SLED vectors, EGFP is translated when cell-specific exons are spliced-in. (b) SLED.RAB, packaged in AAV2.7m8, was intravertically injected into P0 mouse retinas and processed at P30. EGFP is highly enriched in retinal photoreceptors while inner retinal neurons are strongly positive for dsRed. ONL = outer nuclear layer, INL = inner nuclear layer. (c-e) SLED.NPL, packaged in AAV9, was used to transduce primary rat hippocampal cultures (c), mouse cortex (d), and human iPSC-derived neurons (e). In all cases, EGFP is highly enriched in HuC/D+ </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[124, 92, 866, 234]]<|/det|>
+(rat, mouse) or NeuN+ (human) neurons, while non- neuronal cells are strongly positive for dsRed. (f- h) SLED.ENS, packaged in AAV9, was used to transduce primary rat hippocampal cultures (f), mouse hippocampus (g), and ferret cortex (h). In all cases, EGFP is highly enriched in GAD67- excitatory neurons, while GAD67+ inhibitory neurons are strongly positive for dsRed. \*\*\* indicate \(p < 0.0001\) , two- tailed t- test. For ratio calculations in panels b to h, \(n = 205\) (b), \(n = 94\) (c), \(n = 78\) (d), \(n = 55\) (e), \(n = 121\) (f), \(n = 198\) (g), and \(n = 154\) (h). Scale bars \(= 50\mu \mathrm{m}\) .  
+
+<|ref|>text<|/ref|><|det|>[[114, 242, 875, 404]]<|/det|>
+stereotactic injection of AAV9. SLED.NPL into the mouse hippocampal region. Efficient and widespread infection was observed and the neuronal specificity of EGFP expression was maintained (Fig. 2d). AAV- based gene therapies are being explored as treatment options for neurological disorders and SLED vectors may improve the safety and efficacy of these methods. To determine whether AAV9. SLED.NPL maintains neuron- specific expression in human cells, we transduced human iPSC- derived mixed neuronal and glial cultures (Fig. 2e). Here too, we observed strong EGFP expression in NeuN- positive neurons but only dsRed expression in NeuN- negative glial cells.  
+
+<|ref|>text<|/ref|><|det|>[[114, 404, 880, 622]]<|/det|>
+Lastly, we tested the specificity of the excitatory neuron- specific AAV9. SLED.ENS construct. Primary rat hippocampal cultures show strong expression of EGFP in Gad67- negative excitatory neurons, but little or no expression of EGFP in Gad67- positive GABAergic interneurons (Fig. 2f). Stereotactic injection of AAV9. SLED.ENS into mouse hippocampus likewise resulted in strong and broad neuronal expression of EGFP, but exclusion of EGFP signal from dsRed- positive, Gad67- positive interneurons (Fig. 2g). Finally, a similar pattern of exclusion from Gad67- positive interneurons was observed following transduction of primary ferret visual cortex (Fig. 2h). Together, these findings demonstrate that SLED cell specificity is maintained in vitro and in vivo across multiple species, and that mutually exclusive splicing events can be simultaneously monitored using AAV- based SLED tools.  
+
+<|ref|>sub_title<|/ref|><|det|>[[172, 643, 759, 663]]<|/det|>
+## Generation of unique splicing-based tools using SLED vectors  
+
+<|ref|>text<|/ref|><|det|>[[114, 664, 870, 903]]<|/det|>
+We next sought to test whether SLED- based AAV vectors could be adapted to study in vivo patterns of flip/flop splicing in the AMPA- type glutamate receptor subunit GluA2. Flip/flop alternative splicing occurs within the ligand binding domain, and influences AMPA receptors assembly and channel kinetics \(^{34 - 37}\) . The short lengths and high sequence similarity of the mutually exclusive flip/flop exons precludes the use of immunostaining or in situ hybridization to detect their localization in situ, which has effectively restricted previous efforts investigating flip/flop splicing in learning and plasticity to using qRT- PCR analysis \(^{38,39}\) . To address this, we generated an hSyn- driven AAV vector which expresses EGFP when the flop exon is incorporated, and dsRed when the flip exon is incorporated (SLED.GluA2, Fig. 3a). To validate that AAV9. SLED.GluA2 splicing reflected endogenous flip/flop splicing, we electroporated primary rat cortical neurons and FACS- isolated high EGFP- expressing cells. We
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[170, 55, 825, 925]]<|/det|>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[123, 93, 800, 113]]<|/det|>
+## Figure 3: Generation of unique splicing-based tools using SLED vectors  
+
+<|ref|>text<|/ref|><|det|>[[121, 113, 867, 597]]<|/det|>
+(a) Diagramatic sketch of GluA2 (Gria2) flip/flop SLED vector design (SLED.GluA2). 
+(b) To validate that SLED.GluA2 reflected endogenous GluA2 flip/flop splicing patterns, we designed endogenous mRNA-specific primers to PCR amplify the GluA2 flip/flop locus. Although the mutually exclusive flip and flop exons are identical in length and highly similar in sequence, Hpal will selectively digest the flop PCR product into two fragments. SLED.GluA2, packaged into AAV9, was used to electroporate primary rat neuronal cultures and EGFPhigh/mCherrylow cells were isolated using FACS (Fig. S4). RNA was extracted from EGFPhigh/mCherrylow cells and bulk rat neuronal cultures and primers (b) were used to amplify PCR products. 
+(c) Hpal incubation yielded digestion products in the EGFPhigh/mCherrylow cells, which was further confirmed using Sanger sequencing (d, Fig. S4, n=36, \*\*\* indicate p < 0.0001, two-tailed t-test.). (e) Primary rat neuronal cultures transduced with SLED.GluA2 show significantly different EGFP/mCherry ratios between excitatory (GAD67-) and inhibitory (GAD67+) neurons. \*\*\* indicate p < 0.0001, two-tailed t-test. 
+(f) Diagramatic sketch of bicistronic jRGECO1a (inhibitory neurons) and jGCaMP7b (excitatory neurons) SLED vector design (SLED.CaRPv1). (g) Transfection of primary rat neuronal cultures yielded divergent ratios in jGCaMP7b and RGECO1a intensities in excitatory (mDlx-Azurite-) and inhibitory (mDlx-Azurite+) neurons. Data presented represent jGCaMP7b (top row) and jRGECO1a (middle row) intensity values over a 60s time-lapse (4Hz). Bottom row represents a normalized representation of total Ca intensity scaled by the delta between jGCaMP7b and jRGECO1a pixel values. Individual Ca indicator traces are demonstrated in the bottom panels. For ratio calculations in panel e, n=266 (cortex) and n=316 (hippocampus). Scale bars = 50μm (panel e) and 20μm (panel g).  
+
+<|ref|>text<|/ref|><|det|>[[113, 605, 874, 707]]<|/det|>
+designed primers to selectively amplify spliced flip/flop exons in GluA2 mRNA (Fig. 3b). Digestion using Hpal, which selectively cleaves the flop exon into two smaller fragments, revealed expected enrichment of flop exon fragments in the EGFP- enriched fraction (Fig. 3c). This was further confirmed using Sanger sequencing of amplified products (Fig. 3d, Fig. S4).  
+
+<|ref|>text<|/ref|><|det|>[[113, 707, 872, 847]]<|/det|>
+Transduction of AAV9.SLED.GluA2 into both primary rat hippocampal and cortical cultures (Fig. 3e) revealed a variety of different cellular patterns of reporter expression, with EGFP- dominant, dsRed- dominant and mixed cells all present. However, we observe that Gad67- positive hippocampal neurons are enriched for EGFP- dominant expression, matching previous observations obtained using single- cell SMART- Seq analysis and bulk RNA- Seq analysis of RiboTRAP- expressing interneurons \(^{40 - 42}\) (Fig. S5).  
+
+<|ref|>text<|/ref|><|det|>[[114, 847, 864, 907]]<|/det|>
+The translational frameshifting used in SLED vectors also offers the potential to deliver multiple functional payloads, such as genetically encoded calcium sensors or optogenetic actuators, using a single viral vector. As proof of concept, we created a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 875, 388]]<|/det|>
+bicistronic calcium indicator vector based on the excitatory neuron- specific SLED.ENS. We termed this calcium reporter plasmid version 1 (CaRPv1) (Fig. 3f). In CaRPv1, GCaMP7b is expressed in excitatory neurons, while RGECO1a is expressed in the default translational reading frame (inhibitory neurons) \(^{43,44}\) . Identification of excitatory vs inhibitory neurons was established using the \(\Delta\) pixel intensity of normalized fluorescence values, due to differences in dynamic range and baseline fluorescence at resting calcium concentrations for GCaMP7b and RGECO1a (see methods). Transfection of CaRPv1 into primary rat hippocampal cultures revealed the expected patterns of calcium transients (Fig. S6, Supplemental Videos 1 & 2). Furthermore, large and synchronous calcium transients were observed following the addition of bicuculline (a GABA receptor antagonist, used to induce disinhibition), indicating that CaRPv1 was reporting cellular activity as expected (Fig. 3g). In its current design, CaRPv1 is unable to be packaged into AAV due to the size of the ENS intron ( \(\sim 1600bp\) ). However, future deletion mutagenesis and sequence optimization should enable AAV packaging of CaRPv1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[173, 410, 622, 430]]<|/det|>
+## Adapting SLED vectors for translational studies  
+
+<|ref|>text<|/ref|><|det|>[[113, 430, 877, 667]]<|/det|>
+Mutations in the photoreceptor outer segment structural gene PRPH2 cause human retinal degeneration \(^{45}\) and null mutations in Prph2 lead to slow- onset photoreceptor degeneration in mice \(^{46 - 49}\) . AAV- based constructs driven by the photoreceptor- specific minipromoter mOps have been previously used to rescue Prph2 expression in rds/rdr Prph2- deficient mice (Prph2 \(^{rds/rdr}\) ), although only modest photoreceptor preservation, and no long- term recovery of visual function was observed due to weak promoter efficiency \(^{50}\) . We modified the photoreceptor- specific SLED.RAB to selectively express PRPH2 under control of the ubiquitous CBh promoter. In parallel, we generated mOps minipromoter- driven rescue vectors that were used in previous studies \(^{50 - 52}\) . CMV- driven EGFP vectors were also obtained as controls (Fig. 4a). These were all packaged into AAV2.7m8 capsids and injected subretinally at P28 into Prph2 \(^{rds/rdr}\) mice.  
+
+<|ref|>text<|/ref|><|det|>[[113, 669, 882, 870]]<|/det|>
+Mice were injected with AAV at one month of age, and three months later were then analyzed using optical coherence tomography (OCT) to measure the relative thickness of the retinal outer nuclear layer (ONL), where photoreceptors reside. We observed that ONL thickness was similar in both SLED and mOps- regulated AAV constructs, and significantly greater than mice injected with CMV.GFP control virus (Fig. 4b). The amplitude of the light- adapted, cone- mediated, full- field electroretinogram (ERG) was larger in SLED relative to mOps- based rescue constructs, with both showing significantly higher b- wave responses relative to CMV.GFP controls (Fig. 4c). Immunostaining for Prph2 in transduced retina showed no detectable expression in CMV.GFP controls (Fig. 4d), but Prph2 signal was detected in photoreceptor inner
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[66, 55, 930, 940]]<|/det|>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[123, 95, 662, 115]]<|/det|>
+## Figure 4: Adapting SLED vectors for translational studies  
+
+<|ref|>text<|/ref|><|det|>[[120, 113, 870, 780]]<|/det|>
+(a) Diagramatic sketch of GFP under the control of the constitutive CMV promoter (CMV.GFP), PRPH2 under the control of the photoreceptor specific mOps promoter (mOps.PRPH2), and PRPH2 regulated by the photoreceptor-specific SLED.RAB (SLED.PRPH2) vectors designed for AAV packaging. CMV.GFP, mOps.PRPH2, and SLED.PRPH2 were packaged into AAV2.7m8 for testing in Prph2<sup>rds</sup>/<sup>rds</sup> animals. For experimental design, n=6 for each AAV treatment. (b) Average ONL/INL ratios and (c) average light-adapted ERG b-wave amplitudes in three month post-injected Prph2<sup>rds</sup>/<sup>rds</sup> animals treated with CMV.GFP, SLED.PRPH2 or mOps.PRPH2 viral constructs (asterisks indicate p < 0.05, two-tailed t-test, comparison between mOps.PRPH2 and SLED.PRPH2). (d-f) Immunofluorescence staining of retinal sections from Prph2<sup>rds</sup>/<sup>rds</sup> eyes injected with CMV.GFP (d), mOps.PRPH2 (e), and SLED.PRPH2 (f). ONL = outer nuclear layer, INL = inner nuclear layer. EGFP is only detected in CMV.GFP treated controls, but Prph2 signal (yellow arrow) is detected in photoreceptor inner segments of retinas transduced with both mOps.PRPH2 and SLED.PRPH2. (g) UCSC genome browser view of a cryptic exon in UBA1 (green arrow) that is present in cancers with oncogenic SF3B1 mutations (TCGA 76) and absent in all normal human tissues sequenced by the GTEx consortium 77. (h) Diagramatic sketch of bichromatic fluorescent reporter based on the SF3B1<sup>mut</sup> associated exon (SLED.SFUv1) and a similar vector where an inducible iCaspase9 kill switch is coupled to incorporation of the SF3B1<sup>mut</sup>-associated exon. (i) As a proof of concept, SLED.SFUv1 was transfected into uveal melanoma cell lines with (Mel-202) and without (92-1) SF3B1 mutations. EGFP was highly enriched in only Mel-202 cells while dsRed was strongly expressed in 92-1 cells, which was validated using FACS (j). Isogenic cell lines derived from Mel-202 with the SF3B1<sup>R625G</sup> mutation genetically inactivated (PC76B6) and maintained (MR5) showed similarly concordance, with strong EGFP expression only present in the SF3B1<sup>R625G</sup> MR5 cell line. Likewise, strong EGFP expression was only present in SF3B1<sup>K700E</sup> K562 leukemia cells, as compared to wildtype K562 cells. **** indicate p < 0.0001, two-tailed t-test. For ratio calculations, n=245 (92-1), n=1158 (Mel-202), n=377 (PC76B6), n=526 (MR5), n=49 (K562<sup>WT</sup>), n=677 (K562<sup>MUT</sup>). (k) Transfection of wildtype K562 cells and mutant SF3B1<sup>K700E</sup> K562 cells with SLED.SFUv2 revealed strong apoptosis only in SF3B1<sup>K700E</sup> K562 cells treated with the iCaspase9 activating dimerizer (n=3 FACS replicates). Scale bars = 50μm.  
+
+<|ref|>text<|/ref|><|det|>[[115, 789, 839, 828]]<|/det|>
+segments in retinas transduced with both mOps (Fig. 4e) and SLED- based (Fig. 4f) Prph2 rescue constructs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 828, 876, 909]]<|/det|>
+Finally, because oncolytic virotherapy is now an approved treatment modality in oncology 53, we sought to leverage tumor- specific RNA splicing patterns to generate SLED- based oncolytic vectors. Specifically, we identified a cryptic alternative exon in the constitutively expressed ubiquitin- like modifier activating enzyme 1 (UBA1) that was
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 879, 469]]<|/det|>
+observed exclusively in \(SF3B1\) mutant cancer cells (Fig. 4g). We designed two SLED- based vectors incorporating this alternative exon: one that express a bichromatic fluorescent reporter that selectively expresses EGFP in \(SF3B1\) mutant cells (SLED.SFUv1, where EGFP will only express in mutant \(SF3B1\) cells), and one that selectively expresses an oncolytic inducible Caspase 9 \(54,55\) in \(SF3B1\) mutant cells (SLED.SFUv2) (Fig. 4h). We first tested SFUv1 specificity by transfected 92- 1 and Mel- 202 uveal melanoma cell lines. We observed that EGFP expression was present in \(SF3B1^{R625G}\) Mel- 202 cell lines \(^{56}\) , but absent in the 92- 1 uveal melanoma cell line, which is wildtype for \(SF3B1\) (Fig. 4i) \(^{57}\) . We next quantified this using FACS analysis and analyzed four additional cell lines, two of which were isogenic to Mel- 202: PC76B6, in which AAV- based gene targeting was used to revert the mutant \(SF3B1\) status to wildtype through inactivation of the \(SF3B1^{R625G}\) allele, and MR5, a gene targeting control clone of Mel- 202 that retains the \(SF3B1^{R625G}\) mutation (Fig. S7). The other two cell lines analyzed by FACS were wildtype and \(SF3B1^{K700E}\) K562 leukemia cells. FACS analysis revealed that the EGFP/dsRed ratio is strongly dependent on the presence of either the \(SF3B1^{R625G}\) or \(SF3B1^{K700E}\) mutation (Fig. 4j). Finally, we tested the efficacy of SLED.SFUv2 by transfecting the wildtype and \(SF3B1^{K700E}\) K562 cells and observed efficient and selective induction of apoptosis in cells carrying the \(SF3B1^{K700E}\) following induction of Caspase9 dimerization (Fig. 4k).  
+
+<|ref|>sub_title<|/ref|><|det|>[[174, 490, 282, 508]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[113, 510, 881, 890]]<|/det|>
+In this study, we demonstrate that SLED- based vectors can produce cell- specific expression in a variety of constructs, and that SLED- based approaches are both orthogonal and complementary to existing methodology. SLED- based alternative splicing can be combined in a Boolean fashion with minipromoters to achieve higher levels of cell type specificity, as demonstrated by the integration of the pan- neuronal hSyn minipromoter and the excitatory neuron and glial- specific SYNRG exon to generate the excitatory neuron- specific SLED.ENS vector. SLED- based AAV vectors can also be used to study previously intractable problems without the use of complex transgenics, such as the in vivo dynamics of GluA2 flip/flop splicing. The use of splicing- related frameshifting allows efficient cell type- specific expression of multiple reporter or effector constructs in a single vector. SLED- based vectors also enable new strategies to improve gene therapies. For instance, SLED vectors can use any promoter, potentially allowing for stronger and more sustained levels of expression relative to conventional minipromoters and enabling more consistent patterns of cell- specific expression across multiple model organisms \(^{58,59}\) . This is critical, as photoreceptor minipromoters have encountered complex issues when tested in various mammalian species. For instance, the hRK1 minipromoter, which is widely used to drive expression in both rods and cones in rodents, is unable to drive efficient expression in cones of other model organisms such as dogs and pigs unless used at very high titers, and expresses at lower levels in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 879, 169]]<|/det|>
+rods than the rod- specific mOps promoter \(^{59 - 61}\) . Finally, SLED vectors can also selectively target disease states associated with abnormal splicing that would not be accessible using minipromoters. The use of photoreceptor minipromoters in AAV vectors can lead to long- term toxicity, but this is avoided using constitutive promoters \(^{62}\) .  
+
+<|ref|>text<|/ref|><|det|>[[114, 170, 881, 409]]<|/det|>
+Identification of evolutionarily- conserved patterns of cell- specific alternative splicing is straightforward, provided that good quality full- length RNA- Seq data is available. As transcriptomes from more tissues and cell types are profiled and deposited in public archives, our ability to identify highly cell- specific patterns of alternative splicing will increase and these datasets will guide the design of the next generation of SLED vectors. While transcriptome analysis has increasingly shifted towards 3'- directed short read single cell RNA- Seq platforms in recent years, emerging techniques such as long- read nanopore sequencing \(^{63}\) and economical full- length scRNA- Seq techniques such as SMART- Seq v3 will continue to improve our knowledge of splicing patterns \(^{64,65}\) . Recent compendia of splice- junction and transcript- level expression have surveyed 100,000s to millions of datasets \(^{28,29,66}\) , making these patterns easier to discover computationally.  
+
+<|ref|>text<|/ref|><|det|>[[114, 409, 872, 525]]<|/det|>
+With detailed characterization of alternative splicing patterns in the tremendous diversity of cell types, particularly in the human central nervous system, an intersectional approach combining SLED and cell- specific minipromoters may generate vectors that can selectively target to date untargetable cell types. Indeed, alternative splicing generates another layer of transcriptional complexity to the nervous system \(^{25,67,68}\) .  
+
+<|ref|>text<|/ref|><|det|>[[114, 529, 875, 708]]<|/det|>
+SLED- based vectors are still intrinsically limited by the size of the genomic intronic sequences used to control alternative splicing, which are generally substantially larger than minipromoters. While this is a less severe obstacle for transfection- or nanoparticle- based gene delivery, it is still a substantial limitation for AAV- based delivery. While the effects of deletion mutagenesis on cell- specific splicing can be unpredictable, recently developed machine learning algorithms may help facilitate rational design of smaller SLED vectors \(^{69,70}\) . Drug- inducible approaches to regulate splicing \(^{71 - 73}\) and the inclusion of miRNA target sites \(^{74,75}\) may enable further control of SLED- based constructs.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 334, 108]]<|/det|>
+## Materials and Methods:  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 129, 548, 148]]<|/det|>
+## Molecular cloning and cancer cell line culture:  
+
+<|ref|>text<|/ref|><|det|>[[114, 150, 881, 470]]<|/det|>
+Materials and Methods:Molecular cloning and cancer cell line culture:To generate SLED plasmids, gene fragments were commercially synthesized using Twist Biosciences and ThermoFisher GeneArt and cloned into an AAV vector backbone (Addgene #105922) using restriction enzyme cloning. HEK293, HepG2, and N2a cells were cultured in Dulbecco's Modified Eagle's Medium (Corning, 10- 017- CV) supplemented with 1x GlutaMAX (ThermoFisher Scientific, 35050061), 10% FBS (Corning, 35- 010- CV). Human uveal melanoma cell lines 92- 1 (generously provided by Charles Eberhart, Johns Hopkins University), MP41 (ATCC), and Mel- 202 (Sigma) were cultured in RPMI medium with 10% fetal bovine serum (FBS), penicillin/streptomycin, and l- glutamine. SF3B1<sup>K700E</sup> and control K562 cells were obtained from Horizon Discovery and cultured in RPMI with 20% FBS. The isolation, early characterization and further genetic and molecular characterization of the cell lines have been described elsewhere<sup>78- 80</sup>. Transfection of SLED vectors in uveal melanoma cells was achieved using Lipofectamine 3000 (ThermoFisher Scientific, L3000008) and with the 4D- Nucleofector X (Lonza) for K562 cells. The SF3B1<sup>R702R</sup> AAV targeting vector as described<sup>81</sup> was applied to SF3B1<sup>R625G</sup> Mel202 cells. iCaspase9 dimerization was induced by 100nM AP21087 (Sigma- Aldrich).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 490, 218, 507]]<|/det|>
+## Antibodies  
+
+<|ref|>text<|/ref|><|det|>[[114, 509, 879, 729]]<|/det|>
+AntibodiesThe following antibodies were used for primary culture neurons (supplier, catalog number and working dilution are indicated): anti- Gad67 ms IgG2a (Millipore MAB5406, 1:500); anti- somatostatin rat (Millipore MAB354, 1:400); anti- HuCD IgG2B (Thermo 16A11, 1:200); anti- NeuN mouse IgG1 (Thermo MAB377, 1:500). The following antibodies were used for brain sections (supplier, catalog number and working dilution are indicated): anti- GFP chicken polyclonal (Abcam ab13970, 1:2000); anti- dsRED rabbit polyclonal (Tanaka LivingColors 632496, 1:1000) (this antibody also detects mCherry); anti- NeuN mouse monoclonal IgG1 (Thermo MAB377, 1:500); anti- Gad67 mouse monoclonal IgG2a (Millipore MAB5406, 1:200); anti- PV mouse monoclonal IgG1 (Swant PV235, 1:2000); anti- somatostatin rat monoclonal IgG2b (Millipore MAB354, 1:400).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 750, 580, 768]]<|/det|>
+## Preparation and treatment of rat primary cultures:  
+
+<|ref|>text<|/ref|><|det|>[[114, 770, 879, 889]]<|/det|>
+Preparation and treatment of rat primary cultures:Hippocampi and cortices were dissected from embryonic day 18 rats, incubated with papain (Worthington Biochemical) and gently triturated with polished glass pipettes. Hippocampal neurons were plated on 18mm glass coverslips precoated with poly- L- Lysine (1mg/ml) in NeuroBasal media (Gibco) supplemented with 2% B27 (Gibco), 50 U/ml penicillin, 50 mg/ml streptomycin, 2mM GlutaMax (Gibco) and 5% horse serum (Hyclone). Hippocampal cells were plated at a density of 150K/coverslip (in a 12- well
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 880, 309]]<|/det|>
+plate). At days in vitro 1 (DIV1) media was replaced with NMO consisting of the above plating media without the addition of serum. Cells were then fed every 7 days with NMO. Hippocampal cultures were transduced with viral vectors at DIV1 and fixed at DIV15 for immunofluorescence analysis. SLED constructs were also transfected into hippocampal neurons using Lipofectamine 2000 (Invitrogen) following the manufacturer's instruction. Cortical neurons were used to evaluate SLED.GluA2. Before plating, cortical cells (6M per reaction) were electroporated with SLED.GluA2 plasmid DNA (3- 5ug) using a Rat Neurofection kit (Amaxa) and split evenly between 3 wells of a 6- well plate. As a comparison group, neurons were plated at equivalent density without electroporation. Cortical cells were harvested at (DIV4- 6) for FACS and downstream evaluation of Gria2 flip/flop splicing.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 330, 501, 348]]<|/det|>
+## Quantification of SLED.ENS with IMARIS:  
+
+<|ref|>text<|/ref|><|det|>[[114, 350, 868, 530]]<|/det|>
+To calculate the green/red ratios of SLED.ENS in excitatory and inhibitory neurons, Z- stacks were imported into IMARIS (Bitplane version 9.7.0) and 3D surfaces created around the cell bodies of transduced cells. The average EGFR and mCherry signal within the 3D surface was extracted for each cell. The Gad67 immunohistochemical signal was used to classify the cells into Gad67 positive (inhibitory interneuron) or Gad67 negative (putative excitatory neurons). To determine the statistical specificity of SLED vectors, log base 2 transformed ratios of green/red fluorescence were compared between sample groups using unpaired t- tests (GraphPad).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 550, 709, 569]]<|/det|>
+## Quantification of SLED.ENS, SLED.NPL and SLED.GluA2 in Fiji:  
+
+<|ref|>text<|/ref|><|det|>[[114, 570, 879, 809]]<|/det|>
+Confocal Z- stack images acquired with a 40x objective were processed using Fiji \(^{82}\) . Maximum intensity projections were generated, and the EGFR or mCherry/dsRed channel used to draw circle/oval ROIs around the cell body of transduced neurons without looking at signal in the 405 or 647 channels that contained cell- specific immunofluorescent markers. The raw EGFR and mCherry/dsRED average ROI intensities were then extracted for each cell. Subsequently, the cell- specific identification of each cell was observed from immunofluorescence using the 405 and 647 channel (for SLED.NPL this was HuC/D expression, for SLED.ENS this was Gad67 expression and for SLED.GluA2 this was Gad67). This enables clustering of the individual cells for comparisons of the green/red ratios. To determine the statistical specificity of SLED vectors, log base 2 transformed ratios of green/red fluorescence were compared between sample groups using unpaired t- tests (GraphPad).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 830, 437, 849]]<|/det|>
+## Live-cell imaging of SLED.CaRPv1  
+
+<|ref|>text<|/ref|><|det|>[[115, 851, 810, 889]]<|/det|>
+Rat primary hippocampal neurons at DIV11 were transfected with \(1\mu \mathrm{g}\) SLED.CaRPv1 alongside \(1\mu \mathrm{g}\) of mDlx- Azurite (as an enhancer based marker of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 881, 428]]<|/det|>
+interneurons \(^{4,83)}\) per coverslip in a 12- well plate. At DIV13 coverslips were imaged live in ACSF (NaCl 120mM, KCl 5mM, HEPES 10mM, D- Glucose 10mM, CaCl2.2H2O 2mM, MgCl2 1mM) at pH 7.4 using a Zeiss LSM 880 confocal microscope in a temperature \((37^{\circ}C)\) and humidity controlled chamber. Interneurons were identified by presence of the mDlx- Azurite signal, and mDlx- Azurite- negative cells were considered putative excitatory neurons. Time series were collected using 20x or 10x objectives for single cell or multiple cell imaging, respectively. Images were acquired at baseline, and also following addition of Bicuculline (20μM) to promote network activity through disinhibition. Files were processed and analyzed using Fiji \(^{82}\) . Fluorescence signals from jGCaMP7b and jRGECO1a were normalized to maximize variation between mDlx- Azurite positive and negative cells and the difference (Δintensity) between jGCaMP7b and jRGECO1a was calculated across each pixel and image frame (a gaussian blur (1px) was applied to each image before Δintensity to avoid pixelation artifacts). The sum of jGCaMP7b and jRGECO1a pixel values (sumGR) were calculated across each pixel and image frame. To generate the CaGreen- CaRed heatmap in Figure 3g and Supplemental Videos 1 and 2, sumGR was multiplied by Δintensity and colored using a custom lookup table.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 450, 670, 469]]<|/det|>
+## Immunofluorescence analysis of primary cultured neurons:  
+
+<|ref|>text<|/ref|><|det|>[[114, 470, 882, 750]]<|/det|>
+Unless otherwise stated, hippocampal neurons were fixed at DIV15. Media was aspirated and cells were washed with PBS at RT once before being incubated with \(4\%\) PFA (Electron Microscopy Sciences) made up in PBS with the addition of \(4\%\) sucrose at RT for 15 mins. Coverslips were washed 4 times with PBS then immunofluorescence commenced using a gelatin- based buffer (15 mM phosphate buffer (pH 7.4) containing \(0.1\%\) gelatin, \(0.3\%\) Triton X- 100, and \(0.25M\) NaCl) for combined blocking/antibody incubation. Primary antibodies (see below) were incubated with coverslips O/N at \(4^{\circ}C\) Secondary antibodies (Invitrogen for 488, 568 and 647 and Jackson labs and Abcam for 405; all at 1:500 dilution) were incubated for 1hr at RT. Between antibody incubations cells were washed with PBS, with some experiments including a brief DAPI incubation to label cell nuclei. Coverslips were mounted on slides with PermaFluor (Thermo Fisher Scientific). Samples were imaged on a Zeiss LSM 880 confocal microscope. Care was taken to ensure pixels in each channel were not over saturated. Images were analyzed using Fiji. In cultured neurons the SLED- driven fluorophores were not antibody boosted.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 771, 273, 789]]<|/det|>
+## AAV production:  
+
+<|ref|>text<|/ref|><|det|>[[115, 790, 867, 869]]<|/det|>
+SLED.ENS (1E+12 vg/ml, mouse and ferret cortex) and SLED.GluA2 (1E+12 vg/ml) were generated by the UNC Vector Core and SLED.NPL (2E+13 vg/ml) and SLED.RAB (2E+13 vg/ml) were generated by Virovek. SLED.ENS used for primary rat neuronal cultures was generated by Virovek (2E+13 vg/ml)
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 497, 109]]<|/det|>
+## Stereotaxic surgery and virus injections:  
+
+<|ref|>text<|/ref|><|det|>[[114, 110, 881, 529]]<|/det|>
+All animals were treated in accordance with the Johns Hopkins University Animal Care and Use Committee (IACUC) guidelines. Adult Blk6/J mice were used to evaluate SLED.AAV vectors. Animals were anesthetized with isofluorane (Baxter) using a SomnoSuite system (Kent Scientific) and secured in a stereotaxic frame (Kopf). The animal's temperature was controlled with a closed- loop system (RightTemp, Kent Scientific). The animal's scalp skin was cleaned with an ethanol wipe, and the hair removed. Animals were injected with \(0.5ml\) sterile saline (VetOne) to maintain hydration, buprenorphine (ZooPharm; \(1mg / ml\) ) and lidocaine (VetOne; \(2\%\) ) subcutaneously. The lidocaine was injected under the scalp as a local anesthetic. An incision was made to expose the skull surface, and to enable a small craniotomy to be made (see below for coordinates) exposing the brain surface. A glass pipette (Drummond Science Company; Wiretrol II) was pulled (Sutter Instruments) and sharpened to a \(30^{\circ}\) angle (Medical System Corp) and controlled by a pneumatic injector (Narishige) to enable controlled virus injection. Pipettes loaded with SLED viruses were slowly lowered to the desired stereotaxic coordinate, and after a delay of 2 minutes, virus was injected at a rate of \(100\mathrm{ml / min}\) . The pipette was left in position for 5 minutes after virus injection to reduce backflow up the injection tract. After pipette removal, the skin was sutured (Ethicon) and sealed with glue (Vetbond). Mice were closely monitored during the recovery phase, and placed in a clean cage on a warmed surface with access to a softened chow diet. Animals were given 2 weeks to recover, and for the virus to express, before being euthanized for perfusion/fixation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 530, 855, 609]]<|/det|>
+Stereotaxic coordinates: Targeting of dorsal hippocampus (all with respect to Bregma): [AP: - 2 | ML: 1.5 | Z: - 1.5, - 1.3, - 1.1 (from pia, 300nl at each site)]. Note, for SLED.NPL deep cortical layers above the hippocampus were imaged (with overflow virus injection into the dorsal hippocampus).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 630, 488, 648]]<|/det|>
+## Perfusions and immunohistochemistry:  
+
+<|ref|>text<|/ref|><|det|>[[114, 650, 877, 890]]<|/det|>
+Mice were terminally anesthetized and transcardially perfused with PBS followed by \(4\%\) PFA (Electron Microscopy Sciences), both ice cold. Brains were then postfixed for 2 hours at \(4^{\circ}C\) and then washed with PBS. Brains were then either sliced on a vibratome (Leica; VT- 1000; 60 \(\mu \mathrm{m}\) thick) or incubated overnight in \(30\%\) sucrose and cut into \(40\mu \mathrm{m}\) sections using a cryostat (Leica Biosystems). For slices that required Gad67 staining the following IHC protocol was followed as previously described4. Sections were washed x3 in PBS (10 minutes each) and permeabilized with PBS containing \(0.1\%\) Triton- X (Sigma) for 30 minutes at RT. Slices were blocked with PBS containing \(3\%\) BSA and \(5\%\) normal goat serum (Vector Laboratories) for 1 hour at RT. Primary antibodies were made up in the same blocking buffer and incubated at RT for 24hrs. Slices were washed 4x with PBS and then incubated with fluorescently conjugated secondary antibodies (Invitrogen, all at 1:500) ON at \(4^{\circ}C\) . Slices were then washed x4
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 866, 328]]<|/det|>
+with PBS and mounted on slides with PermaFluor (Thermo Fisher Scientific). For other antibody combinations the same overall protocol was followed with the following differences. Slices were permeabilized with PBS containing \(0.3\%\) Triton- X for 20 minutes. Slices were blocked with PBS containing \(5\%\) normal goat serum with the addition of \(0.15\%\) Trixon- X. Primary antibodies were incubated in the same blocking buffer but at \(4^{\circ}C\) overnight. Secondary antibodies were made up in the same blocking buffer and incubated ON at \(4^{\circ}C\) . In some instances, slices were washed with PBS containing DAPI to label nuclei after the secondary antibody incubation. For SLED.MEv2 the fluorophores were not antibody boosted. For evaluation of SLED.NPL and SLED.ENS the EGFP and mCherry/dsRed was antibody boosted. Slides were imaged on a confocal microscope (as described above), or on an Apotome epifluorescence scope (Zeiss) and analyzed further in Fiji.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 349, 390, 368]]<|/det|>
+## Ferret Cortex AAV injections:  
+
+<|ref|>text<|/ref|><|det|>[[114, 369, 879, 629]]<|/det|>
+An adult female ferret (Mustela putoris furo, Marshall Farms) was used for the virus injection. Anesthesia was induced with ketamine \((40 \mathrm{mg / kg})\) and maintained with isoflurane \((1.5 - 3\%)\) . Atropine \((0.05 \mathrm{mg / kg})\) was given at the start of the surgery. Burenorphine \((0.01 - 0.03 \mathrm{mg / kg})\) was administered pre- and post- operatively for analgesia in combination with a subcutaneous injection of lidocaine during the surgery, and post- operative administration of meloxicam \((0.1 - 0.2 \mathrm{mg / kg})\) . Animals were maintained at normal body temperature during the surgery using a heating pad. Skin and muscle over primary visual cortex were reflexed, and a small craniotomy was made over the brain region of interest. Virus was then injected through a pulled glass pipette sharpened to a tip angle of about 60 deg. Approximately 1 uL of virus was then injected, distributed across multiple depths at a single site in the craniotomy. After the injection, muscle and skin were closed and the animal was recovered and returned to its home cage. The animal was perfused 3 months after the virus injection.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 650, 330, 669]]<|/det|>
+## Ocular AAV injections:  
+
+<|ref|>text<|/ref|><|det|>[[114, 670, 870, 889]]<|/det|>
+For subretinal injections, AAVs were injected into the subretinal space of 28 day old Prph2rds/rds mice (#001979 Jackson Laboratory, Bar Harbor, ME). Briefly, mice were anesthetized by intraperitoneal injection of ketamine \((100 \mathrm{mg / kg})\) and xylazine hydrochloride \((20 \mathrm{mg / kg})\) . The pupils were dilated with \(1\%\) tropicamide (Alcon, Ft. Worth, TX). The corneas were covered with Healon GV sodium hyaluronate solution (Abbott Medical Optics Inc., Santa Ana, CA) and cover glass to facilitate transpupillary visualization. 1uL of AAV \((10^{\wedge}13\) viral genomes/mL) were loaded into a 33G needle micro- syringe (Hamilton Company, Reno, NV), then tangentially injected into the subretinal space through the sclera of the mice. A successful injection was verified by direct visualization through the dilated pupil of the recipient under the surgical microscope (Leica, Wetzlar, Germany).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 876, 228]]<|/det|>
+For intravitreal injections, AAVs were injected into the vitreous cavity of 0 day old CD1 pups (Charles River Laboratories, Wilmington, MA). The neonatal animals were anesthetized by placing them on a waterproof surface over crushed ice until the pup was no longer responsive to touch. The eyelids were surgically separated before injecting \(1 \mu \mathrm{L}\) of AAV ( \(10^{\wedge}13\) viral genomes/mL) into the vitreous space using a custom 33G sharp needle micro- syringe (Hamilton Company, Reno, NV). The needle was held in place for 10 seconds to avoid outflow before being gently removed.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 250, 582, 269]]<|/det|>
+## Spectral Domain-Optical Coherence Tomography:  
+
+<|ref|>text<|/ref|><|det|>[[114, 270, 874, 490]]<|/det|>
+For in vivo retinal imaging, Spectral Domain- OCT images were obtained and analyzed as previously described \(^{84}\) . Mice were first anesthetized with ketamine (100 mg/kg) and xylazine hydrochloride (20 mg/kg), followed by dilation with \(1\%\) tropicamide and \(2.5\%\) phenylephrine. The clarity of the cornea and lens was maintained using GenTeal lubricating eye gel (Novartis Pharmaceuticals, Basel, Switzerland). The mice were secured using a bite bar to a movable stage. The stage was adjusted manually to center the image of the retina at the optic nerve head. Cross- sectional images were generated using 1000 rectangular volume scans using the Envisu OCT system (Leica Microsystems, Wetzlar, Germany). Outer nuclear layer and inner nuclear layer thickness was measured using the linear caliper function in the software by a masked observer using a pre- established uniform grid.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 510, 370, 529]]<|/det|>
+## Electrotinography (ERG):  
+
+<|ref|>text<|/ref|><|det|>[[113, 530, 880, 870]]<|/det|>
+Full- field flash ERGs were performed as previously described \(^{85}\) . In brief, mice were dark adapted overnight and anesthetized with ketamine (80 mg/kg) and xylazine hydrochloride (16 mg/kg) prior to recording. The pupils were dilated with \(1\%\) tropicamide, \(1\%\) cyclopentolate and \(2.5\%\) phenylephrine and the corneal surface was anesthetized with \(0.5\%\) proparacaine HCl eye drops. For recording retinal electrical responses, stainless- steel wire electrodes were placed on the corneas as the active electrodes, contacting the center of the corneal surface through a thin layer of artificial tear. Needle electrodes were subcutaneously inserted into the cheek and the tail as reference and ground electrodes, respectively. To maintain body temperature during the procedure, the animals were placed on a temperature- controlled heating pad. Using the UTAS Bigshot ERG system (LKC Technologies, Gaithersburg, MD), ERG responses were differentially amplified (0.3- 300 Hz), digitized at 1,000 Hz, averaged and stored. The recording epoch was 512 ms, with a 20 ms pre- stimulation baseline. After 7 min of light adaptation, ERGs were obtained to strobe flashes (- 0.8 to 1.9 log cd.s/m \(^2\) ) superimposed upon a steady 30 cd.s/m \(^2\) white background. in response to a series of flashes ranging from - 0.8 to 1.9 log cd.s/m \(^2\) . The b- wave amplitude was measured from the a- wave trough to the peak of the b- wave.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 235, 108]]<|/det|>
+## Disclosures:  
+
+<|ref|>text<|/ref|><|det|>[[115, 109, 879, 189]]<|/det|>
+Disclosures:SB receives research support from Genentech, is a co- founder and shareholder in CDI Labs LLC, and was a consultant to Third Rock Ventures. JPL receives research support from Takeda Pharmaceuticals. SB and JPL have filed a patent application covering SLED technology.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 210, 306, 228]]<|/det|>
+## Acknowledgements:  
+
+<|ref|>text<|/ref|><|det|>[[113, 230, 881, 448]]<|/det|>
+Acknowledgements:This work was supported by NIH grants RF1MH123237 (S.B.), R24EY027283 (S.B.), K08EY027093 (F.R.), and R01EY033103 (M.S.S.), a Stein Innovation Award from Research to Prevent Blindness to S.B., an unrestricted departmental grant to the Wilmer Eye Institute from Research to Prevent Blindness awarded to F.R., an NSF NeuroNex grant #1934288 awarded to K.J.N., a Visual Sciences Training grant 2T32EY007143 awarded to C.P.S., a Johns Hopkins Kavli NDI fellowship awarded to J.P.L., and a Johns Hopkins IDIES Seed Fund awarded to J.P.L. We thank the Ross Flow Cytometry Core (JHMI), the Wilmer Microscopy module supported by the EY001765 core grant for flow cytometry, the Single Cell & Transcriptomics Core (JHMI), and the Cleveland Clinic core grant (EY025585). We thank W. Yap, W. Xin, and R. Roth for comments on the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 470, 317, 488]]<|/det|>
+## Author contributions:  
+
+<|ref|>text<|/ref|><|det|>[[113, 489, 877, 690]]<|/det|>
+Author contributions:S.B. and J.P.L. conceived and oversaw all aspects of the study. J.P.L., A.M.B., C.P.S., R.P.C.O., V.T., M.Y.G., R.D., L.Z., T.A.B., K.T., J.H., Y.Li and P.J.L. analyzed cellular specificity of SLED vectors. Y.Liu and M.S.S. assisted with subretinal injections. Y.Li and S.S. assisted with human iPSC neuronal culture. D.B., P.P., P.O.K., and T.A.B. assisted with development of calcium indicator SLED vectors. M.Y. and N.P. conducted ERG analysis. B.L. assisted with computational efforts. R.L.H. supervised analysis of SLED.ENS and SLED.NPL vectors. W.B.D., F.R., B.D., K.T. and J.H. carried out studies of oncolytic SLED vectors. K.J.N., J.J.S., and R.V. carried out all ferret studies. J.P.L., A.M.B, C.P.S. and S.B. drafted the manuscript. All authors approved the final manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 129, 223, 149]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[110, 163, 884, 900]]<|/det|>
+1. Liu, Y., Hegarty, S., Winter, C., Wang, F. & He, Z. Viral vectors for neuronal cell type-specific visualization and manipulations. Curr. Opin. Neurobiol. 63, 67–76 (2020).2. Weissleder, R., Kelly, K., Sun, E. Y., Shtatland, T. & Josephson, L. Cell-specific targeting of nanoparticles by multivalent attachment of small molecules. Nat. Biotechnol. 23, 1418–1423 (2005).3. Bedbrook, C. N., Deverman, B. E. & Gradinaru, V. Viral Strategies for Targeting the Central and Peripheral Nervous Systems. Annu. Rev. Neurosci. 41, 323–348 (2018).4. Dimidschstein, J. et al. A viral strategy for targeting and manipulating interneurons across vertebrate species. Nat. Neurosci. 19, 1743–1749 (2016).5. de Leeuw, C. N. et al. rAAV-compatible MiniPromoters for restricted expression in the brain and eye. Mol. Brain 9, 52 (2016).6. Mich, J. K. et al. Functional enhancer elements drive subclass-selective expression from mouse to primate neocortex. Cell Rep. 34, 108754 (2021).7. Nair, R. R., Blankvoort, S., Lagartos, M. J. & Kentros, C. Enhancer-Driven Gene Expression (EDGE) Enables the Generation of Viral Vectors Specific to Neuronal Subtypes. iScience 23, 100888 (2020).8. Doronin, K. et al. Tissue-specific, tumor-selective, replication-competent adenovirus vector for cancer gene therapy. J. Virol. 75, 3314–3324 (2001).9. Srivastava, A. In vivo tissue-tropism of adeno-associated viral vectors. Curr. Opin. Virol. 21, 75–80 (2016).10. Ortiz, J. L. S. Engineering of Novel Adeno-Associated Virus Vectors for Gene Therapy Applications. (2016).11. Dhungel, B., Ramlogan-Steel, C. A. & Steel, J. C. MicroRNA-Regulated Gene Delivery Systems for Research and Therapeutic Purposes. Molecules 23, (2018).12. Singh, H. M. et al. MicroRNA-sensitive oncolytic measles virus for chemovirotherapy of pancreatic cancer. Mol Ther Oncolytics 21, 340–355 (2021).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 880, 129]]<|/det|>
+13. Keaveney, M. K. et al. A MicroRNA-Based Gene-Targeting Tool for Virally Labeling Interneurons in the Rodent Cortex. Cell Rep. 24, 294–303 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[115, 142, 877, 205]]<|/det|>
+14. Buck, T. M. & Wijnholds, J. Recombinant Adeno-Associated Viral Vectors (rAAV)-Vector Elements in Ocular Gene Therapy Clinical Trials and Transgene Expression and Bioactivity Assays. Int. J. Mol. Sci. 21, (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 219, 870, 260]]<|/det|>
+15. Orengo, J. P., Bundman, D. & Cooper, T. A. A bichromatic fluorescent reporter for cell-based screens of alternative splicing. Nucleic Acids Res. 34, e148 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[115, 274, 868, 316]]<|/det|>
+16. Hu, N. et al. Non-invasive monitoring of alternative splicing outcomes to identify candidate therapies for myotonic dystrophy type 1. Nat. Commun. 9, 5227 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[115, 329, 827, 391]]<|/det|>
+17. Star, E. et al. A drug-repositioning screen using splicing-sensitive fluorescent reporters identifies novel modulators of VEGF-A splicing with anti-angiogenic properties. Oncogenesis 10, 1–12 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 404, 838, 465]]<|/det|>
+18. Takeuchi, A., Hosokawa, M., Nojima, T. & Hagiwara, M. Splicing reporter mice revealed the evolutionally conserved switching mechanism of tissue-specific alternative exon selection. PLoS One 5, e10946 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[115, 480, 840, 520]]<|/det|>
+19. Massaro, G. et al. Fetal gene therapy for neurodegenerative disease of infants. Nat. Med. 24, 1317–1323 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[115, 534, 879, 575]]<|/det|>
+20. Tabebordbar, M. et al. In vivo gene editing in dystrophic mouse muscle and muscle stem cells. Science 351, 407–411 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 589, 847, 630]]<|/det|>
+21. Nathwani, A. C. et al. Long-term safety and efficacy of factor IX gene therapy in hemophilia B. N. Engl. J. Med. 371, 1994–2004 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[115, 644, 880, 705]]<|/det|>
+22. Bennett, J. et al. Safety and durability of effect of contralateral-eye administration of AAV2 gene therapy in patients with childhood-onset blindness caused by RPE65 mutations: a follow-on phase 1 trial. Lancet 388, 661–672 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 719, 868, 760]]<|/det|>
+23. Wang, D., Tai, P. W. L. & Gao, G. Adeno-associated virus vector as a platform for gene therapy delivery. Nat. Rev. Drug Discov. 18, 358–378 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 774, 827, 815]]<|/det|>
+24. Sakharkar, M. K., Chow, V. T. K. & Kanguene, P. Distributions of exons and introns in the human genome. In Silico Biol. 4, 387–393 (2004).  
+
+<|ref|>text<|/ref|><|det|>[[115, 830, 870, 870]]<|/det|>
+25. Raj, B. & Blencowe, B. J. Alternative Splicing in the Mammalian Nervous System: Recent Insights into Mechanisms and Functional Roles. Neuron 87, 14–27 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[111, 884, 840, 905]]<|/det|>
+26. Hallegger, M., Llorian, M. & Smith, C. W. J. Alternative splicing: global insights.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 90, 422, 108]]<|/det|>
+FEBS J. 277, 856- 866 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[113, 123, 830, 163]]<|/det|>
+27. Ling, J. P. et al. ASCOT identifies key regulators of neuronal subtype-specific splicing. Nat. Commun. 11, 137 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[113, 179, 854, 238]]<|/det|>
+28. Wilks, C., Gaddipati, P., Nellore, A. & Langmead, B. Snaptron: querying splicing patterns across tens of thousands of RNA-seq samples. Bioinformatics 34, 114-116 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[113, 253, 795, 293]]<|/det|>
+29. Wilks, C. et al. recount3: summaries and queries for large-scale RNA-seq expression and splicing. Genome Biol. 22, 323 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 308, 875, 348]]<|/det|>
+30. Kim, J. H. et al. High cleavage efficiency of a 2A peptide derived from porcine teschovirus-1 in human cell lines, zebrafish and mice. PLoS One 6, e18556 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[113, 364, 820, 404]]<|/det|>
+31. Tremblay, R. G. et al. Differentiation of mouse Neuro 2A cells into dopamine neurons. J. Neurosci. Methods 186, 60-67 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[113, 419, 872, 479]]<|/det|>
+32. Kugler, S., Kilic, E. & Bahr, M. Human synapsin 1 gene promoter confers highly neuron-specific long-term transgene expression from an adenoviral vector in the adult rat brain depending on the transduced area. Gene Ther. 10, 337-347 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[113, 494, 860, 534]]<|/det|>
+33. de Melo, J. & Blackshaw, S. In Vivo Electroporation of Developing Mouse Retina. Methods Mol. Biol. 1715, 101-111 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[113, 550, 875, 609]]<|/det|>
+34. Swanson, G. T., Kamboj, S. K. & Cull-Candy, S. G. Single-channel properties of recombinant AMPA receptors depend on RNA editing, splice variation, and subunit composition. J. Neurosci. 17, 58-69 (1997).  
+
+<|ref|>text<|/ref|><|det|>[[113, 625, 866, 664]]<|/det|>
+35. Mosbacher, J. et al. A molecular determinant for submillisecond desensitization in glutamate receptors. Science 266, 1059-1062 (1994).  
+
+<|ref|>text<|/ref|><|det|>[[113, 680, 857, 740]]<|/det|>
+36. Koike, M., Tsukada, S., Tsuzuki, K., Kijima, H. & Ozawa, S. Regulation of kinetic properties of GluR2 AMPA receptor channels by alternative splicing. J. Neurosci. 20, 2166-2174 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[113, 755, 819, 795]]<|/det|>
+37. Sommer, B. et al. Flip and flop: a cell-specific functional switch in glutamate-operated channels of the CNS. Science 249, 1580-1585 (1990).  
+
+<|ref|>text<|/ref|><|det|>[[113, 811, 870, 870]]<|/det|>
+38. Penn, A. C., Balik, A., Wozny, C., Cais, O. & Greger, I. H. Activity-mediated AMPA receptor remodeling, driven by alternative splicing in the ligand-binding domain. Neuron 76, 503-510 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[110, 886, 844, 905]]<|/det|>
+39. Wright, A. & Vissel, B. The essential role of AMPA receptor GluR2 subunit RNA
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[150, 89, 813, 108]]<|/det|>
+editing in the normal and diseased brain. Front. Mol. Neurosci. 5, 34 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[115, 123, 835, 163]]<|/det|>
+40. Tasic, B. et al. Shared and distinct transcriptomic cell types across neocortical areas. Nature 563, 72–78 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[115, 178, 850, 239]]<|/det|>
+41. Furlanis, E., Traunmüller, L., Fucile, G. & Scheiffele, P. Landscape of ribosome-engaged transcript isoforms reveals extensive neuronal-cell-class-specific alternative splicing programs. Nat. Neurosci. 22, 1709–1717 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 878, 314]]<|/det|>
+42. Huntley, M. A. et al. Genome-Wide Analysis of Differential Gene Expression and Splicing in Excitatory Neurons and Interneuron Subtypes. J. Neurosci. 40, 958–973 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 329, 864, 370]]<|/det|>
+43. Dana, H. et al. High-performance calcium sensors for imaging activity in neuronal populations and microcompartments. Nat. Methods 16, 649–657 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 384, 863, 424]]<|/det|>
+44. Dana, H. et al. Sensitive red protein calcium indicators for imaging neural activity. Elife 5, (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 439, 874, 480]]<|/det|>
+45. Boon, C. J. F. et al. The spectrum of retinal dystrophies caused by mutations in the peripherin/RDS gene. Prog. Retin. Eye Res. 27, 213–235 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[115, 494, 840, 534]]<|/det|>
+46. Cheng, T. et al. The effect of peripherin/rds haploinsufficiency on rod and cone photoreceptors. J. Neurosci. 17, 8118–8128 (1997).  
+
+<|ref|>text<|/ref|><|det|>[[115, 549, 866, 590]]<|/det|>
+47. Jansen, H. G. & Sanyal, S. Development and degeneration of retina in rds mutant mice: electron microscopy. J. Comp. Neurol. 224, 71–84 (1984).  
+
+<|ref|>text<|/ref|><|det|>[[115, 605, 839, 645]]<|/det|>
+48. Sohocki, M. M. et al. Prevalence of mutations causing retinitis pigmentosa and other inherited retinopathies. Hum. Mutat. 17, 42–51 (2001).  
+
+<|ref|>text<|/ref|><|det|>[[115, 660, 866, 720]]<|/det|>
+49. Stuck, M. W., Conley, S. M. & Naash, M. I. PRPH2/RDS and ROM-1: Historical context, current views and future considerations. Prog. Retin. Eye Res. 52, 47–63 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 735, 878, 796]]<|/det|>
+50. Sarra, G. M. et al. Gene replacement therapy in the retinal degeneration slow (rds) mouse: the effect on retinal degeneration following partial transduction of the retina. Hum. Mol. Genet. 10, 2353–2361 (2001).  
+
+<|ref|>text<|/ref|><|det|>[[115, 811, 875, 871]]<|/det|>
+51. Flannery, J. G. et al. Efficient photoreceptor-targeted gene expression in vivo by recombinant adeno-associated virus. Proc. Natl. Acad. Sci. U. S. A. 94, 6916–6921 (1997).  
+
+<|ref|>text<|/ref|><|det|>[[111, 886, 848, 906]]<|/det|>
+52. Ali, R. R. et al. Restoration of photoreceptor ultrastructure and function in retinal
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[150, 89, 802, 109]]<|/det|>
+degeneration slow mice by gene therapy. Nat. Genet. 25, 306–310 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[113, 123, 870, 164]]<|/det|>
+53. Andtbacka, R. H. I. et al. Talimogene Laherparepvec Improves Durable Response Rate in Patients With Advanced Melanoma. J. Clin. Oncol. 33, 2780–2788 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[113, 179, 880, 220]]<|/det|>
+54. Straathof, K. C. et al. An inducible caspase 9 safety switch for T-cell therapy. Blood vol. 105 4247–4254 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[114, 234, 864, 295]]<|/det|>
+55. Wu, C. et al. Development of an inducible caspase-9 safety switch for pluripotent stem cell–based therapies. Molecular Therapy - Methods & Clinical Development vol. 1 14053 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[114, 310, 825, 350]]<|/det|>
+56. Inoue, D. et al. Spliceosomal disruption of the non-canonical BAF complex in cancer. Nature 574, 432–436 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[114, 364, 848, 425]]<|/det|>
+57. Jager, M. J., Magner, J. A. B., Ksander, B. R. & Dubovy, S. R. Uveal Melanoma Cell Lines: Where do they come from? (An American Ophthalmological Society Thesis). Trans. Am. Ophthalmol. Soc. 114, T5 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[114, 439, 877, 500]]<|/det|>
+58. Boye, S. E. et al. The human rhodopsin kinase promoter in an AAV5 vector confers rod- and cone-specific expression in the primate retina. Hum. Gene Ther. 23, 1101–1115 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[114, 514, 872, 555]]<|/det|>
+59. Manfredi, A. et al. Combined rod and cone transduction by adeno-associated virus 2/8. Hum. Gene Ther. 24, 982–992 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[114, 569, 875, 630]]<|/det|>
+60. Beltran, W. A. et al. rAAV2/5 gene-targeting to rods: dose-dependent efficiency and complications associated with different promoters. Gene Ther. 17, 1162–1174 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[114, 644, 867, 705]]<|/det|>
+61. Khani, S. C. et al. AAV-mediated expression targeting of rod and cone photoreceptors with a human rhodopsin kinase promoter. Invest. Ophthalmol. Vis. Sci. 48, 3954–3961 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[114, 720, 880, 760]]<|/det|>
+62. Xiong, W. et al. AAV -regulatory sequences are correlated with ocular toxicity. Proc. Natl. Acad. Sci. U. S. A. 116, 5785–5794 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[114, 775, 790, 816]]<|/det|>
+63. Philpott, M. et al. Nanopore sequencing of single-cell transcriptomes with scCOLOR-seq. Nat. Biotechnol. 39, 1517–1520 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[114, 830, 808, 871]]<|/det|>
+64. Hagemann-Jensen, M. et al. Single-cell RNA counting at allele and isoform resolution using Smart-seq3. Nat. Biotechnol. 38, 708–714 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[111, 885, 833, 906]]<|/det|>
+65. Hagemann-Jensen, M., Ziegenhain, C. & Sandberg, R. Scalable full-transcript
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[150, 90, 765, 128]]<|/det|>
+coverage single cell RNA sequencing with Smart- seq3xpress. bioRxiv 2021.07.10.451889 (2021) doi:10.1101/2021.07.10.451889.  
+
+<|ref|>text<|/ref|><|det|>[[113, 144, 881, 184]]<|/det|>
+66. Lachmann, A. et al. Massive mining of publicly available RNA-seq data from human and mouse. Nat. Commun. 9, 1366 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[113, 199, 850, 238]]<|/det|>
+67. Zheng, S. Alternative splicing programming of axon formation. Wiley Interdiscip. Rev. RNA 11, e1585 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[113, 254, 839, 294]]<|/det|>
+68. Gomez, A. M., Traunmüller, L. & Scheiffele, P. Neurexins: molecular codes for shaping neuronal synapses. Nat. Rev. Neurosci. 22, 137–151 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 309, 880, 369]]<|/det|>
+69. Cheng, J., Çelik, M. H., Kundaje, A. & Gagneur, J. Publisher Correction: MTSplice predicts effects of genetic variants on tissue-specific splicing. Genome Biol. 22, 107 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 384, 809, 424]]<|/det|>
+70. Jaganathan, K. et al. Predicting Splicing from Primary Sequence with Deep Learning. Cell 176, 535–548.e24 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[113, 439, 872, 479]]<|/det|>
+71. Monteys, A. M. et al. Regulated control of gene therapies by drug-induced splicing. Nature 596, 291–295 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[113, 495, 794, 535]]<|/det|>
+72. Mathur, M. et al. Programmable mutually exclusive alternative splicing for generating RNA and protein diversity. Nat. Commun. 10, 2673 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[113, 550, 842, 590]]<|/det|>
+73. Jiang, K. et al. Programmable eukaryotic protein expression with RNA sensors. bioRxiv 2022.01.26.477951 (2022) doi:10.1101/2022.01.26.477951.  
+
+<|ref|>text<|/ref|><|det|>[[113, 605, 827, 645]]<|/det|>
+74. Jain, R. et al. MicroRNAs Enable mRNA Therapeutics to Selectively Program Cancer Cells to Self-Destruct. Nucleic Acid Ther. 28, 285–296 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[113, 660, 876, 720]]<|/det|>
+75. Sago, C. D. et al. Augmented lipid-nanoparticle-mediated in vivo genome editing in the lungs and spleen by disrupting Cas9 activity in the liver. Nat Biomed Eng 6, 157–167 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[113, 736, 875, 775]]<|/det|>
+76. Weinstein, J. N. et al. The Cancer Genome Atlas Pan-Cancer analysis project. Nat. Genet. 45, 1113–1120 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[113, 791, 875, 830]]<|/det|>
+77. GTEx Consortium. The GTEx Consortium atlas of genetic regulatory effects across human tissues. Science 369, 1318–1330 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[113, 846, 816, 886]]<|/det|>
+78. De Waard-Siebinga, I. et al. Establishment and characterization of an uveal-melanoma cell line. Int. J. Cancer 62, 155–161 (1995).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 87, 884, 504]]<|/det|>
+79. Skander, B. R., Rubsamen, P. E., Olsen, K. R., Cousins, S. W. & Streilein, J. W. Studies of tumor-infiltrating lymphocytes from a human choroidal melanoma. Invest. Ophthalmol. Vis. Sci. 32, 3198–3208 (1991).80. Griewank, K. G. et al. Genetic and molecular characterization of uveal melanoma cell lines. Pigment Cell Melanoma Res. 25, 182–187 (2012).81. Dalton, W. B. et al. Hotspot SF3B1 mutations induce metabolic reprogramming and vulnerability to serine deprivation. J. Clin. Invest. 129, 4708–4723 (2019).82. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat. Methods 9, 676–682 (2012).83. Bygrave, A. M. et al. Btbd11 is an inhibitory interneuron specific synaptic scaffolding protein that supports excitatory synapse structure and function. bioRxiv 2021.11.01.466782 (2021) doi:10.1101/2021.11.01.466782.84. Santiago, C. P. et al. A Drug-Tunable Gene Therapy for Broad-Spectrum Protection against Retinal Degeneration. Mol. Ther. 26, 2407–2417 (2018).85. Kinoshita, J. & Peachey, N. S. Noninvasive Electroretinographic Procedures for the Study of the Mouse Retina. Curr. Protoc. Mouse Biol. 8, 1–16 (2018).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 43, 312, 71]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 419, 203]]<|/det|>
+- SLEDSupplementalFigures031422.pdf- SupplementalVideo1CaRPv110x.mp4- SupplementalVideo2CaRPv120x.mp4
+
+<--- Page Split --->

preprint/preprint__b367bf4492c30047229c142d659b318a8f5859981a4e643b55522b53a7f727fa/images_list.json

@@ -0,0 +1,77 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        50,
+        950,
+        682
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        49,
+        953,
+        350
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        37,
+        40,
+        763,
+        787
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        39,
+        106,
+        960,
+        768
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5",
+    "footnote": [],
+    "bbox": [
+      [
+        44,
+        144,
+        955,
+        440
+      ]
+    ],
+    "page_idx": 17
+  }
+]

preprint/preprint__b367bf4492c30047229c142d659b318a8f5859981a4e643b55522b53a7f727fa/preprint__b367bf4492c30047229c142d659b318a8f5859981a4e643b55522b53a7f727fa.mmd

@@ -0,0 +1,285 @@
+
+# Identification of a conserved drug binding pocket in TMEM16 proteins  
+
+Yifan Cheng ( \(\boxed{\times}\) ycheng@ucsf.edu) University of California San Francisco  
+
+Shengjie Feng UCSF  
+
+Cristina Puchades UCSF  
+
+Juyeon Ko UCSF  
+
+Eric Figueroa UCSF https://orcid.org/0000- 0001- 8562- 6552  
+
+Yifei Chen UCSF  
+
+Hao Wu University of California San Francisco  
+
+Shuo Gu UCSF  
+
+Tina Han UCSF  
+
+Junrui Li UCSF  
+
+Brandon Ho UCSF  
+
+Brian Shoichet UCSF https://orcid.org/0000- 0002- 6098- 7367  
+
+Yuh Nung Jan UCSF  
+
+Lily Jan  
+
+University of California, San Francisco https://orcid.org/0000- 0003- 3938- 8498  
+
+Article  
+
+Keywords:
+
+<--- Page Split --->
+
+Posted Date: February 10th, 2022  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 1296933/v1  
+
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Version of Record: A version of this preprint was published at Nature Communications on August 12th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 40410- x.
+
+<--- Page Split --->
+
+## Abstract  
+
+AbstractThe TMEM16 family of calcium- activated membrane proteins includes ten mammalian paralogs (TMEM16A- K) playing distinct physiological roles with some implicated in cancer and airway diseases. Their modulators with therapeutic potential include 1PBC, a potent inhibitor with anti- tumoral properties, and the FDA- approved drug nicosamide that targets TMEM16F to inhibit syncytia formation induced by SARS- CoV- 2 infection. Here, we report cryo- EM structures of TMEM16F associated with 1PBC and nicosamide, revealing that both molecules bind the same drug binding pocket. We functionally and computationally validate this binding pocket in TMEM16A as well as TMEM16F, thereby showing that drug modulation also involves residues that are not conserved between TMEM16A and TMEM16F. This study establishes a much- needed structural framework for the development of more potent and more specific drug molecules targeting TMEM16 proteins.  
+
+## Main Text  
+
+TMEM16 proteins are a family of transmembrane proteins conserved across eukaryotes, encompassing 10 paralogs (TMEM16A- K) in mammals. Despite high sequence similarity, TMEM16 proteins present remarkable functional diversity (1, 2). For instance, TMEM16A is a calcium \((\mathrm{Ca}^{2 + })\) - activated chloride channel that opens in response to increased intracellular \(\mathrm{Ca}^{2 + }\) levels, enabling chloride ions to move across the plasma membrane (3- 5). In contrast, TMEM16F functions as both a \(\mathrm{Ca}^{2 + }\) - activated ion channel and a \(\mathrm{Ca}^{2 + }\) - activated lipid scramblase. TMEM16F channels permeate cations including \(\mathrm{Ca}^{2 + }\) ions (6- 8), however, its selectivity for cations versus anions may vary with the electrostatic field of the permeant pathway (9). Through its lipid scrambling activity, TMEM16F allows diverse lipids, including phosphatidylcholine (PC), phosphatidylethanolamine (PE) and phosphatidylserine (PS), to passively move between the inner and outer leaflets of the plasma membrane (1, 2, 6, 7, 10, 11). Both TMEM16A and TMEM16F play critical roles in numerous physiological processes and have emerged as important targets for therapeutic intervention in multiple diseases, including cancer, asthma, and in particular COVID- 19 (1, 2, 12, 13).  
+
+TMEM16A is required for airway and secretory gland secretion (1, 2). Notably, a TMEM16A activator is under consideration for the treatment of cystic fibrosis (14), whereas TMEM16A inhibitors with potent bronchodilator activities are being tested as anti- asthma drugs (15, 16). Additionally, TMEM16A activity is upregulated via gene amplification or enhanced expression in many types of cancers and is linked to increased cell migration and proliferation as well as metastatic progression (12). Therefore, antagonists of TMEM16A, such as 1PBC and nicosamide, provide a promising new avenue for the treatment of diverse cancers (16).  
+
+TMEM16F activity is important for blood coagulation (6, 17- 19) and mutations in TMEM16F are linked to the Scott syndrome bleeding disorder (20). TMEM16F also plays a critical role in extracellular vesicle generation and release (21- 24) and membrane repair as protection against bacterial infection (25). Importantly, nicosamide, an FDA approved drug, has recently been shown to block SARS- CoV- 2- induced
+
+<--- Page Split --->
+
+syncytia formation and virus replication by inhibiting TMEM16F (13). The repurposing of nicosamide for treatment of severe COVID- 19 is currently under examination in more than a dozen clinical trials (26) (clinicaltrials.gov). Nicosamide is also a potent inhibitor of TMEM16A and robustly mitigates the symptoms of airway diseases in mice (15). Like nicosamide, most small molecule inhibitors identified to date affect multiple TMEM16 paralogs, making off- target effects a major concern for clinical applications. Elucidation of the ligand binding site is important not only for understanding the mechanism of action of these molecules but also for designing more specific drugs for pharmacological targeting of TMEM16 proteins.  
+
+TMEM16 proteins form dimers (1, 2); each subunit comprises 10 transmembrane helices (TMs) and contains its own ion conduction pore enclosed and surrounded by TM3- 7 (Fig. 1). \(\mathrm{Ca^{2 + }}\) - dependent activation involves direct binding of \(\mathrm{Ca^{2 + }}\) ions in two contiguous \(\mathrm{Ca^{2 + }}\) - binding sites that are formed between TM6- TM8 (Fig. 1). Structures of both TMEM16A and TMEM16F in \(\mathrm{Ca^{2 + }}\) - free and \(\mathrm{Ca^{2 + }}\) - bound states reveal that TM6 undergoes major \(\mathrm{Ca^{2 + }}\) - dependent conformational rearrangements, whereby \(\mathrm{Ca^{2 + }}\) binding stabilizes an extended conformation of TM6 (8, 27, 28). While numerous studies have established a critical role for TM6 in binding \(\mathrm{Ca^{2 + }}\) for channel activation (1, 2), it is an intriguing open question as to how TMEM16A and TMEM16F functions might be modulated by small molecule drugs.  
+
+We determined cryo- EM structures of TMEM16F in three distinct unliganded states that reveal structural asymmetry and shed light into the mechanisms underlying \(\mathrm{Ca^{2 + }}\) - activated lipid scrambling. We also determined structures of TMEM16F bound to nicosamide and 1PBC, revealing that both molecules bind the same hydrophobic groove. We validated the binding site using computational docking and mutagenesis analyses and our data also indicate that both nicosamide and 1PBC bind to the equivalent site in TMEM16A. Our work establishes a structural foundation for designing more potent and specific antagonists against TMEM16 proteins with critical implications for the treatment of cancer, asthma and COVID- 19.  
+
+## Results  
+
+## Cryo-EM analysis reveals an asymmetric state of the TMEM16F dimer  
+
+A plethora of genetic, biochemical and electrophysiological studies show that binding of phosphatidylinositol 4,5- biphosphate (PIP2) is important for activation of both TMEM16A and TMEM16F (29- 31). Combination of lipid nanodisc technology with single particle cryo- EM allows structural analysis of membrane proteins embedded in a lipid bilayer (32, 33), which is critical for TMEM16 proteins and other membrane proteins that are modulated by lipids. However, TMEM16 proteins in nanodiscs present strong preferred orientation in particle distribution, severely limiting the attainable resolution of cryo- EM structures of TMEM16 proteins and hampering the study of these proteins in the context of a lipid bilayer (8, 27). We overcame this limitation by collecting data from tilted specimen and implementing an image
+
+<--- Page Split --->
+
+processing pipeline that allowed us to systematically determine sub 3.5 Å structures of TMEM16F in lipid nanodiscs in the presence or absence of different ligands (see Materials and Methods, figs. S1, S2, S3, S4 and table S1).  
+
+First, we determined multiple structures of TMEM16F in the presence of \(\mathrm{Ca^{2 + }}\) and \(\mathrm{PIP}_2\) . These structures represent different conformations of TMEM16F in unliganded states. The quality of these reconstructions enables atomic model building of the TM helices, most of the extracellular and intracellular domains, as well as \(\mathrm{Ca^{2 + }}\) ions and dozens of lipid densities associated with the protein (Fig. 1 and table S1). Whereas all previously reported TMEM16F structures were determined with the assumption of C2 symmetry, we did not impose symmetry and identified 3 distinct states with major differences in the conformation of TM6 and the number of \(\mathrm{Ca^{2 + }}\) atoms bound in each monomer (fig. S4). In State A, both monomers are bound to 2 \(\mathrm{Ca^{2 + }}\) ions and present a clear density for an extended TM6 (fig. S4). In State B, one monomer has 2 \(\mathrm{Ca^{2 + }}\) ions and a straight TM6, whereas the other monomer appears to contain a single \(\mathrm{Ca^{2 + }}\) ion, as density for the second \(\mathrm{Ca^{2 + }}\) ion is significantly weaker. In this single \(\mathrm{Ca^{2 + }}\) - bound monomer, TM6 presents a kink at P628 (Fig. 1). Thus, this structure represents an asymmetric state of the dimer (Fig. 1). In State C, both monomers contain only 1 \(\mathrm{Ca^{2 + }}\) ion and TM6 is bent in both (fig. S4). Comparison between these 3 classes reveals that straightening of TM6 correlates with binding of the second \(\mathrm{Ca^{2 + }}\) ion, whereas kinking of TM6 is associated with an outward rigid body motion of the intracellular domain that brings it closer to the nanodisc (Fig. 1A and fig. S4). Moreover, bending of TM6 directly correlates with distortion of the nanodisc and significant thinning of the membrane at the kinking position (Fig. 1A and fig. S4). Consistent with our previous study of TMEM16F (8), these observations support the notion that kinking of TM6 at P628 causes membrane distortion.  
+
+Our reconstructions also reveal previously unobserved features, including glycans and conserved disulfide bonds in the extracellular region (Fig. 1 and fig. S5, A and B), as well as the presence of a third \(\mathrm{Ca^{2 + }}\) ion coordinated by E395 on TM2 as well as S854 and D859 on TM10, near the dimer interface in the intracellular region of the protein (fig. S5C). These features are likely present in previous reconstructions but not detected due to limited resolution. In fact, a similar \(\mathrm{Ca^{2 + }}\) - binding site has been recently found in TMEM16F (10) as well as TMEM16K (34), and biochemical studies indicate that an equivalent third \(\mathrm{Ca^{2 + }}\) - binding site allosterically regulates channel activity in TMEM16A (35).  
+
+We are also able to unambiguously assign the residues of TM4 and precisely determine the pore- lining residues on TM4 (fig. S5D). These residues form a network of OH- containing side chains along the hydrophilic pore that constitutes an ideal environment for ion conduction across the membrane (Fig. 1C). However, the ion conduction pore is closed in all states resolved in this study and its hydrophilic interior is not accessible to lipids from the surrounding membrane (fig. S5E).  
+
+## TM1 and TM6 form a hydrophobic groove that can be occupied by lipids
+
+<--- Page Split --->
+
+In all three classes, we noticed a trail of densities that appear to correspond to a mixture of multiple lipids extending across the entire lipid bilayer along the membrane- facing surface of each TMEM16F monomer (Fig. 1B). A hydrophobic groove formed between TM1 and TM6 near the extracellular edge of the membrane appears to play a major role in accommodating these lipids. Intriguingly, this area corresponds to the position where membrane thinning occurs. To further investigate these lipid densities, we combined particles from all three States and carried out focused classification around this groove in a single monomer (See Materials and Methods, fig. S1). The particles clustered primarily to 2 classes that each contained approximately \(40\%\) of the particles and rendered 3.1 Å resolution structures (figs. S1 and S2). The overall organization of Class 1 and 2 is essentially indistinguishable (Fig. 2). However, Class 1 almost entirely lacks lipid densities in the TM1- TM6 groove, whereas Class 2 has strong density for numerous lipids in this area (Fig. 2, A and B). This indicates that our dataset contains a mixture of monomers in lipid- free and lipid- bound states.  
+
+## Niclosamide binds the hydrophobic groove formed between TM1 and TM6  
+
+Niclosamide is an FDA- approved drug that has recently emerged as a promising drug for treating severe cases of COVID- 19 (26) (clinicaltrials.gov), and its propensity to inhibit syncytia formation has been attributed to its ability to inhibit TMEM16F (13). Seeking to determine the binding site of this antagonist, we added 50 mM niclosamide to our biochemical preparation and imaged this sample following identical image processing pipeline as in the apo dataset presented above (figs. S1 and S2). In this case, however, focused classification around the TM1- TM6 groove rendered 3 classes. Like in our control sample, Classes 1 and 2 are distinguished by the absence or presence of lipids in the groove. Meanwhile, Class 3 contains a well- defined density in the TM1- TM6 groove that fits niclosamide well while no trail of lipid densities is found in the hydrophobic pocket (Fig. 2C). The niclosamide- like density contacts F321 on TM1, K370 on the TM1- TM2 loop, T606, T607 and T610 on TM6, and F685 and L687 on the TM7- TM8 loop (Fig. 2C). The resolution of our reconstruction is insufficient to unambiguously determine the precise pose of the molecule within the density. To gain some insight into how niclosamide may be oriented within TMEM16F, the compound was computationally docked using the Glide docking software. Using only the atomic model of TMEM16F (without access to our cryo- EM density map), the software identified this pocket as the most likely binding site and the highest- ranking pose fits our cryo- EM density well (Fig. 2C and fig. S6). Notably, this pose had the lowest binding energy and predicts formation of a hydrogen bond with T610. Taken together, our structural and computational data show that niclosamide binds TMEM16F at the hydrophobic groove formed between TM1 and TM6 and that binding of niclosamide prevents lipids from occupying this pocket.  
+
+## 1PBC is a potent inhibitor of TMEM16F  
+
+Niclosamide is known to inhibit both TMEM16F and TMEM16A channels (15). Given the structural similarities between both paralogs, we reasoned that 1PBC, a potent inhibitor of TMEM16A, might also modulate TMEM16F. To test this hypothesis, we first measured \(\mathrm{Ca^{2 + }}\) influx using Fluo8 as a small
+
+<--- Page Split --->
+
+molecule \(\mathrm{Ca^{2 + }}\) reporter dye. Application of 1PBC led to a significant decrease in TMEM16F- dependent \(\mathrm{Ca^{2 + }}\) influx upon chemical induction (Fig. 3 and fig. S7). This indicates that 1PBC is a potent inhibitor of TMEM16F ion channel activity. Next, we explored whether TMEM16F lipid scramblase activity is also inhibited by 1PBC by imaging PS exposure using pSIVA, a fluorescent annexin derivative. Upon chemical induction, the average onset for PS exposure in vehicle controls was 17.23 min (Fig. 3 and fig. S7). 1PBC robustly delayed the onset of TMEM16F- dependent PS exposure to 32.06 min (Fig. 3). We conclude that, like nicosamide, 1PBC potently inhibits TMEM16F function by reducing both ion conduction and lipid scrambling activity.  
+
+## 1PBC and nicosamide target the same site in TMEM16F  
+
+To elucidate the binding site of 1PBC, we supplemented our TMEM16F sample with 100 mM 1PBC. Here too we identified 3 distinct classes that closely resemble the 3 states observed in our drug- free sample. However, lipid densities along the membrane- facing surface of each monomer are absent. Instead, in all three classes we found a strong oval- shaped density in the same hydrophobic groove identified as the drug binding site in our nicosamide- supplemented dataset (Fig. 2D). This density, which is remarkably different from the lipid- like and nicosamide- like densities in our ligand- free and nicosamide- bound structures, fits 1PBC well. Overlay of the 1PBC- bound structure with our control revealed subtle side chain rearrangements of the residues surrounding this density. More specifically, K370 appears to shift from interacting with E366 to establishing a hydrogen bond with the compound (fig. S5F). Consistent with these observations, computational docking using Glide independently predicts formation of a hydrogen bond between K370 and 1PBC and identifies a pose for the ligand that fits our density map well (Fig. 2D and fig. S6). Together, our data show that 1PBC and nicosamide target the same site in TMEM16F and appear to replace bound lipids in the hydrophobic groove formed between TM1 and TM6.  
+
+## Functional validation of the drug binding site in TMEM16F  
+
+We previously showed that chemical induction of giant plasma membrane vesicle formation involves TMEM16F- dependent \(\mathrm{Ca^{2 + }}\) influx as well as TMEM16F- dependent PS exposure in HEK293 cells (7, 8), so it is a robust assay for evaluating TMEM16F activity. We thus generated stable cell lines expressing wildtype or mutant TMEM16F- mScarlet containing alanine substitutions of the residues surrounding the inhibitor densities: F321 on TM1, K370 and F374 on the TM1- TM2 loop, T606 on TM6, and F685 on the TM7- TM8 loop. Interestingly, mutation of these residues altered the basal activity of TMEM16F.  
+
+Compared to the wildtype control, F321A shortened the onset of \(\mathrm{Ca^{2 + }}\) influx by nearly twofold and reduced the onset latency of the PS exposure from 17.23 min to 11.61 min (Fig. 3). In contrast, K370A significantly delayed the onset of PS exposure to 30.16 min. These results indicate that this pocket and its endogenous lipids are critical for scramblase activity of TMEM16F (Fig. 3 and fig. S7). Importantly, in wild type controls, both 1PBC and nicosamide significantly delayed the onset of internal \(\mathrm{Ca^{2 + }}\) rise and PS exposure (Fig. 3). The inhibitory effect of both antagonists was significantly decreased by all the mutations, confirming that these residues are important for binding these inhibitors (Fig. 3). In fact, the F321A mutation almost completely obliterated the inhibitory effect on the onset of both \(\mathrm{Ca^{2 + }}\) rise and PS
+
+<--- Page Split --->
+
+exposure (Fig. 3). In summary, we show that residues in the TM1- TM6 groove are important for nicosamide- and 1PBC- mediated inhibition of TMEM16F and this area is critical for scramblase activity. Functional and computational validation of the nicosamide and 1PBC binding site in TMEM16A  
+
+Nicosamide and 1PBC are potent inhibitors of both TMEM16A and TMEM16F (15). Since the binding pocket we identify in TMEM16F presents a high degree of conservation in TMEM16A, we reasoned that both inhibitors may bind equivalent sites in TMEM16A and TMEM16F. To investigate this hypothesis, we tested whether mutations of residues in the putative binding pocket affect 1PBC- or nicosamide- mediated inhibition of TMEM16A (Fig. 4). We used whole cell patch clamp electrophysiology to measure \(\mathrm{Ca}^{2 + }\) - activated \(\mathrm{Cl}^-\) currents from HEK293 cells expressing either wildtype or mutant TMEM16A and tested the effects of alanine substitutions of F353 on TM1, R399 and F404 on the TM1- TM2 loop or F720 on the TM7- TM8 loop. In the absence of antagonists, these mutations did not alter the \(\mathrm{Cl}^-\) current induced by \(\mathrm{Ca}^{2 + }\) activation of TMEM16A. R399A and F720A significantly reduced the inhibitory effects of both 1PBC and nicosamide while F353A affected the inhibitory effects of nicosamide but not 1PBC, confirming that these residues are important for the interaction of these drugs with TMEM16A (Fig. 4, A to D). Notably, F404A did not alter the efficiency of either of the inhibitors, whereas the equivalent mutation in TMEM16F, F374A, decreased inhibition (Fig. 3). It thus appears that the functional relevance of the specific residues within the binding site might vary between TMEM16A and TMEM16F.  
+
+We further used Glide to computationally dock nicosamide and 1PBC into the \(\mathrm{Ca}^{2 + }\) - bound TMEM16A structure following identical procedures as in TMEM16F, for docking into a cube of 30 Å length on each side. In both cases, the software found binding in this pocket to be most energetically favorable (fig. S6). Taken together, our data indicate that 1PBC and nicosamide bind the same binding pocket in a hydrophobic groove formed between TM1 and TM6 in both TMEM16A and TMEM16F.  
+
+## Discussion  
+
+TMEM16 proteins assemble as dimers and whether the two monomers function independently or cooperatively is unclear. Unlike all previously solved structures of TMEM16 proteins, our reconstructions of TMEM16F in the presence of \(\mathrm{PIP}_2\) with or without drug molecules reveal a high degree of asymmetry within the dimer (Fig. 1). It is important to note that previous studies of TMEM16 proteins in lipid nanodiscs imposed C2 symmetry during cryo- EM data processing. The asymmetry of TMEM16F dimers we observe in our C1 reconstructions is likely linked to the mechanism(s) underlying TMEM16F function and \(\mathrm{PIP}_2\) - mediated activation.  
+
+Structural analysis of TMEM16F in lipid nanodiscs supplemented with \(\mathrm{PIP}_2\) reveals 3 distinct coexisting states and a direct correlation between kinking of TM6 and membrane distortion (fig. S4). A continuous trail of lipids connects the intra- and extracellular sides of TMEM16F at the membrane distortion site (Fig. 1B). These findings are consistent with our previous structural and mutagenesis data (8) and
+
+<--- Page Split --->
+
+support a model for TMEM16F- mediated scrambling of lipids, whereby TMEM16F distorts the membrane, minimizing the distance between the inner and outer leaflets of the lipid bilayer (Fig. 5A). We further find that residues along this lipid trail, such as K370 and F321, are important for scramblase activity. In fact, K370 is a positively charged residue that is ideally positioned for interacting with negatively charged phospholipid heads at the membrane interface. The fact that TMEM16A, which cannot scramble lipids, contains an alanine in this position reinforces the notion that this basic residue is critical for lipid scrambling (Fig. 5B). Together, our data suggest that the lipid trail we identify on TMEM16F might correspond to the pathway for lipid scrambling. We propose that the lipids "surf" along this membrane- facing groove, crossing between the inner and outer leaflets through a path that does not directly involve the hydrophilic ion conduction pore (Fig. 5A).  
+
+TMEM16 proteins have emerged as important pharmacological targets for the treatment of cancer, asthma and more recently COVID- 19 (12, 13) (26) (clinicaltrials.gov). Our data indicate that nicoslamide and 1PBC bind the same, conserved site in both TMEM16A and TMEM16F (Fig. 5B and fig. S6). With both antagonists directly contacting the extracellular- proximal end of TM6, our mutagenesis studies suggest that residues on this helix are critical for drug binding (Figs. 3 and 4). TM6 is the main gating element of the channel as well as part of the ion conduction pore in both TMEM16A and TMEM16F (1, 2). We thus speculate that, by binding to the upper part of TM6, these antagonists simultaneously lock the ion conduction pore and the gating element in a closed configuration.  
+
+Interestingly, the binding site of 1PBC and nicoslamide coincides with the position where we observe the maximum degree of membrane distortion and thinning in the TMEM16F structures. In our model for TMEM16F activity, this site corresponds precisely with the entry and exit point of the lipids as they transition between the inner and outer leaflets of the plasma membrane (Fig. 5A). Our structures show that both antagonists replace the lipids found in this pocket in our drug- free sample (Fig. 2). This suggests that 1PBC and nicoslamide might directly inhibit TMEM16F scramblase activity by physically occluding the path of the lipids across the membrane (Fig. 5A). In fact, lipid densities along this path are significantly reduced in our 1PBC- and nicoslamide- bound structures (Fig. 2). Consistent with a critical role of this region for lipid scrambling, alanine substitutions of residues within the drug binding pocket significantly alter the lipid scrambling activity of TMEM16F in the absence of inhibitors, whereas equivalent mutations in TMEM16A do not affect \(\mathrm{Ca^{2 + }}\) - activated Cl- channel activity (Figs. 3 and 4).  
+
+Like 1PBC and nicoslamide, most drug molecules identified to date are not specific for any particular TMEM16 paralog and instead broadly target TMEM16 family members. This raises concerns about potential off target effects in the clinic. Additionally, our whole cell patch clamp electrophysiology experiments, as well as previous studies (36), show that nicoslamide may activate an ion channel. Although both nicoslamide and 1PBC appear to bind the same, conserved hydrophobic pocket in TMEM16A and TMEM16F, our data also show that the specific contribution of different residues to the interaction is distinct in TMEM16A and TMEM16F (Figs. 3 and 4). We further demonstrate that non- conserved residues within this region, such T606 and K370 in TMEM16F, are important for the inhibitory effects of 1PBC and/or nicoslamide (Fig. 3). In fact, comparison between the two pockets reveals that the
+
+<--- Page Split --->
+
+TMEM16A binding pocket consists almost exclusively of hydrophobic residues, whereas the equivalent site in TMEM16F includes several charged and OH- containing side chains (Fig. 5B). Nicolasamide is a highly hydrophobic molecule that presents extremely poor solubility in aqueous solutions (37). Thus, the identification of non- conserved hydrophilic residues within the drug binding pocket in TMEM16F opens the door for the development of nicosamide analogs with better pharmacological properties that exclusively target TMEM16F for the treatment of severe COVID- 19. Taken together, our work establishes a much- needed structural framework for designing more potent and more specific antagonists against individual members of the TMEM16 family with critical implications for the treatment of asthma, cancer and COVID- 19.  
+
+## Declarations  
+
+## Acknowledgments  
+
+We thank D. Bulkley and Zanlin Yu at UCSF cryo- EM facility for help with grid screening. We also thank Z. Yu, Rui Yan and Shixin Yang at the HHMI Janelia Cryo- EM Facility for help with data acquisition and colleagues in our laboratories for discussion. This work was supported by grants from the NIH (1R35GM140847, S100D0020054, and S100D021741 to Y.C.; R01NS069229 and R35NS097229 to L.Y.J.). C.P. is supported by a Damon Runyon Postdoctoral Fellowship Award. T.W.H. is supported by the Jane Coffin Childs Memorial Fund for Medical Research. L.Y.J., Y.N.J., and Y.C. are Investigators with the Howard Hughes Medical Institute.  
+
+## References  
+
+1. M. E. Falzone, M. Malvezzi, B. C. Lee, A. Accardi, Known structures and unknown mechanisms of TMEM16 scramblases and channels. J Gen Physiol 150, 933-947 (2018).  
+2. V. Kalienkova, V. Clerico Mosina, C. Paulino, The Groovy TMEM16 Family: Molecular Mechanisms of Lipid Scrambling and Ion Conduction. J Mol Biol 433, 166941 (2021).  
+3. A. Caputo et al., TMEM16A, a membrane protein associated with calcium-dependent chloride channel activity. Science 322, 590-594 (2008).  
+4. Y. D. Yang et al., TMEM16A confers receptor-activated calcium-dependent chloride conductance. Nature 455, 1210-1215 (2008).  
+5. B. C. Schroeder, T. Cheng, Y. N. Jan, L. Y. Jan, Expression cloning of TMEM16A as a calcium-activated chloride channel subunit. Cell 134, 1019-1029 (2008).  
+6. H. Yang et al., TMEM16F forms a \(\mathrm{Ca^{2 + }}\) -activated cation channel required for lipid scrambling in platelets during blood coagulation. Cell 151, 111-122 (2012).  
+7. T. W. Han et al., Chemically induced vesiculation as a platform for studying TMEM16F activity. Proc Natl Acad Sci U S A 116, 1309-1318 (2019).
+
+<--- Page Split --->
+
+8. S. Feng et al., Cryo-EM studies of TMEM16F calcium-activated ion channel suggest features important for lipid scrambling. Cell Rep 28, 567-579 e564 (2019).  
+
+9. W. Ye, T. W. Han, M. He, Y. N. Jan, L. Y. Jan, Dynamic change of electrostatic field in TMEM16F permeation pathway shifts its ion selectivity. Elife 8, (2019).  
+
+10. C. Alvadia et al., Cryo-EM structures and functional characterization of the murine lipid scramblase TMEM16F. Elife 8, (2019).  
+
+11. R. Watanabe, T. Sakuragi, H. Noji, S. Nagata, Single-molecule analysis of phospholipid scrambling by TMEM16F. Proc Natl Acad Sci U S A 115, 3066-3071 (2018).  
+
+12. W. Chen et al., The Prognostic Value and Mechanisms of TMEM16A in Human Cancer. Front Mol Biosci 8, 542156 (2021).  
+
+13. L. Braga et al., Drugs that inhibit TMEM16 proteins block SARS-CoV-2 spike-induced syncytia. Nature 594, 88-93 (2021).  
+
+14. H. L. Danahay et al., TMEM16A Potentiation: A Novel Therapeutic Approach for the Treatment of Cystic Fibrosis. Am J Respir Crit Care Med 201, 946-954 (2020).  
+
+15. I. Cabrita, R. Benedetto, R. Schreiber, K. Kunzelmann, Niclosamide repurposed for the treatment of inflammatory airway disease. JCI Insight 4, (2019).  
+
+16. K. Miner et al., Drug Repurposing: The Anthelmintics Niclosamide and Nitazoxanide Are Potent TMEM16A Antagonists That Fully Bronchodilate Airways. Front Pharmacol 10, 51 (2019).  
+
+17. T. Fujii, A. Sakata, S. Nishimura, K. Eto, S. Nagata, TMEM16F is required for phosphatidylserine exposure and microparticle release in activated mouse platelets. Proc Natl Acad Sci U S A 112, 12800-12805 (2015).  
+
+18. P. Wolf, The nature and significance of platelet products in human plasma. Br J Haematol 13, 269-288 (1967).  
+
+19. R. F. Zwaal, P. Comfurius, E. M. Bevers, Scott syndrome, a bleeding disorder caused by defective scrambling of membrane phospholipids. Biochim Biophys Acta 1636, 119-128 (2004).  
+
+20. J. Suzuki, M. Umeda, P. J. Sims, S. Nagata, Calcium-dependent phospholipid scrambling by TMEM16F. Nature 468, 834-838 (2010).  
+
+21. B. Gyorgy et al., Membrane vesicles, current state-of-the-art: emerging role of extracellular vesicles. Cell Mol Life Sci 68, 2667-2688 (2011).  
+
+22. G. Raposo, W. Stoorvogel, Extracellular vesicles: exosomes, microvesicles, and friends. J Cell Biol 200, 373-383 (2013).  
+
+23. P. J. Sims, T. Wiedmer, C. T. Esmon, H. J. Weiss, S. J. Shattil, Assembly of the platelet prothrombinase complex is linked to vesiculation of the platelet plasma membrane. Studies in Scott syndrome: an isolated defect in platelet procoagulant activity. J Biol Chem 264, 17049-17057 (1989).  
+
+24. J. M. Whitlock, H. C. Hartzell, Anoctamins/TMEM16 Proteins: Chloride Channels Flirting with Lipids and Extracellular Vesicles. Annu Rev Physiol 79, 119-143 (2017).
+
+<--- Page Split --->
+
+25. N. Wu et al., Critical Role of Lipid Scramblase TMEM16F in Phosphatidylserine Exposure and Repair of Plasma Membrane after Pore Formation. Cell Rep 30, 1129-1140 e1125 (2020).  
+
+26. A. S. Abdulamir et al., A randomised controlled trial of effectiveness and safety of Nicolasamide as add on therapy to the standard of care measures in COVID-19 management. Ann Med Surg (Lond) 69, 102779 (2021).  
+
+27. S. Dang et al., Cryo-EM structures of the TMEM16A calcium-activated chloride channel. Nature 552, 426-429 (2017).  
+
+28. C. Paulino, V. Kalienkova, A. K. M. Lam, Y. Neldner, R. Dutzler, Activation mechanism of the calcium-activated chloride channel TMEM16A revealed by cryo-EM. Nature 552, 421-425 (2017).  
+
+29. J. Arreola, H. C. Hartzell, Wasted TMEM16A channels are rescued by phosphatidylinositol 4,5-bisphosphate. Cell Calcium 84, 102103 (2019).  
+
+30. S. C. Le, Z. Jia, J. Chen, H. Yang, Molecular basis of PIP2-dependent regulation of the Ca(2+)-activated chloride channel TMEM16A. Nat Commun 10, 3769 (2019).  
+
+31. W. Ye et al., Phosphatidylinositol-(4, 5)-bisphosphate regulates calcium gating of small-conductance cation channel TMEM16F. Proc Natl Acad Sci U S A 115, E1667-E1674 (2018).  
+
+32. Y. Gao, E. Cao, D. Julius, Y. Cheng, TRPV1 structures in nanodiscs reveal mechanisms of ligand and lipid action. Nature 534, 347-351 (2016).  
+
+33. M. A. Schuler, I. G. Denisov, S. G. Sligar, Nanodiscs as a new tool to examine lipid-protein interactions. Methods Mol Biol 974, 415-433 (2013).  
+
+34. S. R. Bushell et al., The structural basis of lipid scrambling and inactivation in the endoplasmic reticulum scramblase TMEM16K. Nat Commun 10, 3956 (2019).  
+
+35. S. C. Le, H. Yang, An Additional Ca(2+) Binding Site Allosterically Controls TMEM16A Activation. Cell Rep 33, 108570 (2020).  
+
+36. B. Biton et al., The antipsychotic drug loxapine is an opener of the sodium-activated potassium channel slack (Slo2.2). J Pharmacol Exp Ther 340, 706-715 (2012).  
+
+37. W. Chen, R. A. Mook, Jr., R. T. Premont, J. Wang, Niclosamide: Beyond an antihelmintic drug. Cell Signal 41, 89-96 (2018).  
+
+38. D. N. Mastronarde, Automated electron microscope tomography using robust prediction of specimen movements. J Struct Biol 152, 36-51 (2005).  
+
+39. Y. Z. Tan et al., Addressing preferred specimen orientation in single-particle cryo-EM through tilting. Nat Methods 14, 793-796 (2017).  
+
+40. S. Q. Zheng et al., MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat Methods 14, 331-332 (2017).  
+
+41. A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat Methods 14, 290-296 (2017).  
+
+42. P. Emsley, K. Cowtan, Coot: model-building tools for molecular graphics. Acta Crystallogr D Biol Crystallogr 60, 2126-2132 (2004).
+
+<--- Page Split --->
+
+43. P. V. Afonine et al., Towards automated crystallographic structure refinement with phenix.refine. Acta Crystallogr D Biol Crystallogr 68, 352–367 (2012).44. R. A. Friesner et al., Glide: a new approach for rapid, accurate docking and scoring. 1. Method and assessment of docking accuracy. J Med Chem 47, 1739–1749 (2004).45. E. Harder et al., OPLS3: A Force Field Providing Broad Coverage of Drug-like Small Molecules and Proteins. J Chem Theory Comput 12, 281–296 (2016).46. J. C. Shelley et al., Epik: a software program for pK(a) prediction and protonation state generation for drug-like molecules. J Comput Aided Mol Des 21, 681–691 (2007).47. R. A. Friesner et al., Extra precision glide: docking and scoring incorporating a model of hydrophobic enclosure for protein-ligand complexes. J Med Chem 49, 6177–6196 (2006).  
+
+## Figures
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1 </center>  
+
+Cryo- EM analysis reveals asymmetry within the TMEM16F dimer. (A) Cryo- EM density of the asymmetric state of the TMEM16F dimer with the monomers colored blue and light blue, respectively, and the lipid densities in grey. The gaussian filtered cryo- EM density (semitransparent) reveals distortion of the lipid nanodisc. (B) Side view of a TMEM16F monomer (blue) highlighting the trail of lipids (grey) covering the TM region. (C) Front and side view of the atomic model of the asymmetric state of TMEM16F with \(\mathrm{Ca^{2 + }}\) atoms and glycans shown in green and red, respectively. The ion conduction channel identified by HOLE is represented by spheres colored in rainbow scale based on the local width of the channel, where red \(< 1.5 \text{Å}\) and blue \(> 7.5 \text{Å}\) .
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2 </center>  
+
+Nicosamide and 1PBC bind the same hydrophobic groove in TMEM16F. Atomic model of the TM1- TM6 region of (A) Class1 and (B) Class 2 of the drug- free control, (C) nicosamide- and (D) 1PBC- supplemented datasets. In each case, the additional cryo- EM densities found in the area are shown. Below, zoom into the TM1- TM6 groove with the residues shown as sticks and colored by heteroatom and the additional density found within the pocket shown in semitransparent. Structures of nicosamide and 1PBC as determined by computational docking using Glide are shown in purple and green, respectively.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3 </center>  
+
+Functional validation of the drug binding site in TMEM16F. Representative curves of live imaging of TMEM16F- dependent PS exposure (A) and (C); and \(\mathrm{Ca^{2 + }}\) influx (E) and (G). Data are represented as mean \(\pm\) SEM. Scattered dot plots of time of onset of TMEM16F- dependent PS exposure [(B) and (D)] and \(\mathrm{Ca^{2 + }}\) influx [(F) and (H)]. Time of onset could not be determined for time courses with a linear rather than sigmoidal rise. The mean \(\pm\) SEM is shown along with the statistical significance determined by unpaired
+
+<--- Page Split --->
+
+t- test for each mutant as compared to vehicle controls ( \(^{*}p < 0.05\) ; \(^{**}p < 0.01\) ; \(^{***}p < 0.0001\) ; \(^{****}p < 0.0001\) ).  
+
+![](images/Figure_4.jpg)
+
+<center>Figure 4 </center>  
+
+Electrophysiology- based validation of the binding site in TMEM16A. Representative voltage clamp current traces at \(+70 \text{mV}\) with holding potential at - 5 mV of wildtype mTMEM16A and alanine substitution mutants in the absence or presence of (A) 3 μM nicosamide or (B) 30 μM 1PBC recorded in 500 nM \([Ca^{2 + }]_i\) or 12 mM \([Ca^{2 + }]_i\) , respectively. Graphs showing the coefficient of the steady- state current measured for the wildtype control and each mutant upon addition of (C) nicosamide and (D) 1PBC divided by the
+
+<--- Page Split --->
+
+maximum Intensity \((1 / 1_{\mathrm{max}})\) in each case. The mean \(\pm\) SEM is shown along with the statistical significance determined by unpaired t- test for each mutant as compared to wildtype controls ( \(*p<\) 0.05; \(**p< 0.01\) ; \(***p< 0.001\) ).  
+
+![](images/Figure_5.jpg)
+
+<center>Figure 5 </center>  
+
+Comparison of the drug binding pocket in TMEM16A and TMEM16F. (A) Schematic representation of the TMEM16F dimer (light blue and blue) embedded in a lipid bilayer (grey), where \(\mathrm{Ca^{2 + }}\) atoms are shown as green circles and the inhibitors as a purple polygon and dotted black lines represent the closed ion conduction pore. (B) Structure of the drug binding pocket in TMEM16F (blue, left) and TMEM16A (grey, right) with the side chains of the surrounding residues shown as sticks and the non- conserved residues highlighted in orange. Computationally docked structures of nicosamide and 1PBC are shown in purple and green, respectively. All atoms are colored by heteroatom.  
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+Supplementary materials.docx
+
+<--- Page Split --->

preprint/preprint__b367bf4492c30047229c142d659b318a8f5859981a4e643b55522b53a7f727fa/preprint__b367bf4492c30047229c142d659b318a8f5859981a4e643b55522b53a7f727fa_det.mmd

@@ -0,0 +1,383 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 943, 175]]<|/det|>
+# Identification of a conserved drug binding pocket in TMEM16 proteins  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 384, 234]]<|/det|>
+Yifan Cheng ( \(\boxed{\times}\) ycheng@ucsf.edu) University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 241, 170, 280]]<|/det|>
+Shengjie Feng UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 289, 204, 327]]<|/det|>
+Cristina Puchades UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 336, 140, 373]]<|/det|>
+Juyeon Ko UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 382, 463, 421]]<|/det|>
+Eric Figueroa UCSF https://orcid.org/0000- 0001- 8562- 6552  
+
+<|ref|>text<|/ref|><|det|>[[44, 428, 140, 465]]<|/det|>
+Yifei Chen UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 474, 384, 513]]<|/det|>
+Hao Wu University of California San Francisco  
+
+<|ref|>text<|/ref|><|det|>[[44, 520, 120, 557]]<|/det|>
+Shuo Gu UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 567, 125, 602]]<|/det|>
+Tina Han UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 612, 122, 648]]<|/det|>
+Junrui Li UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 658, 150, 694]]<|/det|>
+Brandon Ho UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 704, 174, 740]]<|/det|>
+Brian Shoichet UCSF https://orcid.org/0000- 0002- 6098- 7367  
+
+<|ref|>text<|/ref|><|det|>[[44, 749, 170, 785]]<|/det|>
+Yuh Nung Jan UCSF  
+
+<|ref|>text<|/ref|><|det|>[[44, 795, 115, 812]]<|/det|>
+Lily Jan  
+
+<|ref|>text<|/ref|><|det|>[[50, 818, 745, 838]]<|/det|>
+University of California, San Francisco https://orcid.org/0000- 0003- 3938- 8498  
+
+<|ref|>text<|/ref|><|det|>[[44, 880, 101, 897]]<|/det|>
+Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 917, 135, 935]]<|/det|>
+Keywords:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[44, 45, 336, 64]]<|/det|>
+Posted Date: February 10th, 2022  
+
+<|ref|>text<|/ref|><|det|>[[44, 83, 474, 102]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 1296933/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 120, 911, 163]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 199, 930, 242]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on August 12th, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 40410- x.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 159, 68]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[41, 82, 944, 310]]<|/det|>
+AbstractThe TMEM16 family of calcium- activated membrane proteins includes ten mammalian paralogs (TMEM16A- K) playing distinct physiological roles with some implicated in cancer and airway diseases. Their modulators with therapeutic potential include 1PBC, a potent inhibitor with anti- tumoral properties, and the FDA- approved drug nicosamide that targets TMEM16F to inhibit syncytia formation induced by SARS- CoV- 2 infection. Here, we report cryo- EM structures of TMEM16F associated with 1PBC and nicosamide, revealing that both molecules bind the same drug binding pocket. We functionally and computationally validate this binding pocket in TMEM16A as well as TMEM16F, thereby showing that drug modulation also involves residues that are not conserved between TMEM16A and TMEM16F. This study establishes a much- needed structural framework for the development of more potent and more specific drug molecules targeting TMEM16 proteins.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 332, 175, 358]]<|/det|>
+## Main Text  
+
+<|ref|>text<|/ref|><|det|>[[40, 370, 955, 675]]<|/det|>
+TMEM16 proteins are a family of transmembrane proteins conserved across eukaryotes, encompassing 10 paralogs (TMEM16A- K) in mammals. Despite high sequence similarity, TMEM16 proteins present remarkable functional diversity (1, 2). For instance, TMEM16A is a calcium \((\mathrm{Ca}^{2 + })\) - activated chloride channel that opens in response to increased intracellular \(\mathrm{Ca}^{2 + }\) levels, enabling chloride ions to move across the plasma membrane (3- 5). In contrast, TMEM16F functions as both a \(\mathrm{Ca}^{2 + }\) - activated ion channel and a \(\mathrm{Ca}^{2 + }\) - activated lipid scramblase. TMEM16F channels permeate cations including \(\mathrm{Ca}^{2 + }\) ions (6- 8), however, its selectivity for cations versus anions may vary with the electrostatic field of the permeant pathway (9). Through its lipid scrambling activity, TMEM16F allows diverse lipids, including phosphatidylcholine (PC), phosphatidylethanolamine (PE) and phosphatidylserine (PS), to passively move between the inner and outer leaflets of the plasma membrane (1, 2, 6, 7, 10, 11). Both TMEM16A and TMEM16F play critical roles in numerous physiological processes and have emerged as important targets for therapeutic intervention in multiple diseases, including cancer, asthma, and in particular COVID- 19 (1, 2, 12, 13).  
+
+<|ref|>text<|/ref|><|det|>[[41, 688, 957, 847]]<|/det|>
+TMEM16A is required for airway and secretory gland secretion (1, 2). Notably, a TMEM16A activator is under consideration for the treatment of cystic fibrosis (14), whereas TMEM16A inhibitors with potent bronchodilator activities are being tested as anti- asthma drugs (15, 16). Additionally, TMEM16A activity is upregulated via gene amplification or enhanced expression in many types of cancers and is linked to increased cell migration and proliferation as well as metastatic progression (12). Therefore, antagonists of TMEM16A, such as 1PBC and nicosamide, provide a promising new avenue for the treatment of diverse cancers (16).  
+
+<|ref|>text<|/ref|><|det|>[[42, 863, 942, 953]]<|/det|>
+TMEM16F activity is important for blood coagulation (6, 17- 19) and mutations in TMEM16F are linked to the Scott syndrome bleeding disorder (20). TMEM16F also plays a critical role in extracellular vesicle generation and release (21- 24) and membrane repair as protection against bacterial infection (25). Importantly, nicosamide, an FDA approved drug, has recently been shown to block SARS- CoV- 2- induced
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 44, 951, 224]]<|/det|>
+syncytia formation and virus replication by inhibiting TMEM16F (13). The repurposing of nicosamide for treatment of severe COVID- 19 is currently under examination in more than a dozen clinical trials (26) (clinicaltrials.gov). Nicosamide is also a potent inhibitor of TMEM16A and robustly mitigates the symptoms of airway diseases in mice (15). Like nicosamide, most small molecule inhibitors identified to date affect multiple TMEM16 paralogs, making off- target effects a major concern for clinical applications. Elucidation of the ligand binding site is important not only for understanding the mechanism of action of these molecules but also for designing more specific drugs for pharmacological targeting of TMEM16 proteins.  
+
+<|ref|>text<|/ref|><|det|>[[41, 241, 936, 431]]<|/det|>
+TMEM16 proteins form dimers (1, 2); each subunit comprises 10 transmembrane helices (TMs) and contains its own ion conduction pore enclosed and surrounded by TM3- 7 (Fig. 1). \(\mathrm{Ca^{2 + }}\) - dependent activation involves direct binding of \(\mathrm{Ca^{2 + }}\) ions in two contiguous \(\mathrm{Ca^{2 + }}\) - binding sites that are formed between TM6- TM8 (Fig. 1). Structures of both TMEM16A and TMEM16F in \(\mathrm{Ca^{2 + }}\) - free and \(\mathrm{Ca^{2 + }}\) - bound states reveal that TM6 undergoes major \(\mathrm{Ca^{2 + }}\) - dependent conformational rearrangements, whereby \(\mathrm{Ca^{2 + }}\) binding stabilizes an extended conformation of TM6 (8, 27, 28). While numerous studies have established a critical role for TM6 in binding \(\mathrm{Ca^{2 + }}\) for channel activation (1, 2), it is an intriguing open question as to how TMEM16A and TMEM16F functions might be modulated by small molecule drugs.  
+
+<|ref|>text<|/ref|><|det|>[[41, 447, 949, 629]]<|/det|>
+We determined cryo- EM structures of TMEM16F in three distinct unliganded states that reveal structural asymmetry and shed light into the mechanisms underlying \(\mathrm{Ca^{2 + }}\) - activated lipid scrambling. We also determined structures of TMEM16F bound to nicosamide and 1PBC, revealing that both molecules bind the same hydrophobic groove. We validated the binding site using computational docking and mutagenesis analyses and our data also indicate that both nicosamide and 1PBC bind to the equivalent site in TMEM16A. Our work establishes a structural foundation for designing more potent and specific antagonists against TMEM16 proteins with critical implications for the treatment of cancer, asthma and COVID- 19.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 653, 145, 677]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 692, 840, 750]]<|/det|>
+## Cryo-EM analysis reveals an asymmetric state of the TMEM16F dimer  
+
+<|ref|>text<|/ref|><|det|>[[41, 768, 955, 952]]<|/det|>
+A plethora of genetic, biochemical and electrophysiological studies show that binding of phosphatidylinositol 4,5- biphosphate (PIP2) is important for activation of both TMEM16A and TMEM16F (29- 31). Combination of lipid nanodisc technology with single particle cryo- EM allows structural analysis of membrane proteins embedded in a lipid bilayer (32, 33), which is critical for TMEM16 proteins and other membrane proteins that are modulated by lipids. However, TMEM16 proteins in nanodiscs present strong preferred orientation in particle distribution, severely limiting the attainable resolution of cryo- EM structures of TMEM16 proteins and hampering the study of these proteins in the context of a lipid bilayer (8, 27). We overcame this limitation by collecting data from tilted specimen and implementing an image
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 955, 112]]<|/det|>
+processing pipeline that allowed us to systematically determine sub 3.5 Å structures of TMEM16F in lipid nanodiscs in the presence or absence of different ligands (see Materials and Methods, figs. S1, S2, S3, S4 and table S1).  
+
+<|ref|>text<|/ref|><|det|>[[40, 130, 958, 552]]<|/det|>
+First, we determined multiple structures of TMEM16F in the presence of \(\mathrm{Ca^{2 + }}\) and \(\mathrm{PIP}_2\) . These structures represent different conformations of TMEM16F in unliganded states. The quality of these reconstructions enables atomic model building of the TM helices, most of the extracellular and intracellular domains, as well as \(\mathrm{Ca^{2 + }}\) ions and dozens of lipid densities associated with the protein (Fig. 1 and table S1). Whereas all previously reported TMEM16F structures were determined with the assumption of C2 symmetry, we did not impose symmetry and identified 3 distinct states with major differences in the conformation of TM6 and the number of \(\mathrm{Ca^{2 + }}\) atoms bound in each monomer (fig. S4). In State A, both monomers are bound to 2 \(\mathrm{Ca^{2 + }}\) ions and present a clear density for an extended TM6 (fig. S4). In State B, one monomer has 2 \(\mathrm{Ca^{2 + }}\) ions and a straight TM6, whereas the other monomer appears to contain a single \(\mathrm{Ca^{2 + }}\) ion, as density for the second \(\mathrm{Ca^{2 + }}\) ion is significantly weaker. In this single \(\mathrm{Ca^{2 + }}\) - bound monomer, TM6 presents a kink at P628 (Fig. 1). Thus, this structure represents an asymmetric state of the dimer (Fig. 1). In State C, both monomers contain only 1 \(\mathrm{Ca^{2 + }}\) ion and TM6 is bent in both (fig. S4). Comparison between these 3 classes reveals that straightening of TM6 correlates with binding of the second \(\mathrm{Ca^{2 + }}\) ion, whereas kinking of TM6 is associated with an outward rigid body motion of the intracellular domain that brings it closer to the nanodisc (Fig. 1A and fig. S4). Moreover, bending of TM6 directly correlates with distortion of the nanodisc and significant thinning of the membrane at the kinking position (Fig. 1A and fig. S4). Consistent with our previous study of TMEM16F (8), these observations support the notion that kinking of TM6 at P628 causes membrane distortion.  
+
+<|ref|>text<|/ref|><|det|>[[41, 568, 955, 732]]<|/det|>
+Our reconstructions also reveal previously unobserved features, including glycans and conserved disulfide bonds in the extracellular region (Fig. 1 and fig. S5, A and B), as well as the presence of a third \(\mathrm{Ca^{2 + }}\) ion coordinated by E395 on TM2 as well as S854 and D859 on TM10, near the dimer interface in the intracellular region of the protein (fig. S5C). These features are likely present in previous reconstructions but not detected due to limited resolution. In fact, a similar \(\mathrm{Ca^{2 + }}\) - binding site has been recently found in TMEM16F (10) as well as TMEM16K (34), and biochemical studies indicate that an equivalent third \(\mathrm{Ca^{2 + }}\) - binding site allosterically regulates channel activity in TMEM16A (35).  
+
+<|ref|>text<|/ref|><|det|>[[41, 749, 955, 860]]<|/det|>
+We are also able to unambiguously assign the residues of TM4 and precisely determine the pore- lining residues on TM4 (fig. S5D). These residues form a network of OH- containing side chains along the hydrophilic pore that constitutes an ideal environment for ion conduction across the membrane (Fig. 1C). However, the ion conduction pore is closed in all states resolved in this study and its hydrophilic interior is not accessible to lipids from the surrounding membrane (fig. S5E).  
+
+<|ref|>sub_title<|/ref|><|det|>[[42, 861, 860, 925]]<|/det|>
+## TM1 and TM6 form a hydrophobic groove that can be occupied by lipids
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[40, 45, 955, 316]]<|/det|>
+In all three classes, we noticed a trail of densities that appear to correspond to a mixture of multiple lipids extending across the entire lipid bilayer along the membrane- facing surface of each TMEM16F monomer (Fig. 1B). A hydrophobic groove formed between TM1 and TM6 near the extracellular edge of the membrane appears to play a major role in accommodating these lipids. Intriguingly, this area corresponds to the position where membrane thinning occurs. To further investigate these lipid densities, we combined particles from all three States and carried out focused classification around this groove in a single monomer (See Materials and Methods, fig. S1). The particles clustered primarily to 2 classes that each contained approximately \(40\%\) of the particles and rendered 3.1 Å resolution structures (figs. S1 and S2). The overall organization of Class 1 and 2 is essentially indistinguishable (Fig. 2). However, Class 1 almost entirely lacks lipid densities in the TM1- TM6 groove, whereas Class 2 has strong density for numerous lipids in this area (Fig. 2, A and B). This indicates that our dataset contains a mixture of monomers in lipid- free and lipid- bound states.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 316, 820, 376]]<|/det|>
+## Niclosamide binds the hydrophobic groove formed between TM1 and TM6  
+
+<|ref|>text<|/ref|><|det|>[[40, 391, 955, 822]]<|/det|>
+Niclosamide is an FDA- approved drug that has recently emerged as a promising drug for treating severe cases of COVID- 19 (26) (clinicaltrials.gov), and its propensity to inhibit syncytia formation has been attributed to its ability to inhibit TMEM16F (13). Seeking to determine the binding site of this antagonist, we added 50 mM niclosamide to our biochemical preparation and imaged this sample following identical image processing pipeline as in the apo dataset presented above (figs. S1 and S2). In this case, however, focused classification around the TM1- TM6 groove rendered 3 classes. Like in our control sample, Classes 1 and 2 are distinguished by the absence or presence of lipids in the groove. Meanwhile, Class 3 contains a well- defined density in the TM1- TM6 groove that fits niclosamide well while no trail of lipid densities is found in the hydrophobic pocket (Fig. 2C). The niclosamide- like density contacts F321 on TM1, K370 on the TM1- TM2 loop, T606, T607 and T610 on TM6, and F685 and L687 on the TM7- TM8 loop (Fig. 2C). The resolution of our reconstruction is insufficient to unambiguously determine the precise pose of the molecule within the density. To gain some insight into how niclosamide may be oriented within TMEM16F, the compound was computationally docked using the Glide docking software. Using only the atomic model of TMEM16F (without access to our cryo- EM density map), the software identified this pocket as the most likely binding site and the highest- ranking pose fits our cryo- EM density well (Fig. 2C and fig. S6). Notably, this pose had the lowest binding energy and predicts formation of a hydrogen bond with T610. Taken together, our structural and computational data show that niclosamide binds TMEM16F at the hydrophobic groove formed between TM1 and TM6 and that binding of niclosamide prevents lipids from occupying this pocket.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 822, 638, 853]]<|/det|>
+## 1PBC is a potent inhibitor of TMEM16F  
+
+<|ref|>text<|/ref|><|det|>[[44, 869, 931, 937]]<|/det|>
+Niclosamide is known to inhibit both TMEM16F and TMEM16A channels (15). Given the structural similarities between both paralogs, we reasoned that 1PBC, a potent inhibitor of TMEM16A, might also modulate TMEM16F. To test this hypothesis, we first measured \(\mathrm{Ca^{2 + }}\) influx using Fluo8 as a small
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 45, 950, 228]]<|/det|>
+molecule \(\mathrm{Ca^{2 + }}\) reporter dye. Application of 1PBC led to a significant decrease in TMEM16F- dependent \(\mathrm{Ca^{2 + }}\) influx upon chemical induction (Fig. 3 and fig. S7). This indicates that 1PBC is a potent inhibitor of TMEM16F ion channel activity. Next, we explored whether TMEM16F lipid scramblase activity is also inhibited by 1PBC by imaging PS exposure using pSIVA, a fluorescent annexin derivative. Upon chemical induction, the average onset for PS exposure in vehicle controls was 17.23 min (Fig. 3 and fig. S7). 1PBC robustly delayed the onset of TMEM16F- dependent PS exposure to 32.06 min (Fig. 3). We conclude that, like nicosamide, 1PBC potently inhibits TMEM16F function by reducing both ion conduction and lipid scrambling activity.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 228, 904, 260]]<|/det|>
+## 1PBC and nicosamide target the same site in TMEM16F  
+
+<|ref|>text<|/ref|><|det|>[[41, 273, 950, 570]]<|/det|>
+To elucidate the binding site of 1PBC, we supplemented our TMEM16F sample with 100 mM 1PBC. Here too we identified 3 distinct classes that closely resemble the 3 states observed in our drug- free sample. However, lipid densities along the membrane- facing surface of each monomer are absent. Instead, in all three classes we found a strong oval- shaped density in the same hydrophobic groove identified as the drug binding site in our nicosamide- supplemented dataset (Fig. 2D). This density, which is remarkably different from the lipid- like and nicosamide- like densities in our ligand- free and nicosamide- bound structures, fits 1PBC well. Overlay of the 1PBC- bound structure with our control revealed subtle side chain rearrangements of the residues surrounding this density. More specifically, K370 appears to shift from interacting with E366 to establishing a hydrogen bond with the compound (fig. S5F). Consistent with these observations, computational docking using Glide independently predicts formation of a hydrogen bond between K370 and 1PBC and identifies a pose for the ligand that fits our density map well (Fig. 2D and fig. S6). Together, our data show that 1PBC and nicosamide target the same site in TMEM16F and appear to replace bound lipids in the hydrophobic groove formed between TM1 and TM6.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 568, 923, 602]]<|/det|>
+## Functional validation of the drug binding site in TMEM16F  
+
+<|ref|>text<|/ref|><|det|>[[41, 615, 949, 752]]<|/det|>
+We previously showed that chemical induction of giant plasma membrane vesicle formation involves TMEM16F- dependent \(\mathrm{Ca^{2 + }}\) influx as well as TMEM16F- dependent PS exposure in HEK293 cells (7, 8), so it is a robust assay for evaluating TMEM16F activity. We thus generated stable cell lines expressing wildtype or mutant TMEM16F- mScarlet containing alanine substitutions of the residues surrounding the inhibitor densities: F321 on TM1, K370 and F374 on the TM1- TM2 loop, T606 on TM6, and F685 on the TM7- TM8 loop. Interestingly, mutation of these residues altered the basal activity of TMEM16F.  
+
+<|ref|>text<|/ref|><|det|>[[41, 754, 955, 940]]<|/det|>
+Compared to the wildtype control, F321A shortened the onset of \(\mathrm{Ca^{2 + }}\) influx by nearly twofold and reduced the onset latency of the PS exposure from 17.23 min to 11.61 min (Fig. 3). In contrast, K370A significantly delayed the onset of PS exposure to 30.16 min. These results indicate that this pocket and its endogenous lipids are critical for scramblase activity of TMEM16F (Fig. 3 and fig. S7). Importantly, in wild type controls, both 1PBC and nicosamide significantly delayed the onset of internal \(\mathrm{Ca^{2 + }}\) rise and PS exposure (Fig. 3). The inhibitory effect of both antagonists was significantly decreased by all the mutations, confirming that these residues are important for binding these inhibitors (Fig. 3). In fact, the F321A mutation almost completely obliterated the inhibitory effect on the onset of both \(\mathrm{Ca^{2 + }}\) rise and PS
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 949, 150]]<|/det|>
+exposure (Fig. 3). In summary, we show that residues in the TM1- TM6 groove are important for nicosamide- and 1PBC- mediated inhibition of TMEM16F and this area is critical for scramblase activity. Functional and computational validation of the nicosamide and 1PBC binding site in TMEM16A  
+
+<|ref|>text<|/ref|><|det|>[[40, 163, 958, 489]]<|/det|>
+Nicosamide and 1PBC are potent inhibitors of both TMEM16A and TMEM16F (15). Since the binding pocket we identify in TMEM16F presents a high degree of conservation in TMEM16A, we reasoned that both inhibitors may bind equivalent sites in TMEM16A and TMEM16F. To investigate this hypothesis, we tested whether mutations of residues in the putative binding pocket affect 1PBC- or nicosamide- mediated inhibition of TMEM16A (Fig. 4). We used whole cell patch clamp electrophysiology to measure \(\mathrm{Ca}^{2 + }\) - activated \(\mathrm{Cl}^-\) currents from HEK293 cells expressing either wildtype or mutant TMEM16A and tested the effects of alanine substitutions of F353 on TM1, R399 and F404 on the TM1- TM2 loop or F720 on the TM7- TM8 loop. In the absence of antagonists, these mutations did not alter the \(\mathrm{Cl}^-\) current induced by \(\mathrm{Ca}^{2 + }\) activation of TMEM16A. R399A and F720A significantly reduced the inhibitory effects of both 1PBC and nicosamide while F353A affected the inhibitory effects of nicosamide but not 1PBC, confirming that these residues are important for the interaction of these drugs with TMEM16A (Fig. 4, A to D). Notably, F404A did not alter the efficiency of either of the inhibitors, whereas the equivalent mutation in TMEM16F, F374A, decreased inhibition (Fig. 3). It thus appears that the functional relevance of the specific residues within the binding site might vary between TMEM16A and TMEM16F.  
+
+<|ref|>text<|/ref|><|det|>[[42, 505, 949, 618]]<|/det|>
+We further used Glide to computationally dock nicosamide and 1PBC into the \(\mathrm{Ca}^{2 + }\) - bound TMEM16A structure following identical procedures as in TMEM16F, for docking into a cube of 30 Å length on each side. In both cases, the software found binding in this pocket to be most energetically favorable (fig. S6). Taken together, our data indicate that 1PBC and nicosamide bind the same binding pocket in a hydrophobic groove formed between TM1 and TM6 in both TMEM16A and TMEM16F.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 641, 191, 666]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[42, 680, 944, 841]]<|/det|>
+TMEM16 proteins assemble as dimers and whether the two monomers function independently or cooperatively is unclear. Unlike all previously solved structures of TMEM16 proteins, our reconstructions of TMEM16F in the presence of \(\mathrm{PIP}_2\) with or without drug molecules reveal a high degree of asymmetry within the dimer (Fig. 1). It is important to note that previous studies of TMEM16 proteins in lipid nanodiscs imposed C2 symmetry during cryo- EM data processing. The asymmetry of TMEM16F dimers we observe in our C1 reconstructions is likely linked to the mechanism(s) underlying TMEM16F function and \(\mathrm{PIP}_2\) - mediated activation.  
+
+<|ref|>text<|/ref|><|det|>[[42, 858, 941, 952]]<|/det|>
+Structural analysis of TMEM16F in lipid nanodiscs supplemented with \(\mathrm{PIP}_2\) reveals 3 distinct coexisting states and a direct correlation between kinking of TM6 and membrane distortion (fig. S4). A continuous trail of lipids connects the intra- and extracellular sides of TMEM16F at the membrane distortion site (Fig. 1B). These findings are consistent with our previous structural and mutagenesis data (8) and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 44, 953, 271]]<|/det|>
+support a model for TMEM16F- mediated scrambling of lipids, whereby TMEM16F distorts the membrane, minimizing the distance between the inner and outer leaflets of the lipid bilayer (Fig. 5A). We further find that residues along this lipid trail, such as K370 and F321, are important for scramblase activity. In fact, K370 is a positively charged residue that is ideally positioned for interacting with negatively charged phospholipid heads at the membrane interface. The fact that TMEM16A, which cannot scramble lipids, contains an alanine in this position reinforces the notion that this basic residue is critical for lipid scrambling (Fig. 5B). Together, our data suggest that the lipid trail we identify on TMEM16F might correspond to the pathway for lipid scrambling. We propose that the lipids "surf" along this membrane- facing groove, crossing between the inner and outer leaflets through a path that does not directly involve the hydrophilic ion conduction pore (Fig. 5A).  
+
+<|ref|>text<|/ref|><|det|>[[41, 286, 955, 468]]<|/det|>
+TMEM16 proteins have emerged as important pharmacological targets for the treatment of cancer, asthma and more recently COVID- 19 (12, 13) (26) (clinicaltrials.gov). Our data indicate that nicoslamide and 1PBC bind the same, conserved site in both TMEM16A and TMEM16F (Fig. 5B and fig. S6). With both antagonists directly contacting the extracellular- proximal end of TM6, our mutagenesis studies suggest that residues on this helix are critical for drug binding (Figs. 3 and 4). TM6 is the main gating element of the channel as well as part of the ion conduction pore in both TMEM16A and TMEM16F (1, 2). We thus speculate that, by binding to the upper part of TM6, these antagonists simultaneously lock the ion conduction pore and the gating element in a closed configuration.  
+
+<|ref|>text<|/ref|><|det|>[[41, 483, 949, 733]]<|/det|>
+Interestingly, the binding site of 1PBC and nicoslamide coincides with the position where we observe the maximum degree of membrane distortion and thinning in the TMEM16F structures. In our model for TMEM16F activity, this site corresponds precisely with the entry and exit point of the lipids as they transition between the inner and outer leaflets of the plasma membrane (Fig. 5A). Our structures show that both antagonists replace the lipids found in this pocket in our drug- free sample (Fig. 2). This suggests that 1PBC and nicoslamide might directly inhibit TMEM16F scramblase activity by physically occluding the path of the lipids across the membrane (Fig. 5A). In fact, lipid densities along this path are significantly reduced in our 1PBC- and nicoslamide- bound structures (Fig. 2). Consistent with a critical role of this region for lipid scrambling, alanine substitutions of residues within the drug binding pocket significantly alter the lipid scrambling activity of TMEM16F in the absence of inhibitors, whereas equivalent mutations in TMEM16A do not affect \(\mathrm{Ca^{2 + }}\) - activated Cl- channel activity (Figs. 3 and 4).  
+
+<|ref|>text<|/ref|><|det|>[[41, 750, 951, 954]]<|/det|>
+Like 1PBC and nicoslamide, most drug molecules identified to date are not specific for any particular TMEM16 paralog and instead broadly target TMEM16 family members. This raises concerns about potential off target effects in the clinic. Additionally, our whole cell patch clamp electrophysiology experiments, as well as previous studies (36), show that nicoslamide may activate an ion channel. Although both nicoslamide and 1PBC appear to bind the same, conserved hydrophobic pocket in TMEM16A and TMEM16F, our data also show that the specific contribution of different residues to the interaction is distinct in TMEM16A and TMEM16F (Figs. 3 and 4). We further demonstrate that non- conserved residues within this region, such T606 and K370 in TMEM16F, are important for the inhibitory effects of 1PBC and/or nicoslamide (Fig. 3). In fact, comparison between the two pockets reveals that the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 44, 951, 248]]<|/det|>
+TMEM16A binding pocket consists almost exclusively of hydrophobic residues, whereas the equivalent site in TMEM16F includes several charged and OH- containing side chains (Fig. 5B). Nicolasamide is a highly hydrophobic molecule that presents extremely poor solubility in aqueous solutions (37). Thus, the identification of non- conserved hydrophilic residues within the drug binding pocket in TMEM16F opens the door for the development of nicosamide analogs with better pharmacological properties that exclusively target TMEM16F for the treatment of severe COVID- 19. Taken together, our work establishes a much- needed structural framework for designing more potent and more specific antagonists against individual members of the TMEM16 family with critical implications for the treatment of asthma, cancer and COVID- 19.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 270, 213, 296]]<|/det|>
+## Declarations  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 310, 331, 341]]<|/det|>
+## Acknowledgments  
+
+<|ref|>text<|/ref|><|det|>[[41, 355, 950, 512]]<|/det|>
+We thank D. Bulkley and Zanlin Yu at UCSF cryo- EM facility for help with grid screening. We also thank Z. Yu, Rui Yan and Shixin Yang at the HHMI Janelia Cryo- EM Facility for help with data acquisition and colleagues in our laboratories for discussion. This work was supported by grants from the NIH (1R35GM140847, S100D0020054, and S100D021741 to Y.C.; R01NS069229 and R35NS097229 to L.Y.J.). C.P. is supported by a Damon Runyon Postdoctoral Fellowship Award. T.W.H. is supported by the Jane Coffin Childs Memorial Fund for Medical Research. L.Y.J., Y.N.J., and Y.C. are Investigators with the Howard Hughes Medical Institute.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 535, 196, 560]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[55, 577, 950, 920]]<|/det|>
+1. M. E. Falzone, M. Malvezzi, B. C. Lee, A. Accardi, Known structures and unknown mechanisms of TMEM16 scramblases and channels. J Gen Physiol 150, 933-947 (2018).  
+2. V. Kalienkova, V. Clerico Mosina, C. Paulino, The Groovy TMEM16 Family: Molecular Mechanisms of Lipid Scrambling and Ion Conduction. J Mol Biol 433, 166941 (2021).  
+3. A. Caputo et al., TMEM16A, a membrane protein associated with calcium-dependent chloride channel activity. Science 322, 590-594 (2008).  
+4. Y. D. Yang et al., TMEM16A confers receptor-activated calcium-dependent chloride conductance. Nature 455, 1210-1215 (2008).  
+5. B. C. Schroeder, T. Cheng, Y. N. Jan, L. Y. Jan, Expression cloning of TMEM16A as a calcium-activated chloride channel subunit. Cell 134, 1019-1029 (2008).  
+6. H. Yang et al., TMEM16F forms a \(\mathrm{Ca^{2 + }}\) -activated cation channel required for lipid scrambling in platelets during blood coagulation. Cell 151, 111-122 (2012).  
+7. T. W. Han et al., Chemically induced vesiculation as a platform for studying TMEM16F activity. Proc Natl Acad Sci U S A 116, 1309-1318 (2019).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 45, 884, 92]]<|/det|>
+8. S. Feng et al., Cryo-EM studies of TMEM16F calcium-activated ion channel suggest features important for lipid scrambling. Cell Rep 28, 567-579 e564 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[55, 96, 896, 140]]<|/det|>
+9. W. Ye, T. W. Han, M. He, Y. N. Jan, L. Y. Jan, Dynamic change of electrostatic field in TMEM16F permeation pathway shifts its ion selectivity. Elife 8, (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 145, 940, 189]]<|/det|>
+10. C. Alvadia et al., Cryo-EM structures and functional characterization of the murine lipid scramblase TMEM16F. Elife 8, (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 194, 951, 238]]<|/det|>
+11. R. Watanabe, T. Sakuragi, H. Noji, S. Nagata, Single-molecule analysis of phospholipid scrambling by TMEM16F. Proc Natl Acad Sci U S A 115, 3066-3071 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 242, 920, 286]]<|/det|>
+12. W. Chen et al., The Prognostic Value and Mechanisms of TMEM16A in Human Cancer. Front Mol Biosci 8, 542156 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[50, 291, 951, 335]]<|/det|>
+13. L. Braga et al., Drugs that inhibit TMEM16 proteins block SARS-CoV-2 spike-induced syncytia. Nature 594, 88-93 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[50, 340, 920, 384]]<|/det|>
+14. H. L. Danahay et al., TMEM16A Potentiation: A Novel Therapeutic Approach for the Treatment of Cystic Fibrosis. Am J Respir Crit Care Med 201, 946-954 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[50, 389, 923, 433]]<|/det|>
+15. I. Cabrita, R. Benedetto, R. Schreiber, K. Kunzelmann, Niclosamide repurposed for the treatment of inflammatory airway disease. JCI Insight 4, (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 438, 904, 481]]<|/det|>
+16. K. Miner et al., Drug Repurposing: The Anthelmintics Niclosamide and Nitazoxanide Are Potent TMEM16A Antagonists That Fully Bronchodilate Airways. Front Pharmacol 10, 51 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 486, 904, 554]]<|/det|>
+17. T. Fujii, A. Sakata, S. Nishimura, K. Eto, S. Nagata, TMEM16F is required for phosphatidylserine exposure and microparticle release in activated mouse platelets. Proc Natl Acad Sci U S A 112, 12800-12805 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[50, 560, 936, 603]]<|/det|>
+18. P. Wolf, The nature and significance of platelet products in human plasma. Br J Haematol 13, 269-288 (1967).  
+
+<|ref|>text<|/ref|><|det|>[[50, 608, 911, 653]]<|/det|>
+19. R. F. Zwaal, P. Comfurius, E. M. Bevers, Scott syndrome, a bleeding disorder caused by defective scrambling of membrane phospholipids. Biochim Biophys Acta 1636, 119-128 (2004).  
+
+<|ref|>text<|/ref|><|det|>[[50, 657, 875, 701]]<|/det|>
+20. J. Suzuki, M. Umeda, P. J. Sims, S. Nagata, Calcium-dependent phospholipid scrambling by TMEM16F. Nature 468, 834-838 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[50, 706, 930, 750]]<|/det|>
+21. B. Gyorgy et al., Membrane vesicles, current state-of-the-art: emerging role of extracellular vesicles. Cell Mol Life Sci 68, 2667-2688 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[50, 756, 920, 800]]<|/det|>
+22. G. Raposo, W. Stoorvogel, Extracellular vesicles: exosomes, microvesicles, and friends. J Cell Biol 200, 373-383 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[50, 805, 953, 873]]<|/det|>
+23. P. J. Sims, T. Wiedmer, C. T. Esmon, H. J. Weiss, S. J. Shattil, Assembly of the platelet prothrombinase complex is linked to vesiculation of the platelet plasma membrane. Studies in Scott syndrome: an isolated defect in platelet procoagulant activity. J Biol Chem 264, 17049-17057 (1989).  
+
+<|ref|>text<|/ref|><|det|>[[50, 877, 936, 922]]<|/det|>
+24. J. M. Whitlock, H. C. Hartzell, Anoctamins/TMEM16 Proteins: Chloride Channels Flirting with Lipids and Extracellular Vesicles. Annu Rev Physiol 79, 119-143 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[45, 44, 945, 94]]<|/det|>
+25. N. Wu et al., Critical Role of Lipid Scramblase TMEM16F in Phosphatidylserine Exposure and Repair of Plasma Membrane after Pore Formation. Cell Rep 30, 1129-1140 e1125 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[47, 95, 936, 163]]<|/det|>
+26. A. S. Abdulamir et al., A randomised controlled trial of effectiveness and safety of Nicolasamide as add on therapy to the standard of care measures in COVID-19 management. Ann Med Surg (Lond) 69, 102779 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[47, 166, 940, 211]]<|/det|>
+27. S. Dang et al., Cryo-EM structures of the TMEM16A calcium-activated chloride channel. Nature 552, 426-429 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[47, 216, 940, 261]]<|/det|>
+28. C. Paulino, V. Kalienkova, A. K. M. Lam, Y. Neldner, R. Dutzler, Activation mechanism of the calcium-activated chloride channel TMEM16A revealed by cryo-EM. Nature 552, 421-425 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[47, 264, 894, 309]]<|/det|>
+29. J. Arreola, H. C. Hartzell, Wasted TMEM16A channels are rescued by phosphatidylinositol 4,5-bisphosphate. Cell Calcium 84, 102103 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[47, 313, 888, 358]]<|/det|>
+30. S. C. Le, Z. Jia, J. Chen, H. Yang, Molecular basis of PIP2-dependent regulation of the Ca(2+)-activated chloride channel TMEM16A. Nat Commun 10, 3769 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[47, 362, 950, 408]]<|/det|>
+31. W. Ye et al., Phosphatidylinositol-(4, 5)-bisphosphate regulates calcium gating of small-conductance cation channel TMEM16F. Proc Natl Acad Sci U S A 115, E1667-E1674 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[47, 411, 945, 457]]<|/det|>
+32. Y. Gao, E. Cao, D. Julius, Y. Cheng, TRPV1 structures in nanodiscs reveal mechanisms of ligand and lipid action. Nature 534, 347-351 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[47, 460, 953, 506]]<|/det|>
+33. M. A. Schuler, I. G. Denisov, S. G. Sligar, Nanodiscs as a new tool to examine lipid-protein interactions. Methods Mol Biol 974, 415-433 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[47, 510, 907, 555]]<|/det|>
+34. S. R. Bushell et al., The structural basis of lipid scrambling and inactivation in the endoplasmic reticulum scramblase TMEM16K. Nat Commun 10, 3956 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[47, 559, 950, 604]]<|/det|>
+35. S. C. Le, H. Yang, An Additional Ca(2+) Binding Site Allosterically Controls TMEM16A Activation. Cell Rep 33, 108570 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[47, 608, 905, 653]]<|/det|>
+36. B. Biton et al., The antipsychotic drug loxapine is an opener of the sodium-activated potassium channel slack (Slo2.2). J Pharmacol Exp Ther 340, 706-715 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[47, 657, 930, 702]]<|/det|>
+37. W. Chen, R. A. Mook, Jr., R. T. Premont, J. Wang, Niclosamide: Beyond an antihelmintic drug. Cell Signal 41, 89-96 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[47, 706, 950, 752]]<|/det|>
+38. D. N. Mastronarde, Automated electron microscope tomography using robust prediction of specimen movements. J Struct Biol 152, 36-51 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[47, 756, 936, 801]]<|/det|>
+39. Y. Z. Tan et al., Addressing preferred specimen orientation in single-particle cryo-EM through tilting. Nat Methods 14, 793-796 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[47, 805, 925, 851]]<|/det|>
+40. S. Q. Zheng et al., MotionCor2: anisotropic correction of beam-induced motion for improved cryo-electron microscopy. Nat Methods 14, 331-332 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[47, 855, 950, 901]]<|/det|>
+41. A. Punjani, J. L. Rubinstein, D. J. Fleet, M. A. Brubaker, cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat Methods 14, 290-296 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[47, 905, 904, 950]]<|/det|>
+42. P. Emsley, K. Cowtan, Coot: model-building tools for molecular graphics. Acta Crystallogr D Biol Crystallogr 60, 2126-2132 (2004).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[46, 44, 950, 285]]<|/det|>
+43. P. V. Afonine et al., Towards automated crystallographic structure refinement with phenix.refine. Acta Crystallogr D Biol Crystallogr 68, 352–367 (2012).44. R. A. Friesner et al., Glide: a new approach for rapid, accurate docking and scoring. 1. Method and assessment of docking accuracy. J Med Chem 47, 1739–1749 (2004).45. E. Harder et al., OPLS3: A Force Field Providing Broad Coverage of Drug-like Small Molecules and Proteins. J Chem Theory Comput 12, 281–296 (2016).46. J. C. Shelley et al., Epik: a software program for pK(a) prediction and protonation state generation for drug-like molecules. J Comput Aided Mol Des 21, 681–691 (2007).47. R. A. Friesner et al., Extra precision glide: docking and scoring incorporating a model of hydrophobic enclosure for protein-ligand complexes. J Med Chem 49, 6177–6196 (2006).  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 308, 144, 333]]<|/det|>
+## Figures
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[50, 50, 950, 682]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 720, 115, 739]]<|/det|>
+<center>Figure 1 </center>  
+
+<|ref|>text<|/ref|><|det|>[[40, 758, 949, 943]]<|/det|>
+Cryo- EM analysis reveals asymmetry within the TMEM16F dimer. (A) Cryo- EM density of the asymmetric state of the TMEM16F dimer with the monomers colored blue and light blue, respectively, and the lipid densities in grey. The gaussian filtered cryo- EM density (semitransparent) reveals distortion of the lipid nanodisc. (B) Side view of a TMEM16F monomer (blue) highlighting the trail of lipids (grey) covering the TM region. (C) Front and side view of the atomic model of the asymmetric state of TMEM16F with \(\mathrm{Ca^{2 + }}\) atoms and glycans shown in green and red, respectively. The ion conduction channel identified by HOLE is represented by spheres colored in rainbow scale based on the local width of the channel, where red \(< 1.5 \text{Å}\) and blue \(> 7.5 \text{Å}\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[55, 49, 953, 350]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 371, 118, 390]]<|/det|>
+<center>Figure 2 </center>  
+
+<|ref|>text<|/ref|><|det|>[[41, 411, 949, 550]]<|/det|>
+Nicosamide and 1PBC bind the same hydrophobic groove in TMEM16F. Atomic model of the TM1- TM6 region of (A) Class1 and (B) Class 2 of the drug- free control, (C) nicosamide- and (D) 1PBC- supplemented datasets. In each case, the additional cryo- EM densities found in the area are shown. Below, zoom into the TM1- TM6 groove with the residues shown as sticks and colored by heteroatom and the additional density found within the pocket shown in semitransparent. Structures of nicosamide and 1PBC as determined by computational docking using Glide are shown in purple and green, respectively.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[37, 40, 763, 787]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 800, 117, 820]]<|/det|>
+<center>Figure 3 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 841, 951, 958]]<|/det|>
+Functional validation of the drug binding site in TMEM16F. Representative curves of live imaging of TMEM16F- dependent PS exposure (A) and (C); and \(\mathrm{Ca^{2 + }}\) influx (E) and (G). Data are represented as mean \(\pm\) SEM. Scattered dot plots of time of onset of TMEM16F- dependent PS exposure [(B) and (D)] and \(\mathrm{Ca^{2 + }}\) influx [(F) and (H)]. Time of onset could not be determined for time courses with a linear rather than sigmoidal rise. The mean \(\pm\) SEM is shown along with the statistical significance determined by unpaired
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 46, 904, 90]]<|/det|>
+t- test for each mutant as compared to vehicle controls ( \(^{*}p < 0.05\) ; \(^{**}p < 0.01\) ; \(^{***}p < 0.0001\) ; \(^{****}p < 0.0001\) ).  
+
+<|ref|>image<|/ref|><|det|>[[39, 106, 960, 768]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 781, 118, 800]]<|/det|>
+<center>Figure 4 </center>  
+
+<|ref|>text<|/ref|><|det|>[[40, 821, 958, 939]]<|/det|>
+Electrophysiology- based validation of the binding site in TMEM16A. Representative voltage clamp current traces at \(+70 \text{mV}\) with holding potential at - 5 mV of wildtype mTMEM16A and alanine substitution mutants in the absence or presence of (A) 3 μM nicosamide or (B) 30 μM 1PBC recorded in 500 nM \([Ca^{2 + }]_i\) or 12 mM \([Ca^{2 + }]_i\) , respectively. Graphs showing the coefficient of the steady- state current measured for the wildtype control and each mutant upon addition of (C) nicosamide and (D) 1PBC divided by the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 892, 117]]<|/det|>
+maximum Intensity \((1 / 1_{\mathrm{max}})\) in each case. The mean \(\pm\) SEM is shown along with the statistical significance determined by unpaired t- test for each mutant as compared to wildtype controls ( \(*p<\) 0.05; \(**p< 0.01\) ; \(***p< 0.001\) ).  
+
+<|ref|>image<|/ref|><|det|>[[44, 144, 955, 440]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 459, 118, 479]]<|/det|>
+<center>Figure 5 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 500, 950, 660]]<|/det|>
+Comparison of the drug binding pocket in TMEM16A and TMEM16F. (A) Schematic representation of the TMEM16F dimer (light blue and blue) embedded in a lipid bilayer (grey), where \(\mathrm{Ca^{2 + }}\) atoms are shown as green circles and the inhibitors as a purple polygon and dotted black lines represent the closed ion conduction pore. (B) Structure of the drug binding pocket in TMEM16F (blue, left) and TMEM16A (grey, right) with the side chains of the surrounding residues shown as sticks and the non- conserved residues highlighted in orange. Computationally docked structures of nicosamide and 1PBC are shown in purple and green, respectively. All atoms are colored by heteroatom.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 683, 311, 710]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 733, 765, 753]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[61, 772, 345, 790]]<|/det|>
+Supplementary materials.docx
+
+<--- Page Split --->

preprint/preprint__b38f37bbaa6ba9c98888df06d22bb37e705754f2b008a22e0b0ee6bc28d62cae/images_list.json

@@ -0,0 +1,10 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "E12.5",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 30
+  }
+]

preprint/preprint__b38f37bbaa6ba9c98888df06d22bb37e705754f2b008a22e0b0ee6bc28d62cae/preprint__b38f37bbaa6ba9c98888df06d22bb37e705754f2b008a22e0b0ee6bc28d62cae.mmd

@@ -0,0 +1,445 @@
+
+# Intercellular exchange of Wnt ligands reduces cell population heterogeneity in embryogenesis  
+
+Yudai Hatakeyama National Institute for Basic Biology  
+
+Nen Saito Exploratory Research Center on Life and Living Systems https://orcid.org/0000- 0002- 8317- 9389  
+
+Yusuke Mii National Institute for Basic Biology and Okazaki Institute for Integrative Bioscience https://orcid.org/0000- 0002- 1907- 5665  
+
+Takuma Shinozuka National Institute for Basic Biology  
+
+Tatsuya Takemoto Tokushima University https://orcid.org/0000- 0003- 1860- 0269  
+
+Honda Naoki Hiroshima University Shinji Takada ( \(\boxed{\bullet}\) stakada@nibb.ac.jp) Exploratory Research Center on Life and Living Systems https://orcid.org/0000- 0003- 4125- 6056  
+
+## Article  
+
+Keywords: Wnt, epiblast, tailbud, NMP, paracrine, community effect, retinoic acid  
+
+Posted Date: February 2nd, 2022  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 1271602/v1  
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+<--- Page Split --->
+
+# Intercellular exchange of Wnt ligands reduces cell population heterogeneity in embryogenesis  
+
+Yudai Hatakeyama<sup>1,2,3</sup>, Nen Saito<sup>1,2,3</sup>, Yusuke Mii<sup>1,2,3,4</sup>, Takuma Shinozuka<sup>1,2,3,4</sup>, Tatsuya Takemoto<sup>5</sup>, Honda Naoki<sup>1,6</sup>, & Shinji Takada<sup>1,2,3,</sup>  
+
+1 National Institute for Basic Biology and Exploratory Research Center on Life and Living Systems (ExCELLS), National Institutes of Natural Sciences, 5- 1 Higashiyama, Myodaiji- cho, Okazaki, Aichi 444- 8787, Japan2 National Institute for Basic Biology, National Institutes of Natural Sciences, 5- 1 Higashiyama, Myodaiji- cho, Okazaki, Aichi 444- 8787, Japan3 The Graduate University for Advanced Studies (SOKENDAI), 5- 1 Higashiyama, Myodaiji- cho, Okazaki, Aichi 444- 8787, Japan4 PREST, Japan Science and Technology Agency (JST), Kawaguchi, Saitama, 332- 0012, Japan5 Institute of Advanced Medical Sciences, Tokushima University, 3- 18- 5 Kuramoto- cho, Tokushima, 770- 8503, Japan6 Graduate School of Integrated Sciences for Life, Hiroshima University, 1- 3- 2 Kagamiyama, Higashi- hiroshima, Hiroshima, 739- 8511, Japan  
+
+# Present address: Nara Institute of Science and Technology, 8916- 5 Takayama- cho, Ikoma, Nara 630- 0912, Japan  
+
+\*To whom correspondence should be addressed. Shinji Takada Ph. D.; e- mail: stakada@nibb.ac.jp  
+
+Running title: Compensation of heterogeneity by paracrine Wnt signal  
+
+Key words: Wnt, epiblast, tailbud, NMP, paracrine, community effect, retinoic acid
+
+<--- Page Split --->
+
+## Abstract  
+
+Wnt signaling is required to maintain bipotent progenitors for neural and paraxial mesoderm cells, the neuromesodermal progenitor (NMP) cells that reside in the epiblast and tailbud. Since epiblast/ tailbud cells receive Wnt ligands produced by one another, this exchange may average out the heterogeneity of Wnt signaling levels among these cells. Here, we examined this possibility by replacing endogenous Wnt3a with a receptor- fused form that can activate signaling in producing cells, but not in neighboring cells. Mutant mouse embryos showed a unique phenotype in which maintenance of many NMP cells was impaired, although some cells persisted for long periods. The epiblast cell population of these embryos increased heterogeneity in Wnt signaling levels as embryogenesis progressed and were sensitive to retinoic acid, an endogenous antagonist of NMP maintenance. Thus, mutual intercellular exchange of Wnt ligands in the epiblast cell population reduces heterogeneity and achieves robustness to environmental stress.  
+
+## Main  
+
+The number of stem and progenitor cells is tightly controlled during embryogenesis and homeostasis. As the developmental context or external environment surrounding these cells changes, stem and progenitor cell populations respond to these changes, sometimes robustly, sometimes flexibly, thereby keeping these cells under control. In many cases, secreted signal proteins control maintenance and differentiation of stem/progenitor cells. However, mechanisms by which such signal proteins contribute to the robustness of these cell populations remain to be determined.  
+
+The body axis of vertebrate embryos elongates in an anterior- to- posterior fashion. During this elongation process, cells that constitute tissues in the trunk and tail are continuously generated from progenitor cells 1. These progenitor cells are found in an area at the posterior end of embryos, termed the epiblast in early embryonic stages and the tail bud in later stages. Clonal lineage analysis revealed that both neural and paraxial mesodermal cell types are commonly generated from the same progenitor cells throughout the period of axis elongation 2. These bipotent progenitor cells are called "Neuromesodermal Progenitor" (NMP) cells. NMP cells appear just before the onset of somitogenesis and are maintained until the conclusion of axis elongation. In mouse embryos, NMPs are located in the caudal lateral epiblast (CLE) posterior to the node- streak border of the primitive streak region and the chordoneural hinge (CNH) of the tail bud 2- 7. Population and clonal analyses indicate that these cells behave like stem cells 3,4.
+
+<--- Page Split --->
+
+Cell signaling molecules and transcription factors are implicated in regulation of axis elongation, probably by maintaining NMPs. For instance, in the mouse, at least three Wnt ligands are sequentially expressed in the epiblast and tailbud. Wnt3 expression is first activated in the posterior epiblast at E5.5, followed by Wnt8a and Wnt3a expression \(^{8 - 11}\) , While Wnt3 and Wnt8a expression cease by early somite stage, Wnt3a expression continues until E12.5, when tail elongation is almost completed \(^{12}\) . Along with expression of these Wnt ligands, a T- box transcription factor, T/Brachyury (Bra), is continuously expressed in the same region from the onset to the end of Wnt3 expression \(^{13}\) . Evidence suggests that Wnt signaling and Bra are important for maintenance of NMPs. Genetic studies of null mutant embryos of Wnt3a and Bra, showed their importance for axis elongation \(^{8,14}\) , and lineage tracing of cells that express Bra revealed that both neural and paraxial mesoderm cells are derived from Bra- expressing cells \(^{15 - 18}\) . In addition to NMP maintenance, Wnt signaling and Bra are involved in fate determination between the neural and paraxial- mesodermal lineages \(^{14,19}\) . Of note, Wnt signaling directly activates Bra expression through Tcf transcription factor, while Bra is required for Wnt3a expression \(^{14,20 - 22}\) . Thus, Wnt and Bra form a positive feedback loop in which each actively regulates expression of the other in NMP maintenance. Similarly, positive feedback between Wnt8 and tbxta (ntl), a zebrafish ortholog of Bra, has been reported in zebrafish \(^{23,24}\) .  
+
+During maintenance of NMPs, activation of Wnt signaling and expression of Bra overlap widely in the epiblast and the tail bud, including the area where NMPs exists. Thus, some Wnt ligands may act in an autocrine manner in the epiblast and the tail bud, resulting in self- activation of a Wnt/Bra regulatory loop in each cell. On the other hand, given that cells adjoining NMP cells also express Wnt ligands and Bra, paracrine Wnt ligands supplied by neighboring cells may also be involved in NMP maintenance. To examine the importance of Wnt paracrine function in maintenance of NMPs, we generated knock- in mouse embryos in which endogenous Wnt3a is replaced with a receptor- fused form that lacks paracrine activity, but maintains autocrine activity. Exacting analysis of Wnt paracrine- deficient embryos revealed the significance of the paracrine signal for maintenance of the NMP cell population and resilience to stress from external tissue. To the best of our knowledge, this is the first direct experimental evidence of the significance of intercellular exchange of secreted signal proteins in the emergence of cell population features.
+
+<--- Page Split --->
+
+## Results  
+
+## Wnt3a fused with Frizzled possesses signaling activity, but no paracrine activity  
+
+Wnt3a fused with Frizzled possesses signaling activity, but no paracrine activityTo eliminate paracrine activity of WNT3A, we fused mouse WNT3A to the N- terminus of human FRIZZLED5 via 2 MYC tags (WNT3A- FZD5; Fig. 1a). Activity of WNT3A- FZD5 was examined in cells stably expressing TOP- FLASH reporter (STF293 cells) in comparison with authentic WNT3A, as well as GFP- fused WNT3A (GFP- WNT3A; Fig. 1a). Activity of GFP- WNT3A was lower than that of authentic WNT3A (Fig. 1b), but sufficient to replace endogenous Wnt3a in vivo \(^{25}\) . WNT3A- FZD5 activated canonical Wnt signaling to almost same extent as authentic WNT3A and more strongly than GFP- WNT3A, 48 h after transfection, and this activity was nearly saturated even after longer incubation (Fig. 1b). In contrast, whereas Wnt activity was activated in STF293 cells co- cultured with cells expressing intact WNT3A or GFP- WNT3A, almost no activation was detected in co- culture with WNT3A- FZD5- expressing cells (Fig. 1c). Consistent with this result, WNT3A- FZD5 was not detected in culture supernatant (Extended Data Fig. 1). These results show that as expected, WNT3A- FZD5 possesses sufficient signaling activity, but almost no paracrine activity.  
+
+Axis elongation is impaired, but partially maintained in Wnt3a- Fzd5 homozygous embryosWe next generated mouse embryos in which endogenous Wnt3a is substituted for Wnt3a- Fzd5, using a CRISPR/Cas9- mediated knock- in approach (Extended Data Fig. 2a- f). Mice heterozygous for Wnt3a- Fzd5 were morphologically normal and fertile (Fig. 1d- j, Extended Data Fig. 2g- k). As expected, Wnt3a- Fzd5 exhibited an expression pattern identical to that of endogenous Wnt3a in these embryos (Fig. 1d- g, Extended Data Fig. 2g- j). In addition, Western blotting analysis revealed that Wnt3a- Fzd5 heterozygous and homozygous embryos expressed WNT3A- FZD5 proteins at the expense of authentic Wnt3a in the posterior region (Fig. 1h). Thus, Wnt3a- Fzd5 properly replaced endogenous Wnt3a, being expressed in the same spatial pattern as endogenous Wnt3a.  
+
+While Wnt3a- Fzd5 heterozygotes (Wnt3a \(^{+/Fzd5}\) ) showed no obvious embryonic abnormality (Fig. 1i, j), adult Wnt3a- Fzd5 homozygotes (Wnt3a \(^{Fzd5 / Fzd5}\) ), were embryonically lethal and die after E12.5 (Extended Data Fig. 2k and data not shown). However, the phenotype of Wnt3a- Fzd5 homozygotes was milder than that of Wnt3a null mutant embryos, which die around E9.5 with posterior truncation \(^{8}\) . Thus, even though WNT3A- FZD5 has sufficient signaling activity in vitro (Fig. 1b, c), Wnt3a- Fzd5 partially, but not completely, substitutes for endogenous Wnt3a.  
+
+To better understand this puzzling phenotype of Wnt3a- Fzd5 homozygous embryos, we examined their morphology. While the gross morphology of Wnt3a- Fzd5 homozygous embryos
+
+<--- Page Split --->
+
+appeared normal in the anterior trunk, it was highly disorganized posterior to the hindlimbs (Fig. 1i- k). This posterior defect became evident no later than E8.75 (Fig. 4a, b and 5k, l). Transverse images of E11.5 embryos stained with DAPI showed that neural tube morphology was gradually disturbed along the anterior- posterior axis in these embryos (Fig. 11- n). This disruption was evident in the intermediate region between fore- and hindlimbs (Fig. 1n') and pronounced in the more posterior region, resulting in an opened neural tube at the hindlimb level (Fig. 1n'). However, in spite of this severe defect in posterior morphogenesis, a thin, kinked tail- like structure was found at the posterior end of these embryos (Fig. 1k: red arrowhead).  
+
+Whole mount in situ hybridization analyses also revealed that the neural tube, marked by Sox2 expression, was abnormally opened posterior to the hindlimb at E10.5 in Wnt3a- Fzd5 homozygous embryos (Fig. 2r, s). Somites, stained with the Uncx 4.1 probe, were normally formed in the anterior trunk, but their size is reduced posterior to the hindlimb (Fig. 2m, n). Of note, Brachyury (Bra), which is expressed in the tailbud and notochord of normal embryos, was expressed at the tip of the thin and kinked tail, although the number of Bra- positive cells was decreased (Fig. 2a, b). In addition, Tbx6, expression of which is turned on immediately after specification to the paraxial mesoderm lineage, was also expressed at this posterior end (Fig. 2g, h). Notably, expression of Bra and Tbx6 at the posterior tip was maintained even at E12.5, when tail elongation is nearly arrested in normal embryos (Fig. 2w- ab) Thus, in Wnt3a- Fzd5 homozygous embryos, trunk morphogenesis was disrupted at the hindlimb level, accompanied by reduction of tailbud size, but differentiation from the tailbud appears to be maintained throughout the period of axis elongation.  
+
+## Wnt-positive progenitor cells are responsible for abnormal neural and somite development in Wnt3a-Fzd5 homozygous embryos  
+
+To examine the impact of tailbud reduction in posterior morphogenesis of Wnt3a- Fzd5 homozygous embryos, we followed cells that had activated Wnt signaling, because Wnt signaling is activated in progenitor cells in the epiblast and tailbud region. To this end, Axin2- creERT2 and floxed tdTomato alleles were introduced into Wnt3a- Fzd5 homozygous embryos. Tamoxifen was injected into pregnant female mice at 7.5 or 8.5 days post coitus (dpc) and embryos were fixed at E10.5 (Fig. 3a and Extended Data Fig. 3a). In control, wild- type, and Wnt3a- Fzd5 heterozygous, embryos, labelled cells were detected in most tissues at the hindlimb level, regardless of the timing of tamoxifen administration (Extended Data Fig. 3b- e). However, when tamoxifen was injected at 8.5 dpc, but not at E7.5 dpc, labelled cells were rarely detected in the ventral neural tube at the hindlimb level (Extended Data Fig. 3d, e), showing that the
+
+<--- Page Split --->
+
+origin of ventral neural cells loses Wnt signaling after E7.5. As in control littermates, in Wnt3a- Fzd5 homozygous embryos labelled at 8.5 dpc, labelled cells were similarly distributed in most tissues, except neural tube (Fig.s 3b and c). Labelling efficiency monitored in somites (Fig. 3d) and the nephric duct (Fig. 3e) was not significantly changed between littermates. However, the number of labelled cells was specifically reduced in somites (Fig. 3f) and neural tube (Fig. 3g) in Wnt3a- Fzd5 homozygous embryos. Thus, the number of cells derived from Wnt-positive progenitors at E8.5 was decreased in dorsal neural tube and somites of Wnt3a- Fzd5 homozygous embryos. It is plausible that the decrease of dorsal neural cells results in the opened neural tube in Wnt3a- Fzd5 homozygous embryos.  
+
+As described above, Wnt3a is expressed in the roof plate of the neural tube, in addition to the epiblast and tailbud (Fig.s 1d- g). Thus, it also seems probable that Wnt3a- Fzd5 expression in the roof plate region causes the morphological abnormality in Wnt3a- Fzd5 homozygous embryos. To test this possibility, we examined the contribution of Bra to this phenotype, because Bra interacts specifically with Wnt3a in development of the epiblast/ tailbud, but not the roof plate. While Wnt3a- Fz5 heterozygotes (Extended Data Fig. 4a) and Bra single heterozygotes (Extended Data Fig. 4b) appeared normal, Wnt3a- Fz5 and Bra compound heterozygous embryos (Wnt3a \(^{+/Fz5}\) ; Bra \(^{+/}\) ) had open neural tubes and bent tails.(Extended Data Fig. 4c). Thus, interaction of Wnt3a- Fz5 and Bra in the epiblast and tailbud region is responsible for the phenotype of Wnt3a- Fzd5 homozygous embryos.  
+
+## Evidence indicates that the phenotype of Wnt3a-Fzd5 homozygotes is due to the lack of paracrine activity  
+
+Reduction of the Wnt3a signal impairs maintenance of the tailbud, including NMP cells \(^{8,16}\) . This defect results in truncation of A- P elongation in a manner dependent on Wnt3a activity \(^{26}\) . Since Wnt3a \(^{wt / - }\) (vt/- ) embryos, which possess one copy of a hypomorphic (vt) allele of Wnt3a with reduced Wnt3a expression in the tailbud (Extended Data Fig. 5a- f: Greco et al., 1996) impairs trunk development at the same level as in Wnt3a- Fzd5 homozygous embryos (Fig. 2c, d, i, j, o, p, t, and u), we compared the morphology of vt/- embryos with Wnt3a- Fzd5 homozygous embryos. In contrast to Wnt3a- Fzd5 homozygotes, vt/- embryos did not exhibit thin, kinked tails and open neural tubes, but they failed to maintain the tailbud, marked by Bra and Tbx6 expression, at E10.5 (Fig. 2d, j) and E12.5 (Extended Data Fig. 5g, h). Therefore, the phenotype of Wnt3a- Fzd5 homozygotes is unique, compared with other Wnt3a hypomorphic mutants and is not simply due to decreased Wnt3a activity.
+
+<--- Page Split --->
+
+This characteristic phenotype of Wnt3a- Fzd5 homozygous embryos was also observed in Wnt3aFzd5/ embryos (Extended Data Fig. 6a- c, e- g). Because Wnt3a- Fzd5 heterozygous embryos (Wnt3a+/Fzd5) appeared normal, as previously described (Fig. 1j, Extended Data Fig. 6a, e), Wnt3a- Fzd5 seems to cause this phenotype in the absence of wild- type Wnt3a. Furthermore, this phenotype can be rescued depending on the expression level of Wnt3a, because the phenotype of Wnt3aFzd5/ embryos was partially rescued by replacing the null allele to vt (Wnt3aFzd5/vt; Extended Data Fig. 6d, h). Based on the results of these analyses using various Wnt3a mutants, the characteristic phenotype of Wnt3a- Fzd5 homozygotes is due to some property lost in Wnt3a- Fzd5, most likely paracrine activity.  
+
+## NMP cells are reduced, but maintained in Wnt3a-Fzd5 homozygous embryos  
+
+In spite of improper development in the posterior neural tube and somites, our analyses with molecular markers revealed that differentiation of paraxial mesoderm and neural cells was partially maintained in Wnt3a- Fzd5 homozygous embryos (Fig. 2b, h, n, s, x, aa). Furthermore, the tailbud marked by expression of Bra, was maintained at the posterior tip of the tail of these embryos(Fig. 2a, b, x, y, aa, ab). These data suggest that a small number of NMP cells persist in Wnt3a- Fzd5 homozygous embryos.  
+
+Since one of the characteristics of NMP cells is expression of Bra and Sox2<sup>27</sup>, we compared the number of Bra and Sox2 double- positive cells using immunohistochemistry in Wnt3a- Fzd5 homozygous embryos and control littermates (Fig. 4a- x and Extended Data Fig. 7). The number of Bra and Sox2 double- positive cells started to diminish at E8.75 in Wnt3a- Fzd5 homozygous embryos (Fig. 4a- p), but a small number of double- positive cells were still maintained at E11.5 (Fig. 4q, r, u, v). In contrast, double- positive cells disappeared in vt/- embryos at E11.5 (Fig. 4s, w, t, x). These results support the idea that a small number of NMP cells are specifically maintained in Wnt3a- Fzd5 homozygous embryos, even after trunk development is impaired.  
+
+## Wnt signaling activity persists in a small number of epiblast cells in Wnt3a-Fzd5 homozygous embryos  
+
+To further investigate the defect of Wnt3a- Fzd5 homozygous embryos, we directly examined Wnt signaling activity in individual cells in the epiblast. To this end, we utilized the R26 WntVis reporter, expression of which is driven by heptameric TCF/LEF1 binding sequences combined with a viral minimal promoter in the Rosa26 locus<sup>28</sup>. This reporter responds in a graded fashion to a wide range of Wnt signal strengths. In addition, the histone H2B- EGFP protein, used as a fluorescent reporter, facilitates single- cell resolution analysis under confocal microscopy. In this study, fluorescence of this reporter was measured in individual cells in a
+
+<--- Page Split --->
+
+photon- counting mode. Epiblast cells in the areas lateral to the node were individually analyzed in a single confocal plane.  
+
+During development of mouse epiblast, three Wnt ligands, Wnt3, Wnt8, and Wnt3a, sequentially activate Wnt signaling \(^{8,10,13,29}\) . Because Wnt3 and Wnt8 are expressed prior to Wnt3a, Wnt activity was detected even in Wnt3a null mutant embryos at early headfold (EHF) stage (E7.0; Takemoto et al., 2016). At this stage, no obvious change in Wnt activity was detected in Wnt3a- Fzd5 homozygous embryos, as predicted (Fig. 5a- c). Then, Wnt3a expression was activated, and Wnt signaling level subsequently increased in both control and Wnt3a- Fzd5 homozygous embryos (Fig. 5d, e and Extended Data Fig. 8a), but not in Wnt3a null embryos (Extended Data Fig. 8b), at the late headfold (LHF) stage (E7.5). Thus, Wnt signaling was properly activated in the initial phase of Wnt3a- dependent activation, even in Wnt3a- Fzd5 homozygous embryos. Of note, in these stages, the level of the fluorescent reporter differed among epiblast cells in both control and Wnt3a- Fzd5 homozygous embryos.  
+
+From early somite stage (E8.0), Wnt signaling began to be perturbed in Wnt3a- Fzd5 homozygous embryos. At E8.0, around 2- 3- somite stage, the number of Wnt signaling- positive cells was reduced in the anterior and lateral epiblast regions of Wnt3a- Fzd5 homozygous embryos (Fig. 5f, g, h, o). The reduction in Wnt signaling- positive cells was pronounced in most of the epiblast region of Wnt3a- Fzd5 homozygous embryos at E8.75, where Wnt- positive and negative cells were distributed in a patch- work pattern (Fig. 5f, k, l, p). This reduction was enhanced by E9.5, but some Wnt- positive cells remained at the posterior end of Wnt3a- Fzd5 homozygous embryos (Fig. 5q, r). These posterior Wnt- positive cells were further maintained until E11.5 (data not shown). On the other hand, in \(vt\) - embryos, the decrease of Wnt signaling started at E8.75 (Fig. 5i, j, m, n). In contrast to Wnt3a- Fzd5 homozygous embryos, Wnt signaling was almost abolished around E9.5 in epiblast (Fig. 5s, t). Furthermore, the activity level appeared to decrease gradually in most Wnt- positive epiblast cells and the deviation of Wnt- activity in epiblast population was smaller than in Wnt3a- Fzd5 homozygous embryos.  
+
+Taken together, in Wnt3a- Fzd5 homozygous embryos, reduction of Wnt activity occurred from early somite stage, but a small number of Wnt- positive cells remain longer. Notably, Wnt activity appeared to fluctuate between adjacent cells even in control embryos, but in Wnt3a- Fzd5 homozygous embryos this heterogeneity was enhanced (Fig. 5g- p). Probably, the accelerated reduction of Wnt activity in many epiblast cells reduces the number of NMP cells, as well as of neural and somite cells produced by NMP cells in Wnt3a- Fzd5 homozygous embryos. In contrast, persistent activation of Wnt signaling in the other epiblast cells probably
+
+<--- Page Split --->
+
+contributes to maintenance of a small number of NMP cells, resulting in formation of thin, kinked tails.  
+
+## Retinoic acid enhances the phenotype of Wnt3a-Fzd5 homozygous embryos  
+
+Since impairment of Wnt activation was observed from early somite stage, it seems probable that somite formation affects the Wnt3a- Fzd5- specific reduction of Wnt- active cells. Interestingly, retinoic acid (RA), which is synthesized in somite cells, antagonizes the function of Bra in zebrafish embryos \(^{24}\) . Consistently, mouse embryos with mutated cyp26a, which encodes an enzyme to degrade RA, exhibit an axis truncation phenotype, similar to Wnt3a and Bra mutant embryos \(^{30}\) . Thus, we hypothesized that epiblast cells of Wnt3a- Fzd5 homozygous embryos are more sensitive to RA in maintenance of Wnt activity.  
+
+To address this issue, female Wnt3a- Fzd5 heterozygous mice intercrossed with Wnt3a- Fzd5 heterozygous males were treated with RA 7.5 days post coitum and effects of RA on the phenotype of mutant embryos were examined. RA treatment specifically enhanced the abnormality in gross morphology of Wnt3a- Fzd5 homozygous embryos (Fig. 6a- h). Furthermore, Wnt reporter analysis revealed that RA treatment enhanced the specifically reduced pattern of Wnt activity in Wnt3a- Fzd5 homozygous embryos, showing enhancement of a patch- work pattern, irrespective of cell position along the anterior- posterior axis (Fig. 6i- m). This result suggests that the epiblast cell population of Wnt3a- Fzd5 homozygous embryos is specifically susceptible to RA.  
+
+## Mathematical modeling supports the importance of paracrine function in maintaining Wnt-positive epiblast cell populations  
+
+The results described above strongly suggest that lack of paracrine signaling enhances heterogeneity of Wnt activity in the epiblast cell population and that a cell population with such enhanced heterogeneity is more sensitive to antagonists, like RA. Thus, we also tested the validity of these ideas by creating a mathematical model (Fig. 7a- g, Extended Data Fig. 9 and Movie1). In this model, spatiotemporal changes in Wnt activity were compared in a hypothetical epiblast plane with and without intercellular exchange of Wnt ligands. The temporal increase or decrease of Wnt activity in each cell is defined by the production rate regulated by autocatalysis, which represents a positive feedback loop of Wnt3a/Bra, in addition to the basic rate of production and degradation of Wnt ligands. The stochastic increase/decrease in Wnt activity is also incorporated as a noise term. In this virtual plane, we assume that each cell divides stochastically and that a newly produced daughter cell locates laterally or anteriorly to the original cell. It is also assumed that an RA gradient from anterior to posterior is imposed
+
+<--- Page Split --->
+
+at a specific time, which represents anteroposterior diffusion of RA. However, in this virtual space, cells that are aligned along the left- right axis were treated as if there is no difference in their distance from the RA source (Fig. 7a).  
+
+By using adjusted parameters, we first simulated spatiotemporal patterns of Wnt activity in the hypothetical epiblast plane under conditions in which intercellular exchange of Wnt ligands is present (Fig. 7b, d, e) and absent (Fingers 7c, f). Wnt activity levels decreased after the addition of RA (t>0 shown in Fig. 7g, Extended Data Fig. 9a, and movie Extended Data Movie1) and the number of Wnt- low cells increased in the absence of intercellular Wnt exchange. However, a small number of Wnt- high cells remained for a while (until t=3 shown in Fig. 7c, g, Extended Data Movie 1B). On the other hand, if the production rate was reduced, mimicking the situation of vt/- embryos, Wnt- low cells gradually increased and few Wnt- high cell remained (Fig. 7D, G and Extended Data Movie 1C). These simulations showed that a lack of intercellular Wnt exchange reproduced the spatio- temporal pattern of Wnt activity observed in Wnt3a- Fzd5 homozygous embryos. Furthermore, we reproduced the sensitivity of Wnt3a- Fzd5 homozygote cells when the RA concentration was uniformly increased in the hypothetical epiblast plane (Fig. 7e, f, g, and Extended Data Movies 1d, e). Taken together, these simulations based on our mathematical model support the idea that the Wnt paracrine signal reduces heterogeneity in Wnt activity in the epiblast cell population and increases robustness to RA.  
+
+## Discussion  
+
+It is widely believed that secreted signal proteins act on cells in the vicinity of the source cells, and in some cases, more distally \(^{31,32}\) . In contrast, in the epiblast and tailbud, most cells both produce and receive Wnt ligands \(^{16,33 - 35}\) . As a result, Wnt ligands secreted from each cell into the extracellular space activate the intracellular Wnt signaling pathway in cells in the population. As a result, Wnt3a ligands secreted extracellularly from each cell activate the intracellular Wnt signaling pathway in cells within the population. Thus, in contrast to unidirectional transfer from Wnt- producing cells to receiving cells, Wnt ligands seem to be reciprocally exchanged between epiblast and tailbud cells.  
+
+To understand the biological significance of reciprocal ligand exchange within a cell population, we generated Wnt3a- Fzd5 homozygous embryos, in which Wnt3a- mediated intercellular communication, or paracrine function, is specifically impaired. In these embryos, the number of Wnt- positive cells decreases rapidly from the anterior and lateral sides of the epiblast after RA begins to be synthesized in the somite, but a small number of Wnt- positive cells, including
+
+<--- Page Split --->
+
+NMP cells, remain at the posterior end for a long time. Precise examination of Wnt3a- Fzd5 homozygous embryos and mathematical simulation support a model in which Wnt3a- mediated intercellular communication is required for maintenance of the NMP population (Figure 7h).  
+
+In epiblast and tailbud regions, including NMPs, Wnt3a- expressing cells also express Bra \(^{14}\) . Bra is a direct transcriptional target of the Wnt signaling pathway, whereas Wnt3a expression is also dependent on Bra \(^{14,18}\) . Thus, Wnt3a and Bra mutually activate one another, forming a positive feedback regulatory loop. Because positive feedback amplifies small changes, this regulatory system can rapidly increase or decrease the amount of Wnt3a and Bra in a cell \(^{36}\) .  
+
+It seems plausible that the epiblast/tailbud cells vary with respect to Wnt production and degradation rates, efficiency of feedback amplification, and/or resistance to environmental factors that reduce Wnt activity. Thus, cells that are prone to losing Wnt activity rapidly lose this activity due to the positive feedback, resulting in an increased disparity generated by the fluctuations. In the epiblast of Wnt3a- Fzd5 homozygotes, the number of cells with little or no Wnt activity rapidly increases from E8.0. In contrast, a small, but significant number of cells maintain high Wnt activity for a long time in these embryos. A probable reason for persistence of Wnt- high cells is that the change of Wnt activity in these cells is below the threshold to trigger a rapid decrease by positive feedback. Actually, our mathematical model, which assumes fluctuation in Wnt activity and positive feedback regulation, produces similar spatial patterns of Wnt activity in Wnt3a- Fzd5 homozygous cell populations under conditions without Wnt- mediated intercellular communication.  
+
+In contrast, in control embryos, including Wnt3a- Fzd5 heterozygotes, the number of Wnt3a positive cells slowly decreased and disappeared around E13.5, when axis elongation was terminated. Probably, in these embryos, Wnt ligands supplied by neighboring Wnt- high cells compensate to some extent for the decrease of Wnt activity in Wnt- low cells. Thus, the exchange of Wnt3a ligands appears to compensate for the rapid decrease in Wnt activity in individual cells. Taken together, positive feedback regulation can amplify heterogeneity among members of the cell population, but our results suggest that sharing of intercellular components of the feedback loop, such as Wnt ligands, inhibits amplification of this heterogeneity.  
+
+In the epiblast of Wnt3a- Fzd5 homozygotes, Wnt activity rapidly decreases in many cells at E8.0, when several anterior somites are generated. In these embryos, the decrease in Wnt activity was more pronounced anteriorly and laterally in the epiblast. Of note, in early mouse embryos, RA synthesis requires retinaldehyde dehydrogenase 2 (RALDH2/ALDH1a2), which is
+
+<--- Page Split --->
+
+activated in somites and lateral plate mesoderm \(^{37,38}\) . A line of evidence has shown that RA signaling antagonizes Wnt/Bra activity gradually from the anterior side of the epiblast and tailbud region in the mouse embryo \(^{24,30,37,39,40}\) . Thus, we speculated that the epiblast cell population in Wnt3a- Fzd5 homozygotes is sensitive to RA stress originating from tissues developed anteriorly and laterally to the epiblast.  
+
+Actually, RA treatment reduces Wnt signaling in epiblast cells specifically in Wnt3a- Fzd5 homozygous embryos, indicating that the epiblast cell population in these embryos is more susceptible to RA. Probably due to a failure to reduce heterogeneity in these embryos, an RA- triggered decrease in Wnt activity may be amplified rapidly via a positive feedback loop in individual cells. Thus, maintaining cooperativity among members of the epiblast/tailbud cell population and reducing the disparity in Wnt signaling among members may render the cell population more resilient to external stress.  
+
+It has been shown that intercellular communication within a cell population is critical to regulate cell differentiation. In Xenopus gastrulas, muscle progenitor cells communicate with each other as they differentiate. In such a case, more than one hundred Xenopus muscle precursor cells transplanted into ectoderm sandwiches can differentiate, while smaller groups and single cells cannot \(^{41,42}\) . This cell number- dependent differentiation was described as a "community effect." It is caused by an intercellular interaction among precursor cells and such an interaction is necessary for the cells to differentiate. Theoretical studies have suggested that the positive feedback mediated by intercellular communication is the mechanism underlying this cell number- dependent differentiation \(^{43}\) .  
+
+In this study, we showed that Wnt- mediated intercellular communication is actually involved in maintenance of the cell population in the epiblast/tailbud region. In this case, intercellular exchange of Wnt ligands is important to compensate for the disparity amplified via positive feedback from Wnt3a and Bra. An interesting question is whether a similar molecular network is involved in other events in which a community effect is exerted. Differences in the efficiency of cell signaling or the amplification efficiency of positive feedback loops probably generate differences in the features of cell populations. If this is the case, it will be important in future studies to identify key parameters in the molecular network to produce each of these events.
+
+<--- Page Split --->
+
+## Methods  
+
+## Mice  
+
+Animal care and experiments were performed in accordance with guidelines for animal experimentation of the National Institutes for Natural Sciences. All animal experiments were approved by the Animal Research Committee of National Institutes for Natural Sciences. Mice were maintained in a light- and temperature- controlled room using a \(12\mathrm{h:}12\mathrm{h}\) light:dark cycle at \(21\pm 2^{\circ}\mathrm{C}\) . Embryos derived from timed matings were genotyped by PCR with DNA from yolk sacs or embryos. PCR conditions and primer sequences for Wnt3a KO \(^8\) and Wnt vs reporter \(^{28}\mathrm{mice}\) have been previously described.  
+
+## Cell culture and transfection  
+
+HEK293T cells and STF293 cells, which are HEK293 cells stably expressing Super 7x TOPFlash \(^{44}\) , were cultured at \(37^{\circ}\mathrm{C}\) in a 1:1 mixture of DMEM and Ham's F12 medium supplemented with \(8.3\%\) fetal bovine serum.  
+
+Plasmids were transfected into HEK 293T cells or STF293 cells using FuGENE6 transfection reagent (Roche). Culture medium was changed 6 h after transfection. At 24, 48, and 72 h after transfection, cells and culture medium were harvested for Western blotting and luciferase reporter assay. In co- culture experiments, HEK 293T cells transfected with each plasmid were collected 24 h after transfection and mixed 1:1 with STF293 cells. The luciferase reporter assay was performed 24 or 48 h after co- culture. Details of Western blotting and the luciferase reporter assay are described below.  
+
+## Plasmid construction  
+
+To generate plasmid constructs from which Wnt3a fused with human Frizzled 5 (hFzd5) is expressed, a DNA fragment encoding the full length of mouse Wnt3a protein fused to the N- terminus of hFzd5 mediated with 2xMyc tag (TSEQKLISEEDLNEMEQKLISEEDLRS) (Extended Data Fig.1a), was integrated between the ClaI and XbaI sites of pCSf107 plasmid vector, which carries the CMV IE94 promoter. This fusion protein was designed to remove the signal peptide of hFz5, resulting in direct fusion of the N- terminus of hFzd5 to the 2xMyc tag. DNA encoding the full length of mouse Wnt3a and EGFP fused Wnt3a(GFP- Wnt3a) was integrated into pCS2 plasmid vector.  
+
+## Western blotting and luciferase reporter assay  
+
+For detection of proteins in cultured cells and culture supernatant, samples were collected at the time points described in "Cell Culture and Transfection." To detect Wnt3a- Fzd5 proteins in
+
+<--- Page Split --->
+
+embryos, the area posterior to the newly formed somite of E8.5 embryos was cut and collected. SDS- PAGE was carried out according to a standard protocol<sup>45</sup>. Briefly, each sample was mixed with 2x sample buffer [4% SDS, 20% glycerol, 0.001% bromophenol blue and 0.125 M Tris HCl (pH 6.8)] and heated at 37°C for 1 h. Samples were electrophoresed using 10% polyacrylamide gels. After electroporation, proteins on the gel were transferred to a polyvinylidene difluoride (PVDF) membrane (Millipore). These membranes were treated overnight at 4°C with primary antibody (mouse anti-mouse Wnt3a antibody: Takada et al., Dev. Cell, 2006), followed by treatment with secondary antibodies (goat anti-mouse IgG: HRP conjugated, Promega W402B) for 1 h at room temperature. Finally, these proteins were visualized using an Enhanced Chemiluminescent Detection System (Amersham).  
+
+Luciferase reporter assay was performed according to the manufacturer's protocol (Dual- Glo Luciferase Assay System: Promega). Since STF293 cells contain a firefly Luciferase cDNA driven by eight tandem repeats of the TCF binding site, Wnt activity was quantified by monitoring activity of firefly Luciferase. Therefore, Luciferase is expressed depending on the strength of Wnt signaling. Renilla luciferase was used as an internal control to compensate for the mosaic nature of gene transfection. The activity of Luciferase was detected using a Luminometer (Turner Designs).  
+
+## Generation of Wnt3a-Fzd5 knock-in mice  
+
+In Wnt3a- Fzd5 knock- in mice, a DNA fragment encoding the MYC- hFZD5 fragment was designed to be integrated just before the stop codon in exon4 of the mouse Wnt3a gene (Extended Data Fig. 1). The resulting protein expressed from this recombined locus is the same as that expressed in the cell culture experiment described above. To generate this knock- in allele, a pLSODN- 3- based plasmid containing a DNA fragment of myc- Fzd5 was co- injected with plasmids to express gRNA and Cas9 in fertilized eggs. The sequence of the gRNA is as follows: 5'- TTAGGAGCTCTCCTACTTGC- 3'. This gRNA was inserted into pX330. Genotyping was carried out by PCR using the following primers:  
+
+Wnt3a- Fzd5 5F, 5'- TGGTGCTTATCTGCCATTC- 3'; Wnt3a- Fzd5 WTF2, 5'- GTCACATGCACCTCAAGTGC- 3'; Wnt3a- Fzd5 7F, 5'- GGTGTGCCAGGAAATCACGG- 3'; Wnt3a- Fzd5 7R, 5'- GGACACCTGCTTGTGGTAGG- 3'; Wnt3a- Fzd5 WTR2, 5'- AGGATCCTTCCTAGCAGTCC- 3'; Wnt3a- Fzd5 4R, 5'- TTTCTACAGTTGACCGGCCTC- 3'.  
+
+The combination of primers used for PCR is shown in Extended Data Fig. 1. Fragments of 2,564- bp and 3,338- bp were expected from the 5'- region of wild- type Wnt3a and Wnt3a- Fzd5
+
+<--- Page Split --->
+
+alleles, respectively. On the other hand, a 2,368- bp fragment and a 3,564- bp fragment were expected in the \(3^{\prime}\) - region of wild- type Wnt3a and Wnt3a- Fzd5 alleles, respectively.  
+
+## In situ hybridization  
+
+Whole- mount in situ hybridization was performed using digoxigenin- labeled, antisense RNA probes. Briefly, embryos collected at the indicated stages were fixed with \(4\%\) paraformadhyde (PFA) overnight at \(4^{\circ}\mathrm{C}\) , washed with PBS, and treated with \(20\mu \mathrm{g / mL}\) of proteinase K for 5 min. These embryos were incubated in hybridization buffer ( \(50\%\) formamide, \(5\times\) SSC, \(1\%\) SDS, \(50\mu \mathrm{g / mL}\) tRNA) overnight at \(55^{\circ}\mathrm{C}\) . The next day, embryos were washed consecutively with \(5\times\) SSC, \(2\times\) SSC, and Tris- buffered saline with Tween 20 (TBST). Next, embryos were incubated with \(1\%\) sheep serum (Sigma) in TBST for \(1\mathrm{h}\) and then treated with a 1:500 dilution of antidigoxigenin- AP Fab fragments (Roche) overnight at \(4^{\circ}\mathrm{C}\) . The following day, embryos were washed with TBST and alkaline phosphatase buffer [ \(100\mathrm{mMNaCl}\) , \(100\mathrm{mM}\) Tris- HCl (pH 9.5), \(50\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(0.5\%\) Tween 20], and signals were developed using BM Purple (Roche). Wild- type and mutant embryos were stained for the same period in individual experiments. The following probes that have been reported previously were used: mouse Wnt3a<sup>46</sup>, mouse Brachyury and mouse Tbx6<sup>14</sup>, mouse Uncx4.1<sup>47</sup>, and human FRIZZLED5<sup>48</sup>. To generate a Sox2 probe, the first exon of mouse Sox2 was amplified from mouse genomic DNA and cloned to generate an antisense probe.  
+
+## Immunofluorescence  
+
+Whole- mount immunofluorescence was performed on embryos collected at the indicated stages. These embryos were fixed with \(4\%\) PFA overnight at \(4^{\circ}\mathrm{C}\) , and washed with PBS. Embryos were incubated overnight at \(4^{\circ}\mathrm{C}\) with the following primary antibodies: rabbit anti- Sox2 (polyclonal, Millipore, AB5603, 1:200) and goat anti- Brachyury (polyclonal, Santacruz, 17745, 1:1000). After washing with PBS, embryos were incubated overnight at \(4^{\circ}\mathrm{C}\) with DAPI (Dojindo) and following secondary antibodies: donkey anti- rabbit IgG (Alexa fluor 647 conjugated, 1:500) and donkey anti- goat IgG (Alexa fluor 647 conjugated, 1:500). Then, embryos were rinsed with PBS again and mounted on \(0.8\%\) LMP agarose for observation. Fluorescent images were acquired using an inverted confocal microscope (Leica SP8).  
+
+## Retinoic acid treatment  
+
+To detect sensitivity to RA signaling in Wnt3a- Fz5 homozygotes, we crossed Wnt3a- Fz5 heterozygotes. RA or DMSO was injected into the abdominal cavities of female mice at E7.5, and embryos were harvested at E8.5.
+
+<--- Page Split --->
+
+## Wnt vis reporter detection  
+
+Wnt vis reporter detectionEmbryos were harvested at each stage and fixed overnight in \(4\%\) PFA. Then, they were stained with DAPI in \(1\%\) Triton X- 100 solution (1:2,000) for several hours or overnight and mounted in \(0.8\%\) LMP agarose. Fluorescent images were acquired using an inverted confocal microscope (Leica SP8). Fluorescence of this reporter in the nucleus was measured in individual cells in a photon- counting mode. Epiblast cells in the region lateral to the node were individually analyzed in a single confocal plane. The area of the nuclei in each cell was identified by DAPI staining. The GFP intensity in the identified area was counted. For the cells with the top \(10\%\) GFP intensity in region II, the relative GFP intensities of the cells in contact with those high- GFP cells were summarized.  
+
+## Quantification and statistical analysis  
+
+Quantification and statistical analysisStatistical analyses were performed using Excel and R software. Differences were assessed for statistical significance using T- tests. p- values \(< 0.05\) were considered statistically significant. Error bars in graphs indicate the standard deviation of each group.  
+
+## Mathematical Model  
+
+We considered an ideal situation in which \(L_{x}\times L_{y}\) epiblast cells align on a regular square lattice such that the x and y axes coincide with the left- right and the anterior- posterior axes (Fig. 7a). Each cell has its Wnt signaling activity as a variable \(W_{ij}\) , where \(i\) and \(j\) are the indexes of the cell on the x- y coordinate. The activity \(W_{ij}\) changes in time through the production, degradation, and intercellular diffusion of Wnt molecules (Fig. 7a), and its time evolution is given by Langevin equation:  
+
+\[\frac{d}{dt} W_{ij} = a + \alpha \frac{W_{ij}^{2}}{K^{2} + W_{ij}} -\beta W + D\sum_{k,k'}\left(W_{k,k'} - W_{ij}\right) - d_{RA}(j,t)W_{ij} + \gamma \xi_{ij}(t) \quad (1)\]  
+
+The first and second terms represent the production of Wnt molecules, a constant production ( \(1^{\mathrm{st}}\) term), and autocatalytic reaction with half saturation concentration \(K\) (2nd term), while the third term represents degradation with a degradation constant \(\beta\) . The fourth term denotes intercellular diffusion of Wnt molecules with diffusion constant \(D\) . The summation \(\sum_{k,k'\in n.n.}(\cdot)\) indicates the summation with respect to the nearest- neighbor cells around \(W_{ij}\) (i.e., \(W_{i\pm 1,j}\) and \(W_{i,j\pm 1}\) ). RA- dependent repression is given by the fifth term. As long as the RA diffusion does not reach the region of epiblast cells (t< t_RA), the value of d_RA(j,t) is set to 0; \(d_{RA}(j,t) = 0\) . After RA diffusion reaches this region ( \(t\geq t_{RA}\) ), \(d_{RA}(j,t)\) creates the following
+
+<--- Page Split --->
+
+time- independent spatial gradient that increases linearly along y- axis (the anterior- posterior axis) with a slope \(\Delta\) :  
+
+\[d_{RA}(j) = \beta_{1} - \beta +j \Delta .\]  
+
+The last term in Eq.(1) denotes the noise term: \(\gamma\) and \(\xi_{ij}(t)\) indicate the noise strength and independent Gaussian white noise with zero mean and variance \(\langle \xi_{ij}(t)\xi_{i^{\prime}j^{\prime}}(t^{\prime}) \rangle =\)  
+
+\(\delta \left(t - t^{'}\right)\delta_{i,i^{\prime}}\delta_{j,j^{\prime}}\) , respectively. Cell division events are also considered a stochastic  
+
+Poisson process in the model. Each cell randomly divides at a constant rate \(\lambda\) . By the cell division, surrounding cells are pushed stochastically toward either the left (in the negative direction of the x- axis with probability 1/4), right (along the positive direction of x- axis with probability 1/4) or upward (along the positive direction of y- axis with probability 1/2). For instance, when the left is chosen in the division of cell with \(x = i\) and \(y = j\) , all cells with \(x < i\) and \(y = j\) move toward the left. The extruded cell outside the \(L_{x} \times L_{y}\) lattice is removed.  
+
+Equation (1) is numerically solved by the Euler- Maruyama method with \(\Delta t = 10^{- 5}\) , while presence or absence of a cell division event is determined at each time step following the probability \(\lambda L_{x}L_{y}\Delta t\) . The time unit is normalized so that \(\lambda = 1\) , while the length scale is normalized by cell length (the lattice size). The diffusion constant \(D\) is set to \(D = 8.0\) for paracrine (+) cells (Fig. 7b, d and e) and \(D = 0\) for paracrine (- ) cells (Fig. 7c and f). The production rate of Wnt molecules was chosen as \(a = 30.0\) for paracrine \((\pm)\) cells (Fig. 7b, c, e and f) and \(a = 26.8\) for paracrine (+)- production \((\downarrow)\) cells (Fig.7d). The parameter \(\beta_{1}\) that represents basal degradation for \(t \geq t_{RA}\) was chosen as \(\beta_{1} = 250.0\) for RA(- ) situation (Fig.7b- d) and \(\beta_{1} = 252.0\) for RA(+) situation (Fig.7e and f). For other parameters, the following values were used: \(L_{x} = L_{y} = 50\) , \(\alpha = 522.0\) , \(K = 1.12\) , \(\beta = 220.0\) , \(\Delta = 0.2\) , \(\gamma = 3.0\) .  
+
+## Acknowledgements  
+
+We thank the Spectrography and Bioimaging Facility of the NIBB Core Research Facilities for their technical support. We also thank Drs. Takahashi and Mizuno at the University of Tsukuba for generating Wnt3a- Fzd5 knock- in mice using CRISPR/Cas9- mediated genome editing and Dr. Fujimori at NIBB for providing mice and technical support. Dr. Aoki at NIBB and ExCELLS and all members of S.T.'s laboratory are gratefully acknowledged for helpful discussions.
+
+<--- Page Split --->
+
+## Author contributions  
+
+Y.H. performed the majority of experiments, participated in their planning, and wrote the text. N.S. formulated the mathematical concept, conducted computer modeling, and wrote the text. Y.M. participated in the planning of the experiments and discussion of the results. T.S. participated in generation of mouse mutants. T.T. generated Wnt reporter mice. H.N. participate in modeling. S.T. formulated the initial key hypothesis, organized all the work, planned experiments, and wrote the text. All authors reviewed and approved the manuscript.  
+
+## Funding  
+
+This work was supported by the following programs: Grants- in- Aid for Scientific Research (B), 18H02454 and 21H02498 to ST, Grants- in- aid for Scientific Research on Innovative Areas, 24111002, 17H05782, 19H04797 to ST, from the Japan Society for the Promotion of Science. Additional support came from grants from National Institutes of Natural Sciences (NINS Joint Research Program to ST) and the Cooperative Study Program of Exploratory Research Center on Life and Living Systems (ExCELLS; program Nos. 21- 102 to HN and NS).  
+
+## Competing financial interests  
+
+The authors declare no competing or financial interests.
+
+<--- Page Split --->
+
+## References  
+
+1. Stern, C. D. et al. Head-tail patterning of the vertebrate embryo: One, two or many unresolved problems? International Journal of Developmental Biology 50, 3-15 (2006).  
+2. Tzouanacou, E., Wegener, A., Wymeersch, F. J., Wilson, V. & Nicolas, J. F. Redefining the Progression of Lineage Segregations during Mammalian Embryogenesis by Clonal Analysis. Developmental Cell 17, 365-376 (2009).  
+3. Cambray, N. & Wilson, V. Two distinct sources for a population of maturing axial progenitors. Development 134, 2829-2840 (2007).  
+4. Cambray, N. & Wilson, V. Axial progenitors with extensive potency are localised to the mouse chordoneural hinge. Development 129, 4855-4866 (2002).  
+5. Tam, P. P. & Beddington, R. S. The formation of mesodermal tissues in the mouse embryo during gastrulation and early organogenesis. Development (Cambridge, England) 99, 109-26 (1987).  
+6. Wilson, V., Olivera-Martinez, I. & Storey, K. G. Stem cells signals and vertebrate body axis extension. Development vol. 136 1591-1604 (2009).  
+7. Wilson, V. & Beddington, R. S. Cell fate and morphogenetic movement in the late mouse primitive streak. Mechanisms of development 55, 79-89 (1996).  
+8. Takada, S. et al. Wnt-3a regulates somite and tailbud formation in the mouse embryo. Genes & development 8, 174-89 (1994).  
+9. Cunningham, T. J., Kumar, S., Yamaguchi, T. P. & Duester, G. Wnt8a and Wnt3a cooperate in the axial stem cell niche to promote mammalian body axis extension. Developmental dynamics: an official publication of the American Association of Anatomists 244, 797-807 (2015).  
+10. Zhao, X. & Duester, G. Effect of retinoic acid signaling on Wnt/ \(\beta\) -catenin and FGF signaling during body axis extension. Gene Expression Patterns 9, 430-435 (2009).  
+11. Liu, P. et al. Requirement for Wnt3 in vertebrate axis formation. Nature genetics 22, 361-5 (1999).  
+12. Tenin, G. et al. The chick somitogenesis oscillator is arrested before all paraxial mesoderm is segmented into somites. BMC Developmental Biology 10, (2010).  
+13. Rivera-Perez, J. A. & Magnuson, T. Primitive streak formation in mice is preceded by localized activation of Brachyury and Wnt3. Developmental Biology 288, 363-371 (2005).
+
+<--- Page Split --->
+
+14. Yamaguchi, T. P., Takada, S., Yoshikawa, Y., Wu, N. & McMahon, A. P. T (Brachyury) is a direct target of Wnt3a during paraxial mesoderm specification. Genes & development 13, 3185-90 (1999).15. Anderson, M. J. et al. TCreERT2, a Transgenic Mouse Line for Temporal Control of Cre-Mediated Recombination in Lineages Emerging from the Primitive Streak or Tail Bud. PLoS ONE 8, (2013).16. Garriock, R. J. et al. Lineage tracing of neuromesodermal progenitors reveals novel wnt-dependent roles in trunk progenitor cell maintenance and differentiation. Development 142, 1628-1638 (2015).17. Imuta, Y., Kiyonari, H., Jang, C. W., Behringer, R. R. & Sasaki, H. Generation of knock-in mice that express nuclear enhanced green fluorescent protein and tamoxifen-inducible Cre recombinase in the notochord from Foxa2 and T loci. Genesis 51, 210-218 (2013).18. Koch, F. et al. Antagonistic Activities of Sox2 and Brachyury Control the Fate Choice of Neuro-Mesodermal Progenitors. Developmental Cell 42, 514-526.e7 (2017).19. Yoshikawa, Y., Fujimori, T., McMahon, A. P. & Takada, S. Evidence that absence of Wnt-3a signaling promotes neuralization instead of paraxial mesoderm development in the mouse. Developmental biology 183, 234-42 (1997).20. Dunt, W. C. et al. Wnt3a/ \(\beta\) -catenin signaling controls posterior body development by coordinating mesoderm formation and segmentation. Development 135, 85-94 (2008).21. Dunt, W. C., Kennedy, M. W. L., Chalamalasetty, R. B., Campbell, K. & Yamaguchi, T. P. Transcriptional profiling of Wnt3a mutants identifies Sp transcription factors as essential effectors of the Wnt/ \(\beta\) -catenin pathway in neuromesodermal stem cells. PLoS ONE 9, e87018 (2014).22. Galceran, J., Hsu, S. C. & Grosschedl, R. Rescue of a Wnt mutation by an activated form of LEF-1: regulation of maintenance but not initiation of Brachyury expression. Proceedings of the National Academy of Sciences of the United States of America 98, 8668-73 (2001).23. Martin, B. L. & Kimelman, D. Regulation of Canonical Wnt Signaling by Brachyury Is Essential for Posterior Mesoderm Formation. Developmental Cell 15, 121-133 (2008).24. Martin, B. L. & Kimelman, D. Brachyury establishes the embryonic mesodermal progenitor niche. Genes and Development 24, 2778-2783 (2010).
+
+<--- Page Split --->
+
+25. Shinozuka, T., Takada, R., Yoshida, S., Yonemura, S. & Takada, S. Wnt produced by stretched roof-plate cells is required for the promotion of cell proliferation around the central canal of the spinal cord. Development 146, (2019).26. Greco, T. L. et al. Analysis of the vestigial tail mutation demonstrates that Wnt-3a gene dosage regulates mouse axial development. Genes & development 10, 313-24 (1996).27. Tsakiridis, A. et al. Distinct Wnt-driven primitive streak-like populations reflect in vivo lineage precursors. Development 141, 1209-1221 (2014).28. Takemoto, T. et al. R26-WntVis reporter mice showing graded response to Wnt signal levels. Genes to Cells 21, 661-669 (2016).29. Yamaguchi, T. P. Genetics of Wnt signaling during early mammalian development. Methods in Molecular Biology 468, 287-305 (2008).30. Abu-Abed, S. et al. The retinoic acid-metabolizing enzyme, CYP26A1, is essential for normal hindbrain patterning, vertebral identity, and development of posterior structures. Genes & development 15, 226-40 (2001).31. Briscoe, J. & Small, S. Morphogen rules: Design principles of gradient-mediated embryo patterning. Development (Cambridge) vol. 142 3996-4009 (2015).32. Gurdon, J. B. & Bourillot, P. Y. Morphogen gradient interpretation. Nature 413, 797-803 (2001).33. Christian, J. L. & Moon, R. T. Interactions between Xwnt-8 and Spemann organizer signaling pathways generate dorsoventral pattern in the embryonic mesoderm of Xenopus. Genes and development 7, 13-28 (1993).34. Petersen, C. P. & Reddien, P. W. Wnt Signaling and the Polarity of the Primary Body Axis. Cell vol. 139 1056-1068 (2009).35. Shimizu, T., Bae, Y. K., Muraoka, O. & Hibi, M. Interaction of Wnt and caudal-related genes in zebrafish posterior body formation. Developmental Biology 279, 125-141 (2005).36. Mitrophanov, A. Y. & Groisman, E. A. Positive feedback in cellular control systems. BioEssays: news and reviews in molecular, cellular and developmental biology 30, 542-55 (2008).37. Abu-Abed, S. et al. Developing with lethal RA levels: Genetic ablation of Rarg can restore the viability of mice lacking Cyp26a1. Development vol. 130 1449-1459 (2003).
+
+<--- Page Split --->
+
+38. Niederreither, K., McCaffery, P., Dräger, U. C., Chambon, P. & Dollé, P. Restricted expression and retinoic acid-induced downregulation of the retinaldehyde dehydrogenase type 2 (RALDH-2) gene during mouse development. Mechanisms of development 62, 67-78 (1997).39. MacLean, G. et al. Cloning of a novel retinoic-acid metabolizing cytochrome P450, Cyp26B1, and comparative expression analysis with Cyp26A1 during early murine development. Mechanisms of development 107, 195-201 (2001).40. Sakai, Y. et al. The retinoic acid-inactivating enzyme CYP26 is essential for establishing an uneven distribution of retinoic acid along the anterio-posterior axis within the mouse embryo. Genes & development 15, 213-25 (2001).41. Gurdon, J. B. A community effect in animal development. Nature 336, 772-4 (1988).42. Gurdon, J. B., Lemaire, P. & Kato, K. Community effects and related phenomena in development. Cell 75, 831-4 (1993).43. Saka, Y., Lhoussaine, C., Kuttler, C., Ullner, E. & Thiel, M. Theoretical basis of the community effect in development. BMC Systems Biology 5, (2011).44. Tsukiyama, T. et al. Molecular Role of RNF43 in Canonical and Noncanonical Wnt Signaling. Molecular and Cellular Biology 35, 2007-2023 (2015).45. Laemmli, U. K. Cleavage of structural proteins during the assembly of the head of bacteriophage T4. Nature 227, 680-5 (1970).46. Roelink, H. & Nusse, R. Expression of two members of the Wnt family during mouse development—restricted temporal and spatial patterns in the developing neural tube. Genes & development 5, 381-8 (1991).47. Mansouri, A. et al. Paired-related murine homeobox gene expressed in the developing sclerotome, kidney, and nervous system. Developmental dynamics: an official publication of the American Association of Anatomists 210, 53-65 (1997).48. Liu, C., Wang, Y., Smallwood, P. M. & Nathans, J. An essential role for Frizzled5 in neuronal survival in the parafascicular nucleus of the thalamus. Journal of Neuroscience 28, 5641-5653 (2008).
+
+<--- Page Split --->
+
+## Figure Legends  
+
+## Fig. 1. In vitro activity of WNT3A-FZD5 and generation of Wnt3a-Fzd5 knock-in mice  
+
+(a) Schematic fig. of WNT3A-FZD5 protein, comparing it with WNT3A and GFP-WNT3A. (b, c) Wnt signaling activity of each construct shown in A. Wnt signaling activity was monitored in HEK293T (STF293) cells stably expressing the SuperTOPFLASH reporter. In (b), Wnt activity was monitored at 24, 48, and 72 h after transfection of each plasmid into STF293 cells. In (c), paracrine Wnt activity was monitored in co-cultures of Wnt-expressing HEK293T cells with STF293 cells at 24 and 48 h after transfection. Differences were assessed for statistical significance using a T-test; *** P < 0.001; ** P < 0.01; * P < 0.05; P > 0.05; NS (not statistically significant). Error bars in the graph indicate the standard deviation of each group. (d-g) Expression of mouse Wnt3a and human Fzd5 in Wnt3a-Fzd5 heterozygous (Wnt3a<sup>+/Fzd5</sup>) embryos. Whole-mount in situ hybridization was carried out using probes of mouse Wnt3a (d, e) or human Fzd5 (f, g) in wild type (d, f) and Wnt3a<sup>+/Fzd5</sup> (e, g) embryos at E10.5. Images are highlighted on the tailbud (d, e, f, g) and the roof plate of neural tube (d', e', f', g'). Numbers of stained embryos are indicated by "n=" in the images. (h) Western blotting analysis of proteins prepared from E8.5 embryos with anti-mouse Wnt3a antibody. Samples prepared from two embryos were applied to each lane. Note that bands with the predicted molecular weight of WNT3A-FZD5 were detected both in Wnt3a-Fzd5 heterozygous (+/Fzd5) and homozygous (Fzd5/Fzd5) embryos. (i-k) Sagittal views of wt (i), Wnt3a<sup>+/Fzd5</sup> (j) and Wnt3a<sup>Fzd5/Fzd5</sup> (k) embryos at E10.5. i', j', and k' are magnified images of i, j, and k, respectively. i'', j'', and k'' are drawings of the images of i', j', and k', respectively. (l-n) Transverse sections of the neural tube of WT (l, l', l''), Wnt3a<sup>+/Fzd5</sup> (m, m', m'') and Wnt3a<sup>Fzd5/Fzd5</sup> (n, n', n'') embryos at E11.5. Sections at the forelimb (l, m, n), the intermediate between fore and hindlimb (l', m', n') and the hindlimb (l'', m'', n'') levels are shown. Scale bars: 1 mm (d-g', i-k''), 100 μm (l-n''), 200 μm (n''). HL: Hindlimb. FL: forelimb.  
+
+## Fig. 2. Expression of mesoderm and neural marker genes in Wnt3a-Fzd5 homozygous embryos  
+
+(a-v) Expression of mesoderm and neural marker genes in embryos at E10.5. Whole- mount in situ hybridization was carried out using probes of Bra (a-f), Tbx6 (g-l), Uncx4.1 (m-q), and Sox2 (r-v) in wt & Wnt3a<sup>+/Fzd5</sup> (a, g, m, r), Wnt3a<sup>Fzd5/Fzd5</sup> (b, h, n, s), Wnt3a<sup>+/v</sup> (c, i, o, t), Wnt3a<sup>v</sup> (d, j, p, u), Wnt3a<sup>+/</sup> (e, k, q, v) and Wnt3a<sup>+/</sup> (f, l) embryos at E10.5. Red dotted lines indicate tail regions. (w-ad) Expression of mesoderm and neural marker genes in embryos at E12.5. Whole- mount in situ hybridization was carried out using probes of Bra (w, x, y), and Tbx6 (z, aa, ab) in WT & Wnt3a<sup>+/Fzd5</sup> (w, z), Wnt3a<sup>Fzd5/Fzd5</sup> (x, y, aa, ab) embryos at E12.5. Tail regions of stained embryos were cut out and shown in y and ab. x' and aa are drawings of the images of x and aa
+
+<--- Page Split --->
+
+respectively. Numbers of stained embryos are indicated by "n=" in the images. Scale bars: 1 mm. HL: Hindlimb.  
+
+## Fig. 3. Differentiation of Wnt-positive progenitor cells in the neural tube and somites in Wnt3a-Fzd5 homozygous embryos.  
+
+(a) Experimental procedure. Cells once activated by Wnt signaling, which is monitored by Axin2-CreERT2 expression, were eternally labelled by expression of tdTomato. Tamoxifen (TM) was injected into pregnant females at 8.5 dpc and embryos were fixed at E10.5. (b, c) Distribution of tdTomato-labelled cells at the posterior hindlimb level in Wnt3a<sup>+/Fzd5</sup> (b) and Wnt3a<sup>Fzd5/Fzd5</sup> (c) embryos at E10.5. Merged images with DAPI-staining are also indicated (b', c'). Neural tube and dermomyotome, which is derived from somite, are outlined with white and orange dotted lines, respectively. Squares framed by green dotted lines indicate the area around the nephric duct. The percentage of tdTomato-positive cells (d, e) and total cell number (f, g) in somite (d, f), nephric duct (e) and neural tube (g) at the posterior hindlimb level in Wnt3a<sup>+/Fzd5</sup> and Wnt3a<sup>Fzd5/Fzd5</sup> embryos at E10.5. Numbers or percentages of labelled cells (mean±s.d.) per section are shown. Differences were assessed for statistical significance using a T-test; ***, P < 0.001; **, P < 0.01; NS, not statistically significant (P > 0.05). Error bars in the graph mean the standard deviation of each group. Scale bars: 100 μm  
+
+## Fig. 4. NMP cells in Wnt3a-Fzd5 homozygous embryos  
+
+(a- l) Whole-mount staining of Wnt3a- Fzd5 heterozygous (a, c, e, g, j, l, n) and homozygous (b, d, f, h, k, m, o) embryos at E8.75. Maximum intensity projection images of posterior ends of embryos stained with anti- SOX2 (magenta) and anti- BRA (green) antibodies are shown in A and B. To quantify the number of SOX2/BRA double- positive cells, single- plane images of medial (I) and lateral (II) regions lateral to the node were analyzed (c- p). Images of DAPI staining (blue; c, d, j, k), and merged images of staining with anti- SOX2 (magenta) and anti- BRA (green) antibodies (e, f, l, m) are shown. Summarized schematic fig.s (g, h, n, o) and diagrams (i, p) are also shown. The size of the medial and lateral regions is 50 μm x 100 μm. Two embryos were examined for each genotype. (q- x) Whole- mount staining of Wnt3a<sup>+/Fzd5</sup> (q, u), Wnt3a<sup>Fzd5/Fzd5</sup> (r, v), Wnt3a<sup>+/vt</sup> (s, w), and Wnt3a<sup>wt/</sup> (t, x) embryos at E11.5. Maximum intensity projection images of posterior ends of embryos stained with DAPI (blue; q- t), and with anti- SOX2 (magenta) and anti- BRA (green) antibodies (q'- t'), are shown. Single- plane images of the areas indicated with yellow- lined boxes in q'- t' are magnified in u- x, respectively. Images of staining with anti- SOX2 (magenta; u- x) and anti- BRA (green; u'- x') antibodies, as well as merged images (u'- x") are shown. The yellow- lined box is a square with one side = 100 μm. Arrowheads in v" indicate a small number of Sox2/Bra- positive cells. Note that there are no
+
+<--- Page Split --->
+
+SOX2 and BRA double- positive cells in Wnt3a<sup>wt</sup>(t, x). The number of stained embryos is indicated by "n=". Scale bars: \(100 \mu \mathrm{m}\) (a, b, q- t)  
+
+Fig. 5. Wnt signaling in the epiblast cell population of Wnt3a- Fzd5 homozygous embryos Wnt signaling activity in individual epiblast cells was visualized using mouse embryos carrying an EGFP- reporter gene, expression of which is specifically activated by Wnt signaling. The observation scheme of embryos at cylinder stage(E7.0 and E7.5 : a) and post- somiteogenesis stage(E8.5 and E8.75 : f) . Eye marks in (a, f) indicate the direction of observation. Blue boxes in (f) indicate somites. Wnt signaling activity was monitored in Wnt3a- Fzd5 heterozygous (b, d, g, k, q) and homozygous (c, e, h, l, r) embryos at E7.0 (b, c), E7.5 (d, e), E8.0 (g, h), E8.75 (k, l), and E9.5(q, r). Wnt signaling activity was also visualized in \(+ / vt\) (i, m, s) and \(vt / - (j, n, t)\) embryos at E8.0 (i, j), E8.75 (m, n), and E9.5(s, t). Each embryos, magnified images of the areas indicated by boxes. Note that Wnt signaling activity is not obviously changed in Wnt3a- Fzd5 homozygous embryos at E7.0 (b, c) or at E7.5 (d, e). In E8.0 and E8.75 embryos, magnified images of the CLE in (f) are shown in each genotyped embryo. (Areas = \(100 \times 100 \mathrm{mm}\) .) The magnified images were taken at a single confocal plane while the others were processed by maximum intensity projection. GFP intensity in individual cells in CLE was quantified in each genotyped embryo at E8.0 (o) and E8.75 (p). Two embryos were examined for each genotype. Box plots indicate the first and third quartiles and the median. Scale bars: \(100 \mu \mathrm{m}\) . The star in the E9.5 shows the nephric duct.  
+
+## Fig. 6. Effect of retinoic acid on the epiblast cell population of Wnt3a-Fzd5 homozygous embryos  
+
+(a- g) Analysis of retinoic acid (RA)- treated embryos at E8.75. Experimental schemes are shown in (a, b). Dorsal images of DMSO (c, d) or \(10 \mathrm{mM}\) RA (e, f) treated Wnt3a- Fzd5 heterozygous (c, e) and homozygous (d, f) embryos at E8.75 stained with DAPI (blue). Results of quantification of the width at NSB (g) and the length posterior to NSB (h) in each genotyped embryo are shown. Note that RA treatment enhances the abnormality in gross morphology specifically in Wnt3a- Fzd5 homozygous embryos. Red arrows indicate the width at NSB while orange arrows indicate the length posterior to NSB. Differences were assessed for statistical significance using a T- test; \(*** \mathrm{P} < 0.001\) ; \(** \mathrm{P} < 0.01\) ; \(* \mathrm{P} < 0.05\) ; \(\mathrm{P} > 0.05\) ; n.s. (not statistically significant). Error bars in the graph indicate the standard deviation of each group. Scale bars: \(100 \mu \mathrm{m}\) . (i- m) Analysis of retinoic acid (RA)- treated embryos at E8.5. These embryos were treated with RA 7.5 days post coitum. Wnt signaling activity in individual epiblast cells was visualized as shown in Fig. 5(f). Dorsal images, processed by maximum intensity projection of DMSO- (i, k) or \(10 \mathrm{mM}\) RA- (j, l) treated Wnt3a- Fzd5 heterozygous (i, j)
+
+<--- Page Split --->
+
+and homozygous (k, l) embryos at E8.5 are shown. Images of a single confocal plane in CLE. The size of these areas is \(100 \mu \mathrm{m} \times 100 \mu \mathrm{m}\) and their positions in the epiblast are identical to those shown in Fig. 5(f). GFP intensity in individual cells in CLE is summarized in (m). Two embryos were examined for each genotype. Box plots indicate the first and third quartiles and the median. Differences were assessed for statistical significance using a wilcoxon signed- rank test; \(\mathrm{***P< 0.001}\) ; \(\mathrm{**P< 0.01}\) ; \(\mathrm{*P< 0.05}\) ; \(\mathrm{P > 0.05}\) ; n.s. (not statistically significant). In (a) and (h), PS indicates the primitive streak and blue boxes indicate somites. Scale bars: \(100 \mu \mathrm{m}\) .  
+
+## Fig. 7. Mathematical Model for Examining the Effect of Intercellular Communication in Cell Populations  
+
+(a) Schematic diagram showing parameters used in the model. We assumed a virtual space corresponding to the cell sheet of the epiblast. This virtual space is divided into \(50 \times 50\) sections along the antero-posterior and medio-lateral axes. Each section corresponds to a single cell in the epiblast. Wnt activity (W) is determined by parameters such as the rate of production and degradation of Wnt protein, the rate of amplification or reduction by positive feedback, the rate of intercellular exchange of Wnt protein, the rate of inhibition by RA, and fluctuating noise that affects Wnt activity. It is assumed that cell division occurs randomly and that dividing daughter cells are extruded in one section in either the left, right, or anterior direction in a 1:1:2 ratio. (b-f) Spatial patterns of Wnt activity in a virtual sheet of cells. Examples of the spatial pattern in the presence (b, d, e) or absence (c, f) of the paracrine function of Wnt are shown at the same time point (mean division time \(t = 3.00\) ) after addition of RA \((t = 0)\) . In the condition of D, the Wnt production rate is reduced (see Method). The spatial patterns of Wnt activity were calculated in the absence (b-d) and presence (e, f) of uniformly supplied RA. (g) Time course of Wnt-positive cells in a virtual sheet of cells. The time course of the proportion of Wnt-positive cells ( \(>50\%\) of maximum activity) at the same spatial level \((y = 35\) in b-d) along the anterior-posterior axis in a virtual sheet of cells is shown. Orange and blue lines indicate the result with and without the paracrine function of Wnt, respectively. A green line indicates the result obtained in the condition where the Wnt production rate is reduced in the presence of the paracrine function of Wnt. Solid and dashed lines indicate results obtained in the absence and presence of uniformly supplied RA, respectively. (h) Schematic representation showing the effect of Wnt paracrine in the epiblast cell population. Prior to somite formation (E7.0-E8.0), Wnt activity in each epiblast cell is dramatically increased by positive feedback regulation mediated by Wnt3a and Bra. During this period, no obvious difference was observed between control and embryos lacking paracrine Wnt signaling (Wnt3a-Fzd5
+
+<--- Page Split --->
+
+homozygous embryos). After the onset of somite formation (after E8.0), the number of Wnt weak cells was increased by the antagonistic effect of RA, which is synthesized in somites, but a small number of Wnt- strong cells remain for a long period in embryos lacking paracrine Wnt signaling (paracrine (- )). This increased heterogeneity in Wnt signaling is compensated for by intercellular exchange of Wnt ligands between epiblast cells (paracrine (+)).  
+
+## Extended Data Fig. 1. WNT3A-FZD5 was not detected in culture supernatant.  
+
+Western blotting analysis of cell lysate (a) and culture supernatant (b) prepared from WNT3A and WNT3A- FZD5 expressing HEK293T cells at 24, 48, and 72 h after transfection. While the expression level of WNT3A- FZD5 was similar to WNT3A in the cell lysate, WNT3A- FZD5 was not detectable in culture supernatant. Red and blue arrowheads indicate bands corresponding to the predicted molecular weights of WNT3A- FZD5 and WNT3A, respectively.  
+
+## Extended Data Fig. 2. Generation of Wnt3a-Fz5 knock-in mice.  
+
+Extended Data Fig. 2. Generation of Wnt3a- Fz5 knock- in mice.(a- f) Generation the Wnt3a- Fz5 knock- in allele. A schematic fig. indicates the mouse Wnt3a locus and the Wnt3a- Fzd5 knock- in allele is shown in (a). In the knock- in allele, human Frizzled5 (blue) fused with 2 myc tags (green) is inserted at the C- terminus of mouse Wnt3a. The knock- in event was confirmed by PCR analysis using the primer sets indicated in b. The results of PCR analyses are shown (c- f). Prime sets are indicated on the upper side of each fig.. Band sizes indicated by colored arrowheads correspond to the predicted sizes shown in b. (g- j) Whole image of wt (g, i) and \(Wnt3a^{+ / Fzd5}\) (h, j) embryos at E10.5 hybridized with Wnt3a (g, h) or \(hFzd5\) (i, j) probes. Magnified images of posterior bodies and dorsal views of these embryos are shown in Fig. 1d- g. Scale bars: 1 mm. (k) The proportion of individuals of each genotype during embryonic development and immediately after birth.  
+
+## Extended Data Fig. 3. The source of ventral neural cells loses Wnt signaling after E7.5 during development of Wnt3a-Fzd5 heterozygous embryos.  
+
+(a) Experimental procedure. Cells once activated by Wnt signaling, which is monitored by Axin2-CreERT2 expression, were eternally labelled by expression of tdTomato. Tamoxifen (TM) was injected to pregnant females at 7.5 dpc (b, d) or 8.5 dpc (c, e) and embryos were fixed at E10.5. Whole-mount bright field images (b, c) and tdTomato staining are also indicated (b', c'). Distribution of tdTomato-labelled cells at the posterior hindlimb level in \(Wnt3a^{+ / Fzd5}\) embryos at E10.5(d, e). Scale bars: \(100 \mu \mathrm{m}\)
+
+<--- Page Split --->
+
+## Extended Data Fig. 4. Synergistic effect of Wnt3a-Frizzled and Bra on the posterior development of body axis elongation  
+
+Wnt3a+/Fzd5; Bra+/+(a), Wnt3a+/+; Bra+/-(b) and Wnt3a+/Fzd5; Bra+/-(c) embryos stained by whole- mount in situ hybridization using the Sox2 probe are shown. Embryos were fixed at E11.5. a', b', and c' are magnified images of a, b, and c, respectively. Note that Wnt3a- Fzd5 and Bra compound heterozygous embryos (Wnt3a+/Fzd5; Bra+/-(c)) impair the posterior development of body axis elongation while embryos heterozygous for either of them appear normal. Scale bars: 1 mm.  
+
+## Extended Data Fig. 5. Characteristics of Wnt3a<sup>w/- </sup>embryos  
+
+(a- c) Sagittal views of wt (a), Wnt3a+/+(b) and Wnt3a<sup>w/- </sup> (c) embryos at the E11.5. a', b', and c' are magnified images of a, b, and c, respectively. a", b", and c" are drawings of the images of a', b', and c', respectively. (d- f) Wnt3a expression was detected by whole- mount in situ hybridization of wt (d), Wnt3a+/+(e) and Wnt3a<sup>w/- </sup> (f) embryos at E9.5. Dorsal views of the posterior region of each embryo are indicated. Red dotted lines indicate the outer edge of the tail. In Wnt3a<sup>w/- </sup> embryos, Wnt3a expression is highly decreased at this stage. (g, h) Whole- mount in situ hybridization of Wnt3a<sup>w/+ </sup> and Wnt3a<sup>w/- </sup> embryos at E12.5 using Bra(g) and Tbx6(h) probes. In contrast to Wnt3a- Fzd5 homozygous embryos, the expression of Bra nad Tbx6 is not detectable in Wnt3a<sup>w/- </sup> embryos. Scale bars: 1 mm.  
+
+## Extended Data Fig. 6. The phenotype of Wnt3a-Fzd5 homozygous embryos can be rescued, depending on intercellular signaling of Wnt3a  
+
+(a- d) Sagittal views of Wnt3a<sup>+/Fz5</sup> (a, a', a"), Wnt3a<sup>Fz5/Fz5</sup> (b, b', b"), Wnt3a<sup>Fz5</sup> (c, c', c") and Wnt3a<sup>w/Fz5</sup> (d, d', d") embryos at the E10.5. a', b', c', and d' are magnified images of a, b, c, and d, respectively. a", b", c", and d" are drawings of the images of a', b', c', and d', respectively. (e- h) Whole- mount in situ hybridization of Wnt3a<sup>+/Fz5</sup> (e), Wnt3a<sup>Fz5/Fz5</sup> (f), Wnt3a<sup>Fz5</sup> (g) and Wnt3a<sup>w/Fz5</sup> (h) embryos at E10.5 with Bra probe. Red dotted lines indicate the edge of the body posterior to the hindlimb. Scale bars: 1 mm.  
+
+## Extended Data Fig. 7. Summary plots of Bra and Sox2 signal intensities examined by immunohistochemistry.  
+
+(a) Schematic fig. showing the area examined. (b) Summary plots of Bra and Sox2 signal intensities in medial (I) and lateral (II) areas at the node-streak border in Wnt3a-Fzd5 heterozygous and homozygous embryos. Two embryos were examined for each genotype. Measurements for each cell are plotted according to levels of Bra (x-axis) and Sox (y-axis). Levels of Bra and Sox2 in each cell were normalized by the average of levels of Bra and Sox2
+
+<--- Page Split --->
+
+level in the medial area of Wnt3a- Fzd5 heterozygous embryos. Cells located between the two dashed lines were defined as Bra and Sox2 double- positive cells.  
+
+## Extended Data Fig. 8. Wnt signaling in the epiblast cell population of Wnt3a knock-out embryos  
+
+Wnt signaling activity in individual epiblast cells was visualized using mouse embryos carrying an EGFP- reporter gene, expression of which is specifically activated by Wnt signaling. Wnt signaling activity was monitored in Wnt3a knock- out (b, d) and WT (a, c) embryos at E7.5 (a, b) and E8.5 (c, d). Note that Wnt signaling is drastically reduced at E7.5 (e) and completely lost at E8.5 (f) in Wnt3a null embryos, suggesting that Wnt activity at and after E8.5 epiblast is dependent on only Wnt3a ligand. Scale bars: \(100 \mu \mathrm{m}\) .  
+
+## Extended Data Fig. 9. Temporal changes of the spatial profile of Wnt-positive cells as simulated by our mathematical model.  
+
+Spatial profiles of the proportion of Wnt- positive cells ( \(>50\%\) of maximum activity) in the virtual space are indicated at 0 (T=TR), 0.2 (T=TR+0.2), 1 (T=TR+1), 2 (T=TR+2), 3 (T=TR+3) mean division time after addition of RA. Spatial profiles under combined conditions with and without Wnt- mediated intercellular communication and with and without uniformly supplied RA are shown. Also shown is the spatial profile under the condition of Wnt- mediated intercellular communication and a reduced rate of Wnt production. In Fig. 7G, the time course of the proportion of Wnt- positive cells at the same spatial level ( \(y = 35\) ) is summarized in a single graph. A and P indicate anterior and posterior, respectively.  
+
+## Extended Data Movie 1 Simulation of the time course of the spatial pattern of Wnt signaling activity in a hypothetical epiblast using our mathematical model.  
+
+(a) Time course of Wnt activity in the hypothetical epiblast in the presence of intercellular  
+
+exchange of Wnt ligands.  
+
+(b) Time course of Wnt activity in the hypothetical epiblast in the absence of intercellular exchange of Wnt ligands.  
+
+(c) Time course of Wnt activity in the hypothetical epiblast in the presence of intercellular exchange of Wnt ligands, but reduced Wnt production.  
+
+(d) Time course of Wnt activity in the hypothetical epiblast with uniform addition of RA in the presence of intercellular exchange of Wnt ligands.  
+
+(e) Time course of Wnt activity in the hypothetical epiblast with uniform addition of RA in the absence of intercellular exchange of Wnt ligands.
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+
+<--- Page Split --->
+
+<center>Fig. 2</center> 
+
+![PLACEHOLDER_31_0]
+
+ 
+
+<center>E12.5</center> 
+
+![PLACEHOLDER_31_1]
+
+
+<--- Page Split --->
+![PLACEHOLDER_32_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_33_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_34_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_35_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_36_0]
+
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+aDiffusionRA.mp4  NCBformatExtendedDataFig.pdf  dDiffusionRA.mp4  cProduction.mp4  bDiffusionRA.mp4  eDiffusionRA.mp4
+
+<--- Page Split --->

preprint/preprint__b38f37bbaa6ba9c98888df06d22bb37e705754f2b008a22e0b0ee6bc28d62cae/preprint__b38f37bbaa6ba9c98888df06d22bb37e705754f2b008a22e0b0ee6bc28d62cae_det.mmd

@@ -0,0 +1,574 @@
+<|ref|>title<|/ref|><|det|>[[44, 106, 916, 178]]<|/det|>
+# Intercellular exchange of Wnt ligands reduces cell population heterogeneity in embryogenesis  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 360, 238]]<|/det|>
+Yudai Hatakeyama National Institute for Basic Biology  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 901, 285]]<|/det|>
+Nen Saito Exploratory Research Center on Life and Living Systems https://orcid.org/0000- 0002- 8317- 9389  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 772, 353]]<|/det|>
+Yusuke Mii National Institute for Basic Biology and Okazaki Institute for Integrative Bioscience https://orcid.org/0000- 0002- 1907- 5665  
+
+<|ref|>text<|/ref|><|det|>[[44, 358, 360, 400]]<|/det|>
+Takuma Shinozuka National Institute for Basic Biology  
+
+<|ref|>text<|/ref|><|det|>[[44, 405, 602, 448]]<|/det|>
+Tatsuya Takemoto Tokushima University https://orcid.org/0000- 0003- 1860- 0269  
+
+<|ref|>text<|/ref|><|det|>[[44, 453, 900, 540]]<|/det|>
+Honda Naoki Hiroshima University Shinji Takada ( \(\boxed{\bullet}\) stakada@nibb.ac.jp) Exploratory Research Center on Life and Living Systems https://orcid.org/0000- 0003- 4125- 6056  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 580, 102, 598]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 617, 735, 638]]<|/det|>
+Keywords: Wnt, epiblast, tailbud, NMP, paracrine, community effect, retinoic acid  
+
+<|ref|>text<|/ref|><|det|>[[44, 656, 330, 676]]<|/det|>
+Posted Date: February 2nd, 2022  
+
+<|ref|>text<|/ref|><|det|>[[44, 694, 474, 714]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 1271602/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 731, 910, 775]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[140, 160, 760, 202]]<|/det|>
+# Intercellular exchange of Wnt ligands reduces cell population heterogeneity in embryogenesis  
+
+<|ref|>text<|/ref|><|det|>[[140, 224, 833, 265]]<|/det|>
+Yudai Hatakeyama<sup>1,2,3</sup>, Nen Saito<sup>1,2,3</sup>, Yusuke Mii<sup>1,2,3,4</sup>, Takuma Shinozuka<sup>1,2,3,4</sup>, Tatsuya Takemoto<sup>5</sup>, Honda Naoki<sup>1,6</sup>, & Shinji Takada<sup>1,2,3,</sup>  
+
+<|ref|>text<|/ref|><|det|>[[138, 290, 856, 545]]<|/det|>
+1 National Institute for Basic Biology and Exploratory Research Center on Life and Living Systems (ExCELLS), National Institutes of Natural Sciences, 5- 1 Higashiyama, Myodaiji- cho, Okazaki, Aichi 444- 8787, Japan2 National Institute for Basic Biology, National Institutes of Natural Sciences, 5- 1 Higashiyama, Myodaiji- cho, Okazaki, Aichi 444- 8787, Japan3 The Graduate University for Advanced Studies (SOKENDAI), 5- 1 Higashiyama, Myodaiji- cho, Okazaki, Aichi 444- 8787, Japan4 PREST, Japan Science and Technology Agency (JST), Kawaguchi, Saitama, 332- 0012, Japan5 Institute of Advanced Medical Sciences, Tokushima University, 3- 18- 5 Kuramoto- cho, Tokushima, 770- 8503, Japan6 Graduate School of Integrated Sciences for Life, Hiroshima University, 1- 3- 2 Kagamiyama, Higashi- hiroshima, Hiroshima, 739- 8511, Japan  
+
+<|ref|>text<|/ref|><|det|>[[140, 568, 828, 606]]<|/det|>
+# Present address: Nara Institute of Science and Technology, 8916- 5 Takayama- cho, Ikoma, Nara 630- 0912, Japan  
+
+<|ref|>text<|/ref|><|det|>[[140, 653, 508, 692]]<|/det|>
+\*To whom correspondence should be addressed. Shinji Takada Ph. D.; e- mail: stakada@nibb.ac.jp  
+
+<|ref|>text<|/ref|><|det|>[[140, 738, 664, 756]]<|/det|>
+Running title: Compensation of heterogeneity by paracrine Wnt signal  
+
+<|ref|>text<|/ref|><|det|>[[140, 781, 761, 799]]<|/det|>
+Key words: Wnt, epiblast, tailbud, NMP, paracrine, community effect, retinoic acid
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 121, 212, 136]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[139, 141, 860, 393]]<|/det|>
+Wnt signaling is required to maintain bipotent progenitors for neural and paraxial mesoderm cells, the neuromesodermal progenitor (NMP) cells that reside in the epiblast and tailbud. Since epiblast/ tailbud cells receive Wnt ligands produced by one another, this exchange may average out the heterogeneity of Wnt signaling levels among these cells. Here, we examined this possibility by replacing endogenous Wnt3a with a receptor- fused form that can activate signaling in producing cells, but not in neighboring cells. Mutant mouse embryos showed a unique phenotype in which maintenance of many NMP cells was impaired, although some cells persisted for long periods. The epiblast cell population of these embryos increased heterogeneity in Wnt signaling levels as embryogenesis progressed and were sensitive to retinoic acid, an endogenous antagonist of NMP maintenance. Thus, mutual intercellular exchange of Wnt ligands in the epiblast cell population reduces heterogeneity and achieves robustness to environmental stress.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 419, 185, 434]]<|/det|>
+## Main  
+
+<|ref|>text<|/ref|><|det|>[[139, 439, 860, 585]]<|/det|>
+The number of stem and progenitor cells is tightly controlled during embryogenesis and homeostasis. As the developmental context or external environment surrounding these cells changes, stem and progenitor cell populations respond to these changes, sometimes robustly, sometimes flexibly, thereby keeping these cells under control. In many cases, secreted signal proteins control maintenance and differentiation of stem/progenitor cells. However, mechanisms by which such signal proteins contribute to the robustness of these cell populations remain to be determined.  
+
+<|ref|>text<|/ref|><|det|>[[139, 610, 850, 842]]<|/det|>
+The body axis of vertebrate embryos elongates in an anterior- to- posterior fashion. During this elongation process, cells that constitute tissues in the trunk and tail are continuously generated from progenitor cells 1. These progenitor cells are found in an area at the posterior end of embryos, termed the epiblast in early embryonic stages and the tail bud in later stages. Clonal lineage analysis revealed that both neural and paraxial mesodermal cell types are commonly generated from the same progenitor cells throughout the period of axis elongation 2. These bipotent progenitor cells are called "Neuromesodermal Progenitor" (NMP) cells. NMP cells appear just before the onset of somitogenesis and are maintained until the conclusion of axis elongation. In mouse embryos, NMPs are located in the caudal lateral epiblast (CLE) posterior to the node- streak border of the primitive streak region and the chordoneural hinge (CNH) of the tail bud 2- 7. Population and clonal analyses indicate that these cells behave like stem cells 3,4.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[138, 118, 855, 480]]<|/det|>
+Cell signaling molecules and transcription factors are implicated in regulation of axis elongation, probably by maintaining NMPs. For instance, in the mouse, at least three Wnt ligands are sequentially expressed in the epiblast and tailbud. Wnt3 expression is first activated in the posterior epiblast at E5.5, followed by Wnt8a and Wnt3a expression \(^{8 - 11}\) , While Wnt3 and Wnt8a expression cease by early somite stage, Wnt3a expression continues until E12.5, when tail elongation is almost completed \(^{12}\) . Along with expression of these Wnt ligands, a T- box transcription factor, T/Brachyury (Bra), is continuously expressed in the same region from the onset to the end of Wnt3 expression \(^{13}\) . Evidence suggests that Wnt signaling and Bra are important for maintenance of NMPs. Genetic studies of null mutant embryos of Wnt3a and Bra, showed their importance for axis elongation \(^{8,14}\) , and lineage tracing of cells that express Bra revealed that both neural and paraxial mesoderm cells are derived from Bra- expressing cells \(^{15 - 18}\) . In addition to NMP maintenance, Wnt signaling and Bra are involved in fate determination between the neural and paraxial- mesodermal lineages \(^{14,19}\) . Of note, Wnt signaling directly activates Bra expression through Tcf transcription factor, while Bra is required for Wnt3a expression \(^{14,20 - 22}\) . Thus, Wnt and Bra form a positive feedback loop in which each actively regulates expression of the other in NMP maintenance. Similarly, positive feedback between Wnt8 and tbxta (ntl), a zebrafish ortholog of Bra, has been reported in zebrafish \(^{23,24}\) .  
+
+<|ref|>text<|/ref|><|det|>[[139, 503, 856, 779]]<|/det|>
+During maintenance of NMPs, activation of Wnt signaling and expression of Bra overlap widely in the epiblast and the tail bud, including the area where NMPs exists. Thus, some Wnt ligands may act in an autocrine manner in the epiblast and the tail bud, resulting in self- activation of a Wnt/Bra regulatory loop in each cell. On the other hand, given that cells adjoining NMP cells also express Wnt ligands and Bra, paracrine Wnt ligands supplied by neighboring cells may also be involved in NMP maintenance. To examine the importance of Wnt paracrine function in maintenance of NMPs, we generated knock- in mouse embryos in which endogenous Wnt3a is replaced with a receptor- fused form that lacks paracrine activity, but maintains autocrine activity. Exacting analysis of Wnt paracrine- deficient embryos revealed the significance of the paracrine signal for maintenance of the NMP cell population and resilience to stress from external tissue. To the best of our knowledge, this is the first direct experimental evidence of the significance of intercellular exchange of secreted signal proteins in the emergence of cell population features.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 120, 201, 135]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 141, 772, 158]]<|/det|>
+## Wnt3a fused with Frizzled possesses signaling activity, but no paracrine activity  
+
+<|ref|>text<|/ref|><|det|>[[139, 161, 856, 438]]<|/det|>
+Wnt3a fused with Frizzled possesses signaling activity, but no paracrine activityTo eliminate paracrine activity of WNT3A, we fused mouse WNT3A to the N- terminus of human FRIZZLED5 via 2 MYC tags (WNT3A- FZD5; Fig. 1a). Activity of WNT3A- FZD5 was examined in cells stably expressing TOP- FLASH reporter (STF293 cells) in comparison with authentic WNT3A, as well as GFP- fused WNT3A (GFP- WNT3A; Fig. 1a). Activity of GFP- WNT3A was lower than that of authentic WNT3A (Fig. 1b), but sufficient to replace endogenous Wnt3a in vivo \(^{25}\) . WNT3A- FZD5 activated canonical Wnt signaling to almost same extent as authentic WNT3A and more strongly than GFP- WNT3A, 48 h after transfection, and this activity was nearly saturated even after longer incubation (Fig. 1b). In contrast, whereas Wnt activity was activated in STF293 cells co- cultured with cells expressing intact WNT3A or GFP- WNT3A, almost no activation was detected in co- culture with WNT3A- FZD5- expressing cells (Fig. 1c). Consistent with this result, WNT3A- FZD5 was not detected in culture supernatant (Extended Data Fig. 1). These results show that as expected, WNT3A- FZD5 possesses sufficient signaling activity, but almost no paracrine activity.  
+
+<|ref|>text<|/ref|><|det|>[[140, 461, 856, 671]]<|/det|>
+Axis elongation is impaired, but partially maintained in Wnt3a- Fzd5 homozygous embryosWe next generated mouse embryos in which endogenous Wnt3a is substituted for Wnt3a- Fzd5, using a CRISPR/Cas9- mediated knock- in approach (Extended Data Fig. 2a- f). Mice heterozygous for Wnt3a- Fzd5 were morphologically normal and fertile (Fig. 1d- j, Extended Data Fig. 2g- k). As expected, Wnt3a- Fzd5 exhibited an expression pattern identical to that of endogenous Wnt3a in these embryos (Fig. 1d- g, Extended Data Fig. 2g- j). In addition, Western blotting analysis revealed that Wnt3a- Fzd5 heterozygous and homozygous embryos expressed WNT3A- FZD5 proteins at the expense of authentic Wnt3a in the posterior region (Fig. 1h). Thus, Wnt3a- Fzd5 properly replaced endogenous Wnt3a, being expressed in the same spatial pattern as endogenous Wnt3a.  
+
+<|ref|>text<|/ref|><|det|>[[140, 696, 858, 820]]<|/det|>
+While Wnt3a- Fzd5 heterozygotes (Wnt3a \(^{+/Fzd5}\) ) showed no obvious embryonic abnormality (Fig. 1i, j), adult Wnt3a- Fzd5 homozygotes (Wnt3a \(^{Fzd5 / Fzd5}\) ), were embryonically lethal and die after E12.5 (Extended Data Fig. 2k and data not shown). However, the phenotype of Wnt3a- Fzd5 homozygotes was milder than that of Wnt3a null mutant embryos, which die around E9.5 with posterior truncation \(^{8}\) . Thus, even though WNT3A- FZD5 has sufficient signaling activity in vitro (Fig. 1b, c), Wnt3a- Fzd5 partially, but not completely, substitutes for endogenous Wnt3a.  
+
+<|ref|>text<|/ref|><|det|>[[140, 846, 856, 885]]<|/det|>
+To better understand this puzzling phenotype of Wnt3a- Fzd5 homozygous embryos, we examined their morphology. While the gross morphology of Wnt3a- Fzd5 homozygous embryos
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 118, 839, 308]]<|/det|>
+appeared normal in the anterior trunk, it was highly disorganized posterior to the hindlimbs (Fig. 1i- k). This posterior defect became evident no later than E8.75 (Fig. 4a, b and 5k, l). Transverse images of E11.5 embryos stained with DAPI showed that neural tube morphology was gradually disturbed along the anterior- posterior axis in these embryos (Fig. 11- n). This disruption was evident in the intermediate region between fore- and hindlimbs (Fig. 1n') and pronounced in the more posterior region, resulting in an opened neural tube at the hindlimb level (Fig. 1n'). However, in spite of this severe defect in posterior morphogenesis, a thin, kinked tail- like structure was found at the posterior end of these embryos (Fig. 1k: red arrowhead).  
+
+<|ref|>text<|/ref|><|det|>[[139, 333, 858, 606]]<|/det|>
+Whole mount in situ hybridization analyses also revealed that the neural tube, marked by Sox2 expression, was abnormally opened posterior to the hindlimb at E10.5 in Wnt3a- Fzd5 homozygous embryos (Fig. 2r, s). Somites, stained with the Uncx 4.1 probe, were normally formed in the anterior trunk, but their size is reduced posterior to the hindlimb (Fig. 2m, n). Of note, Brachyury (Bra), which is expressed in the tailbud and notochord of normal embryos, was expressed at the tip of the thin and kinked tail, although the number of Bra- positive cells was decreased (Fig. 2a, b). In addition, Tbx6, expression of which is turned on immediately after specification to the paraxial mesoderm lineage, was also expressed at this posterior end (Fig. 2g, h). Notably, expression of Bra and Tbx6 at the posterior tip was maintained even at E12.5, when tail elongation is nearly arrested in normal embryos (Fig. 2w- ab) Thus, in Wnt3a- Fzd5 homozygous embryos, trunk morphogenesis was disrupted at the hindlimb level, accompanied by reduction of tailbud size, but differentiation from the tailbud appears to be maintained throughout the period of axis elongation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 632, 752, 670]]<|/det|>
+## Wnt-positive progenitor cells are responsible for abnormal neural and somite development in Wnt3a-Fzd5 homozygous embryos  
+
+<|ref|>text<|/ref|><|det|>[[139, 674, 857, 885]]<|/det|>
+To examine the impact of tailbud reduction in posterior morphogenesis of Wnt3a- Fzd5 homozygous embryos, we followed cells that had activated Wnt signaling, because Wnt signaling is activated in progenitor cells in the epiblast and tailbud region. To this end, Axin2- creERT2 and floxed tdTomato alleles were introduced into Wnt3a- Fzd5 homozygous embryos. Tamoxifen was injected into pregnant female mice at 7.5 or 8.5 days post coitus (dpc) and embryos were fixed at E10.5 (Fig. 3a and Extended Data Fig. 3a). In control, wild- type, and Wnt3a- Fzd5 heterozygous, embryos, labelled cells were detected in most tissues at the hindlimb level, regardless of the timing of tamoxifen administration (Extended Data Fig. 3b- e). However, when tamoxifen was injected at 8.5 dpc, but not at E7.5 dpc, labelled cells were rarely detected in the ventral neural tube at the hindlimb level (Extended Data Fig. 3d, e), showing that the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 118, 853, 308]]<|/det|>
+origin of ventral neural cells loses Wnt signaling after E7.5. As in control littermates, in Wnt3a- Fzd5 homozygous embryos labelled at 8.5 dpc, labelled cells were similarly distributed in most tissues, except neural tube (Fig.s 3b and c). Labelling efficiency monitored in somites (Fig. 3d) and the nephric duct (Fig. 3e) was not significantly changed between littermates. However, the number of labelled cells was specifically reduced in somites (Fig. 3f) and neural tube (Fig. 3g) in Wnt3a- Fzd5 homozygous embryos. Thus, the number of cells derived from Wnt-positive progenitors at E8.5 was decreased in dorsal neural tube and somites of Wnt3a- Fzd5 homozygous embryos. It is plausible that the decrease of dorsal neural cells results in the opened neural tube in Wnt3a- Fzd5 homozygous embryos.  
+
+<|ref|>text<|/ref|><|det|>[[139, 332, 860, 544]]<|/det|>
+As described above, Wnt3a is expressed in the roof plate of the neural tube, in addition to the epiblast and tailbud (Fig.s 1d- g). Thus, it also seems probable that Wnt3a- Fzd5 expression in the roof plate region causes the morphological abnormality in Wnt3a- Fzd5 homozygous embryos. To test this possibility, we examined the contribution of Bra to this phenotype, because Bra interacts specifically with Wnt3a in development of the epiblast/ tailbud, but not the roof plate. While Wnt3a- Fz5 heterozygotes (Extended Data Fig. 4a) and Bra single heterozygotes (Extended Data Fig. 4b) appeared normal, Wnt3a- Fz5 and Bra compound heterozygous embryos (Wnt3a \(^{+/Fz5}\) ; Bra \(^{+/}\) ) had open neural tubes and bent tails.(Extended Data Fig. 4c). Thus, interaction of Wnt3a- Fz5 and Bra in the epiblast and tailbud region is responsible for the phenotype of Wnt3a- Fzd5 homozygous embryos.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 568, 827, 606]]<|/det|>
+## Evidence indicates that the phenotype of Wnt3a-Fzd5 homozygotes is due to the lack of paracrine activity  
+
+<|ref|>text<|/ref|><|det|>[[139, 610, 860, 843]]<|/det|>
+Reduction of the Wnt3a signal impairs maintenance of the tailbud, including NMP cells \(^{8,16}\) . This defect results in truncation of A- P elongation in a manner dependent on Wnt3a activity \(^{26}\) . Since Wnt3a \(^{wt / - }\) (vt/- ) embryos, which possess one copy of a hypomorphic (vt) allele of Wnt3a with reduced Wnt3a expression in the tailbud (Extended Data Fig. 5a- f: Greco et al., 1996) impairs trunk development at the same level as in Wnt3a- Fzd5 homozygous embryos (Fig. 2c, d, i, j, o, p, t, and u), we compared the morphology of vt/- embryos with Wnt3a- Fzd5 homozygous embryos. In contrast to Wnt3a- Fzd5 homozygotes, vt/- embryos did not exhibit thin, kinked tails and open neural tubes, but they failed to maintain the tailbud, marked by Bra and Tbx6 expression, at E10.5 (Fig. 2d, j) and E12.5 (Extended Data Fig. 5g, h). Therefore, the phenotype of Wnt3a- Fzd5 homozygotes is unique, compared with other Wnt3a hypomorphic mutants and is not simply due to decreased Wnt3a activity.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 119, 860, 308]]<|/det|>
+This characteristic phenotype of Wnt3a- Fzd5 homozygous embryos was also observed in Wnt3aFzd5/ embryos (Extended Data Fig. 6a- c, e- g). Because Wnt3a- Fzd5 heterozygous embryos (Wnt3a+/Fzd5) appeared normal, as previously described (Fig. 1j, Extended Data Fig. 6a, e), Wnt3a- Fzd5 seems to cause this phenotype in the absence of wild- type Wnt3a. Furthermore, this phenotype can be rescued depending on the expression level of Wnt3a, because the phenotype of Wnt3aFzd5/ embryos was partially rescued by replacing the null allele to vt (Wnt3aFzd5/vt; Extended Data Fig. 6d, h). Based on the results of these analyses using various Wnt3a mutants, the characteristic phenotype of Wnt3a- Fzd5 homozygotes is due to some property lost in Wnt3a- Fzd5, most likely paracrine activity.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 333, 750, 351]]<|/det|>
+## NMP cells are reduced, but maintained in Wnt3a-Fzd5 homozygous embryos  
+
+<|ref|>text<|/ref|><|det|>[[140, 355, 856, 479]]<|/det|>
+In spite of improper development in the posterior neural tube and somites, our analyses with molecular markers revealed that differentiation of paraxial mesoderm and neural cells was partially maintained in Wnt3a- Fzd5 homozygous embryos (Fig. 2b, h, n, s, x, aa). Furthermore, the tailbud marked by expression of Bra, was maintained at the posterior tip of the tail of these embryos(Fig. 2a, b, x, y, aa, ab). These data suggest that a small number of NMP cells persist in Wnt3a- Fzd5 homozygous embryos.  
+
+<|ref|>text<|/ref|><|det|>[[139, 503, 857, 670]]<|/det|>
+Since one of the characteristics of NMP cells is expression of Bra and Sox2<sup>27</sup>, we compared the number of Bra and Sox2 double- positive cells using immunohistochemistry in Wnt3a- Fzd5 homozygous embryos and control littermates (Fig. 4a- x and Extended Data Fig. 7). The number of Bra and Sox2 double- positive cells started to diminish at E8.75 in Wnt3a- Fzd5 homozygous embryos (Fig. 4a- p), but a small number of double- positive cells were still maintained at E11.5 (Fig. 4q, r, u, v). In contrast, double- positive cells disappeared in vt/- embryos at E11.5 (Fig. 4s, w, t, x). These results support the idea that a small number of NMP cells are specifically maintained in Wnt3a- Fzd5 homozygous embryos, even after trunk development is impaired.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 696, 777, 734]]<|/det|>
+## Wnt signaling activity persists in a small number of epiblast cells in Wnt3a-Fzd5 homozygous embryos  
+
+<|ref|>text<|/ref|><|det|>[[140, 738, 842, 884]]<|/det|>
+To further investigate the defect of Wnt3a- Fzd5 homozygous embryos, we directly examined Wnt signaling activity in individual cells in the epiblast. To this end, we utilized the R26 WntVis reporter, expression of which is driven by heptameric TCF/LEF1 binding sequences combined with a viral minimal promoter in the Rosa26 locus<sup>28</sup>. This reporter responds in a graded fashion to a wide range of Wnt signal strengths. In addition, the histone H2B- EGFP protein, used as a fluorescent reporter, facilitates single- cell resolution analysis under confocal microscopy. In this study, fluorescence of this reporter was measured in individual cells in a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 119, 852, 158]]<|/det|>
+photon- counting mode. Epiblast cells in the areas lateral to the node were individually analyzed in a single confocal plane.  
+
+<|ref|>text<|/ref|><|det|>[[139, 183, 858, 415]]<|/det|>
+During development of mouse epiblast, three Wnt ligands, Wnt3, Wnt8, and Wnt3a, sequentially activate Wnt signaling \(^{8,10,13,29}\) . Because Wnt3 and Wnt8 are expressed prior to Wnt3a, Wnt activity was detected even in Wnt3a null mutant embryos at early headfold (EHF) stage (E7.0; Takemoto et al., 2016). At this stage, no obvious change in Wnt activity was detected in Wnt3a- Fzd5 homozygous embryos, as predicted (Fig. 5a- c). Then, Wnt3a expression was activated, and Wnt signaling level subsequently increased in both control and Wnt3a- Fzd5 homozygous embryos (Fig. 5d, e and Extended Data Fig. 8a), but not in Wnt3a null embryos (Extended Data Fig. 8b), at the late headfold (LHF) stage (E7.5). Thus, Wnt signaling was properly activated in the initial phase of Wnt3a- dependent activation, even in Wnt3a- Fzd5 homozygous embryos. Of note, in these stages, the level of the fluorescent reporter differed among epiblast cells in both control and Wnt3a- Fzd5 homozygous embryos.  
+
+<|ref|>text<|/ref|><|det|>[[139, 439, 858, 715]]<|/det|>
+From early somite stage (E8.0), Wnt signaling began to be perturbed in Wnt3a- Fzd5 homozygous embryos. At E8.0, around 2- 3- somite stage, the number of Wnt signaling- positive cells was reduced in the anterior and lateral epiblast regions of Wnt3a- Fzd5 homozygous embryos (Fig. 5f, g, h, o). The reduction in Wnt signaling- positive cells was pronounced in most of the epiblast region of Wnt3a- Fzd5 homozygous embryos at E8.75, where Wnt- positive and negative cells were distributed in a patch- work pattern (Fig. 5f, k, l, p). This reduction was enhanced by E9.5, but some Wnt- positive cells remained at the posterior end of Wnt3a- Fzd5 homozygous embryos (Fig. 5q, r). These posterior Wnt- positive cells were further maintained until E11.5 (data not shown). On the other hand, in \(vt\) - embryos, the decrease of Wnt signaling started at E8.75 (Fig. 5i, j, m, n). In contrast to Wnt3a- Fzd5 homozygous embryos, Wnt signaling was almost abolished around E9.5 in epiblast (Fig. 5s, t). Furthermore, the activity level appeared to decrease gradually in most Wnt- positive epiblast cells and the deviation of Wnt- activity in epiblast population was smaller than in Wnt3a- Fzd5 homozygous embryos.  
+
+<|ref|>text<|/ref|><|det|>[[139, 739, 848, 885]]<|/det|>
+Taken together, in Wnt3a- Fzd5 homozygous embryos, reduction of Wnt activity occurred from early somite stage, but a small number of Wnt- positive cells remain longer. Notably, Wnt activity appeared to fluctuate between adjacent cells even in control embryos, but in Wnt3a- Fzd5 homozygous embryos this heterogeneity was enhanced (Fig. 5g- p). Probably, the accelerated reduction of Wnt activity in many epiblast cells reduces the number of NMP cells, as well as of neural and somite cells produced by NMP cells in Wnt3a- Fzd5 homozygous embryos. In contrast, persistent activation of Wnt signaling in the other epiblast cells probably
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 120, 813, 157]]<|/det|>
+contributes to maintenance of a small number of NMP cells, resulting in formation of thin, kinked tails.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 184, 733, 201]]<|/det|>
+## Retinoic acid enhances the phenotype of Wnt3a-Fzd5 homozygous embryos  
+
+<|ref|>text<|/ref|><|det|>[[140, 205, 845, 350]]<|/det|>
+Since impairment of Wnt activation was observed from early somite stage, it seems probable that somite formation affects the Wnt3a- Fzd5- specific reduction of Wnt- active cells. Interestingly, retinoic acid (RA), which is synthesized in somite cells, antagonizes the function of Bra in zebrafish embryos \(^{24}\) . Consistently, mouse embryos with mutated cyp26a, which encodes an enzyme to degrade RA, exhibit an axis truncation phenotype, similar to Wnt3a and Bra mutant embryos \(^{30}\) . Thus, we hypothesized that epiblast cells of Wnt3a- Fzd5 homozygous embryos are more sensitive to RA in maintenance of Wnt activity.  
+
+<|ref|>text<|/ref|><|det|>[[139, 375, 855, 565]]<|/det|>
+To address this issue, female Wnt3a- Fzd5 heterozygous mice intercrossed with Wnt3a- Fzd5 heterozygous males were treated with RA 7.5 days post coitum and effects of RA on the phenotype of mutant embryos were examined. RA treatment specifically enhanced the abnormality in gross morphology of Wnt3a- Fzd5 homozygous embryos (Fig. 6a- h). Furthermore, Wnt reporter analysis revealed that RA treatment enhanced the specifically reduced pattern of Wnt activity in Wnt3a- Fzd5 homozygous embryos, showing enhancement of a patch- work pattern, irrespective of cell position along the anterior- posterior axis (Fig. 6i- m). This result suggests that the epiblast cell population of Wnt3a- Fzd5 homozygous embryos is specifically susceptible to RA.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 590, 820, 628]]<|/det|>
+## Mathematical modeling supports the importance of paracrine function in maintaining Wnt-positive epiblast cell populations  
+
+<|ref|>text<|/ref|><|det|>[[139, 632, 855, 885]]<|/det|>
+The results described above strongly suggest that lack of paracrine signaling enhances heterogeneity of Wnt activity in the epiblast cell population and that a cell population with such enhanced heterogeneity is more sensitive to antagonists, like RA. Thus, we also tested the validity of these ideas by creating a mathematical model (Fig. 7a- g, Extended Data Fig. 9 and Movie1). In this model, spatiotemporal changes in Wnt activity were compared in a hypothetical epiblast plane with and without intercellular exchange of Wnt ligands. The temporal increase or decrease of Wnt activity in each cell is defined by the production rate regulated by autocatalysis, which represents a positive feedback loop of Wnt3a/Bra, in addition to the basic rate of production and degradation of Wnt ligands. The stochastic increase/decrease in Wnt activity is also incorporated as a noise term. In this virtual plane, we assume that each cell divides stochastically and that a newly produced daughter cell locates laterally or anteriorly to the original cell. It is also assumed that an RA gradient from anterior to posterior is imposed
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 118, 837, 180]]<|/det|>
+at a specific time, which represents anteroposterior diffusion of RA. However, in this virtual space, cells that are aligned along the left- right axis were treated as if there is no difference in their distance from the RA source (Fig. 7a).  
+
+<|ref|>text<|/ref|><|det|>[[139, 204, 857, 523]]<|/det|>
+By using adjusted parameters, we first simulated spatiotemporal patterns of Wnt activity in the hypothetical epiblast plane under conditions in which intercellular exchange of Wnt ligands is present (Fig. 7b, d, e) and absent (Fingers 7c, f). Wnt activity levels decreased after the addition of RA (t>0 shown in Fig. 7g, Extended Data Fig. 9a, and movie Extended Data Movie1) and the number of Wnt- low cells increased in the absence of intercellular Wnt exchange. However, a small number of Wnt- high cells remained for a while (until t=3 shown in Fig. 7c, g, Extended Data Movie 1B). On the other hand, if the production rate was reduced, mimicking the situation of vt/- embryos, Wnt- low cells gradually increased and few Wnt- high cell remained (Fig. 7D, G and Extended Data Movie 1C). These simulations showed that a lack of intercellular Wnt exchange reproduced the spatio- temporal pattern of Wnt activity observed in Wnt3a- Fzd5 homozygous embryos. Furthermore, we reproduced the sensitivity of Wnt3a- Fzd5 homozygote cells when the RA concentration was uniformly increased in the hypothetical epiblast plane (Fig. 7e, f, g, and Extended Data Movies 1d, e). Taken together, these simulations based on our mathematical model support the idea that the Wnt paracrine signal reduces heterogeneity in Wnt activity in the epiblast cell population and increases robustness to RA.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 570, 226, 584]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[140, 589, 857, 757]]<|/det|>
+It is widely believed that secreted signal proteins act on cells in the vicinity of the source cells, and in some cases, more distally \(^{31,32}\) . In contrast, in the epiblast and tailbud, most cells both produce and receive Wnt ligands \(^{16,33 - 35}\) . As a result, Wnt ligands secreted from each cell into the extracellular space activate the intracellular Wnt signaling pathway in cells in the population. As a result, Wnt3a ligands secreted extracellularly from each cell activate the intracellular Wnt signaling pathway in cells within the population. Thus, in contrast to unidirectional transfer from Wnt- producing cells to receiving cells, Wnt ligands seem to be reciprocally exchanged between epiblast and tailbud cells.  
+
+<|ref|>text<|/ref|><|det|>[[140, 781, 858, 885]]<|/det|>
+To understand the biological significance of reciprocal ligand exchange within a cell population, we generated Wnt3a- Fzd5 homozygous embryos, in which Wnt3a- mediated intercellular communication, or paracrine function, is specifically impaired. In these embryos, the number of Wnt- positive cells decreases rapidly from the anterior and lateral sides of the epiblast after RA begins to be synthesized in the somite, but a small number of Wnt- positive cells, including
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 118, 848, 180]]<|/det|>
+NMP cells, remain at the posterior end for a long time. Precise examination of Wnt3a- Fzd5 homozygous embryos and mathematical simulation support a model in which Wnt3a- mediated intercellular communication is required for maintenance of the NMP population (Figure 7h).  
+
+<|ref|>text<|/ref|><|det|>[[140, 204, 853, 308]]<|/det|>
+In epiblast and tailbud regions, including NMPs, Wnt3a- expressing cells also express Bra \(^{14}\) . Bra is a direct transcriptional target of the Wnt signaling pathway, whereas Wnt3a expression is also dependent on Bra \(^{14,18}\) . Thus, Wnt3a and Bra mutually activate one another, forming a positive feedback regulatory loop. Because positive feedback amplifies small changes, this regulatory system can rapidly increase or decrease the amount of Wnt3a and Bra in a cell \(^{36}\) .  
+
+<|ref|>text<|/ref|><|det|>[[139, 332, 857, 586]]<|/det|>
+It seems plausible that the epiblast/tailbud cells vary with respect to Wnt production and degradation rates, efficiency of feedback amplification, and/or resistance to environmental factors that reduce Wnt activity. Thus, cells that are prone to losing Wnt activity rapidly lose this activity due to the positive feedback, resulting in an increased disparity generated by the fluctuations. In the epiblast of Wnt3a- Fzd5 homozygotes, the number of cells with little or no Wnt activity rapidly increases from E8.0. In contrast, a small, but significant number of cells maintain high Wnt activity for a long time in these embryos. A probable reason for persistence of Wnt- high cells is that the change of Wnt activity in these cells is below the threshold to trigger a rapid decrease by positive feedback. Actually, our mathematical model, which assumes fluctuation in Wnt activity and positive feedback regulation, produces similar spatial patterns of Wnt activity in Wnt3a- Fzd5 homozygous cell populations under conditions without Wnt- mediated intercellular communication.  
+
+<|ref|>text<|/ref|><|det|>[[139, 610, 852, 779]]<|/det|>
+In contrast, in control embryos, including Wnt3a- Fzd5 heterozygotes, the number of Wnt3a positive cells slowly decreased and disappeared around E13.5, when axis elongation was terminated. Probably, in these embryos, Wnt ligands supplied by neighboring Wnt- high cells compensate to some extent for the decrease of Wnt activity in Wnt- low cells. Thus, the exchange of Wnt3a ligands appears to compensate for the rapid decrease in Wnt activity in individual cells. Taken together, positive feedback regulation can amplify heterogeneity among members of the cell population, but our results suggest that sharing of intercellular components of the feedback loop, such as Wnt ligands, inhibits amplification of this heterogeneity.  
+
+<|ref|>text<|/ref|><|det|>[[140, 803, 860, 886]]<|/det|>
+In the epiblast of Wnt3a- Fzd5 homozygotes, Wnt activity rapidly decreases in many cells at E8.0, when several anterior somites are generated. In these embryos, the decrease in Wnt activity was more pronounced anteriorly and laterally in the epiblast. Of note, in early mouse embryos, RA synthesis requires retinaldehyde dehydrogenase 2 (RALDH2/ALDH1a2), which is
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 118, 816, 223]]<|/det|>
+activated in somites and lateral plate mesoderm \(^{37,38}\) . A line of evidence has shown that RA signaling antagonizes Wnt/Bra activity gradually from the anterior side of the epiblast and tailbud region in the mouse embryo \(^{24,30,37,39,40}\) . Thus, we speculated that the epiblast cell population in Wnt3a- Fzd5 homozygotes is sensitive to RA stress originating from tissues developed anteriorly and laterally to the epiblast.  
+
+<|ref|>text<|/ref|><|det|>[[139, 247, 840, 394]]<|/det|>
+Actually, RA treatment reduces Wnt signaling in epiblast cells specifically in Wnt3a- Fzd5 homozygous embryos, indicating that the epiblast cell population in these embryos is more susceptible to RA. Probably due to a failure to reduce heterogeneity in these embryos, an RA- triggered decrease in Wnt activity may be amplified rapidly via a positive feedback loop in individual cells. Thus, maintaining cooperativity among members of the epiblast/tailbud cell population and reducing the disparity in Wnt signaling among members may render the cell population more resilient to external stress.  
+
+<|ref|>text<|/ref|><|det|>[[139, 417, 847, 607]]<|/det|>
+It has been shown that intercellular communication within a cell population is critical to regulate cell differentiation. In Xenopus gastrulas, muscle progenitor cells communicate with each other as they differentiate. In such a case, more than one hundred Xenopus muscle precursor cells transplanted into ectoderm sandwiches can differentiate, while smaller groups and single cells cannot \(^{41,42}\) . This cell number- dependent differentiation was described as a "community effect." It is caused by an intercellular interaction among precursor cells and such an interaction is necessary for the cells to differentiate. Theoretical studies have suggested that the positive feedback mediated by intercellular communication is the mechanism underlying this cell number- dependent differentiation \(^{43}\) .  
+
+<|ref|>text<|/ref|><|det|>[[139, 631, 854, 800]]<|/det|>
+In this study, we showed that Wnt- mediated intercellular communication is actually involved in maintenance of the cell population in the epiblast/tailbud region. In this case, intercellular exchange of Wnt ligands is important to compensate for the disparity amplified via positive feedback from Wnt3a and Bra. An interesting question is whether a similar molecular network is involved in other events in which a community effect is exerted. Differences in the efficiency of cell signaling or the amplification efficiency of positive feedback loops probably generate differences in the features of cell populations. If this is the case, it will be important in future studies to identify key parameters in the molecular network to produce each of these events.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 120, 212, 136]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 142, 182, 156]]<|/det|>
+## Mice  
+
+<|ref|>text<|/ref|><|det|>[[139, 162, 852, 308]]<|/det|>
+Animal care and experiments were performed in accordance with guidelines for animal experimentation of the National Institutes for Natural Sciences. All animal experiments were approved by the Animal Research Committee of National Institutes for Natural Sciences. Mice were maintained in a light- and temperature- controlled room using a \(12\mathrm{h:}12\mathrm{h}\) light:dark cycle at \(21\pm 2^{\circ}\mathrm{C}\) . Embryos derived from timed matings were genotyped by PCR with DNA from yolk sacs or embryos. PCR conditions and primer sequences for Wnt3a KO \(^8\) and Wnt vs reporter \(^{28}\mathrm{mice}\) have been previously described.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 334, 369, 350]]<|/det|>
+## Cell culture and transfection  
+
+<|ref|>text<|/ref|><|det|>[[140, 355, 790, 414]]<|/det|>
+HEK293T cells and STF293 cells, which are HEK293 cells stably expressing Super 7x TOPFlash \(^{44}\) , were cultured at \(37^{\circ}\mathrm{C}\) in a 1:1 mixture of DMEM and Ham's F12 medium supplemented with \(8.3\%\) fetal bovine serum.  
+
+<|ref|>text<|/ref|><|det|>[[139, 439, 850, 586]]<|/det|>
+Plasmids were transfected into HEK 293T cells or STF293 cells using FuGENE6 transfection reagent (Roche). Culture medium was changed 6 h after transfection. At 24, 48, and 72 h after transfection, cells and culture medium were harvested for Western blotting and luciferase reporter assay. In co- culture experiments, HEK 293T cells transfected with each plasmid were collected 24 h after transfection and mixed 1:1 with STF293 cells. The luciferase reporter assay was performed 24 or 48 h after co- culture. Details of Western blotting and the luciferase reporter assay are described below.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 611, 310, 627]]<|/det|>
+## Plasmid construction  
+
+<|ref|>text<|/ref|><|det|>[[139, 632, 848, 799]]<|/det|>
+To generate plasmid constructs from which Wnt3a fused with human Frizzled 5 (hFzd5) is expressed, a DNA fragment encoding the full length of mouse Wnt3a protein fused to the N- terminus of hFzd5 mediated with 2xMyc tag (TSEQKLISEEDLNEMEQKLISEEDLRS) (Extended Data Fig.1a), was integrated between the ClaI and XbaI sites of pCSf107 plasmid vector, which carries the CMV IE94 promoter. This fusion protein was designed to remove the signal peptide of hFz5, resulting in direct fusion of the N- terminus of hFzd5 to the 2xMyc tag. DNA encoding the full length of mouse Wnt3a and EGFP fused Wnt3a(GFP- Wnt3a) was integrated into pCS2 plasmid vector.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 825, 509, 841]]<|/det|>
+## Western blotting and luciferase reporter assay  
+
+<|ref|>text<|/ref|><|det|>[[140, 846, 852, 884]]<|/det|>
+For detection of proteins in cultured cells and culture supernatant, samples were collected at the time points described in "Cell Culture and Transfection." To detect Wnt3a- Fzd5 proteins in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 118, 855, 330]]<|/det|>
+embryos, the area posterior to the newly formed somite of E8.5 embryos was cut and collected. SDS- PAGE was carried out according to a standard protocol<sup>45</sup>. Briefly, each sample was mixed with 2x sample buffer [4% SDS, 20% glycerol, 0.001% bromophenol blue and 0.125 M Tris HCl (pH 6.8)] and heated at 37°C for 1 h. Samples were electrophoresed using 10% polyacrylamide gels. After electroporation, proteins on the gel were transferred to a polyvinylidene difluoride (PVDF) membrane (Millipore). These membranes were treated overnight at 4°C with primary antibody (mouse anti-mouse Wnt3a antibody: Takada et al., Dev. Cell, 2006), followed by treatment with secondary antibodies (goat anti-mouse IgG: HRP conjugated, Promega W402B) for 1 h at room temperature. Finally, these proteins were visualized using an Enhanced Chemiluminescent Detection System (Amersham).  
+
+<|ref|>text<|/ref|><|det|>[[139, 354, 844, 500]]<|/det|>
+Luciferase reporter assay was performed according to the manufacturer's protocol (Dual- Glo Luciferase Assay System: Promega). Since STF293 cells contain a firefly Luciferase cDNA driven by eight tandem repeats of the TCF binding site, Wnt activity was quantified by monitoring activity of firefly Luciferase. Therefore, Luciferase is expressed depending on the strength of Wnt signaling. Renilla luciferase was used as an internal control to compensate for the mosaic nature of gene transfection. The activity of Luciferase was detected using a Luminometer (Turner Designs).  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 525, 468, 542]]<|/det|>
+## Generation of Wnt3a-Fzd5 knock-in mice  
+
+<|ref|>text<|/ref|><|det|>[[139, 546, 850, 710]]<|/det|>
+In Wnt3a- Fzd5 knock- in mice, a DNA fragment encoding the MYC- hFZD5 fragment was designed to be integrated just before the stop codon in exon4 of the mouse Wnt3a gene (Extended Data Fig. 1). The resulting protein expressed from this recombined locus is the same as that expressed in the cell culture experiment described above. To generate this knock- in allele, a pLSODN- 3- based plasmid containing a DNA fragment of myc- Fzd5 was co- injected with plasmids to express gRNA and Cas9 in fertilized eggs. The sequence of the gRNA is as follows: 5'- TTAGGAGCTCTCCTACTTGC- 3'. This gRNA was inserted into pX330. Genotyping was carried out by PCR using the following primers:  
+
+<|ref|>text<|/ref|><|det|>[[140, 714, 600, 844]]<|/det|>
+Wnt3a- Fzd5 5F, 5'- TGGTGCTTATCTGCCATTC- 3'; Wnt3a- Fzd5 WTF2, 5'- GTCACATGCACCTCAAGTGC- 3'; Wnt3a- Fzd5 7F, 5'- GGTGTGCCAGGAAATCACGG- 3'; Wnt3a- Fzd5 7R, 5'- GGACACCTGCTTGTGGTAGG- 3'; Wnt3a- Fzd5 WTR2, 5'- AGGATCCTTCCTAGCAGTCC- 3'; Wnt3a- Fzd5 4R, 5'- TTTCTACAGTTGACCGGCCTC- 3'.  
+
+<|ref|>text<|/ref|><|det|>[[140, 847, 840, 886]]<|/det|>
+The combination of primers used for PCR is shown in Extended Data Fig. 1. Fragments of 2,564- bp and 3,338- bp were expected from the 5'- region of wild- type Wnt3a and Wnt3a- Fzd5
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 118, 819, 158]]<|/det|>
+alleles, respectively. On the other hand, a 2,368- bp fragment and a 3,564- bp fragment were expected in the \(3^{\prime}\) - region of wild- type Wnt3a and Wnt3a- Fzd5 alleles, respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 184, 304, 200]]<|/det|>
+## In situ hybridization  
+
+<|ref|>text<|/ref|><|det|>[[139, 205, 857, 521]]<|/det|>
+Whole- mount in situ hybridization was performed using digoxigenin- labeled, antisense RNA probes. Briefly, embryos collected at the indicated stages were fixed with \(4\%\) paraformadhyde (PFA) overnight at \(4^{\circ}\mathrm{C}\) , washed with PBS, and treated with \(20\mu \mathrm{g / mL}\) of proteinase K for 5 min. These embryos were incubated in hybridization buffer ( \(50\%\) formamide, \(5\times\) SSC, \(1\%\) SDS, \(50\mu \mathrm{g / mL}\) tRNA) overnight at \(55^{\circ}\mathrm{C}\) . The next day, embryos were washed consecutively with \(5\times\) SSC, \(2\times\) SSC, and Tris- buffered saline with Tween 20 (TBST). Next, embryos were incubated with \(1\%\) sheep serum (Sigma) in TBST for \(1\mathrm{h}\) and then treated with a 1:500 dilution of antidigoxigenin- AP Fab fragments (Roche) overnight at \(4^{\circ}\mathrm{C}\) . The following day, embryos were washed with TBST and alkaline phosphatase buffer [ \(100\mathrm{mMNaCl}\) , \(100\mathrm{mM}\) Tris- HCl (pH 9.5), \(50\mathrm{mM}\) \(\mathrm{MgCl}_2\) , \(0.5\%\) Tween 20], and signals were developed using BM Purple (Roche). Wild- type and mutant embryos were stained for the same period in individual experiments. The following probes that have been reported previously were used: mouse Wnt3a<sup>46</sup>, mouse Brachyury and mouse Tbx6<sup>14</sup>, mouse Uncx4.1<sup>47</sup>, and human FRIZZLED5<sup>48</sup>. To generate a Sox2 probe, the first exon of mouse Sox2 was amplified from mouse genomic DNA and cloned to generate an antisense probe.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 547, 308, 562]]<|/det|>
+## Immunofluorescence  
+
+<|ref|>text<|/ref|><|det|>[[139, 567, 855, 756]]<|/det|>
+Whole- mount immunofluorescence was performed on embryos collected at the indicated stages. These embryos were fixed with \(4\%\) PFA overnight at \(4^{\circ}\mathrm{C}\) , and washed with PBS. Embryos were incubated overnight at \(4^{\circ}\mathrm{C}\) with the following primary antibodies: rabbit anti- Sox2 (polyclonal, Millipore, AB5603, 1:200) and goat anti- Brachyury (polyclonal, Santacruz, 17745, 1:1000). After washing with PBS, embryos were incubated overnight at \(4^{\circ}\mathrm{C}\) with DAPI (Dojindo) and following secondary antibodies: donkey anti- rabbit IgG (Alexa fluor 647 conjugated, 1:500) and donkey anti- goat IgG (Alexa fluor 647 conjugated, 1:500). Then, embryos were rinsed with PBS again and mounted on \(0.8\%\) LMP agarose for observation. Fluorescent images were acquired using an inverted confocal microscope (Leica SP8).  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 782, 329, 797]]<|/det|>
+## Retinoic acid treatment  
+
+<|ref|>text<|/ref|><|det|>[[140, 803, 838, 862]]<|/det|>
+To detect sensitivity to RA signaling in Wnt3a- Fz5 homozygotes, we crossed Wnt3a- Fz5 heterozygotes. RA or DMSO was injected into the abdominal cavities of female mice at E7.5, and embryos were harvested at E8.5.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 120, 352, 136]]<|/det|>
+## Wnt vis reporter detection  
+
+<|ref|>text<|/ref|><|det|>[[139, 140, 857, 328]]<|/det|>
+Wnt vis reporter detectionEmbryos were harvested at each stage and fixed overnight in \(4\%\) PFA. Then, they were stained with DAPI in \(1\%\) Triton X- 100 solution (1:2,000) for several hours or overnight and mounted in \(0.8\%\) LMP agarose. Fluorescent images were acquired using an inverted confocal microscope (Leica SP8). Fluorescence of this reporter in the nucleus was measured in individual cells in a photon- counting mode. Epiblast cells in the region lateral to the node were individually analyzed in a single confocal plane. The area of the nuclei in each cell was identified by DAPI staining. The GFP intensity in the identified area was counted. For the cells with the top \(10\%\) GFP intensity in region II, the relative GFP intensities of the cells in contact with those high- GFP cells were summarized.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 376, 440, 392]]<|/det|>
+## Quantification and statistical analysis  
+
+<|ref|>text<|/ref|><|det|>[[140, 397, 845, 458]]<|/det|>
+Quantification and statistical analysisStatistical analyses were performed using Excel and R software. Differences were assessed for statistical significance using T- tests. p- values \(< 0.05\) were considered statistically significant. Error bars in graphs indicate the standard deviation of each group.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 484, 309, 499]]<|/det|>
+## Mathematical Model  
+
+<|ref|>text<|/ref|><|det|>[[139, 504, 857, 628]]<|/det|>
+We considered an ideal situation in which \(L_{x}\times L_{y}\) epiblast cells align on a regular square lattice such that the x and y axes coincide with the left- right and the anterior- posterior axes (Fig. 7a). Each cell has its Wnt signaling activity as a variable \(W_{ij}\) , where \(i\) and \(j\) are the indexes of the cell on the x- y coordinate. The activity \(W_{ij}\) changes in time through the production, degradation, and intercellular diffusion of Wnt molecules (Fig. 7a), and its time evolution is given by Langevin equation:  
+
+<|ref|>equation<|/ref|><|det|>[[160, 636, 835, 686]]<|/det|>
+\[\frac{d}{dt} W_{ij} = a + \alpha \frac{W_{ij}^{2}}{K^{2} + W_{ij}} -\beta W + D\sum_{k,k'}\left(W_{k,k'} - W_{ij}\right) - d_{RA}(j,t)W_{ij} + \gamma \xi_{ij}(t) \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[139, 717, 857, 885]]<|/det|>
+The first and second terms represent the production of Wnt molecules, a constant production ( \(1^{\mathrm{st}}\) term), and autocatalytic reaction with half saturation concentration \(K\) (2nd term), while the third term represents degradation with a degradation constant \(\beta\) . The fourth term denotes intercellular diffusion of Wnt molecules with diffusion constant \(D\) . The summation \(\sum_{k,k'\in n.n.}(\cdot)\) indicates the summation with respect to the nearest- neighbor cells around \(W_{ij}\) (i.e., \(W_{i\pm 1,j}\) and \(W_{i,j\pm 1}\) ). RA- dependent repression is given by the fifth term. As long as the RA diffusion does not reach the region of epiblast cells (t< t_RA), the value of d_RA(j,t) is set to 0; \(d_{RA}(j,t) = 0\) . After RA diffusion reaches this region ( \(t\geq t_{RA}\) ), \(d_{RA}(j,t)\) creates the following
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 118, 820, 158]]<|/det|>
+time- independent spatial gradient that increases linearly along y- axis (the anterior- posterior axis) with a slope \(\Delta\) :  
+
+<|ref|>equation<|/ref|><|det|>[[406, 161, 590, 181]]<|/det|>
+\[d_{RA}(j) = \beta_{1} - \beta +j \Delta .\]  
+
+<|ref|>text<|/ref|><|det|>[[139, 204, 812, 254]]<|/det|>
+The last term in Eq.(1) denotes the noise term: \(\gamma\) and \(\xi_{ij}(t)\) indicate the noise strength and independent Gaussian white noise with zero mean and variance \(\langle \xi_{ij}(t)\xi_{i^{\prime}j^{\prime}}(t^{\prime}) \rangle =\)  
+
+<|ref|>text<|/ref|><|det|>[[139, 275, 780, 300]]<|/det|>
+\(\delta \left(t - t^{'}\right)\delta_{i,i^{\prime}}\delta_{j,j^{\prime}}\) , respectively. Cell division events are also considered a stochastic  
+
+<|ref|>text<|/ref|><|det|>[[139, 310, 830, 457]]<|/det|>
+Poisson process in the model. Each cell randomly divides at a constant rate \(\lambda\) . By the cell division, surrounding cells are pushed stochastically toward either the left (in the negative direction of the x- axis with probability 1/4), right (along the positive direction of x- axis with probability 1/4) or upward (along the positive direction of y- axis with probability 1/2). For instance, when the left is chosen in the division of cell with \(x = i\) and \(y = j\) , all cells with \(x < i\) and \(y = j\) move toward the left. The extruded cell outside the \(L_{x} \times L_{y}\) lattice is removed.  
+
+<|ref|>text<|/ref|><|det|>[[139, 481, 848, 693]]<|/det|>
+Equation (1) is numerically solved by the Euler- Maruyama method with \(\Delta t = 10^{- 5}\) , while presence or absence of a cell division event is determined at each time step following the probability \(\lambda L_{x}L_{y}\Delta t\) . The time unit is normalized so that \(\lambda = 1\) , while the length scale is normalized by cell length (the lattice size). The diffusion constant \(D\) is set to \(D = 8.0\) for paracrine (+) cells (Fig. 7b, d and e) and \(D = 0\) for paracrine (- ) cells (Fig. 7c and f). The production rate of Wnt molecules was chosen as \(a = 30.0\) for paracrine \((\pm)\) cells (Fig. 7b, c, e and f) and \(a = 26.8\) for paracrine (+)- production \((\downarrow)\) cells (Fig.7d). The parameter \(\beta_{1}\) that represents basal degradation for \(t \geq t_{RA}\) was chosen as \(\beta_{1} = 250.0\) for RA(- ) situation (Fig.7b- d) and \(\beta_{1} = 252.0\) for RA(+) situation (Fig.7e and f). For other parameters, the following values were used: \(L_{x} = L_{y} = 50\) , \(\alpha = 522.0\) , \(K = 1.12\) , \(\beta = 220.0\) , \(\Delta = 0.2\) , \(\gamma = 3.0\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[141, 719, 296, 735]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[139, 739, 852, 864]]<|/det|>
+We thank the Spectrography and Bioimaging Facility of the NIBB Core Research Facilities for their technical support. We also thank Drs. Takahashi and Mizuno at the University of Tsukuba for generating Wnt3a- Fzd5 knock- in mice using CRISPR/Cas9- mediated genome editing and Dr. Fujimori at NIBB for providing mice and technical support. Dr. Aoki at NIBB and ExCELLS and all members of S.T.'s laboratory are gratefully acknowledged for helpful discussions.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 120, 311, 136]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[139, 141, 857, 266]]<|/det|>
+Y.H. performed the majority of experiments, participated in their planning, and wrote the text. N.S. formulated the mathematical concept, conducted computer modeling, and wrote the text. Y.M. participated in the planning of the experiments and discussion of the results. T.S. participated in generation of mouse mutants. T.T. generated Wnt reporter mice. H.N. participate in modeling. S.T. formulated the initial key hypothesis, organized all the work, planned experiments, and wrote the text. All authors reviewed and approved the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 292, 211, 307]]<|/det|>
+## Funding  
+
+<|ref|>text<|/ref|><|det|>[[140, 312, 861, 435]]<|/det|>
+This work was supported by the following programs: Grants- in- Aid for Scientific Research (B), 18H02454 and 21H02498 to ST, Grants- in- aid for Scientific Research on Innovative Areas, 24111002, 17H05782, 19H04797 to ST, from the Japan Society for the Promotion of Science. Additional support came from grants from National Institutes of Natural Sciences (NINS Joint Research Program to ST) and the Cooperative Study Program of Exploratory Research Center on Life and Living Systems (ExCELLS; program Nos. 21- 102 to HN and NS).  
+
+<|ref|>sub_title<|/ref|><|det|>[[141, 492, 375, 508]]<|/det|>
+## Competing financial interests  
+
+<|ref|>text<|/ref|><|det|>[[140, 513, 554, 530]]<|/det|>
+The authors declare no competing or financial interests.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 120, 230, 135]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[137, 140, 860, 888]]<|/det|>
+1. Stern, C. D. et al. Head-tail patterning of the vertebrate embryo: One, two or many unresolved problems? International Journal of Developmental Biology 50, 3-15 (2006).  
+2. Tzouanacou, E., Wegener, A., Wymeersch, F. J., Wilson, V. & Nicolas, J. F. Redefining the Progression of Lineage Segregations during Mammalian Embryogenesis by Clonal Analysis. Developmental Cell 17, 365-376 (2009).  
+3. Cambray, N. & Wilson, V. Two distinct sources for a population of maturing axial progenitors. Development 134, 2829-2840 (2007).  
+4. Cambray, N. & Wilson, V. Axial progenitors with extensive potency are localised to the mouse chordoneural hinge. Development 129, 4855-4866 (2002).  
+5. Tam, P. P. & Beddington, R. S. The formation of mesodermal tissues in the mouse embryo during gastrulation and early organogenesis. Development (Cambridge, England) 99, 109-26 (1987).  
+6. Wilson, V., Olivera-Martinez, I. & Storey, K. G. Stem cells signals and vertebrate body axis extension. Development vol. 136 1591-1604 (2009).  
+7. Wilson, V. & Beddington, R. S. Cell fate and morphogenetic movement in the late mouse primitive streak. Mechanisms of development 55, 79-89 (1996).  
+8. Takada, S. et al. Wnt-3a regulates somite and tailbud formation in the mouse embryo. Genes & development 8, 174-89 (1994).  
+9. Cunningham, T. J., Kumar, S., Yamaguchi, T. P. & Duester, G. Wnt8a and Wnt3a cooperate in the axial stem cell niche to promote mammalian body axis extension. Developmental dynamics: an official publication of the American Association of Anatomists 244, 797-807 (2015).  
+10. Zhao, X. & Duester, G. Effect of retinoic acid signaling on Wnt/ \(\beta\) -catenin and FGF signaling during body axis extension. Gene Expression Patterns 9, 430-435 (2009).  
+11. Liu, P. et al. Requirement for Wnt3 in vertebrate axis formation. Nature genetics 22, 361-5 (1999).  
+12. Tenin, G. et al. The chick somitogenesis oscillator is arrested before all paraxial mesoderm is segmented into somites. BMC Developmental Biology 10, (2010).  
+13. Rivera-Perez, J. A. & Magnuson, T. Primitive streak formation in mice is preceded by localized activation of Brachyury and Wnt3. Developmental Biology 288, 363-371 (2005).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 118, 856, 888]]<|/det|>
+14. Yamaguchi, T. P., Takada, S., Yoshikawa, Y., Wu, N. & McMahon, A. P. T (Brachyury) is a direct target of Wnt3a during paraxial mesoderm specification. Genes & development 13, 3185-90 (1999).15. Anderson, M. J. et al. TCreERT2, a Transgenic Mouse Line for Temporal Control of Cre-Mediated Recombination in Lineages Emerging from the Primitive Streak or Tail Bud. PLoS ONE 8, (2013).16. Garriock, R. J. et al. Lineage tracing of neuromesodermal progenitors reveals novel wnt-dependent roles in trunk progenitor cell maintenance and differentiation. Development 142, 1628-1638 (2015).17. Imuta, Y., Kiyonari, H., Jang, C. W., Behringer, R. R. & Sasaki, H. Generation of knock-in mice that express nuclear enhanced green fluorescent protein and tamoxifen-inducible Cre recombinase in the notochord from Foxa2 and T loci. Genesis 51, 210-218 (2013).18. Koch, F. et al. Antagonistic Activities of Sox2 and Brachyury Control the Fate Choice of Neuro-Mesodermal Progenitors. Developmental Cell 42, 514-526.e7 (2017).19. Yoshikawa, Y., Fujimori, T., McMahon, A. P. & Takada, S. Evidence that absence of Wnt-3a signaling promotes neuralization instead of paraxial mesoderm development in the mouse. Developmental biology 183, 234-42 (1997).20. Dunt, W. C. et al. Wnt3a/ \(\beta\) -catenin signaling controls posterior body development by coordinating mesoderm formation and segmentation. Development 135, 85-94 (2008).21. Dunt, W. C., Kennedy, M. W. L., Chalamalasetty, R. B., Campbell, K. & Yamaguchi, T. P. Transcriptional profiling of Wnt3a mutants identifies Sp transcription factors as essential effectors of the Wnt/ \(\beta\) -catenin pathway in neuromesodermal stem cells. PLoS ONE 9, e87018 (2014).22. Galceran, J., Hsu, S. C. & Grosschedl, R. Rescue of a Wnt mutation by an activated form of LEF-1: regulation of maintenance but not initiation of Brachyury expression. Proceedings of the National Academy of Sciences of the United States of America 98, 8668-73 (2001).23. Martin, B. L. & Kimelman, D. Regulation of Canonical Wnt Signaling by Brachyury Is Essential for Posterior Mesoderm Formation. Developmental Cell 15, 121-133 (2008).24. Martin, B. L. & Kimelman, D. Brachyury establishes the embryonic mesodermal progenitor niche. Genes and Development 24, 2778-2783 (2010).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 119, 855, 867]]<|/det|>
+25. Shinozuka, T., Takada, R., Yoshida, S., Yonemura, S. & Takada, S. Wnt produced by stretched roof-plate cells is required for the promotion of cell proliferation around the central canal of the spinal cord. Development 146, (2019).26. Greco, T. L. et al. Analysis of the vestigial tail mutation demonstrates that Wnt-3a gene dosage regulates mouse axial development. Genes & development 10, 313-24 (1996).27. Tsakiridis, A. et al. Distinct Wnt-driven primitive streak-like populations reflect in vivo lineage precursors. Development 141, 1209-1221 (2014).28. Takemoto, T. et al. R26-WntVis reporter mice showing graded response to Wnt signal levels. Genes to Cells 21, 661-669 (2016).29. Yamaguchi, T. P. Genetics of Wnt signaling during early mammalian development. Methods in Molecular Biology 468, 287-305 (2008).30. Abu-Abed, S. et al. The retinoic acid-metabolizing enzyme, CYP26A1, is essential for normal hindbrain patterning, vertebral identity, and development of posterior structures. Genes & development 15, 226-40 (2001).31. Briscoe, J. & Small, S. Morphogen rules: Design principles of gradient-mediated embryo patterning. Development (Cambridge) vol. 142 3996-4009 (2015).32. Gurdon, J. B. & Bourillot, P. Y. Morphogen gradient interpretation. Nature 413, 797-803 (2001).33. Christian, J. L. & Moon, R. T. Interactions between Xwnt-8 and Spemann organizer signaling pathways generate dorsoventral pattern in the embryonic mesoderm of Xenopus. Genes and development 7, 13-28 (1993).34. Petersen, C. P. & Reddien, P. W. Wnt Signaling and the Polarity of the Primary Body Axis. Cell vol. 139 1056-1068 (2009).35. Shimizu, T., Bae, Y. K., Muraoka, O. & Hibi, M. Interaction of Wnt and caudal-related genes in zebrafish posterior body formation. Developmental Biology 279, 125-141 (2005).36. Mitrophanov, A. Y. & Groisman, E. A. Positive feedback in cellular control systems. BioEssays: news and reviews in molecular, cellular and developmental biology 30, 542-55 (2008).37. Abu-Abed, S. et al. Developing with lethal RA levels: Genetic ablation of Rarg can restore the viability of mice lacking Cyp26a1. Development vol. 130 1449-1459 (2003).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[137, 118, 857, 761]]<|/det|>
+38. Niederreither, K., McCaffery, P., Dräger, U. C., Chambon, P. & Dollé, P. Restricted expression and retinoic acid-induced downregulation of the retinaldehyde dehydrogenase type 2 (RALDH-2) gene during mouse development. Mechanisms of development 62, 67-78 (1997).39. MacLean, G. et al. Cloning of a novel retinoic-acid metabolizing cytochrome P450, Cyp26B1, and comparative expression analysis with Cyp26A1 during early murine development. Mechanisms of development 107, 195-201 (2001).40. Sakai, Y. et al. The retinoic acid-inactivating enzyme CYP26 is essential for establishing an uneven distribution of retinoic acid along the anterio-posterior axis within the mouse embryo. Genes & development 15, 213-25 (2001).41. Gurdon, J. B. A community effect in animal development. Nature 336, 772-4 (1988).42. Gurdon, J. B., Lemaire, P. & Kato, K. Community effects and related phenomena in development. Cell 75, 831-4 (1993).43. Saka, Y., Lhoussaine, C., Kuttler, C., Ullner, E. & Thiel, M. Theoretical basis of the community effect in development. BMC Systems Biology 5, (2011).44. Tsukiyama, T. et al. Molecular Role of RNF43 in Canonical and Noncanonical Wnt Signaling. Molecular and Cellular Biology 35, 2007-2023 (2015).45. Laemmli, U. K. Cleavage of structural proteins during the assembly of the head of bacteriophage T4. Nature 227, 680-5 (1970).46. Roelink, H. & Nusse, R. Expression of two members of the Wnt family during mouse development—restricted temporal and spatial patterns in the developing neural tube. Genes & development 5, 381-8 (1991).47. Mansouri, A. et al. Paired-related murine homeobox gene expressed in the developing sclerotome, kidney, and nervous system. Developmental dynamics: an official publication of the American Association of Anatomists 210, 53-65 (1997).48. Liu, C., Wang, Y., Smallwood, P. M. & Nathans, J. An essential role for Frizzled5 in neuronal survival in the parafascicular nucleus of the thalamus. Journal of Neuroscience 28, 5641-5653 (2008).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 120, 266, 136]]<|/det|>
+## Figure Legends  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 140, 825, 157]]<|/det|>
+## Fig. 1. In vitro activity of WNT3A-FZD5 and generation of Wnt3a-Fzd5 knock-in mice  
+
+<|ref|>text<|/ref|><|det|>[[137, 160, 857, 652]]<|/det|>
+(a) Schematic fig. of WNT3A-FZD5 protein, comparing it with WNT3A and GFP-WNT3A. (b, c) Wnt signaling activity of each construct shown in A. Wnt signaling activity was monitored in HEK293T (STF293) cells stably expressing the SuperTOPFLASH reporter. In (b), Wnt activity was monitored at 24, 48, and 72 h after transfection of each plasmid into STF293 cells. In (c), paracrine Wnt activity was monitored in co-cultures of Wnt-expressing HEK293T cells with STF293 cells at 24 and 48 h after transfection. Differences were assessed for statistical significance using a T-test; *** P < 0.001; ** P < 0.01; * P < 0.05; P > 0.05; NS (not statistically significant). Error bars in the graph indicate the standard deviation of each group. (d-g) Expression of mouse Wnt3a and human Fzd5 in Wnt3a-Fzd5 heterozygous (Wnt3a<sup>+/Fzd5</sup>) embryos. Whole-mount in situ hybridization was carried out using probes of mouse Wnt3a (d, e) or human Fzd5 (f, g) in wild type (d, f) and Wnt3a<sup>+/Fzd5</sup> (e, g) embryos at E10.5. Images are highlighted on the tailbud (d, e, f, g) and the roof plate of neural tube (d', e', f', g'). Numbers of stained embryos are indicated by "n=" in the images. (h) Western blotting analysis of proteins prepared from E8.5 embryos with anti-mouse Wnt3a antibody. Samples prepared from two embryos were applied to each lane. Note that bands with the predicted molecular weight of WNT3A-FZD5 were detected both in Wnt3a-Fzd5 heterozygous (+/Fzd5) and homozygous (Fzd5/Fzd5) embryos. (i-k) Sagittal views of wt (i), Wnt3a<sup>+/Fzd5</sup> (j) and Wnt3a<sup>Fzd5/Fzd5</sup> (k) embryos at E10.5. i', j', and k' are magnified images of i, j, and k, respectively. i'', j'', and k'' are drawings of the images of i', j', and k', respectively. (l-n) Transverse sections of the neural tube of WT (l, l', l''), Wnt3a<sup>+/Fzd5</sup> (m, m', m'') and Wnt3a<sup>Fzd5/Fzd5</sup> (n, n', n'') embryos at E11.5. Sections at the forelimb (l, m, n), the intermediate between fore and hindlimb (l', m', n') and the hindlimb (l'', m'', n'') levels are shown. Scale bars: 1 mm (d-g', i-k''), 100 μm (l-n''), 200 μm (n''). HL: Hindlimb. FL: forelimb.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 675, 815, 713]]<|/det|>
+## Fig. 2. Expression of mesoderm and neural marker genes in Wnt3a-Fzd5 homozygous embryos  
+
+<|ref|>text<|/ref|><|det|>[[139, 717, 858, 885]]<|/det|>
+(a-v) Expression of mesoderm and neural marker genes in embryos at E10.5. Whole- mount in situ hybridization was carried out using probes of Bra (a-f), Tbx6 (g-l), Uncx4.1 (m-q), and Sox2 (r-v) in wt & Wnt3a<sup>+/Fzd5</sup> (a, g, m, r), Wnt3a<sup>Fzd5/Fzd5</sup> (b, h, n, s), Wnt3a<sup>+/v</sup> (c, i, o, t), Wnt3a<sup>v</sup> (d, j, p, u), Wnt3a<sup>+/</sup> (e, k, q, v) and Wnt3a<sup>+/</sup> (f, l) embryos at E10.5. Red dotted lines indicate tail regions. (w-ad) Expression of mesoderm and neural marker genes in embryos at E12.5. Whole- mount in situ hybridization was carried out using probes of Bra (w, x, y), and Tbx6 (z, aa, ab) in WT & Wnt3a<sup>+/Fzd5</sup> (w, z), Wnt3a<sup>Fzd5/Fzd5</sup> (x, y, aa, ab) embryos at E12.5. Tail regions of stained embryos were cut out and shown in y and ab. x' and aa are drawings of the images of x and aa
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 118, 823, 157]]<|/det|>
+respectively. Numbers of stained embryos are indicated by "n=" in the images. Scale bars: 1 mm. HL: Hindlimb.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 183, 830, 222]]<|/det|>
+## Fig. 3. Differentiation of Wnt-positive progenitor cells in the neural tube and somites in Wnt3a-Fzd5 homozygous embryos.  
+
+<|ref|>text<|/ref|><|det|>[[139, 225, 852, 501]]<|/det|>
+(a) Experimental procedure. Cells once activated by Wnt signaling, which is monitored by Axin2-CreERT2 expression, were eternally labelled by expression of tdTomato. Tamoxifen (TM) was injected into pregnant females at 8.5 dpc and embryos were fixed at E10.5. (b, c) Distribution of tdTomato-labelled cells at the posterior hindlimb level in Wnt3a<sup>+/Fzd5</sup> (b) and Wnt3a<sup>Fzd5/Fzd5</sup> (c) embryos at E10.5. Merged images with DAPI-staining are also indicated (b', c'). Neural tube and dermomyotome, which is derived from somite, are outlined with white and orange dotted lines, respectively. Squares framed by green dotted lines indicate the area around the nephric duct. The percentage of tdTomato-positive cells (d, e) and total cell number (f, g) in somite (d, f), nephric duct (e) and neural tube (g) at the posterior hindlimb level in Wnt3a<sup>+/Fzd5</sup> and Wnt3a<sup>Fzd5/Fzd5</sup> embryos at E10.5. Numbers or percentages of labelled cells (mean±s.d.) per section are shown. Differences were assessed for statistical significance using a T-test; ***, P < 0.001; **, P < 0.01; NS, not statistically significant (P > 0.05). Error bars in the graph mean the standard deviation of each group. Scale bars: 100 μm  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 526, 571, 544]]<|/det|>
+## Fig. 4. NMP cells in Wnt3a-Fzd5 homozygous embryos  
+
+<|ref|>text<|/ref|><|det|>[[139, 547, 857, 886]]<|/det|>
+(a- l) Whole-mount staining of Wnt3a- Fzd5 heterozygous (a, c, e, g, j, l, n) and homozygous (b, d, f, h, k, m, o) embryos at E8.75. Maximum intensity projection images of posterior ends of embryos stained with anti- SOX2 (magenta) and anti- BRA (green) antibodies are shown in A and B. To quantify the number of SOX2/BRA double- positive cells, single- plane images of medial (I) and lateral (II) regions lateral to the node were analyzed (c- p). Images of DAPI staining (blue; c, d, j, k), and merged images of staining with anti- SOX2 (magenta) and anti- BRA (green) antibodies (e, f, l, m) are shown. Summarized schematic fig.s (g, h, n, o) and diagrams (i, p) are also shown. The size of the medial and lateral regions is 50 μm x 100 μm. Two embryos were examined for each genotype. (q- x) Whole- mount staining of Wnt3a<sup>+/Fzd5</sup> (q, u), Wnt3a<sup>Fzd5/Fzd5</sup> (r, v), Wnt3a<sup>+/vt</sup> (s, w), and Wnt3a<sup>wt/</sup> (t, x) embryos at E11.5. Maximum intensity projection images of posterior ends of embryos stained with DAPI (blue; q- t), and with anti- SOX2 (magenta) and anti- BRA (green) antibodies (q'- t'), are shown. Single- plane images of the areas indicated with yellow- lined boxes in q'- t' are magnified in u- x, respectively. Images of staining with anti- SOX2 (magenta; u- x) and anti- BRA (green; u'- x') antibodies, as well as merged images (u'- x") are shown. The yellow- lined box is a square with one side = 100 μm. Arrowheads in v" indicate a small number of Sox2/Bra- positive cells. Note that there are no
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 118, 808, 159]]<|/det|>
+SOX2 and BRA double- positive cells in Wnt3a<sup>wt</sup>(t, x). The number of stained embryos is indicated by "n=". Scale bars: \(100 \mu \mathrm{m}\) (a, b, q- t)  
+
+<|ref|>text<|/ref|><|det|>[[138, 183, 860, 543]]<|/det|>
+Fig. 5. Wnt signaling in the epiblast cell population of Wnt3a- Fzd5 homozygous embryos Wnt signaling activity in individual epiblast cells was visualized using mouse embryos carrying an EGFP- reporter gene, expression of which is specifically activated by Wnt signaling. The observation scheme of embryos at cylinder stage(E7.0 and E7.5 : a) and post- somiteogenesis stage(E8.5 and E8.75 : f) . Eye marks in (a, f) indicate the direction of observation. Blue boxes in (f) indicate somites. Wnt signaling activity was monitored in Wnt3a- Fzd5 heterozygous (b, d, g, k, q) and homozygous (c, e, h, l, r) embryos at E7.0 (b, c), E7.5 (d, e), E8.0 (g, h), E8.75 (k, l), and E9.5(q, r). Wnt signaling activity was also visualized in \(+ / vt\) (i, m, s) and \(vt / - (j, n, t)\) embryos at E8.0 (i, j), E8.75 (m, n), and E9.5(s, t). Each embryos, magnified images of the areas indicated by boxes. Note that Wnt signaling activity is not obviously changed in Wnt3a- Fzd5 homozygous embryos at E7.0 (b, c) or at E7.5 (d, e). In E8.0 and E8.75 embryos, magnified images of the CLE in (f) are shown in each genotyped embryo. (Areas = \(100 \times 100 \mathrm{mm}\) .) The magnified images were taken at a single confocal plane while the others were processed by maximum intensity projection. GFP intensity in individual cells in CLE was quantified in each genotyped embryo at E8.0 (o) and E8.75 (p). Two embryos were examined for each genotype. Box plots indicate the first and third quartiles and the median. Scale bars: \(100 \mu \mathrm{m}\) . The star in the E9.5 shows the nephric duct.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 567, 830, 606]]<|/det|>
+## Fig. 6. Effect of retinoic acid on the epiblast cell population of Wnt3a-Fzd5 homozygous embryos  
+
+<|ref|>text<|/ref|><|det|>[[138, 610, 855, 885]]<|/det|>
+(a- g) Analysis of retinoic acid (RA)- treated embryos at E8.75. Experimental schemes are shown in (a, b). Dorsal images of DMSO (c, d) or \(10 \mathrm{mM}\) RA (e, f) treated Wnt3a- Fzd5 heterozygous (c, e) and homozygous (d, f) embryos at E8.75 stained with DAPI (blue). Results of quantification of the width at NSB (g) and the length posterior to NSB (h) in each genotyped embryo are shown. Note that RA treatment enhances the abnormality in gross morphology specifically in Wnt3a- Fzd5 homozygous embryos. Red arrows indicate the width at NSB while orange arrows indicate the length posterior to NSB. Differences were assessed for statistical significance using a T- test; \(*** \mathrm{P} < 0.001\) ; \(** \mathrm{P} < 0.01\) ; \(* \mathrm{P} < 0.05\) ; \(\mathrm{P} > 0.05\) ; n.s. (not statistically significant). Error bars in the graph indicate the standard deviation of each group. Scale bars: \(100 \mu \mathrm{m}\) . (i- m) Analysis of retinoic acid (RA)- treated embryos at E8.5. These embryos were treated with RA 7.5 days post coitum. Wnt signaling activity in individual epiblast cells was visualized as shown in Fig. 5(f). Dorsal images, processed by maximum intensity projection of DMSO- (i, k) or \(10 \mathrm{mM}\) RA- (j, l) treated Wnt3a- Fzd5 heterozygous (i, j)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 118, 845, 266]]<|/det|>
+and homozygous (k, l) embryos at E8.5 are shown. Images of a single confocal plane in CLE. The size of these areas is \(100 \mu \mathrm{m} \times 100 \mu \mathrm{m}\) and their positions in the epiblast are identical to those shown in Fig. 5(f). GFP intensity in individual cells in CLE is summarized in (m). Two embryos were examined for each genotype. Box plots indicate the first and third quartiles and the median. Differences were assessed for statistical significance using a wilcoxon signed- rank test; \(\mathrm{***P< 0.001}\) ; \(\mathrm{**P< 0.01}\) ; \(\mathrm{*P< 0.05}\) ; \(\mathrm{P > 0.05}\) ; n.s. (not statistically significant). In (a) and (h), PS indicates the primitive streak and blue boxes indicate somites. Scale bars: \(100 \mu \mathrm{m}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 290, 839, 329]]<|/det|>
+## Fig. 7. Mathematical Model for Examining the Effect of Intercellular Communication in Cell Populations  
+
+<|ref|>text<|/ref|><|det|>[[139, 330, 858, 888]]<|/det|>
+(a) Schematic diagram showing parameters used in the model. We assumed a virtual space corresponding to the cell sheet of the epiblast. This virtual space is divided into \(50 \times 50\) sections along the antero-posterior and medio-lateral axes. Each section corresponds to a single cell in the epiblast. Wnt activity (W) is determined by parameters such as the rate of production and degradation of Wnt protein, the rate of amplification or reduction by positive feedback, the rate of intercellular exchange of Wnt protein, the rate of inhibition by RA, and fluctuating noise that affects Wnt activity. It is assumed that cell division occurs randomly and that dividing daughter cells are extruded in one section in either the left, right, or anterior direction in a 1:1:2 ratio. (b-f) Spatial patterns of Wnt activity in a virtual sheet of cells. Examples of the spatial pattern in the presence (b, d, e) or absence (c, f) of the paracrine function of Wnt are shown at the same time point (mean division time \(t = 3.00\) ) after addition of RA \((t = 0)\) . In the condition of D, the Wnt production rate is reduced (see Method). The spatial patterns of Wnt activity were calculated in the absence (b-d) and presence (e, f) of uniformly supplied RA. (g) Time course of Wnt-positive cells in a virtual sheet of cells. The time course of the proportion of Wnt-positive cells ( \(>50\%\) of maximum activity) at the same spatial level \((y = 35\) in b-d) along the anterior-posterior axis in a virtual sheet of cells is shown. Orange and blue lines indicate the result with and without the paracrine function of Wnt, respectively. A green line indicates the result obtained in the condition where the Wnt production rate is reduced in the presence of the paracrine function of Wnt. Solid and dashed lines indicate results obtained in the absence and presence of uniformly supplied RA, respectively. (h) Schematic representation showing the effect of Wnt paracrine in the epiblast cell population. Prior to somite formation (E7.0-E8.0), Wnt activity in each epiblast cell is dramatically increased by positive feedback regulation mediated by Wnt3a and Bra. During this period, no obvious difference was observed between control and embryos lacking paracrine Wnt signaling (Wnt3a-Fzd5
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[139, 118, 850, 245]]<|/det|>
+homozygous embryos). After the onset of somite formation (after E8.0), the number of Wnt weak cells was increased by the antagonistic effect of RA, which is synthesized in somites, but a small number of Wnt- strong cells remain for a long period in embryos lacking paracrine Wnt signaling (paracrine (- )). This increased heterogeneity in Wnt signaling is compensated for by intercellular exchange of Wnt ligands between epiblast cells (paracrine (+)).  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 269, 758, 287]]<|/det|>
+## Extended Data Fig. 1. WNT3A-FZD5 was not detected in culture supernatant.  
+
+<|ref|>text<|/ref|><|det|>[[140, 290, 852, 394]]<|/det|>
+Western blotting analysis of cell lysate (a) and culture supernatant (b) prepared from WNT3A and WNT3A- FZD5 expressing HEK293T cells at 24, 48, and 72 h after transfection. While the expression level of WNT3A- FZD5 was similar to WNT3A in the cell lysate, WNT3A- FZD5 was not detectable in culture supernatant. Red and blue arrowheads indicate bands corresponding to the predicted molecular weights of WNT3A- FZD5 and WNT3A, respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 418, 640, 436]]<|/det|>
+## Extended Data Fig. 2. Generation of Wnt3a-Fz5 knock-in mice.  
+
+<|ref|>text<|/ref|><|det|>[[139, 439, 848, 650]]<|/det|>
+Extended Data Fig. 2. Generation of Wnt3a- Fz5 knock- in mice.(a- f) Generation the Wnt3a- Fz5 knock- in allele. A schematic fig. indicates the mouse Wnt3a locus and the Wnt3a- Fzd5 knock- in allele is shown in (a). In the knock- in allele, human Frizzled5 (blue) fused with 2 myc tags (green) is inserted at the C- terminus of mouse Wnt3a. The knock- in event was confirmed by PCR analysis using the primer sets indicated in b. The results of PCR analyses are shown (c- f). Prime sets are indicated on the upper side of each fig.. Band sizes indicated by colored arrowheads correspond to the predicted sizes shown in b. (g- j) Whole image of wt (g, i) and \(Wnt3a^{+ / Fzd5}\) (h, j) embryos at E10.5 hybridized with Wnt3a (g, h) or \(hFzd5\) (i, j) probes. Magnified images of posterior bodies and dorsal views of these embryos are shown in Fig. 1d- g. Scale bars: 1 mm. (k) The proportion of individuals of each genotype during embryonic development and immediately after birth.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 675, 820, 713]]<|/det|>
+## Extended Data Fig. 3. The source of ventral neural cells loses Wnt signaling after E7.5 during development of Wnt3a-Fzd5 heterozygous embryos.  
+
+<|ref|>text<|/ref|><|det|>[[139, 717, 858, 842]]<|/det|>
+(a) Experimental procedure. Cells once activated by Wnt signaling, which is monitored by Axin2-CreERT2 expression, were eternally labelled by expression of tdTomato. Tamoxifen (TM) was injected to pregnant females at 7.5 dpc (b, d) or 8.5 dpc (c, e) and embryos were fixed at E10.5. Whole-mount bright field images (b, c) and tdTomato staining are also indicated (b', c'). Distribution of tdTomato-labelled cells at the posterior hindlimb level in \(Wnt3a^{+ / Fzd5}\) embryos at E10.5(d, e). Scale bars: \(100 \mu \mathrm{m}\)
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[140, 119, 802, 157]]<|/det|>
+## Extended Data Fig. 4. Synergistic effect of Wnt3a-Frizzled and Bra on the posterior development of body axis elongation  
+
+<|ref|>text<|/ref|><|det|>[[139, 159, 857, 285]]<|/det|>
+Wnt3a+/Fzd5; Bra+/+(a), Wnt3a+/+; Bra+/-(b) and Wnt3a+/Fzd5; Bra+/-(c) embryos stained by whole- mount in situ hybridization using the Sox2 probe are shown. Embryos were fixed at E11.5. a', b', and c' are magnified images of a, b, and c, respectively. Note that Wnt3a- Fzd5 and Bra compound heterozygous embryos (Wnt3a+/Fzd5; Bra+/-(c)) impair the posterior development of body axis elongation while embryos heterozygous for either of them appear normal. Scale bars: 1 mm.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 311, 607, 328]]<|/det|>
+## Extended Data Fig. 5. Characteristics of Wnt3a<sup>w/- </sup>embryos  
+
+<|ref|>text<|/ref|><|det|>[[139, 331, 856, 522]]<|/det|>
+(a- c) Sagittal views of wt (a), Wnt3a+/+(b) and Wnt3a<sup>w/- </sup> (c) embryos at the E11.5. a', b', and c' are magnified images of a, b, and c, respectively. a", b", and c" are drawings of the images of a', b', and c', respectively. (d- f) Wnt3a expression was detected by whole- mount in situ hybridization of wt (d), Wnt3a+/+(e) and Wnt3a<sup>w/- </sup> (f) embryos at E9.5. Dorsal views of the posterior region of each embryo are indicated. Red dotted lines indicate the outer edge of the tail. In Wnt3a<sup>w/- </sup> embryos, Wnt3a expression is highly decreased at this stage. (g, h) Whole- mount in situ hybridization of Wnt3a<sup>w/+ </sup> and Wnt3a<sup>w/- </sup> embryos at E12.5 using Bra(g) and Tbx6(h) probes. In contrast to Wnt3a- Fzd5 homozygous embryos, the expression of Bra nad Tbx6 is not detectable in Wnt3a<sup>w/- </sup> embryos. Scale bars: 1 mm.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 547, 856, 585]]<|/det|>
+## Extended Data Fig. 6. The phenotype of Wnt3a-Fzd5 homozygous embryos can be rescued, depending on intercellular signaling of Wnt3a  
+
+<|ref|>text<|/ref|><|det|>[[139, 588, 850, 714]]<|/det|>
+(a- d) Sagittal views of Wnt3a<sup>+/Fz5</sup> (a, a', a"), Wnt3a<sup>Fz5/Fz5</sup> (b, b', b"), Wnt3a<sup>Fz5</sup> (c, c', c") and Wnt3a<sup>w/Fz5</sup> (d, d', d") embryos at the E10.5. a', b', c', and d' are magnified images of a, b, c, and d, respectively. a", b", c", and d" are drawings of the images of a', b', c', and d', respectively. (e- h) Whole- mount in situ hybridization of Wnt3a<sup>+/Fz5</sup> (e), Wnt3a<sup>Fz5/Fz5</sup> (f), Wnt3a<sup>Fz5</sup> (g) and Wnt3a<sup>w/Fz5</sup> (h) embryos at E10.5 with Bra probe. Red dotted lines indicate the edge of the body posterior to the hindlimb. Scale bars: 1 mm.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 739, 810, 777]]<|/det|>
+## Extended Data Fig. 7. Summary plots of Bra and Sox2 signal intensities examined by immunohistochemistry.  
+
+<|ref|>text<|/ref|><|det|>[[139, 781, 841, 885]]<|/det|>
+(a) Schematic fig. showing the area examined. (b) Summary plots of Bra and Sox2 signal intensities in medial (I) and lateral (II) areas at the node-streak border in Wnt3a-Fzd5 heterozygous and homozygous embryos. Two embryos were examined for each genotype. Measurements for each cell are plotted according to levels of Bra (x-axis) and Sox (y-axis). Levels of Bra and Sox2 in each cell were normalized by the average of levels of Bra and Sox2
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[140, 119, 836, 158]]<|/det|>
+level in the medial area of Wnt3a- Fzd5 heterozygous embryos. Cells located between the two dashed lines were defined as Bra and Sox2 double- positive cells.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 184, 830, 222]]<|/det|>
+## Extended Data Fig. 8. Wnt signaling in the epiblast cell population of Wnt3a knock-out embryos  
+
+<|ref|>text<|/ref|><|det|>[[139, 227, 855, 352]]<|/det|>
+Wnt signaling activity in individual epiblast cells was visualized using mouse embryos carrying an EGFP- reporter gene, expression of which is specifically activated by Wnt signaling. Wnt signaling activity was monitored in Wnt3a knock- out (b, d) and WT (a, c) embryos at E7.5 (a, b) and E8.5 (c, d). Note that Wnt signaling is drastically reduced at E7.5 (e) and completely lost at E8.5 (f) in Wnt3a null embryos, suggesting that Wnt activity at and after E8.5 epiblast is dependent on only Wnt3a ligand. Scale bars: \(100 \mu \mathrm{m}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 376, 813, 414]]<|/det|>
+## Extended Data Fig. 9. Temporal changes of the spatial profile of Wnt-positive cells as simulated by our mathematical model.  
+
+<|ref|>text<|/ref|><|det|>[[139, 418, 858, 586]]<|/det|>
+Spatial profiles of the proportion of Wnt- positive cells ( \(>50\%\) of maximum activity) in the virtual space are indicated at 0 (T=TR), 0.2 (T=TR+0.2), 1 (T=TR+1), 2 (T=TR+2), 3 (T=TR+3) mean division time after addition of RA. Spatial profiles under combined conditions with and without Wnt- mediated intercellular communication and with and without uniformly supplied RA are shown. Also shown is the spatial profile under the condition of Wnt- mediated intercellular communication and a reduced rate of Wnt production. In Fig. 7G, the time course of the proportion of Wnt- positive cells at the same spatial level ( \(y = 35\) ) is summarized in a single graph. A and P indicate anterior and posterior, respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[140, 611, 794, 649]]<|/det|>
+## Extended Data Movie 1 Simulation of the time course of the spatial pattern of Wnt signaling activity in a hypothetical epiblast using our mathematical model.  
+
+<|ref|>text<|/ref|><|det|>[[139, 653, 808, 664]]<|/det|>
+(a) Time course of Wnt activity in the hypothetical epiblast in the presence of intercellular  
+
+<|ref|>text<|/ref|><|det|>[[140, 670, 330, 686]]<|/det|>
+exchange of Wnt ligands.  
+
+<|ref|>text<|/ref|><|det|>[[140, 692, 806, 731]]<|/det|>
+(b) Time course of Wnt activity in the hypothetical epiblast in the absence of intercellular exchange of Wnt ligands.  
+
+<|ref|>text<|/ref|><|det|>[[140, 736, 810, 775]]<|/det|>
+(c) Time course of Wnt activity in the hypothetical epiblast in the presence of intercellular exchange of Wnt ligands, but reduced Wnt production.  
+
+<|ref|>text<|/ref|><|det|>[[140, 780, 846, 819]]<|/det|>
+(d) Time course of Wnt activity in the hypothetical epiblast with uniform addition of RA in the presence of intercellular exchange of Wnt ligands.  
+
+<|ref|>text<|/ref|><|det|>[[140, 823, 845, 862]]<|/det|>
+(e) Time course of Wnt activity in the hypothetical epiblast with uniform addition of RA in the absence of intercellular exchange of Wnt ligands.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image_caption<|/ref|><|det|>[[10, 5, 100, 32]]<|/det|>
+<center>Fig. 2</center> 
+
+<|ref|>image<|/ref|><|det|>[[30, 45, 970, 415]]<|/det|>
+ 
+
+<|ref|>image_caption<|/ref|><|det|>[[339, 419, 410, 437]]<|/det|>
+<center>E12.5</center> 
+
+<|ref|>image<|/ref|><|det|>[[30, 440, 720, 633]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 540]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 900]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[7, 0, 999, 999]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[20, 37, 968, 780]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[0, 0, 999, 700]]<|/det|>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[59, 130, 370, 283]]<|/det|>
+aDiffusionRA.mp4  NCBformatExtendedDataFig.pdf  dDiffusionRA.mp4  cProduction.mp4  bDiffusionRA.mp4  eDiffusionRA.mp4
+
+<--- Page Split --->

preprint/preprint__b3a21815fc5fa8e7c39f0e5cfc096db93c0e7dec8f3d8fc214234c874fa62efe/images_list.json

@@ -0,0 +1,62 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        90,
+        111,
+        881,
+        570
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        102,
+        75,
+        880,
+        520
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        114,
+        80,
+        614,
+        640
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        100,
+        90,
+        861,
+        475
+      ]
+    ],
+    "page_idx": 12
+  }
+]

preprint/preprint__b3a21815fc5fa8e7c39f0e5cfc096db93c0e7dec8f3d8fc214234c874fa62efe/preprint__b3a21815fc5fa8e7c39f0e5cfc096db93c0e7dec8f3d8fc214234c874fa62efe.mmd

@@ -0,0 +1,220 @@
+
+# Melt-quenched Carboxylate Metal-Organic Framework Glasses  
+
+Minhvuk KimUlsan National Institute of Science and Technology  
+
+Hwa- Sub LeeUniversity of Ulsan  
+
+Dong- Hyun SeoUniversity of Science and Technology  
+
+Eun- chae Jeonhttps://orcid.org/0000- 0002- 6951- 219XHoi Ri Moon ( \(\boxed{\bullet}\) hoirimoon@ewha.ac.kr)Ewha Womans University https://orcid.org/0000- 0002- 6967- 894X  
+
+## Letter  
+
+# Keywords:  
+
+Posted Date: June 12th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 2922761/v1  
+
+License: \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Additional Declarations: There is NO Competing Interest.  
+
+Version of Record: A version of this preprint was published at Nature Communications on February 8th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 45326- 8.
+
+<--- Page Split --->
+
+## Abstract  
+
+AbstractAlthough carboxylate- based frameworks are commonly used architectures in metal- organic frameworks (MOFs), liquid/glass MOFs have thus far mainly been obtained from azole- or weakly coordinating ligand- based frameworks. \(^{1,2}\) This is because strong coordination bonds of carboxylate ligands to metals block the thermal vitrification pathways of carboxylate- based MOFs. \(^{3}\) In this study, we present the first example of carboxylate- based melt- quenched MOF glasses comprising \(\mathrm{Mg^{2 + }}\) or \(\mathrm{Mn^{2 + }}\) and an aliphatic carboxylate ligand, adipate. These MOF have a low melting temperature ( \(T_{\mathrm{m}}\) ) of 284 °C, compared to zeolitic- imidazolate framework (ZIF) glasses, \(^{4,5}\) and superior mechanical properties in terms of hardness and elastic modulus. \(^{6}\) The low \(T_{\mathrm{m}}\) is due to the flexibility and low symmetry of the aliphatic carboxylate ligand (raising entropy of fusion \((\Delta S_{\mathrm{fus}}))^{7,8}\) and the lack of crystal field stabilization energy on metal ions (reducing enthalpy of fusion \((\Delta H_{\mathrm{fus}}))\) . \(^{9}\) This research will serve as a cornerstone for the integration of numerous carboxylate- based MOF into MOF glasses.  
+
+## Full Text  
+
+Metal- organic frameworks (MOFs) are coordination networks with potential pores in a well- ordered structure composed of metal ions and polydentate organic ligands. \(^{10}\) Over the past few decades, the field of MOFs has significantly expanded because of their high designability and tunability. \(^{11}\) Despite their various properties, the practical applications of MOFs are limited because of their crystalline powder nature and low processability. \(^{1}\) To overcome the limitations of crystalline MOFs, metable MOFs have been proposed, as their liquid phase can be molded. Moreover, molten MOFs can generate a novel type of material, glass MOFs via a melt- quenching process. \(^{4,12}\) These glass structures retain the components of the original crystal and exhibit unique properties such as a monolithic manner, transparency, and luminescence. \(^{2,13}\) They also have a distorted pore network distinct from the mother MOFs. \(^{5}\) To enable melting in a MOF, the MOF must have either a low melting temperature ( \(T_{\mathrm{m}}\) ) or a high thermal decomposition temperature ( \(T_{\mathrm{d}}\) ) to satisfy the condition, \(T_{\mathrm{m}} < T_{\mathrm{d}}\) . This requirement arises from the fundamental concern that the average local coordination environment of the structures must be maintained while their long- range order is lost. \(^{7,8}\) So far, studies on metable MOFs have mostly focused on zeolitic- imidazole frameworks (ZIFs) with high \(T_{\mathrm{d}}\) owing to their thermally stable azole ligands, and MOFs composed of phosphates, amides, and sulfonates, which form weak coordination bonds with metals, thereby lowering the \(T_{\mathrm{m}}\) of the framework. \(^{2,14}\)  
+
+Despite recent advancements in this field, an important area that still need attention is the melting and vitrification of carboxylate- based MOFs, which constitute a significant majority of MOFs. \(^{15}\) Most MOFs decompose before vitrification owing to the strong bonds between carboxylate and metal centers, which elevate the \(T_{\mathrm{m}}\) of the framework above its \(T_{\mathrm{d}}\) . \(^{3}\) Recently, there have been a few reports on the synthesis of carboxylate- based MOF glasses. \(^{16,17}\) However, these studies differ from the present work in that the
+
+<--- Page Split --->
+
+starting materials for glasses are hydrogen- bonded networks of the metal complexes or amorphous coordination compounds. While this approach circumvents the thermodynamic challenges of carboxylate- based frameworks, the absence of a solid- liquid phase transition or \(T_{\mathrm{m}}\) in crystalline carboxylate frameworks restricts the variety of reported liquid/glass MOFs. Consequently, it has impeded the establishment of rational design principles for metable MOF structures.  
+
+We introduce a novel class of metable carboxylate- based MOFs consisting of \(\mathrm{Mg^{2 + }}\) or \(\mathrm{Mn^{2 + }}\) ions and an aliphatic carboxylate linker, adipate (adp, \(\mathrm{^7(OO)C(CH_2)_4(COO)^\cdot)}\) . Compared to aromatic carboxylate ligands, aliphatic carboxylate ligands have lower thermal stability and higher degree of conformational freedom. \(^{18,19}\) Based on these properties, we have previously demonstrated the thermal conversion of aliphatic ligand- based MOFs having low \(T_{\mathrm{d}}\) into hierarchically nanoporous metal oxides with nanocrystalline frameworks. \(^{21}\) Here, we utilize the low \(T_{\mathrm{m}}\) of the crystalline MOFs ([Mg4(adipate)4(DMA) \((\mathrm{H}_2\mathrm{O})] = \mathrm{C - Mg - adp}\) and [Mn2(adipate)2(DMA)] = CMn- adp) by controlling the enthalpy of fusion \((\Delta H_{\mathrm{fus}})\) and the entropy of fusion \((\Delta S_{\mathrm{fus}})\) to trigger their transition into the liquid phase and eventually create the carboxylate- based MOF glasses (G- Mg- adp and G- Mn- adp, respectively). X- ray total scattering data and pair distribution functions (PDFs) confirmed that G- Mg- adp retains the connectivity between the carboxylate and metal ions. The mechanical properties of G- Mg- adp were characterized using nanoindentation and exhibited higher hardness \((H)\) and elastic modulus \((E)\) than those of the reported coordination polymer glasses.  
+
+The solvothermal reaction of \(\mathrm{Mg(NO_3)_2\cdot 6H_2O}\) and adipic acid in \(N,N^{\prime}\) - dimethylacetamide (DMA) and methanol (MeOH) yielded block- shaped crystals of [Mg4(adipate)4(DMA)(H2O)]·5DMA·2MeOH·4H2O (Fig. 1a). \(^{21}\) The coordination of carboxylate with \(\mathrm{Mg^{2 + }}\) forms the secondary building units (SBUs) of 1D Mg- O chains, which are bridged by adp ligands to generate the 3D network. As- synthesized crystals were dried under mild conditions, yielding C- Mg- adp. The thermal behavior of C- Mg- adp was monitored using thermogravimetry analysis (TGA) under an inert atmosphere (Supplementary Fig. 1). TGA trace reveals that after the initial weight loss corresponded to the guest molecules and coordinating molecules, the MOF decomposition occurred around 330 °C ( \(T_{\mathrm{d}}\) ). Interestingly, differential scanning calorimetry (DSC) measurements for C- Mg- adp showed an endothermic peak before \(T_{\mathrm{d}}\) , ranging from 283–300 °C (Fig. 1b), indicating the melting transition of C- Mg- adp. The \(T_{\mathrm{m}}\) of C- Mg- adp is 283 °C, which is higher than that of weakly coordinated networks but lower than that of ZIFs. \(^{7,21}\) In the subsequent upscan after cooling, the glass transition of melt- quenched Mg- adp (G- Mg- adp) was observed at 242 °C ( \(T_{\mathrm{g}}\) ). The liquid fragility index (dynamical parameters) of G- Mg- adp calculated with \(T_{\mathrm{g}}\) obtained from various heating rates was 26.3, which implies that its flowing is hardly observed (Supplementary Fig. 3). \(^{4}\) The structural transformations of Mg- adp during this transition were examined using X- ray powder diffraction (XRPD) (Fig. 1c and Supplementary Fig. 4). Figure 1c shows that heating C- Mg- adp to 285 °C resulted in a loss of its crystallinity, whereas the resultant G- Mg- adp only showed diffused peaks. When the block- shaped crystals reach \(T_{\mathrm{m}}\) , C- Mg- adp transformed into glassy monolithic G- Mg- adp with spike- like and puffed
+
+<--- Page Split --->
+
+shapes that supported the existence of Mg- adp in a liquid state (Fig. 1d, Supplementary Fig. 5 and 6). \(^{22}\) Scanning electron microscopy (SEM) analysis showed that after vitrification through cooling to room temperature, shards of G- Mg- adp display a smooth surface (Fig. 1e).  
+
+To confirm that G- Mg- adp had a composition identical to that of C- Mg- adp, nuclear magnetic resonance spectroscopy (NMR) and infrared (IR) spectroscopy were performed (Supplementary Fig. 7). The NMR spectrum showed the presence of adipates after vitrification, and the IR spectrum confirmed that the carboxylate coordination bonds with metal ions were retained. The atomic connectivity and structural correlation in C- Mg- adp and G- Mg- adp were probed using X- ray total scattering data ((Q)) and PDFs (G(r)) (Supplementary Fig. 8- 11 and Fig. 2). \(^{5,23}\) I(Q) shows sharp Bragg peaks for C- Mg- adp, unlike G- Mg- adp, indicating the loss of the highly crystalline structure at G- Mg- adp. However, G(r) revealed that the local coordination environments (r < 6 Å) of G- Mg- adp were nearly identical to those of C- Mg- adp. This corresponds to the short- range bonds and correlations between the ligand and Mg ions. As shown in Figure 2b, peaks 1, 2, and 3 in G(r) correspond to the C- C and C- O bond distances in one adipate ligand, and peak 4 corresponds to the Mg- O coordination bond. Peaks 5 and 6 matched well with the distance of the unconnected C- ·C and C- ·Mg, respectively. Notably, regarding the peaks labeled a- e in Figure 2, some Mg- ·Mg correlations in the same SBU disappeared in the short- range order. However, the Mg- ·Mg correlations between neighboring SBUs were retained in G- Mg- adp. \(^{24}\) Overall, during the transformation from C- Mg- adp to G- Mg- adp, the complex coordination modes of Mg- O in the SBU of a single 1D Mg- O chain shows some variations, but the separations between neighboring SBUs were similar because of the presence of bridging adipate ligands, implying the well- maintained connectivity throughout G- Mg- adp.  
+
+Understanding the mechanical properties of glassy materials is critical for designing and engineering glass- based products and providing insight into their structure- property relationships. \(^{25}\) This has been assessed almost exclusively in ZIF glasses among coordination polymer glasses. Thus, we conducted a nanoindentation test on G- Mg- adp to understand the MOF glasses. This process involves pressing a small indenter into the surface of a sample and measuring the force and displacement during indentation. Using this, the load- depth curves of G- Mg- adp (Supplementary Fig. 12) were determined, resulting in the H and E, as depicted in Figure 3a. G- Mg- adp shows values of \(H \approx 1.18 \pm 0.051\) and \(E \approx 18.29 \pm 0.342\) GPa, recording the highest values above the reported coordination polymer glasses (Fig. 3b). \(^{6,14}\) Notably, while the hardness improves in proportion to \(T_{\mathrm{g}}\) in conventional vitreous materials, G- Mg- adp exhibits higher hardness than ZIFs, even though its \(T_{\mathrm{g}}\) is less than 50 °C compared to that of ZIFs (Table 1 and Supplementary Fig. 13). This result can be interpreted, with caution, as being caused by the stronger coordination bond of G- Mg- adp than ZIF glasses. The results agree well with the higher \(\Delta H_{\mathrm{fus}}\) and \(E\) values of G- Mg- adp. \(^{26 - 28}\) Higher hardness implies that the material is stronger; however, it can be easily fractured. Meanwhile, a higher \(H / E\) ratio indicates greater external stress tolerance until fracture. \(^{29}\) As shown in Table 1, G- Mg- adp demonstrated similar or higher \(H / E\) values than those in previous studies, resulting in its enhanced strength and toughness over existing coordination polymer glasses. Compared to its originated organic ligand crystal of adipic acid, E and H of G- Mg- adp increased
+
+<--- Page Split --->
+
+by about 1.6 and 3.7 times more than \(E(110)\) and \(H(110)\) of adipic acid crystal, \(^{30}\) suggesting that MOF glass could contribute to advancing the yield strain.  
+
+The strong metal- carboxylate bond strength on C- Mg- adp is responsible for its high \(\Delta H_{\mathrm{fus}}\) (134.5 kJ/mol), which is attributed to the strong coordination between metal ions and carboxylate ligands, but the lack of CFSE character of magnesium could relieve the enthalpy gap between the framework and its dissociation form due to kinetically labile bonds. \(^{9,31}\) To better understand the effect of CFSE and metal- ligand bond strength on MOF melting, we studied the thermal behavior of series of C- M- adp (M = Mn \(^{2 + }\) , Co \(^{2 + }\) , and Tb \(^{3 + }\) ) \(^{32}\) upon heating (Fig. 4 and Supplementary Fig. 15- 20). As shown in Fig 4, only amorphization and decomposition were observed in C- Co- adp (Fig. 4b and e) and C- Tb- adp (Fig. 4c and f), while C- Mn- adp melted and transformed into G- Mn- adp during the cooling process (Fig. 4a and d). The DSC data indicated that C- Mn- adp exhibited \(T_{\mathrm{m}}\) of 238 °C and \(T_{\mathrm{g}}\) of 179 °C (Supplementary Fig. 17). The non- melting behavior of C- Co- adp and the low \(T_{\mathrm{m}}\) of C- Mn- adp, as compared to C- Mg- adp, could potentially be attributed to the CFSE \(^{9}\) and ion radius of the metal ions \(^{33}\) , or Irving- Williams series \(^{31,34,35}\) . These findings suggest that the melting properties of MOFs can be controlled by adjusting the strength of the metalligand bond. Moreover, the absence of melting in C- Tb- adp, which has a high oxidation number metal node, may be attributed to its high metal- ligand dissociation energy. \(^{36}\)  
+
+Furthermore, another important driving force for the low \(T_{\mathrm{m}}\) of C- Mg- adp and C- Mn- adp is the larger entropic benefit resulting from the aliphatic moiety. \(^{7}\) Aliphatic carboxylate ligands have many conformation in the liquid phase of the framework than aromatic carboxylates owing to their low symmetry value. \(^{18,19}\) Moreover, the rotationally flexible alkyl chain moiety allows the transformation of the porous framework during heating, resulting in the formation of pore- collapsed structure that can reduce the residual ligand entropy in the solid phase (Supplementary Fig. 4 and 21). \(^{2,7,28}\) The glass- forming ability (GFA) of Mg- adp ( \(T_{g} / T_{m} = 0.92\) ) is much higher than that of most members of the ZIF family as well as the empirical prediction of the Kauzmann "2/3" law, \(^{37}\) despite the fact that structurally comparable aliphatic amide- based networks have low GFAs, leading to recrystallization during cooling. \(^{18,37}\) This feature is due, in part, to the relatively strong coordinate bonds, which partially contribute to the stabilization of local structure in molten phase of MOFs, resulting in a less fragile liquid. \(^{38}\)  
+
+In summary, we report the discovery of the first carboxylate- based MOF glasses obtained via the melt- quenching of a crystalline 3D MOFs. The melting of C- Mg- adp and C- Mn- adp can be explained by the high entropy contribution of the aliphatic ligand and the low CFSE of the magnesium and manganese ions. Furthermore, we demonstrate that G- Mg- adp exhibits unique mechanical properties compared to ZIF glasses, which can be partly attributed to the relatively strong coordination bonds of the carboxylate group. These results provide valuable insights into the structure- property relationship of MOF glasses. Our study not only expands the range of liquid/glass MOF materials but also provides a promising
+
+<--- Page Split --->
+
+approach for the development of meltable MOF structures based on carboxylate linkers, which are widely present in MOFs.  
+
+## Methods  
+
+Thermal vitrification of CM- adp. CM- adp was hand- grinded before heating, to make bulk powder. The powder samples were placed into a crucible or on a slide glass. The prepared samples were put into a tube furnace and then heated at \(10^{\circ}\mathrm{C}\min^{- 1}\) under argon flow of \(100~\mathrm{mL}\) min- 1. After reaching the target temperature of \(T_{\mathrm{m}}\) or \(T_{\mathrm{d}}\) , the heating was stopped, and the samples were cooled down naturally under inert gas flow to room temperature.  
+
+Mechanical properties measurement. All hardness and elastic modulus data were measured by a nanoindenter from Anton- Paar with a Berkovich tip based on the Oliver- Pharr method. Maximum force(load) was \(20~\mathrm{mN}\) and a sinusoidal method (as known as a continuous stiffness measurement method) was applied in order to measure the variations of hardness and elastic modulus with indentation depth. Since hardness and elastic modulus decreased in low indentation depth due to indentation size effect, their saturated values were selected as the representative values in Supplementary Table 1.  
+
+## Declarations  
+
+## Data availability  
+
+The data that support the findings of this work are presented in the Letter and the Supplementary Information.  
+
+## Acknowledgments  
+
+This work was supported by the National Research Foundation of Korea (NRF) grant (NRF- 2020R1A2C3008908 and NRF- 2019M3E6A1103980).  
+
+## Author contributions  
+
+M. K. and H. R. M. conceived the idea for the project. M.K. designed the experiments and performed the synthesis and characterization of the crystalline and glass MOFs. H.- S. L. and D.- H. S. analyzed the mechanical properties of the glass MOF under direction of E.- c. J. All the authors contributed to preparing the manuscript.  
+
+## Competing interests  
+
+The authors declare no competing interests.  
+
+## References
+
+<--- Page Split --->
+
+1. Bennett TD, Horike S (2018) Liquid, glass and amorphous solid states of coordination polymers and metal–organic frameworks. Nat Rev Mater 3:431–440  
+
+2. Ma N, Horike S (2022) Metal–organic network-forming glasses. Chem Rev 122:4163–4203  
+
+3. Healy C et al (2020) The thermal stability of metal-organic frameworks. Coord Chem Rev 419:213388  
+
+4. Bennett TD et al (2015) Hybrid glasses from strong and fragile metal-organic framework liquids. Nat Commun 6:8079  
+
+5. Bennett TD et al (2016) Melt-quenched glasses of metal–organic frameworks. J Am Chem Soc 138:3484–3492  
+
+6. Li S et al (2019) Mechanical properties and processing techniques of bulk metal–organic framework glasses. J Am Chem Soc 141:1027–1034  
+
+7. Liu M et al (2021) Network-forming liquids from metal–Bis (acetamide) frameworks with low melting temperatures. J Am Chem Soc 143:2801–2811  
+
+8. Liu M et al (2022) Designing Glass and Crystalline Phases of Metal–Bis (acetamide) Networks to Promote High Optical Contrast. J Am Chem Soc 144:22262–22271  
+
+9. Ohara Y et al (2018) Formation of coordination polymer glass by mechanical milling: dependence on metal ions and molecular doping for \(\mathsf{H}^{+}\) conductivity. Chem Commun 54:6859–6862  
+
+10. Batten SR et al (2012) Coordination polymers, metal–organic frameworks and the need for terminology guidelines. CrystEngComm 14:3001  
+
+11. Furukawa H et al (2013) The chemistry and applications of metal-organic frameworks. Science 341:1230444  
+
+12. Umeyama D et al (2015) Reversible solid-to-liquid phase transition of coordination polymer crystals. J Am Chem Soc 137:864–870  
+
+13. Horike S et al (2021) Mechanics, Ionics, and Optics of Metal-Organic Framework and Coordination Polymer Glasses. Nano Lett 21:6382–6390  
+
+14. Yu Z et al (2022) Recent progress of amorphous and glassy coordination polymers. Coord Chem Rev 469:214646  
+
+15. Ghasempour H et al (2021) Metal–organic frameworks based on multicarboxylate linkers. Coord Chem Rev 426:213542  
+
+16. Xu W et al (2023) High-Porosity Metal-Organic Framework Glasses. Angew Chem Int Ed 62:e202300003  
+
+17. Wei Y-S et al (2023) De-solvation of metal complexes to construct metal-organic framework glasses. Preprint at https//. 10.26434/chemrxiv-2023-05kv3  
+
+18. Zhao L, Yalkowsky SH (1999) A combined group contribution and molecular geometry approach for predicting melting points of aliphatic compounds. Ind Eng Chem Res 38:3581–3584  
+
+19. Bunn CW (1996) The melting points of chain polymers. J Polym Sci B 34:799–819  
+
+20. Kim TK et al (2013) Nanoporous metal oxides with tunable and nanocrystalline frameworks via conversion of metal–organic frameworks. J Am Chem Soc 135:8940–8946
+
+<--- Page Split --->
+
+21. Zheng Q et al (2019) Understanding glass through differential scanning calorimetry. Chem Rev 119:7848-7939  
+
+22. Bumstead AM et al (2023) Formation of a meltable purinate metal-organic framework and its glass analogue. Chem Commun 59:732-735  
+
+23. Farrow C et al (2007) PDFfit2 and PDFgui: computer programs for studying nanostructure in crystals. J Phys : Condens Matter 19:335219  
+
+24. Meekel EG, Goodwin AL (2021) Correlated disorder in metal-organic frameworks. CrystEngComm 23:2915-2922  
+
+25. Bansal NP, Doremus RH (2013) Handbook of glass properties. Elsevier  
+
+26. Johnson W, Samwer K (2005) A universal criterion for plastic yielding of metallic glasses with a \((T / T_{g})_{2 / 3}\) temperature dependence. Phys Rev Lett 95:195501  
+
+27. Tveryanovich YS (2022) The Relationship between Microhardness and Glass Transition Temperature of Chalcogenide Glasses. Glass Phys Chem 48:243-247  
+
+28. Ma N et al (2022) Exploration of glassy state in Prussian blue analogues. Nat Commun 13:4023  
+
+29. Pintaude G (2013) Introduction of the Ratio of the Hardness to the Reduced Elastic Modulus for Abrasio. Tribology - Fundamentals and Advancements Ch. 7 (InTech,  
+
+30. Mishra MK et al (2013) Odd-even effect in the elastic modulii of \(\alpha , \omega\) -alkanedicarboxylic acids. J Am Chem Soc 135:8121-8124  
+
+31. Smith RM et al (1985) Prediction of stability constants. I. Protonation constants of carboxylates and formation constants of their complexes with class A metal ions. Inorganica Chim Acta 99:207-216  
+
+32. Kim TK et al (2013) Metal-organic frameworks constructed from flexible ditopic ligands: conformational diversity of an aliphatic ligand. New J Chem 37:4130-4139  
+
+33. Shannon RD (1976) Revised effective ionic radii and systematic studies of interatomic distances in halides and chalcogenides. Acta Crystallogr A 32:751-767  
+
+34. Irving H, Williams RJP (1953) 637. The stability of transition-metal complexes. J. Chem. Soc., 3192-3210 https://doi.org/10.1039/JR9530003192  
+
+35. Bunting JW (1970) Thong. K. M. Stability constants for some 1:1 metal-carboxylate complexes. Can J Chem 48:1654-1656  
+
+36. Moltved KA, Keep KP (2019) The chemical bond between transition metals and oxygen: electronegativity, d-orbital effects, and oxophilicity as descriptors of metal-oxygen interactions. J Phys Chem C 123:18432-18444  
+
+37. Qiao A et al (2018) A metal-organic framework with ultrahigh glass-forming ability. Sci Adv 4:eaao6827  
+
+38. Wessels V et al (2011) Rapid chemical and topological ordering in supercooled liquid \(\mathrm{Cu}_{46} \mathrm{Zr}_{54}\) . Phys Rev B Condens Matter 83:094116  
+
+## Table
+
+<--- Page Split --->
+
+Table. 1 | Hardness, modulus, and their derivatives for coordinate polymer glasses and the organic crystal.
+
+<table><tr><td>Samples</td><td>G-Mg-adp</td><td>ZIF-4</td><td>TIF-4</td><td>ZIF-62</td><td>ZIF-76</td><td>Adipic acid [a]</td></tr><tr><td rowspan="2">\(H(GPa)\)</td><td>1.18</td><td>0.92</td><td>0.90</td><td>0.656</td><td>0.682</td><td>0.3</td></tr><tr><td>(±0.051)</td><td>(±0.03)</td><td>(±0.06)</td><td>(±0.005)</td><td>(±0.01)</td><td></td></tr><tr><td rowspan="2">\(E(GPa)\)</td><td>18.29</td><td>8.2</td><td>7.9</td><td>6.58</td><td>6.29</td><td>10.39</td></tr><tr><td>(±0.342)</td><td>(±0.2)</td><td>(±0.3)</td><td>(±0.02)</td><td>(±0.07)</td><td></td></tr><tr><td>\(H/E(arb.u.)\)</td><td>0.065</td><td>0.112</td><td>0.114</td><td>0.0097</td><td>0.108</td><td>0.029</td></tr><tr><td>\(H/E(GPa)\)</td><td>0.076</td><td>0.103</td><td>0.102</td><td>0.065</td><td>0.074</td><td>0.0087</td></tr><tr><td>\(T_{g}(^{\circ }C)\)</td><td>242</td><td>292</td><td>343</td><td>318</td><td>310</td><td>-</td></tr><tr><td>Reference</td><td>This work</td><td>5</td><td>5</td><td>6</td><td>6</td><td>30</td></tr></table>
+
+[a] Values corresponding to the (110) face of adipic acid.
+
+# Figures
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1 </center>  
+
+Characterization of melt-quenching in C-Mg-adp crystals, and G-Mg-adp. a, Single-crystal X- ray structure and an optical microscopy (OM) image of \([Mg_4(\text{adipate})_4(\text{DMA})(H_2O)]\cdot 5\text{DMA.2MeOH.4H}_2\text{O}.\) b, DSC curves of C- Mg- adp heated at a \(10\mathrm{Kmin}^{- 1}\) under argon. The filled area indicates \(\Delta H_{\mathrm{fus}}\) for C- Mg- adp. Inset: The \(2^{\mathrm{nd}}\) upscan in the subsequent cycle under the same conditions. The \(T_{m}\) and \(T_{g}\) were evaluated by the onset point of each peak. c, XRPD patterns for C- Mg- adp and G- Mg- adp. d, OM and e, SEM image of glassy G- Mg- adp.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2 </center>  
+
+Local structure analysis of Mg- adp with PDFs. a, G(r) for samples of CMg- adp (red) and G- Mg- adp (black). b, Local coordination environments of Mg- adp. To represent various bonding modes of adipates with similar correlation distances, the half fragment of one adipate with \(\mu_{3} - \eta_{1}\) : \(\eta_{2}\) mode is shown. c, Two adp ligands connect two neighboring SBUs of Mg- adp. For clarity, hydrogen atoms and other adipates are omitted. The dotted lines in light blue show the Mg- ··Mg correlations. Color scheme: C, gray; O, red; H, white; and Mg, light green.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3 </center>  
+
+Mechanical properties of G- Mg- adp by nanoindentation tests. a, Hardness and elastic modulus of G- Mg- adp as a function of nanoindentation depth under \(20 \text{mN}\) maximum load with three tests. b, Hardness- Modulus correlation of coordination polymer glasses. All data in the chart were evaluated using nanoindentation testing, and the Oliver- Pharr method, except the hardness of PBAs, is the Vickers hardness. More details can be found in Supplementary Fig. 14 and Supplementary Table. 2.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4 </center>  
+
+Variations in the thermal transition of M- adp (M = Mn \(^{2 + }\) , Co \(^{2 + }\) , Tb \(^{3 + }\) ) depending on the type of metal. a- c, OM images of C- Mn- adp (a), C- Co- adp (b) and C- Tb- adp (c). d- f, The products formed upon heating in an inert gas. C- Mn- adp could form G- Mn- adp through a melt- quenching process (d), while C- Co- adp undergoes amorphization by desolvation (e), and C- Tb- adp were maintained the morphology of the crystal until calcination(f).  
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- SINatMaterglassMOF20230509.docx
+
+<--- Page Split --->

preprint/preprint__b3a21815fc5fa8e7c39f0e5cfc096db93c0e7dec8f3d8fc214234c874fa62efe/preprint__b3a21815fc5fa8e7c39f0e5cfc096db93c0e7dec8f3d8fc214234c874fa62efe_det.mmd

@@ -0,0 +1,303 @@
+<|ref|>title<|/ref|><|det|>[[44, 107, 795, 175]]<|/det|>
+# Melt-quenched Carboxylate Metal-Organic Framework Glasses  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 500, 238]]<|/det|>
+Minhvuk KimUlsan National Institute of Science and Technology  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 222, 283]]<|/det|>
+Hwa- Sub LeeUniversity of Ulsan  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 383, 331]]<|/det|>
+Dong- Hyun SeoUniversity of Science and Technology  
+
+<|ref|>text<|/ref|><|det|>[[44, 336, 636, 422]]<|/det|>
+Eun- chae Jeonhttps://orcid.org/0000- 0002- 6951- 219XHoi Ri Moon ( \(\boxed{\bullet}\) hoirimoon@ewha.ac.kr)Ewha Womans University https://orcid.org/0000- 0002- 6967- 894X  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 465, 97, 482]]<|/det|>
+## Letter  
+
+<|ref|>title<|/ref|><|det|>[[44, 502, 136, 520]]<|/det|>
+# Keywords:  
+
+<|ref|>text<|/ref|><|det|>[[44, 539, 302, 559]]<|/det|>
+Posted Date: June 12th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[44, 577, 474, 597]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 2922761/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 614, 910, 658]]<|/det|>
+License: \(\circledcirc\) This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 675, 531, 696]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 731, 935, 775]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on February 8th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 45326- 8.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 159, 68]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[40, 82, 956, 354]]<|/det|>
+AbstractAlthough carboxylate- based frameworks are commonly used architectures in metal- organic frameworks (MOFs), liquid/glass MOFs have thus far mainly been obtained from azole- or weakly coordinating ligand- based frameworks. \(^{1,2}\) This is because strong coordination bonds of carboxylate ligands to metals block the thermal vitrification pathways of carboxylate- based MOFs. \(^{3}\) In this study, we present the first example of carboxylate- based melt- quenched MOF glasses comprising \(\mathrm{Mg^{2 + }}\) or \(\mathrm{Mn^{2 + }}\) and an aliphatic carboxylate ligand, adipate. These MOF have a low melting temperature ( \(T_{\mathrm{m}}\) ) of 284 °C, compared to zeolitic- imidazolate framework (ZIF) glasses, \(^{4,5}\) and superior mechanical properties in terms of hardness and elastic modulus. \(^{6}\) The low \(T_{\mathrm{m}}\) is due to the flexibility and low symmetry of the aliphatic carboxylate ligand (raising entropy of fusion \((\Delta S_{\mathrm{fus}}))^{7,8}\) and the lack of crystal field stabilization energy on metal ions (reducing enthalpy of fusion \((\Delta H_{\mathrm{fus}}))\) . \(^{9}\) This research will serve as a cornerstone for the integration of numerous carboxylate- based MOF into MOF glasses.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 377, 158, 402]]<|/det|>
+## Full Text  
+
+<|ref|>text<|/ref|><|det|>[[40, 415, 955, 802]]<|/det|>
+Metal- organic frameworks (MOFs) are coordination networks with potential pores in a well- ordered structure composed of metal ions and polydentate organic ligands. \(^{10}\) Over the past few decades, the field of MOFs has significantly expanded because of their high designability and tunability. \(^{11}\) Despite their various properties, the practical applications of MOFs are limited because of their crystalline powder nature and low processability. \(^{1}\) To overcome the limitations of crystalline MOFs, metable MOFs have been proposed, as their liquid phase can be molded. Moreover, molten MOFs can generate a novel type of material, glass MOFs via a melt- quenching process. \(^{4,12}\) These glass structures retain the components of the original crystal and exhibit unique properties such as a monolithic manner, transparency, and luminescence. \(^{2,13}\) They also have a distorted pore network distinct from the mother MOFs. \(^{5}\) To enable melting in a MOF, the MOF must have either a low melting temperature ( \(T_{\mathrm{m}}\) ) or a high thermal decomposition temperature ( \(T_{\mathrm{d}}\) ) to satisfy the condition, \(T_{\mathrm{m}} < T_{\mathrm{d}}\) . This requirement arises from the fundamental concern that the average local coordination environment of the structures must be maintained while their long- range order is lost. \(^{7,8}\) So far, studies on metable MOFs have mostly focused on zeolitic- imidazole frameworks (ZIFs) with high \(T_{\mathrm{d}}\) owing to their thermally stable azole ligands, and MOFs composed of phosphates, amides, and sulfonates, which form weak coordination bonds with metals, thereby lowering the \(T_{\mathrm{m}}\) of the framework. \(^{2,14}\)  
+
+<|ref|>text<|/ref|><|det|>[[42, 817, 955, 937]]<|/det|>
+Despite recent advancements in this field, an important area that still need attention is the melting and vitrification of carboxylate- based MOFs, which constitute a significant majority of MOFs. \(^{15}\) Most MOFs decompose before vitrification owing to the strong bonds between carboxylate and metal centers, which elevate the \(T_{\mathrm{m}}\) of the framework above its \(T_{\mathrm{d}}\) . \(^{3}\) Recently, there have been a few reports on the synthesis of carboxylate- based MOF glasses. \(^{16,17}\) However, these studies differ from the present work in that the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 44, 950, 159]]<|/det|>
+starting materials for glasses are hydrogen- bonded networks of the metal complexes or amorphous coordination compounds. While this approach circumvents the thermodynamic challenges of carboxylate- based frameworks, the absence of a solid- liquid phase transition or \(T_{\mathrm{m}}\) in crystalline carboxylate frameworks restricts the variety of reported liquid/glass MOFs. Consequently, it has impeded the establishment of rational design principles for metable MOF structures.  
+
+<|ref|>text<|/ref|><|det|>[[40, 177, 949, 488]]<|/det|>
+We introduce a novel class of metable carboxylate- based MOFs consisting of \(\mathrm{Mg^{2 + }}\) or \(\mathrm{Mn^{2 + }}\) ions and an aliphatic carboxylate linker, adipate (adp, \(\mathrm{^7(OO)C(CH_2)_4(COO)^\cdot)}\) . Compared to aromatic carboxylate ligands, aliphatic carboxylate ligands have lower thermal stability and higher degree of conformational freedom. \(^{18,19}\) Based on these properties, we have previously demonstrated the thermal conversion of aliphatic ligand- based MOFs having low \(T_{\mathrm{d}}\) into hierarchically nanoporous metal oxides with nanocrystalline frameworks. \(^{21}\) Here, we utilize the low \(T_{\mathrm{m}}\) of the crystalline MOFs ([Mg4(adipate)4(DMA) \((\mathrm{H}_2\mathrm{O})] = \mathrm{C - Mg - adp}\) and [Mn2(adipate)2(DMA)] = CMn- adp) by controlling the enthalpy of fusion \((\Delta H_{\mathrm{fus}})\) and the entropy of fusion \((\Delta S_{\mathrm{fus}})\) to trigger their transition into the liquid phase and eventually create the carboxylate- based MOF glasses (G- Mg- adp and G- Mn- adp, respectively). X- ray total scattering data and pair distribution functions (PDFs) confirmed that G- Mg- adp retains the connectivity between the carboxylate and metal ions. The mechanical properties of G- Mg- adp were characterized using nanoindentation and exhibited higher hardness \((H)\) and elastic modulus \((E)\) than those of the reported coordination polymer glasses.  
+
+<|ref|>text<|/ref|><|det|>[[39, 500, 956, 937]]<|/det|>
+The solvothermal reaction of \(\mathrm{Mg(NO_3)_2\cdot 6H_2O}\) and adipic acid in \(N,N^{\prime}\) - dimethylacetamide (DMA) and methanol (MeOH) yielded block- shaped crystals of [Mg4(adipate)4(DMA)(H2O)]·5DMA·2MeOH·4H2O (Fig. 1a). \(^{21}\) The coordination of carboxylate with \(\mathrm{Mg^{2 + }}\) forms the secondary building units (SBUs) of 1D Mg- O chains, which are bridged by adp ligands to generate the 3D network. As- synthesized crystals were dried under mild conditions, yielding C- Mg- adp. The thermal behavior of C- Mg- adp was monitored using thermogravimetry analysis (TGA) under an inert atmosphere (Supplementary Fig. 1). TGA trace reveals that after the initial weight loss corresponded to the guest molecules and coordinating molecules, the MOF decomposition occurred around 330 °C ( \(T_{\mathrm{d}}\) ). Interestingly, differential scanning calorimetry (DSC) measurements for C- Mg- adp showed an endothermic peak before \(T_{\mathrm{d}}\) , ranging from 283–300 °C (Fig. 1b), indicating the melting transition of C- Mg- adp. The \(T_{\mathrm{m}}\) of C- Mg- adp is 283 °C, which is higher than that of weakly coordinated networks but lower than that of ZIFs. \(^{7,21}\) In the subsequent upscan after cooling, the glass transition of melt- quenched Mg- adp (G- Mg- adp) was observed at 242 °C ( \(T_{\mathrm{g}}\) ). The liquid fragility index (dynamical parameters) of G- Mg- adp calculated with \(T_{\mathrm{g}}\) obtained from various heating rates was 26.3, which implies that its flowing is hardly observed (Supplementary Fig. 3). \(^{4}\) The structural transformations of Mg- adp during this transition were examined using X- ray powder diffraction (XRPD) (Fig. 1c and Supplementary Fig. 4). Figure 1c shows that heating C- Mg- adp to 285 °C resulted in a loss of its crystallinity, whereas the resultant G- Mg- adp only showed diffused peaks. When the block- shaped crystals reach \(T_{\mathrm{m}}\) , C- Mg- adp transformed into glassy monolithic G- Mg- adp with spike- like and puffed
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 46, 936, 113]]<|/det|>
+shapes that supported the existence of Mg- adp in a liquid state (Fig. 1d, Supplementary Fig. 5 and 6). \(^{22}\) Scanning electron microscopy (SEM) analysis showed that after vitrification through cooling to room temperature, shards of G- Mg- adp display a smooth surface (Fig. 1e).  
+
+<|ref|>text<|/ref|><|det|>[[39, 128, 951, 520]]<|/det|>
+To confirm that G- Mg- adp had a composition identical to that of C- Mg- adp, nuclear magnetic resonance spectroscopy (NMR) and infrared (IR) spectroscopy were performed (Supplementary Fig. 7). The NMR spectrum showed the presence of adipates after vitrification, and the IR spectrum confirmed that the carboxylate coordination bonds with metal ions were retained. The atomic connectivity and structural correlation in C- Mg- adp and G- Mg- adp were probed using X- ray total scattering data ((Q)) and PDFs (G(r)) (Supplementary Fig. 8- 11 and Fig. 2). \(^{5,23}\) I(Q) shows sharp Bragg peaks for C- Mg- adp, unlike G- Mg- adp, indicating the loss of the highly crystalline structure at G- Mg- adp. However, G(r) revealed that the local coordination environments (r < 6 Å) of G- Mg- adp were nearly identical to those of C- Mg- adp. This corresponds to the short- range bonds and correlations between the ligand and Mg ions. As shown in Figure 2b, peaks 1, 2, and 3 in G(r) correspond to the C- C and C- O bond distances in one adipate ligand, and peak 4 corresponds to the Mg- O coordination bond. Peaks 5 and 6 matched well with the distance of the unconnected C- ·C and C- ·Mg, respectively. Notably, regarding the peaks labeled a- e in Figure 2, some Mg- ·Mg correlations in the same SBU disappeared in the short- range order. However, the Mg- ·Mg correlations between neighboring SBUs were retained in G- Mg- adp. \(^{24}\) Overall, during the transformation from C- Mg- adp to G- Mg- adp, the complex coordination modes of Mg- O in the SBU of a single 1D Mg- O chain shows some variations, but the separations between neighboring SBUs were similar because of the presence of bridging adipate ligands, implying the well- maintained connectivity throughout G- Mg- adp.  
+
+<|ref|>text<|/ref|><|det|>[[39, 533, 955, 940]]<|/det|>
+Understanding the mechanical properties of glassy materials is critical for designing and engineering glass- based products and providing insight into their structure- property relationships. \(^{25}\) This has been assessed almost exclusively in ZIF glasses among coordination polymer glasses. Thus, we conducted a nanoindentation test on G- Mg- adp to understand the MOF glasses. This process involves pressing a small indenter into the surface of a sample and measuring the force and displacement during indentation. Using this, the load- depth curves of G- Mg- adp (Supplementary Fig. 12) were determined, resulting in the H and E, as depicted in Figure 3a. G- Mg- adp shows values of \(H \approx 1.18 \pm 0.051\) and \(E \approx 18.29 \pm 0.342\) GPa, recording the highest values above the reported coordination polymer glasses (Fig. 3b). \(^{6,14}\) Notably, while the hardness improves in proportion to \(T_{\mathrm{g}}\) in conventional vitreous materials, G- Mg- adp exhibits higher hardness than ZIFs, even though its \(T_{\mathrm{g}}\) is less than 50 °C compared to that of ZIFs (Table 1 and Supplementary Fig. 13). This result can be interpreted, with caution, as being caused by the stronger coordination bond of G- Mg- adp than ZIF glasses. The results agree well with the higher \(\Delta H_{\mathrm{fus}}\) and \(E\) values of G- Mg- adp. \(^{26 - 28}\) Higher hardness implies that the material is stronger; however, it can be easily fractured. Meanwhile, a higher \(H / E\) ratio indicates greater external stress tolerance until fracture. \(^{29}\) As shown in Table 1, G- Mg- adp demonstrated similar or higher \(H / E\) values than those in previous studies, resulting in its enhanced strength and toughness over existing coordination polymer glasses. Compared to its originated organic ligand crystal of adipic acid, E and H of G- Mg- adp increased
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 46, 925, 90]]<|/det|>
+by about 1.6 and 3.7 times more than \(E(110)\) and \(H(110)\) of adipic acid crystal, \(^{30}\) suggesting that MOF glass could contribute to advancing the yield strain.  
+
+<|ref|>text<|/ref|><|det|>[[40, 106, 955, 440]]<|/det|>
+The strong metal- carboxylate bond strength on C- Mg- adp is responsible for its high \(\Delta H_{\mathrm{fus}}\) (134.5 kJ/mol), which is attributed to the strong coordination between metal ions and carboxylate ligands, but the lack of CFSE character of magnesium could relieve the enthalpy gap between the framework and its dissociation form due to kinetically labile bonds. \(^{9,31}\) To better understand the effect of CFSE and metal- ligand bond strength on MOF melting, we studied the thermal behavior of series of C- M- adp (M = Mn \(^{2 + }\) , Co \(^{2 + }\) , and Tb \(^{3 + }\) ) \(^{32}\) upon heating (Fig. 4 and Supplementary Fig. 15- 20). As shown in Fig 4, only amorphization and decomposition were observed in C- Co- adp (Fig. 4b and e) and C- Tb- adp (Fig. 4c and f), while C- Mn- adp melted and transformed into G- Mn- adp during the cooling process (Fig. 4a and d). The DSC data indicated that C- Mn- adp exhibited \(T_{\mathrm{m}}\) of 238 °C and \(T_{\mathrm{g}}\) of 179 °C (Supplementary Fig. 17). The non- melting behavior of C- Co- adp and the low \(T_{\mathrm{m}}\) of C- Mn- adp, as compared to C- Mg- adp, could potentially be attributed to the CFSE \(^{9}\) and ion radius of the metal ions \(^{33}\) , or Irving- Williams series \(^{31,34,35}\) . These findings suggest that the melting properties of MOFs can be controlled by adjusting the strength of the metalligand bond. Moreover, the absence of melting in C- Tb- adp, which has a high oxidation number metal node, may be attributed to its high metal- ligand dissociation energy. \(^{36}\)  
+
+<|ref|>text<|/ref|><|det|>[[40, 455, 936, 742]]<|/det|>
+Furthermore, another important driving force for the low \(T_{\mathrm{m}}\) of C- Mg- adp and C- Mn- adp is the larger entropic benefit resulting from the aliphatic moiety. \(^{7}\) Aliphatic carboxylate ligands have many conformation in the liquid phase of the framework than aromatic carboxylates owing to their low symmetry value. \(^{18,19}\) Moreover, the rotationally flexible alkyl chain moiety allows the transformation of the porous framework during heating, resulting in the formation of pore- collapsed structure that can reduce the residual ligand entropy in the solid phase (Supplementary Fig. 4 and 21). \(^{2,7,28}\) The glass- forming ability (GFA) of Mg- adp ( \(T_{g} / T_{m} = 0.92\) ) is much higher than that of most members of the ZIF family as well as the empirical prediction of the Kauzmann "2/3" law, \(^{37}\) despite the fact that structurally comparable aliphatic amide- based networks have low GFAs, leading to recrystallization during cooling. \(^{18,37}\) This feature is due, in part, to the relatively strong coordinate bonds, which partially contribute to the stabilization of local structure in molten phase of MOFs, resulting in a less fragile liquid. \(^{38}\)  
+
+<|ref|>text<|/ref|><|det|>[[41, 758, 953, 917]]<|/det|>
+In summary, we report the discovery of the first carboxylate- based MOF glasses obtained via the melt- quenching of a crystalline 3D MOFs. The melting of C- Mg- adp and C- Mn- adp can be explained by the high entropy contribution of the aliphatic ligand and the low CFSE of the magnesium and manganese ions. Furthermore, we demonstrate that G- Mg- adp exhibits unique mechanical properties compared to ZIF glasses, which can be partly attributed to the relatively strong coordination bonds of the carboxylate group. These results provide valuable insights into the structure- property relationship of MOF glasses. Our study not only expands the range of liquid/glass MOF materials but also provides a promising
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 950, 88]]<|/det|>
+approach for the development of meltable MOF structures based on carboxylate linkers, which are widely present in MOFs.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 110, 163, 136]]<|/det|>
+## Methods  
+
+<|ref|>text<|/ref|><|det|>[[42, 150, 955, 268]]<|/det|>
+Thermal vitrification of CM- adp. CM- adp was hand- grinded before heating, to make bulk powder. The powder samples were placed into a crucible or on a slide glass. The prepared samples were put into a tube furnace and then heated at \(10^{\circ}\mathrm{C}\min^{- 1}\) under argon flow of \(100~\mathrm{mL}\) min- 1. After reaching the target temperature of \(T_{\mathrm{m}}\) or \(T_{\mathrm{d}}\) , the heating was stopped, and the samples were cooled down naturally under inert gas flow to room temperature.  
+
+<|ref|>text<|/ref|><|det|>[[42, 283, 958, 420]]<|/det|>
+Mechanical properties measurement. All hardness and elastic modulus data were measured by a nanoindenter from Anton- Paar with a Berkovich tip based on the Oliver- Pharr method. Maximum force(load) was \(20~\mathrm{mN}\) and a sinusoidal method (as known as a continuous stiffness measurement method) was applied in order to measure the variations of hardness and elastic modulus with indentation depth. Since hardness and elastic modulus decreased in low indentation depth due to indentation size effect, their saturated values were selected as the representative values in Supplementary Table 1.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 442, 212, 468]]<|/det|>
+## Declarations  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 482, 187, 502]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[44, 519, 883, 563]]<|/det|>
+The data that support the findings of this work are presented in the Letter and the Supplementary Information.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 580, 206, 600]]<|/det|>
+## Acknowledgments  
+
+<|ref|>text<|/ref|><|det|>[[44, 617, 826, 661]]<|/det|>
+This work was supported by the National Research Foundation of Korea (NRF) grant (NRF- 2020R1A2C3008908 and NRF- 2019M3E6A1103980).  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 678, 223, 698]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[44, 715, 951, 805]]<|/det|>
+M. K. and H. R. M. conceived the idea for the project. M.K. designed the experiments and performed the synthesis and characterization of the crystalline and glass MOFs. H.- S. L. and D.- H. S. analyzed the mechanical properties of the glass MOF under direction of E.- c. J. All the authors contributed to preparing the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 822, 220, 842]]<|/det|>
+## Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[45, 860, 425, 880]]<|/det|>
+The authors declare no competing interests.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 903, 196, 929]]<|/det|>
+## References
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 45, 951, 96]]<|/det|>
+1. Bennett TD, Horike S (2018) Liquid, glass and amorphous solid states of coordination polymers and metal–organic frameworks. Nat Rev Mater 3:431–440  
+
+<|ref|>text<|/ref|><|det|>[[60, 95, 872, 117]]<|/det|>
+2. Ma N, Horike S (2022) Metal–organic network-forming glasses. Chem Rev 122:4163–4203  
+
+<|ref|>text<|/ref|><|det|>[[57, 121, 955, 143]]<|/det|>
+3. Healy C et al (2020) The thermal stability of metal-organic frameworks. Coord Chem Rev 419:213388  
+
+<|ref|>text<|/ref|><|det|>[[57, 147, 952, 191]]<|/det|>
+4. Bennett TD et al (2015) Hybrid glasses from strong and fragile metal-organic framework liquids. Nat Commun 6:8079  
+
+<|ref|>text<|/ref|><|det|>[[57, 196, 910, 240]]<|/det|>
+5. Bennett TD et al (2016) Melt-quenched glasses of metal–organic frameworks. J Am Chem Soc 138:3484–3492  
+
+<|ref|>text<|/ref|><|det|>[[57, 245, 952, 290]]<|/det|>
+6. Li S et al (2019) Mechanical properties and processing techniques of bulk metal–organic framework glasses. J Am Chem Soc 141:1027–1034  
+
+<|ref|>text<|/ref|><|det|>[[57, 293, 955, 339]]<|/det|>
+7. Liu M et al (2021) Network-forming liquids from metal–Bis (acetamide) frameworks with low melting temperatures. J Am Chem Soc 143:2801–2811  
+
+<|ref|>text<|/ref|><|det|>[[57, 343, 927, 388]]<|/det|>
+8. Liu M et al (2022) Designing Glass and Crystalline Phases of Metal–Bis (acetamide) Networks to Promote High Optical Contrast. J Am Chem Soc 144:22262–22271  
+
+<|ref|>text<|/ref|><|det|>[[57, 392, 953, 438]]<|/det|>
+9. Ohara Y et al (2018) Formation of coordination polymer glass by mechanical milling: dependence on metal ions and molecular doping for \(\mathsf{H}^{+}\) conductivity. Chem Commun 54:6859–6862  
+
+<|ref|>text<|/ref|><|det|>[[52, 442, 872, 488]]<|/det|>
+10. Batten SR et al (2012) Coordination polymers, metal–organic frameworks and the need for terminology guidelines. CrystEngComm 14:3001  
+
+<|ref|>text<|/ref|><|det|>[[52, 492, 909, 536]]<|/det|>
+11. Furukawa H et al (2013) The chemistry and applications of metal-organic frameworks. Science 341:1230444  
+
+<|ref|>text<|/ref|><|det|>[[52, 541, 944, 586]]<|/det|>
+12. Umeyama D et al (2015) Reversible solid-to-liquid phase transition of coordination polymer crystals. J Am Chem Soc 137:864–870  
+
+<|ref|>text<|/ref|><|det|>[[52, 590, 930, 635]]<|/det|>
+13. Horike S et al (2021) Mechanics, Ionics, and Optics of Metal-Organic Framework and Coordination Polymer Glasses. Nano Lett 21:6382–6390  
+
+<|ref|>text<|/ref|><|det|>[[52, 639, 950, 684]]<|/det|>
+14. Yu Z et al (2022) Recent progress of amorphous and glassy coordination polymers. Coord Chem Rev 469:214646  
+
+<|ref|>text<|/ref|><|det|>[[52, 688, 920, 733]]<|/det|>
+15. Ghasempour H et al (2021) Metal–organic frameworks based on multicarboxylate linkers. Coord Chem Rev 426:213542  
+
+<|ref|>text<|/ref|><|det|>[[52, 738, 840, 782]]<|/det|>
+16. Xu W et al (2023) High-Porosity Metal-Organic Framework Glasses. Angew Chem Int Ed 62:e202300003  
+
+<|ref|>text<|/ref|><|det|>[[52, 787, 950, 832]]<|/det|>
+17. Wei Y-S et al (2023) De-solvation of metal complexes to construct metal-organic framework glasses. Preprint at https//. 10.26434/chemrxiv-2023-05kv3  
+
+<|ref|>text<|/ref|><|det|>[[52, 836, 945, 882]]<|/det|>
+18. Zhao L, Yalkowsky SH (1999) A combined group contribution and molecular geometry approach for predicting melting points of aliphatic compounds. Ind Eng Chem Res 38:3581–3584  
+
+<|ref|>text<|/ref|><|det|>[[52, 886, 800, 908]]<|/det|>
+19. Bunn CW (1996) The melting points of chain polymers. J Polym Sci B 34:799–819  
+
+<|ref|>text<|/ref|><|det|>[[52, 912, 910, 958]]<|/det|>
+20. Kim TK et al (2013) Nanoporous metal oxides with tunable and nanocrystalline frameworks via conversion of metal–organic frameworks. J Am Chem Soc 135:8940–8946
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[47, 45, 904, 88]]<|/det|>
+21. Zheng Q et al (2019) Understanding glass through differential scanning calorimetry. Chem Rev 119:7848-7939  
+
+<|ref|>text<|/ref|><|det|>[[47, 94, 950, 138]]<|/det|>
+22. Bumstead AM et al (2023) Formation of a meltable purinate metal-organic framework and its glass analogue. Chem Commun 59:732-735  
+
+<|ref|>text<|/ref|><|det|>[[47, 144, 954, 188]]<|/det|>
+23. Farrow C et al (2007) PDFfit2 and PDFgui: computer programs for studying nanostructure in crystals. J Phys : Condens Matter 19:335219  
+
+<|ref|>text<|/ref|><|det|>[[47, 193, 930, 236]]<|/det|>
+24. Meekel EG, Goodwin AL (2021) Correlated disorder in metal-organic frameworks. CrystEngComm 23:2915-2922  
+
+<|ref|>text<|/ref|><|det|>[[48, 242, 690, 263]]<|/det|>
+25. Bansal NP, Doremus RH (2013) Handbook of glass properties. Elsevier  
+
+<|ref|>text<|/ref|><|det|>[[48, 268, 911, 313]]<|/det|>
+26. Johnson W, Samwer K (2005) A universal criterion for plastic yielding of metallic glasses with a \((T / T_{g})_{2 / 3}\) temperature dependence. Phys Rev Lett 95:195501  
+
+<|ref|>text<|/ref|><|det|>[[48, 318, 953, 363]]<|/det|>
+27. Tveryanovich YS (2022) The Relationship between Microhardness and Glass Transition Temperature of Chalcogenide Glasses. Glass Phys Chem 48:243-247  
+
+<|ref|>text<|/ref|><|det|>[[48, 368, 914, 390]]<|/det|>
+28. Ma N et al (2022) Exploration of glassy state in Prussian blue analogues. Nat Commun 13:4023  
+
+<|ref|>text<|/ref|><|det|>[[48, 395, 914, 440]]<|/det|>
+29. Pintaude G (2013) Introduction of the Ratio of the Hardness to the Reduced Elastic Modulus for Abrasio. Tribology - Fundamentals and Advancements Ch. 7 (InTech,  
+
+<|ref|>text<|/ref|><|det|>[[48, 445, 953, 489]]<|/det|>
+30. Mishra MK et al (2013) Odd-even effect in the elastic modulii of \(\alpha , \omega\) -alkanedicarboxylic acids. J Am Chem Soc 135:8121-8124  
+
+<|ref|>text<|/ref|><|det|>[[48, 494, 950, 539]]<|/det|>
+31. Smith RM et al (1985) Prediction of stability constants. I. Protonation constants of carboxylates and formation constants of their complexes with class A metal ions. Inorganica Chim Acta 99:207-216  
+
+<|ref|>text<|/ref|><|det|>[[48, 544, 856, 588]]<|/det|>
+32. Kim TK et al (2013) Metal-organic frameworks constructed from flexible ditopic ligands: conformational diversity of an aliphatic ligand. New J Chem 37:4130-4139  
+
+<|ref|>text<|/ref|><|det|>[[48, 593, 940, 637]]<|/det|>
+33. Shannon RD (1976) Revised effective ionic radii and systematic studies of interatomic distances in halides and chalcogenides. Acta Crystallogr A 32:751-767  
+
+<|ref|>text<|/ref|><|det|>[[48, 642, 949, 686]]<|/det|>
+34. Irving H, Williams RJP (1953) 637. The stability of transition-metal complexes. J. Chem. Soc., 3192-3210 https://doi.org/10.1039/JR9530003192  
+
+<|ref|>text<|/ref|><|det|>[[48, 691, 953, 735]]<|/det|>
+35. Bunting JW (1970) Thong. K. M. Stability constants for some 1:1 metal-carboxylate complexes. Can J Chem 48:1654-1656  
+
+<|ref|>text<|/ref|><|det|>[[48, 740, 928, 807]]<|/det|>
+36. Moltved KA, Keep KP (2019) The chemical bond between transition metals and oxygen: electronegativity, d-orbital effects, and oxophilicity as descriptors of metal-oxygen interactions. J Phys Chem C 123:18432-18444  
+
+<|ref|>text<|/ref|><|det|>[[48, 813, 870, 856]]<|/det|>
+37. Qiao A et al (2018) A metal-organic framework with ultrahigh glass-forming ability. Sci Adv 4:eaao6827  
+
+<|ref|>text<|/ref|><|det|>[[48, 862, 953, 907]]<|/det|>
+38. Wessels V et al (2011) Rapid chemical and topological ordering in supercooled liquid \(\mathrm{Cu}_{46} \mathrm{Zr}_{54}\) . Phys Rev B Condens Matter 83:094116  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 930, 118, 953]]<|/det|>
+## Table
+
+<--- Page Split --->
+<|ref|>table_caption<|/ref|><|det|>[[42, 46, 894, 88]]<|/det|>
+Table. 1 | Hardness, modulus, and their derivatives for coordinate polymer glasses and the organic crystal.
+
+<|ref|>table<|/ref|><|det|>[[100, 102, 899, 400]]<|/det|>
+
+<table><tr><td>Samples</td><td>G-Mg-adp</td><td>ZIF-4</td><td>TIF-4</td><td>ZIF-62</td><td>ZIF-76</td><td>Adipic acid [a]</td></tr><tr><td rowspan="2">\(H(GPa)\)</td><td>1.18</td><td>0.92</td><td>0.90</td><td>0.656</td><td>0.682</td><td>0.3</td></tr><tr><td>(±0.051)</td><td>(±0.03)</td><td>(±0.06)</td><td>(±0.005)</td><td>(±0.01)</td><td></td></tr><tr><td rowspan="2">\(E(GPa)\)</td><td>18.29</td><td>8.2</td><td>7.9</td><td>6.58</td><td>6.29</td><td>10.39</td></tr><tr><td>(±0.342)</td><td>(±0.2)</td><td>(±0.3)</td><td>(±0.02)</td><td>(±0.07)</td><td></td></tr><tr><td>\(H/E(arb.u.)\)</td><td>0.065</td><td>0.112</td><td>0.114</td><td>0.0097</td><td>0.108</td><td>0.029</td></tr><tr><td>\(H/E(GPa)\)</td><td>0.076</td><td>0.103</td><td>0.102</td><td>0.065</td><td>0.074</td><td>0.0087</td></tr><tr><td>\(T_{g}(^{\circ }C)\)</td><td>242</td><td>292</td><td>343</td><td>318</td><td>310</td><td>-</td></tr><tr><td>Reference</td><td>This work</td><td>5</td><td>5</td><td>6</td><td>6</td><td>30</td></tr></table>
+
+<|ref|>text<|/ref|><|det|>[[42, 419, 531, 437]]<|/det|>
+[a] Values corresponding to the (110) face of adipic acid.
+
+<|ref|>title<|/ref|><|det|>[[42, 463, 145, 486]]<|/det|>
+# Figures
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 111, 881, 570]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 616, 115, 636]]<|/det|>
+<center>Figure 1 </center>  
+
+<|ref|>text<|/ref|><|det|>[[38, 657, 949, 806]]<|/det|>
+Characterization of melt-quenching in C-Mg-adp crystals, and G-Mg-adp. a, Single-crystal X- ray structure and an optical microscopy (OM) image of \([Mg_4(\text{adipate})_4(\text{DMA})(H_2O)]\cdot 5\text{DMA.2MeOH.4H}_2\text{O}.\) b, DSC curves of C- Mg- adp heated at a \(10\mathrm{Kmin}^{- 1}\) under argon. The filled area indicates \(\Delta H_{\mathrm{fus}}\) for C- Mg- adp. Inset: The \(2^{\mathrm{nd}}\) upscan in the subsequent cycle under the same conditions. The \(T_{m}\) and \(T_{g}\) were evaluated by the onset point of each peak. c, XRPD patterns for C- Mg- adp and G- Mg- adp. d, OM and e, SEM image of glassy G- Mg- adp.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[102, 75, 880, 520]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 564, 118, 584]]<|/det|>
+<center>Figure 2 </center>  
+
+<|ref|>text<|/ref|><|det|>[[39, 604, 955, 744]]<|/det|>
+Local structure analysis of Mg- adp with PDFs. a, G(r) for samples of CMg- adp (red) and G- Mg- adp (black). b, Local coordination environments of Mg- adp. To represent various bonding modes of adipates with similar correlation distances, the half fragment of one adipate with \(\mu_{3} - \eta_{1}\) : \(\eta_{2}\) mode is shown. c, Two adp ligands connect two neighboring SBUs of Mg- adp. For clarity, hydrogen atoms and other adipates are omitted. The dotted lines in light blue show the Mg- ··Mg correlations. Color scheme: C, gray; O, red; H, white; and Mg, light green.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[114, 80, 614, 640]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 690, 117, 710]]<|/det|>
+<center>Figure 3 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 731, 940, 844]]<|/det|>
+Mechanical properties of G- Mg- adp by nanoindentation tests. a, Hardness and elastic modulus of G- Mg- adp as a function of nanoindentation depth under \(20 \text{mN}\) maximum load with three tests. b, Hardness- Modulus correlation of coordination polymer glasses. All data in the chart were evaluated using nanoindentation testing, and the Oliver- Pharr method, except the hardness of PBAs, is the Vickers hardness. More details can be found in Supplementary Fig. 14 and Supplementary Table. 2.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[100, 90, 861, 475]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 531, 117, 551]]<|/det|>
+<center>Figure 4 </center>  
+
+<|ref|>text<|/ref|><|det|>[[41, 572, 944, 686]]<|/det|>
+Variations in the thermal transition of M- adp (M = Mn \(^{2 + }\) , Co \(^{2 + }\) , Tb \(^{3 + }\) ) depending on the type of metal. a- c, OM images of C- Mn- adp (a), C- Co- adp (b) and C- Tb- adp (c). d- f, The products formed upon heating in an inert gas. C- Mn- adp could form G- Mn- adp through a melt- quenching process (d), while C- Co- adp undergoes amorphization by desolvation (e), and C- Tb- adp were maintained the morphology of the crystal until calcination(f).  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 709, 311, 736]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 759, 765, 780]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[61, 798, 408, 817]]<|/det|>
+- SINatMaterglassMOF20230509.docx
+
+<--- Page Split --->

preprint/preprint__b3a4ad3c5ba7652df1c77390161d607d74f5f9971a807a4ffd67d2e7e0215f34/images_list.json

@@ -0,0 +1,63 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1: TCR-mediated, rapid methionine consumption governs T cell effector function. a-b, Quantification of intracellular amino acids at 10 min (a) and SAM and SAH up to 60 min (b) in OT-I T cells activated with \\(10\\mathrm{ng / ml}\\) SIINFEKL \\((n = 3)\\) . c, T cell proliferation via cell-trace violet staining of OT-I \\(\\mathrm{CD8^{+}}\\) T cells activated in either \\(0.1\\mathrm{mM}\\) Met or \\(0.03\\mathrm{mM}\\) Met for the indicated times before restoration to \\(0.1\\mathrm{mM}\\) Met in \\(0.03\\mathrm{mM}\\) Met conditions and then analyzed 72 hrs post",
+    "footnote": [],
+    "bbox": [
+      [
+        135,
+        140,
+        844,
+        750
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2: Reduced methionine availability during TCR signaling promotes T cell exhaustion. a, Schematic of experimental design. OT-I CD8+ T cells with different congenic markers were initially activated in \\(0.1\\mathrm{mM}\\) or \\(0.03\\mathrm{mM}\\) Met for \\(30\\mathrm{min}\\) with replenishment of Met in \\(0.03\\mathrm{mM}\\) to \\(0.1\\mathrm{mM}\\) Met for \\(24\\mathrm{hrs}\\) , transferred into B16-OVA tumour-bearing Rag1-/- mouse at a 1:1 ratio, and analyzed day 12 post T cell transfer. b-c, Frequencies (b) and absolute number (c) of",
+    "footnote": [],
+    "bbox": [
+      [
+        140,
+        145,
+        852,
+        760
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3: Extracellular methionine availability regulates \\(\\mathbf{Ca}^{2 + }\\) -mediated NFAT1 activity. a, Fluo-8 AM analysis of \\(\\mathrm{Ca}^{2 + }\\) flux in \\(\\mathrm{CD8}^+\\) T cells activated with anti-CD3 and anti-CD28 by anti-hamster IgG crosslinking in either \\(0.1\\mathrm{mM}\\) or \\(0.03\\mathrm{mM}\\) Met containing \\(\\mathrm{Ca}^{2 + }\\) Ringer solution",
+    "footnote": [],
+    "bbox": [
+      [
+        157,
+        145,
+        835,
+        836
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5. Ablation of KCa3.1 R350 methylation increased Ca \\(^{2 + }\\) -mediated NFAT1 activity, promoting T cell dysfunction.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Extended_Data_Figure_7.jpg",
+    "caption": "Extended Data Fig. 7",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 28
+  }
+]

preprint/preprint__b3c85f4e7eda79d6b312b4b17f3b683b2b0d1fb0bb454f9d6b0b7786fc1b5793/images_list.json

@@ -0,0 +1,122 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1: The phases of a pandemic under pathogen mutation. (a) Pandemic phase. For \\(R_{0} > 1\\) we observe the classic pandemic phase. The prevalence \\(\\rho (t)\\) vs. \\(t\\) (top) grows continuously as the fitness \\(\\bar{F} (t)\\) (bottom) increases due to mutation and natural selection. (b) Lethargic phase. For \\(R_{0}< 1\\) we have \\(\\rho (t)\\) exponentially decaying to zero. The mutation rate \\(\\sigma = 0.01\\) is too slow, \\(\\bar{F} (t)\\) remains almost constant (bottom), and the pathogen fails to reach critical fitness \\(F_{c}\\) (grey dashed line) on time. (c) Mutated phase. We now remain in the sub-pandemic regime \\(R_{0}< 1\\) , but increased the mutation rate to \\(\\sigma = 1\\) . For small \\(t\\) we observe \\(\\rho (t)\\) rapidly decaying (top). However, thanks to the rapid mutations \\(\\bar{F} (t)\\) reaches critical fitness (grey dashed line) within a short time. Following this point the disease reemerges and \\(\\rho (t)\\) changes course, turning pandemic. This is observed in the snapshots at bottom through the appearance of sporadic instances of high fitness pathogens (middle, dark red nodes), which then spread to infect the majority of the population. (d) \\(\\sigma ,R_{0}\\) phase diagram. To systematically observe the different phases we varied \\(R_{0}\\in (0,1.5)\\) and \\(\\sigma \\in (10^{-3},10)\\) , capturing a total of 1,050 epidemiological scenarios, with different \\(\\mu ,\\beta\\) and \\(\\sigma\\) . For each scenario we ran 50 stochastic realizations and measured the probability \\(P\\) to have \\(\\rho (t\\to \\infty) > 0\\) , i.e. pandemic. We observe three phases with sharp boundaries between them. First, the pandemic phase (red) for \\(R_{0} > 1\\) , independent of \\(\\sigma\\) , as predicted by the classic SIS model. In addition to that the sub-pandemic regime \\(R_{0}< 1\\) is split into two phases: Under small \\(\\sigma\\) , \\(P\\) tends to zero (blue) and the pathogen fails to spread, giving rise to the lethargic phase. For large \\(\\sigma\\) , the spreading probability becomes almost certain, as \\(p\\sim 1\\) (green), and we observe a mutation driven contagion. The gap between these phases (grey) indicates an abrupt transition from \\(P\\to 0\\) to \\(P\\to 1\\) , a dramatic shift occurring within a narrow range of \\(R_{0},\\sigma\\) values. This grey range is well-approximated by our theoretical prediction (solid black line) as appears in Eq. (11). All simulations, here and throughout, were done on a random network of \\(N = 5,000\\) nodes and \\(\\bar{k} = 15\\) . The disease parameters were set to \\(\\mu = 0.1\\) and the infection rate was set variably to \\(\\beta = \\mu R_{0} / \\bar{k}\\) , to obtain the different values of \\(R_{0}\\) . The mutation rate \\(\\sigma\\) is specified in each figure. In each scenario we set the initial condition to \\(\\rho (t) = 0.2\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        156,
+        85,
+        836,
+        592
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2: The transition to the mutated phase. To observe a mutated phase a critical mutation must arise before the pathogen is eliminated, namely before \\(\\rho (t)\\) crosses \\(1 / N\\) (grey dashed lines), capturing the unit line in which there is a single infected individual among the \\(N\\) node population. (a) \\(\\rho (t)\\) vs. \\(t\\) (grey solid line) as obtained from Eq. (8) in the lethargic phase ( \\(R_{0} = 0.25\\) , \\(\\sigma = 0.01\\) ). The critical mutation occurs at the minimum point \\((t_{c})\\) , which is below the unit line. Therefore the epidemic decays prior to the appearance of the critical mutation. Indeed, the stochastic simulation (blue solid line) approaches zero prevalence, never reaching the positive branch of \\(\\rho (t)\\) . (b) Setting \\(\\sigma = 0.16\\) the system is at criticality. \\(\\rho (t_{c})\\) is adjacent to the unit line, and hence we observe critical behavior: some realizations decay (blue), whereas others successfully mutate (green). (c) Under \\(\\sigma = 0.5\\) , the system is in the mutated phase, \\(\\rho (t_{c})\\) is sufficiently above the unit line and the critical mutation is reached with probability \\(P \\to 1\\) . (d) The lethargic-mutated phase boundary in Eq. (11) depends on the initial size of the infected population \\(\\mathcal{I}_{0}\\) . Here we show this boundary for \\(\\mathcal{I}_{0} = 10^{2}, \\ldots , 10^{8}\\) (grey solid lines). (e) The long term prevalence \\(\\rho = \\rho (t \\to \\infty)\\) vs. \\(R_{0}\\) under \\(\\sigma = 0.1\\) (yellow dashed path in panel (d)). Approaching from small \\(R_{0}\\) (left to right) we begin with an initial infection of \\(\\mathcal{I}_{0} = 10^{2}\\) and observe a transition to the mutated phase at \\(R_{0} = R_{\\mathrm{High}}\\) . In the opposite direction, however, as we begin with large \\(R_{0}\\) we approach the transition from an already pandemic state with \\(\\mathcal{I}_{0} \\sim 10^{4}\\) . Now the phase boundary traverses through \\(R_{0} = R_{\\mathrm{Low}}\\) . Both transitions are also marked by circles in panel (d). We, therefore arrive at a hysteresis phenomenon, in which the critical transition point depends on the current state of the spread. Consequently, preemptive mitigation, done when the spread is still at its embryonic stage (\\(\\mathcal{I}_{0}\\) small), is more effective than reactive mitigation, applied when \\(\\mathcal{I}_{0}\\) is already large.",
+    "footnote": [],
+    "bbox": [
+      [
+        118,
+        133,
+        880,
+        600
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3: The volatile phase. (a) The \\(\\sigma ,R_{0}\\) phase diagram under the bounded fitness of Eq. (12). We now observe a volatile phase, in which \\(\\rho \\rightarrow 0\\) (blue), when \\(\\sigma\\) is too large. Hence, the mutated phase (green) now only appears in the Goldilocks zone in which the mutation rate in not too high nor too low. The theoretical prediction of (13) is also shown (black solid line on right). (b) \\(\\rho\\) vs. \\(R_{0}\\) under \\(\\sigma = 3\\) (yellow path in panel (a)). As opposed to the lethargic-mutated phase transition, the shift from volatile to mutated follows a continuous second order transition. (c) \\(\\rho (t)\\) vs. \\(t\\) in the volatile phase decaying, as predicted, to the healthy state \\(\\rho = 0\\) . (d) \\(\\overline{F} (t)\\) changes rapidly thanks to the large \\(\\sigma\\) , and crosses the critical \\(F_{c}\\) (grey dashed line) early on. However the rapid mutations prevent the slower natural selection from securing a steady increase in \\(\\overline{F} (t)\\) . Hence, the achieved fitness cannot be stably sustained for the pathogen to continually spread. (e) Indeed, we observe multiple instances of critical fitness (dark red) that fail to reproduce and dominate the pathogen population.",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        230,
+        880,
+        623
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4: Vaccination under the threat of mutation. (a) The SARS-CoV-2 disease cycle. Upon exposure (yellow) individuals enter a pre-symptomatic phase (purple), from which they later develop mild \\((I_{M})\\) , severe \\((I_{S})\\) or critical symptoms \\((I_{C})\\) , determining the duration of their infected phase and their probability to recover (green) or decease (grey). (b) Vaccine resistance is risky under a coexistence of both infected \\((\\rho)\\) and vaccinated \\((V)\\) individuals (center). When \\(\\rho\\) is small, the probability of mutation is marginal (right); when \\((V)\\) is small the selection pressure for resistance is weak (left). (c) Under slow vaccination \\(\\rho (t)\\) increases (red). As a result, when vaccines gain coverage we enter the risky zone (shaded), and become potentially vulnerable to resistance mutation. Indeed, when such mutation occurs (orange line), the trend is reversed, \\(\\rho (t)\\) increases and the vaccine coverage \\(V(t)\\) plummets (blue). (d) We present several snapshots to track the state of the spread. In snapshot 2 we observe a premature mutation (orange node) that fails to spread, since \\(V(t)\\) at that point is still small (blue nodes). Later (snapshot 4), with the system in the risky zone of high \\(\\rho (t)\\) and \\(V(t)\\) , such mutations rapidly take over, as seen by the coverage of the orange nodes in snapshot 5. (e) - (f) Rapid vaccination in and of itself may be insufficient. The system quickly enters the risky zone (shaded) and with \\(R_{0} > 1\\) , a single resistance mutation eventually outruns our vaccination efforts. (g) - (h) Successful eradication of the disease is achieved under a combination of rapid vaccination (blue) and suppression of \\(R_{0}\\) , e.g., through social distancing. Pushing \\(R_{0}\\) down suppresses \\(\\rho (t)\\) , and hence avoids the risky zone by locating the system in the right hand side of panel (b). (i) The probability \\(P\\) to observe a pandemic state as a function of the vaccination rate \\(\\eta\\) for different values of \\(R_{0}\\) . To alleviate the risk of vaccine resistant spread we must remain in the blue zone, in which we not only invest in the vaccine roll-out \\((\\eta)\\) , but also in suppressing the spread (reducing \\(R_{0}\\) ). (j) Similar, albeit less dramatic results are also observed under our low risk scenario \\(\\mathcal{P} = 0.25\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        154,
+        80,
+        840,
+        655
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        80,
+        110,
+        888,
+        787
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        68,
+        60,
+        920,
+        635
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        79,
+        60,
+        916,
+        525
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4",
+    "footnote": [],
+    "bbox": [
+      [
+        84,
+        65,
+        840,
+        770
+      ]
+    ],
+    "page_idx": 24
+  }
+]

preprint/preprint__b3c85f4e7eda79d6b312b4b17f3b683b2b0d1fb0bb454f9d6b0b7786fc1b5793/preprint__b3c85f4e7eda79d6b312b4b17f3b683b2b0d1fb0bb454f9d6b0b7786fc1b5793.mmd

@@ -0,0 +1,453 @@
+
+# Epidemic spreading under pathogen evolution  
+
+Xiyun Zhang Jinan University https://orcid.org/0000- 0002- 7694- 234X  
+
+Ruan Zhongyuan Institute of Cyberspace Security, Zhejiang University of Technology, Hangzhou  
+
+Muhua Zheng East China Normal University https://orcid.org/0000- 0002- 9811- 529X  
+
+Jie Zhou East China Normal University  
+
+Boccaletti Stefano CNR  
+
+Baruch Barzel ( baruchbarzel@gmail.com) Bar- llan University https://orcid.org/0000- 0001- 8862- 4384  
+
+## Article  
+
+Keywords: pathogen evolution, epidemic, COVID- 19, mutant strain  
+
+Posted Date: April 13th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 373402/v1  
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Version of Record: A version of this preprint was published at Nature Communications on October 20th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 34027- 9.
+
+<--- Page Split --->
+
+# Epidemic spreading under pathogen evolution  
+
+Xiyun Zhang, \(^{1,*}\) Zhongyuan Ruan, \(^{2}\) Muhua Zheng, \(^{3}\) Jie Zhou, \(^{4}\)  
+
+Stefano Boccaletti \(^{5,6,7,8,\dagger}\) & Baruch Barzel \(^{9,10,\dagger}\)  
+
+1. Department of Physics, Jinan University, Guangzhou, Guangdong 510632, China  
+2. Institute of Cyberspace Security, Zhejiang University of Technology, Hangzhou, Zhejiang 310023, China  
+3. School of Physics and Electronic Engineering, Jiangsu University, Zhenjiang, Jiangsu, 212013, China  
+4. School of Physics and Electronic Science, East China Normal University, Shanghai 200241, China  
+5. CNR - Institute of Complex Systems, Via Madonna del Piano 10, I-50019 Sesto Fiorentino, Italy  
+6. Unmanned Systems Research Institute, Northwestern Polytechnical University, Xi'an 710072, China  
+7. Moscow Institute of Physics and Technology (National Research University), 9 Institutskiy per., Dolgoprudny, Moscow Region, 141701, Russian Federation  
+8. Universidad Rey Juan Carlos, Calle Tulipán s/n, 28933 Móstoles, Madrid, Spain  
+9. Department of Mathematics, Bar-Ilan University, Ramat-Gan, 5290002, Israel  
+10. Gonda Multidisciplinary Brain Research Center, Bar-Ilan University, Ramat-Gan, 5290002, Israel  
+
+\* Correspondence: xiyunzhang@jnu.edu.cn  
+
+↑ These Authors equally contributed to the manuscript  
+
+Battling a widespread pandemic is an arms race between our mitigation efforts, e.g., social distancing or vaccination, and the pathogen's evolving persistence. This is being observed firsthand during the current COVID- 19 crisis, as novel mutations are constantly challenging our global vaccination race. To address this, we introduce here a general framework for epidemic spreading under pathogen evolution, which shows that mutations can fundamentally alter the projection of the spread. Specifically, we detect a new pandemic phase - the mutated phase - in which, despite the fact that the pathogen is initially non- pandemic ( \(R_{0} < 1\) ), it may still spread due to the emergence of a critical mutation. The boundaries of this phase portray a balance between the epidemic and the evolutionary time- scales. If the mutation rate is too low, the pathogen prevalence decays prior to the appearance of a critical mutation. On the other hand, if mutations are too rapid, the pathogen evolution becomes volatile and, once again, it fails to spread. Between these two extremes, however, a broad range of conditions exists in which an initially sub- pandemic pathogen will eventually gain prevalence. This is especially relevant during vaccination, which creates, as it progresses, increasing selection pressure towards vaccine- resistance. To overcome this, we show that vaccination campaigns must be accompanied by fierce mitigation efforts, to suppress the potential rise of a resistant mutant strain.  
+
+Evolutionary time- scales are often considered to be vast, occurring gradually over the course of millions of years. However, if prevalent enough, a species may undergo even rare mutations at relatively short time- scales. This is especially relevant during the course of a widespread and prolonged pandemic. \(^{1 - 6}\) The global spread ensures a sufficiently large pool of pathogens for mutations to occur, and on top of that, the long duration of the pandemic \(^{7 - 13}\) affords the pathogens sufficient time to evolve.  
+
+Such troubling scenario is currently unfolding in the case of COVID- 19, where novel mutations of the SARS- CoV- 2 virus continue to challenge our mitigation efforts. \(^{14 - 18}\) They are, however, equally relevant in other infections, such as influenza A, forcing us to distribute a dedicated
+
+<--- Page Split --->
+
+vaccine in each yearly cycle. \(^{19 - 24}\) Another notable example is norovirus, whose enhanced transmission, likely due to mutation, led to an observable spike in gastric flu patients in England and Wales from 1991 to 2006, \(^{25}\) and finally, beyond viruses, artemisinin-resistance, a parasite mutation, rendered void the common treatment of malaria in Africa. \(^{26,27}\)  
+
+The common approach for tracking the spread of evolving pathogens is to introduce several competing strains and extract their interacting contagion process. \(^{28 - 32}\) This captures the patterns of spread of already evolved pathogens, overlooking the dynamics, and most importantly, the time- scales, of the evolution itself. Indeed, in an ongoing pandemic, mutations represent a gradual random process, in which an originally unfit pathogen mutates step- by- step via a series of small changes, until reaching a critical mutation that allows it to efficiently spread. Such process may take a significant amount of time, and, in some cases, the disease may taper off before such critical mutation has the opportunity to take over.  
+
+Another crucial aspect, absent when considering pre- mutated strains, is the fact that pathogen evolution is responsive. As we tighten our mitigation, either through prophylactic measures \(^{33 - 35}\) or via pharmaceutical interventions, we induce a selective pressure for mitigation resistance. For example, if one enforces social distancing to push the reproduction rate \(R_{0}\) below the pandemic threshold, the pathogen becomes naturally pressured towards higher transmissibility. Similarly, if one employs therapeutic treatment to expedite recovery, natural selection will push the pathogen to higher drug- persistence.  
+
+To address this, we introduce here an evolving pathogen model, which encompasses the delicate interplay between the pathogen's spread and its developing fitness. The evolution, a random walk in fitness space, is driven by the pathogen's mutation rate. At the same time the natural selection, in which the fitter strains proliferate, is pushed by the epidemiological parameters, characterizing how fast a mutated strain propagates. Together, we identify a rather broad set of conditions - the mutated phase - in which a non- pandemic pathogen will eventually reach an evolved pandemic state.  
+
+We find that besides the classic epidemiological parameters, i.e. infection/recovery rates, two additional components factor in - the mutation rate governing the evolutionary time- scales, and the number of infected individuals, which determines the likelihood of a critical mutation to occur within the relevant time- frame. Therefore, as opposed to classic pandemic transitions, which depend solely on the epidemiological parameters, \(^{1 - 5}\) here the current prevalence \(\rho (t)\) of the pathogen has direct impact on its anticipated spread. This has significant implications pertaining to our two main mitigation strategies \(\bullet\) Social distancing suppresses the reproduction number \(R_{0}\) to below the pandemic threshold. \(^{7 - 13}\) However, if many have already been infected, i.e. \(\rho (t)\) is large, then a stricter suppression may be required to avoid the emergence of a critical mutation. This indicates that the projection of the spread, and hence also its mitigation, depends on its present state \(\rho (t)\) - a hysteresis phenomenon, unobserved in the classic modeling frameworks \(^{36 - 41}\) \(\bullet\) Vaccination campaigns create strong evolutionary pressure towards a vaccine resistant mutation, whose risk, once again, is directly related to the current pathogen prevalence. Hence, to succeed, we show that vaccine roll- out must be coupled with fierce suppression via social distancing.
+
+<--- Page Split --->
+
+## Evolving pathogen model  
+
+Consider a social random network of \(N\) individuals linked through the adjacency matrix \(A\equiv\) \(\{A_{i j}\}\) and with average degree \(\overline{k}\) . At \(t = 0\) the network experiences an outbreak, which then spreads via the susceptible- infected- susceptible42 (SIS) dynamics. In the classic SIS formulation, the projected spread is driven by two time- independent parameters: the recovery rate \(\mu\) and the infection rate \(\beta\) , whose ratio \(R_{0} = \overline{k}\beta /\mu\) , the reproduction number, determines the state of the system - pandemic ( \(R_{0}\geq 1\) ) or healthy ( \(R_{0}< 1\) ). Here, however, the pathogen is allowed to evolve, therefore these parameter may change over the course of the spread. This is captured by the individual recovery rate  
+
+\[\mu_{i}(t) = \frac{1}{F_{i}(t)}\mu , \quad (1)\]  
+
+where the fitness \(F_{i}(t)\) stands for the level of mutation of the pathogen carried by individual \(i\) at time \(t\) , hence the unmutated pathogen has \(F_{i}(0) = 1\) . The above equation models the fact that (i) each individual \(i\) may carry a distinct version of the virus; (ii) this version may gradually change in time \(t\) due to mutations. The smaller is \(\mu_{i}(t)\) , the higher is the transmissibility of the pathogen, as described by the evolving reproduction number  
+
+\[R_{i}(t) = \frac{\overline{k}\beta}{\mu_{i}(t)} = R_{0}F_{i}(t). \quad (2)\]  
+
+Indeed, a low rate of recovery \(\mu_{i}(t)\) extends the duration of the infectious state, providing individual \(i\) with more opportunities to infect their peers. Hence, as the r.h.s. of (2) indicates, pathogens with increased \(F_{i}(t)\) exhibit higher reproduction, and therefore spread more efficiently than their lower fitness competitors.  
+
+Mutation may also impact the transmissibility of the pathogen directly by altering the value of the infection rate \(\beta\) , e.g., by evolving a more infectious strain. However, in the SIS framework, the relevant parameter is not \(\mu\) nor \(\beta\) , but their ratio, as provided by \(R_{i}(t)\) .8,11,12,43,44 Therefore, for simplicity, in (1) we only track the pathogen evolution through \(\mu_{i}(t)\) , and its subsequent \(R_{i}(t)\) , setting \(\beta\) stationary. To complement this analysis, in Supplementary Section 1 we examine the case of \(\beta\) - mutations.  
+
+The spread is driven by the infection, recovery and mutation processes. The process of infection between a pair of individuals \(i\) and \(j\) is modeled by  
+
+\[\begin{array}{r l r}{{S_{i}+I_{j}}}&{{\xrightarrow{A_{i j}\beta}}I_{i}+I_{j}}\\ {{}}&{{}}&{{}}\\ {{F_{i}(t)}}&{{=}}&{{F_{j}(t),}}\end{array} \quad (3)\]  
+
+in which a susceptible \((S)\) individual \(i\) interacts with their infected \((I)\) neighbor \(j\) ( \(A_{i j} = 1\) ) at rate \(\beta\) . This leads to both individuals becoming infected. The newly infected individual \(i\) inherits \(j\) 's pathogen, and hence in (4) we set \(i\) 's fitness at the time of infection equal to that of \(j\) . Both fitness parameters, \(F_{i}(t)\) and \(F_{j}(t)\) may later change via mutation. Next, we consider the process of recovery  
+
+\[I_{i}\xrightarrow{\mu_{i}(t)}S_{i}, \quad (5)\]
+
+<--- Page Split --->
+
+in which an infected individual \(I_{i}\) transitions to \(S_{i}\) at the evolved recovery rate \(\mu_{i}(t)\) of (1). Finally, the process of mutation follows  
+
+\[\left\{ \begin{array}{l l}{F_{i}(0) = 1}\\ {F_{i}(t + 1) = \max \left(F_{i}(t) + \delta_{i}(t)~,~0\right)} \end{array} \right., \quad (6)\]  
+
+capturing a random walk with variable step size \(\delta_{i}(t)\) , i.e. a sequence of random shifts in fitness, caused by small discrepancies in the pathogen's reproduction. Note that \(F_{i}(t)\) is prohibited from becoming negative, as, indeed, a below zero fitness in (2) is meaningless. The case where \(F_{i}(t)\) does approach zero corresponds to \(\mu_{i}(t) \to \infty\) in (1), a limit in which recovery is instantaneous, and hence the pathogen is unfit for reproduction. Such strains will be rapidly eliminated from the pathogen pool.  
+
+The magnitude of each mutation step is extracted from a zero- mean normal distribution, namely \(\delta_{i}(t) \sim \mathcal{N}(0, \sigma^{2})\) . Consequently, in the limit where \(\sigma = 0\) , we have \(\delta_{i}(t) = 0\) at all times, mutations are suppressed, and Eqs. (3) - (6) converge to the classic SIS model, with \(R_{i}(t) = R_{0}\) , a constant reproduction number. In contrast, as \(\sigma\) is increased, significant mutations become more frequent and the pathogens rapidly evolve. We therefore vary \(\sigma\) to control the mutation rate of the pathogens.  
+
+Taken together, our modeling framework accounts for the dynamics of infection and recovery (SIS) under the effect of pathogen mutation. As the spread progresses, pathogens evolve via Eq. (6), blindly altering their epidemiological parameters at random. Natural selection, however, will favor the positive mutations, in which \(\delta_{i}(t) > 0\) . Indeed, such mutations lead to higher fitness, reducing the recovery rate \(\mu_{i}(t)\) , and consequently increasing \(R_{i}(t)\) . Such pathogens, with increased \(R_{i}(t)\) , will proliferate more rapidly, and will eventually dominate the population.  
+
+Critical mutation. Consider an outbreak of a pathogen with \(R_{0} < 1\) , i.e. below the epidemic threshold. This can be either due to the pathogen's initial sub- pandemic parameters, or a result of mitigation, e.g., social distancing to reduce \(\beta\) . In the classic SIS formulation, such pathogen with fail to penetrate the network. However, in the presence of mutations ( \(\sigma > 0\) ) the pathogen may potentially undergo selection, reach \(R_{i}(t) > 1\) , and from that time onward begin to proliferate. This represents a critical mutation, which, using (2), translates to  
+
+\[F_{c} = \frac{1}{R_{0}}, \quad (7)\]  
+
+the critical fitness that, once crossed, may lead an initially non- pandemic pathogen to become pandemic. The smaller is \(R_{0}\) the higher is \(F_{c}\) , as, indeed, weakly transmissible pathogens require a longer evolutionary path to reach pandemic spread.  
+
+Next, we analyze the spreading patterns of our evolving pathogens, seeking the conditions for the appearance of the critical mutation.  
+
+## Phase-diagram of evolving pathogens  
+
+To examine the behavior of (1) - (6) we constructed an Erdős- Rényi (ER) network with \(N = 5,000\) nodes and \(\overline{k} = 15\) , providing a testing ground upon which we incorporate a series of epidemic scenarios (Fig. 1). Each scenario is characterized by a different selection of our model's three epidemiological parameters: \(\mu\) and \(\beta\) , which determine the pathogen's unmutated repro
+
+<--- Page Split --->
+
+duction \(R_{0}\) , and \(\sigma\) , which controls the rate of mutation. We then follow the spread by measuring the prevalence \(\rho (t)\) , which monitors the fraction of infected individuals vs. time. We also track the pathogen's evolution via the population averaged fitness \(\overline{F} (t) = (1 / N)\sum_{i = 1}^{N}F_{i}(t)\) .  
+
+Pandemic (Fig. 1a, red). In our first scenario we set \(\mu = 0.1\) , \(\beta = 8\times 10^{- 3}\) and \(\sigma = 10^{- 2}\) . This captures a pandemic pathogen, which, using \(\overline{k} = 15\) , has \(R_{0} = 1.2 > 1\) , namely it can spread even without mutation. Indeed \(\rho (t)\) rapidly climbs to gain macroscopic coverage, congruent with the prediction of the classic SIS model, but this time constantly growing, due to the gradual, but continuous, increase in fitness \(\overline{F} (t)\) .  
+
+Mutated (Fig. 1c, green). Next we reduce the infection rate to \(\beta = 1.67\times 10^{- 3}\) , an initial reproduction of \(R_{0} = 0.25< 1\) . This describes a pathogen whose transmissibility is significantly below the epidemic threshold, and therefore, following the initial outbreak we observe a decline in \(\rho (t)\) , which by \(t\sim 50\) almost approaches zero, as the disease seems to be tapering off. In this scenario, however, we set a faster mutation rate \(\sigma = 1\) . As a result, despite the initial remission, at around \(t\sim 15\) , the pathogen undergoes a critical mutation as \(\overline{F} (t)\) crosses the critical \(F_{c} =\) \(1 / R_{0} = 4\) (grey dashed line) and transitions into the pandemic regime. Consequently, \(\rho (t)\) changes course, the disease reemerges and the mutated pathogens successfully spreads.  
+
+Lethargic (Fig. 1b, blue). We now remain in the sub- pandemic regime, with \(R_{0} = 0.25\) , but with a much slower mutation rate, set again to \(\sigma = 10^{- 2}\) . As above, \(\rho (t)\) declines, however the pathogen evolution is now too slow - it is lethargic, and cannot reach critical fitness on time. Therefore, the disease fails to penetrate the network, lacking the opportunity for the critical mutation to occur.  
+
+Taken together, the dynamics of the spread are driven by three parameters: the initial epidemiological characteristics of the pathogen, \(\mu\) and \(\beta\) , which determine \(R_{0}\) , and the mutation rate \(\sigma\) , which governs the time- scale for the appearance of the critical mutation. Therefore, to determine the conditions for a mutation- driven contagion, as observed in Fig. 1c, we investigate the balance between the decay in \(\rho (t)\) vs. the gradual increase in \(\overline{F} (t)\) .  
+
+## The mutated phase  
+
+To understand the dynamics of the evolving pathogen model, we show in Supplementary Section 2 that at the initial stages of the spread, the prevalence \(\rho (t)\) follows  
+
+\[\rho (t) = \rho (0)e^{\xi (t)}. \quad (8)\]  
+
+The time- dependent exponential rate \(\xi (t)\) is determined by the epidemiological/mutation rates via  
+
+\[\xi (t) = -\mu (1 - R_{0})t + \frac{1}{2}\sigma^{2}\mu^{2}R_{0}^{2}t^{3}, \quad (9)\]  
+
+whose two terms characterize the pre- mutated vs. post- mutated spread of the pathogen. The first term, linear in \(t\) , represents the initial patterns of spread, which are determined by the original pathogen parameters, \(\mu ,R_{0}\) . For \(R_{0}< 1\) this describes an exponential decay, a la SIS dynamics in the sub- pandemic regime. At later times, however, as \(t^{3}\) becomes large, the second term begins to dominate, and the exponential decay is replaced by a rapid proliferation, now driven by the mutation rate \(\sigma\) . The transition between these two behaviors - decay vs. proliferation - occurs at \(\tau_{c} = \sqrt{2(1 - R_{0}) / 3\mu\sigma^{2}R_{0}^{2}}\) , which provides the anticipated time- scale for the appearance of the critical mutation \(F_{c}\) in (7).
+
+<--- Page Split --->
+
+This analysis portrays the mutated phase as a balance between two competing time- scales: on the one hand the exponential decay of the sub- pandemic pathogen, and on the other hand the evolutionary time- scale \(\tau_{c}\) for the appearance of the critical mutation. For the evolution to win this race the pathogen must not vanish before \(t = \tau_{c}\) . This imposes the condition (Fig. 2a- c)  
+
+\[\rho (\tau_{c})\geq \frac{1}{N}, \quad (10)\]  
+
+ensuring that at \(\tau_{c}\) there are still one or more individuals hosting the pathogen. Indeed, \(\rho (\tau_{c})< 1 / N\) indicates that on average, at \(t = \tau_{c}\) less than a single individual is left in the infected pool. Under this condition, the critical mutation is too late, the spread has already tapered off, and the exponential growth driven by the positive term in (9) is averted.  
+
+Taking \(\rho (\tau_{c})\) from (8), we can now use (10) to express the boundary of the mutated phase, predicting the critical mutation rate as (Supplementary Section 2)  
+
+\[\sigma_{c}\sim \left(\frac{\sqrt{\mu(1 - R_{0})^{3}}}{2R_{0}}\right)\frac{1}{\ln(\mathcal{I}_{0})}, \quad (11)\]  
+
+where \(\mathcal{I}_{0} = N\rho (t = 0)\) is the number of individuals infected at \(t = 0\) . Equation (11) describes the minimal mutation rate required for the pathogen to evolve a pandemic strain. For \(R_{0} = 1\) it predicts \(\sigma_{c} = 0\) , as such pathogen can indeed spread even without mutation. However, as \(R_{0}\) is decreased, for example under mitigation, the pathogen prevalence rapidly declines, and hence it must evolve at an accelerated rate to reach critical fitness. This is expressed in (11) by an increased \(\sigma_{c}\) , which approaches infinity as \(R_{0}\to 0\) .  
+
+To test our predicted phase transition we simulate in Fig. 1d an array of 1,050 realizations of Eqs. (1) - (6), representing different epidemiological scenarios. We varied \(R_{0}\) from 0 to 1.5, i.e. from non- transmissible to highly contagious, and scanned a spectrum of mutation rates from \(\sigma = 10^{- 3}\) to \(\sigma = 10\) , spanning four orders of magnitude. Simulating each scenario 50 times we observe the probability \(P\) for the disease to spread. This is done by tracking the pathogen's long- term prevalence \(\rho = \rho (t\to \infty)\) and counting the realizations in which \(\rho \to 0\) vs. those where \(\rho >0\) . As predicted, we find that the pandemic state, classically observed only at \(R_{0}\geq 1\) , now extends to lower \(R_{0}\) in the presence of sufficiently rapid mutations. This gives rise to the mutated phase (green), in which an initially decaying contagion suddenly turns pandemic. The transition between the lethargic and the mutated states (grey zone) is well- approximated by our theoretical prediction of Eq. (11), as depicted by the black solid line.  
+
+Equation (11) shows that \(\sigma_{c}\) depends not only on the epidemiological characteristics of the pathogen \((\mu ,R_{0})\) , but also on the initial condition, here captured by the number of infected individuals \(\mathcal{I}_{0} = \rho (t = 0)N\) . If \(\mathcal{I}_{0}\) is large the critical rate \(\sigma_{c}\) becomes lower, in effect expanding the bounds of the mutated phase. To understand this consider the evolutionary paths followed by the pathogens as they reproduce. These paths represent random trajectories in fitness space, each starting from \(F_{i}(0) = 1\) , and with a small probability crossing the critical fitness \(F_{c}\) . The more such attempts are made, the higher the chances that at least one of these paths will be successful. Therefore, a higher initial prevalence \(\mathcal{I}_{0}\) of the pathogen increases the probability for the appearance of a critical mutation, enabling a mutated phase even with low \(\sigma\) . In simple words, even rare mutations may occur if the initial pathogen pool \((\mathcal{I}_{0})\) is large enough. Indeed, in Fig. 2d we find that the phase boundary shifts towards lower \(\sigma_{c}\) as the initial prevalence is increased (grey shaded lines). Hence, a greater \(\mathcal{I}_{0}\) , indeed, expands the mutated phase.
+
+<--- Page Split --->
+
+Hysteresis. This dependence on \(\mathcal{I}_{0}\) indicates that the transition of Eq. (11) behaves differently if we approach it from the pandemic state or from the healthy state. To observe this let us fix the mutation rate at \(\sigma = 0.1\) and gradually increase \(R_{0}\) , seeking the critical point where the system shifts to the mutated phase. This is mapped to a vertical trajectory in the \(\sigma ,R_{0}\) plane (Fig. 2d, yellow dashed line). At each value of \(R_{0}\) we instigate an outbreak with \(\rho (0) = 0.2\) , and observe its long- term prevalence \(\rho\) . For small \(R_{0}\) this outbreak decays and the system reverts to the healthy state \(\rho = 0\) . However, as we transition into the mutated phase, here predicted at \(R_{0} = R_{\mathrm{High}}\approx 0.6\) , the pathogen turns pandemic and its prevalence abruptly changes to \(\rho \approx 0.85\) .  
+
+To reverse this transition the naive approach is to push \(R_{0}\) slightly below this critical point, for instance, by practicing social distancing to reduce transmission. The challenge is that now, moving in the opposite direction - from large to small \(R_{0}\) - our initial condition is pandemic, with prevalence of order unity ( \(\sim 85\%\) ), and hence \(\mathcal{I}_{0}\sim N\) . Under these conditions, Eq. (11) predicts that, for our fixed \(\sigma\) , the critical \(R_{0}\) is now lower, at \(R_{\mathrm{Low}} = 0.35\) . This results in a hysteresis phenomenon, in which criticality occurs at different points depending on the state from which we approach the transition (Fig. 2e).  
+
+We find, therefore, that pathogen evolution fundamentally changes the phase space of epidemic spreading. First it predicts a broad range of conditions - the mutated phase - in which a sub- pandemic pathogen can gain prevalence. On top of that, it also predicts that this phase exhibits a discontinuous transition, characterized by hysteresis, a phenomenon unobserved in the classic SIS dynamics, yet congruent with other models \(^{38,41,45 - 49}\) that incorporate feedback between a pathogen's prevalence ( \(\rho (t)\) ) and its potency ( \(R_{i}(t)\) ). These two observations have direct implications on mitigation:  
+
+- Soft mitigation is risky. Most mitigation strategies seek a minimal approach, aiming to drive \(R_{0}\) just below unity. This is understandable as (i) major restrictions on social interactions are costly and difficult to sustain \(^{50}\) for extended periods; (ii) having \(R_{0}< 1\) , even by a small margin, is assumed to naturally suppress the spread, as it leads \(\rho (t)\) to decay exponentially towards zero. Our analysis, however, shows that this is insufficient. For \(R_{0}\lesssim 1\) we have \(\sigma_{c}\to 0\) , indicating that even a relatively stable pathogen, with a low mutation rate, may eventually break through. Using Eq. (11) we can predict for a given \(\sigma\) , the level of tolerable \(R_{0}\) that is sufficient to mitigate the mutated phase risk, providing guidelines for effective mitigation.  
+
+- The sooner the better. Another common assumption, driven by the classic epidemic phase-diagram, is that the projected state \(\rho (t\to \infty)\) depends only on \(R_{0}\) , i.e. the epidemiological parameters. The current state of the spread \(\rho (0)\) at the time we implement our mitigation, plays no role. The observed hysteresis, however, shows that successful mitigation strongly depends on the prevalence at the time of instigation. If the pathogen has already gained sufficient ground, we will need to suppress the reproduction number below \(R_{\mathrm{Low}}\) , namely the lower phase-boundary in Fig. 2e. It is, therefore, crucial to respond early, and initiate our mitigation when \(\rho (t)\) is still small, eradicating the pandemic before mutations may determine a risk for its reemergence.  
+
+Bounded fitness. Our mutation process in Eq. (6) allows the pathogen an unbounded random walk in fitness space. In reality, however, there are practical restrictions on fitness, as \(R_{i}(t)\) cannot grow ad infinitum. Therefore, we now consider our evolving pathogen model, substituting the mutation in (6) with
+
+<--- Page Split --->
+
+\[F_{i}(t + 1) = \min \left(F_{\mathrm{max}},\max \left(F_{i}(t) + \delta_{i}(t),0\right)\right), \quad (12)\]  
+
+in which the pathogen fitness is bounded from above by \(F_{\mathrm{max}}\) and from below by zero. Setting \(F_{\mathrm{max}} = 20\) we now revisit our phase- diagram (Fig. 3a). For small \(\sigma\) , mutations are slow, and the evolution path is unaffected by the upper bound on \(F_{i}(t)\) . Therefore, we continue to observe the same transition as in the unbounded model of Fig. 1d. As we increase \(\sigma\) , however, we witness a second transition, this time back to the healthy state, indicating that now, mutations are too rapid. This captures the final phase of our evolving pathogen model - the volatile phase:  
+
+Volatile (Fig. 3, blue). When the mutation rate is too high the pathogen fitness becomes unstable. On the one hand it can rapidly reach critical fitness, yet, on the other hand, due to the random nature of its frequent mutations, it fails to sustain this fitness - resulting in an irregular \(\overline{F} (t)\) , that fluctuates above and below the critical \(F_{c}\) (Fig. 3c).  
+
+To gain deeper insight into the volatile phase, consider the natural selection process, here driven by the reproduction benefit of the fitter strains. This process is not instantaneous, and requires several reproduction instances, i.e. generations, to gain a sufficient spreading advantage. With \(\sigma\) too high, natural selection is confounded, the pathogen shown no consistent gain in fitness and, as Fig. 3c indicates, \(\rho (t)\) decays exponentially to zero. In Supplementary Section 3 we use a time- scale analysis, similar to the one leading to Eq. (11), to show that the volatile phase occurs when \(\sigma\) exceeds  
+
+\[\sigma_{c}\sim \sqrt{\frac{\mu}{3}}\frac{(F_{\mathrm{max}}R_{0} - 1)^{\frac{3}{2}}}{R_{0}}. \quad (13)\]  
+
+This prediction is, indeed, confirmed by our simulated phase diagram in Fig. 3a (black solid line).  
+
+Our phase- diagram illustrates the different forces governing the spread of pathogens in the presence of mutations. While spread is prohibited classically for \(R_{0}< 1\) , here we observe a new, previously undocumented pandemic phase, in which the disease can successfully permeate despite having an initially low reproduction rate. The conditions for this phase require a balance between three separate time- scales: (i) The time for the initial outbreak \(\rho (0)\) to reach near zero prevalence \(\tau_{r}\) ; (ii) The time for the pathogen to evolve beyond critical fitness \(\tau_{c}\) ; (iii) The time for the natural selection to lock- in the fitter mutations \(\tau_{s}\) . Pathogens with small \(R_{0}\) , we find, can still spread provided that  
+
+\[\tau_{r} > \tau_{c} > \tau_{s}. \quad (14)\]  
+
+The l.h.s. of (14) ensures that the pathogen can reach critical fitness before reaching zero prevalence. This gives rise to the first transition of Eq. (11), between the lethargic and the mutated phases. The r.h.s. of (14) is responsible for the second transition, from mutated to volatile. It ensures that fitter pathogens do not undergo additional mutation before they have time to proliferate via natural selection. Therefore, we observe a Goldilocks zone, in which the mutation rate \(\sigma\) is just right: on the one hand, enabling unfit pathogens to cross the Rubicon towards pandemicity, but on the other hand, avoiding aimless capricious mutations.
+
+<--- Page Split --->
+
+## Mutation risk in vaccine distribution - the case of COVID-19  
+
+Vaccination during an ongoing pandemic is, by nature, a competition between the rate of the vaccine roll- out and the spread of the pathogen. \(^{51 - 53}\) Therefore, naively, to win this race all one has to do is disseminate the vaccine as efficiently as possible, aiming to reach the majority of the population before the pathogen does. This, however, ignores the role of mutations, which may gravely impact even the most efficient vaccination campaign. Such mutations may, generally, be less fit epidemiologically, i.e. have a lower \(F\) and consequently a lower \(R_{i}(t)\) . Therefore, absent a vaccine, they will be rapidly overcome by the faster spreading pathogen strains. However, once the vaccine becomes widespread, resistance, even if less contagious, becomes a highly desirable trait, and a resistant mutation, if occurs, will take over the population.  
+
+To examine this in a realistic setting we consider the spread of SARS- CoV- 2, currently battled by a global vaccination effort. To model the disease dynamics we collected data on the COVID- 19 infection cycle (Fig. 4a), which includes a well- documented and elaborate set of transitions. \(^{54 - 64}\) Upon infection, individuals enter a pre- symptomatic state, which lasts, on average 5 days. During this period, typically within \(2 - 4\) days they begin to shed the virus and infect their network contacts (PS, purple). This continues until the onset of mild \((I_{M})\) , severe \((I_{S})\) or critical \((I_{C})\) symptoms, at which point they enter isolation and cease to spread the virus. A fraction \((\sim 30\%)\) of infected individuals never go on to develop noticeable symptoms (AS, top arrow), and hence they continue to spread the virus until their full recovery \((R)\) , typically within \(\sim 7\) days.  
+
+To evaluate the infection rate \(\beta\) we used empirical data on the observed spread in 12 different countries. \(^{68}\) Focusing on the early stages of the contagion, prior to the instigation of mitigation strategies, we find that \(\beta = 5 \times 10^{- 2}\) best fits the observed spreading dynamics. This corresponds to a reproduction rate of \(R_{0} \approx 2.6\) , congruent with existing valuations of \(R_{0}\) under COVID- 19. \(^{55,69}\) For details on the data analysis see Supplementary Section 4.  
+
+Here we complement this disease cycle by two additional processes  
+
+- Vaccination. The population is vaccinated at a rate \(\nu\) , quantifying the percentage of the (susceptible) population that receives the vaccine per unit time (day).- Resistance. At each time-step, the pathogen may undergo a vaccine-resistant mutation with probability \(p\) . This mutation has no bearing on its epidemiological parameters \(\mu , \beta\) , thus providing no additional spreading advantage, other than being resistant to the vaccine. The larger is the infected population \((\rho (t)N)\) the greater is the risk for such mutation, hence we quantify the mutation risk via \(\mathcal{P} = pN\) , and examine two scenarios: high risk with \(\mathcal{P} = 2.5\) and low risk, setting \(\mathcal{P} = 0.25\) . For a population of \(N \sim 10^{9}\) both cases capture a very rare mutation with \(p \sim 10^{-9}\) and \(10^{-10}\) , respectively.  
+
+Two factors drive the level of risk in this process. The prevalence \(\rho (t)\) determines the size of the pathogen pool, which must be large for the rare mutation to be realized. The vaccine coverage \(V(t)\) determines the selective advantage of the resistant strain, which becomes marginal if only a small fraction of the population is inoculated. Therefore the highest risk occurs under the coexistence of both infected and vaccinated individuals. This enables the interaction between these two populations paving the way for both mutation (large \(\rho (t)\) ) and selection (large \(V(t)\) ), and hence potentially driving the system towards vaccine resistance (Fig. 4b).  
+
+To observe this we simulated three vaccination strategies, under the high risk \(\mathcal{P} = 2.5\) scenario- Slow (Fig. 4c,d). First we assume a slow vaccination rate of \(\eta = 10^{- 3}\) , a \(0.1\%\) daily coverage. Such slow vaccination is insufficient to suppress \(\rho (t)\) , allowing us, after some time to enter the
+
+<--- Page Split --->
+
+risk zone in which \(\rho (t)\) coexists with \(V(t)\) (shaded). Mutations occurring within this window (orange) are likely to proliferate. Indeed, we find that in the long term, vaccination fails, and the resistant strain gains coverage. \(\bullet\) Rapid (Fig. 4e,f). To overcome this we simulate a rapid vaccine roll- out with \(\eta = 10^{- 2}\) , capturing an optimistic scenario, in which \(1\%\) of the population is inoculated per day. Despite these favorable conditions we continue to enter the risk zone, as the pathogen is allowed to spread freely in parallel to our vaccination efforts. The result is, as before, an increased likelihood of a resistant mutation, which, once again, regardless of our efficient dissemination, renders our vaccination void. \(\bullet\) Combined effort (Fig. 4g,h). The only way to avoid the risk zone is to minimize the potential interaction between infected and vaccinated individuals. Since \(V(t)\) will inevitably grow - indeed, this is the goal of vaccine distribution - we must contain \(\rho (t)\) , namely aim for the right- most branch of the risk curve in Fig. 4b. This requires a combined effort of both rapid vaccination ( \(\eta = 10^{- 2}\) ) and fierce mitigation to suppress \(R_{0}\) . The result is a successful elimination of the pathogen with \(V(t) \to 1\) and \(\rho (t) \to 0\) .  
+
+In Fig. 4i,j we systematically plot the spreading probability \(P\) in function of \(\eta , R_{0}\) under our high/low risk scenarios. We find that for COVID- 19, having \(R_{0} \approx 2.6\) (black solid line) the risk of vaccine resistance is significant, even under large \(\eta\) . Reducing \(R_{0}\) via social distancing helps alleviate this risk. For example, for \(\mathcal{P} = 2.5\) , even is we assume a rapid roll- out (large \(\eta\) ), we must reach \(R_{0} \lesssim 2\) to remain within a low mutation risk (blue). Under \(\mathcal{P} = 0.25\) , it is sufficient to aim for \(R_{0} \lesssim 3\) , roughly the natural state of SARS- CoV- 2.  
+
+## Discussion and outlook  
+
+The phase diagram of epidemic spreading is a crucial tool for forecasting and mitigating pandemic risks. First, it identifies the relevant control parameters, such as \(\mu , \beta\) and \(\bar{k}\) in our SIS framework, or additional parameters in more complex contagion processes, whose value determines \(R_{0}\) and hence the expected patterns of spread. The phase boundaries, then, help us assess the state of the system - healthy or pandemic - and provide guidelines for our response. For example, social distancing to reduce \(\bar{k}\) , therapeutic treatment to increase \(\mu\) or mask wearing to suppress \(\beta\) , all aimed to navigate the system's location along the pandemic phase- diagram towards the desired healthy state.  
+
+The common thread binding all of these strategies is the assumption that the epidemiological control parameters themselves are constant in time, and hence our intervention must just push them beyond the static phase- boundary, from which point on the epidemic will decay towards \(\rho \to 0\) spontaneously. This is, indeed, relevant if the temporal evolution of \(\mu , \beta\) is slow compared to the epidemic spreading dynamics - as observed in the case of our lethargic phase. However, once the epidemiological parameters can change at a sufficiently high rate, it fundamentally changes the rules of the game. This is because now, not only are the parameters dynamic, but, thanks to natural selection, they also become responsive. If, for instance, we develop drug- based treatment to increase the recovery rate \(\mu\) , we inevitably also generate selection pressure towards drug persistence. Similarly, if we vaccinate or practice distancing to reduce \(\bar{k}, \beta\) , we initiate an evolutionary race towards higher transmissibility or vaccine resistance. This was clearly observed in our analysis of the COVID- 19 vaccine dissemination.  
+
+The result is a complex interplay between the spreading dynamics ( \(R_{0}\) ), the instantaneous prevalence of the pathogen ( \(\rho (t)\) ), and the dynamic evolution of its parameters ( \(\sigma\) ), which reshapes the pandemic phase diagram. It not only expands the pandemic risk to a range of \(R_{0} < 1\) , but also predicts an explosive transition pattern, i.e. the hysteresis of Fig. 2e, that is
+
+<--- Page Split --->
+
+not observed in standard epidemiological transitions. This altered phase diagram, and its abrupt first- order like transition, we have shown, has crucial implications pertaining to our mitigation strategies. Yet, more broadly, as a physical phenomenon, it offers an interesting mechanism for explosive transitions. Most often, such abrupt phase- shifts are caused by internal suppression rules, that hold back the transition until it breaks through in an explosive fashion. \(^{37,70 - 72}\) In contrast, here what holds back the transition is the waiting time for the critical mutation. Until its appearance the system behaves in one way ( \(R_{0} < 1\) ), but once it occurs, the system suddenly enters the pandemic regime ( \(R_{0} > 1\) ). The explosiveness is therefore traced to a local event, whose probability depends on the system's initial parameters ( \(R_{0}, \mathcal{I}_{0}, \sigma\) ). This local event then changes fundamentally the state of the system - capturing a feedback between the system's phase and its intrinsic control parameters. We believe this describes a unique mechanism, inherent to the basic ingredients of our biological system, reproduction, mutation and selection.  
+
+## Acknowledgements  
+
+X.Z. thanks Dr. Xiaobo Chen, Dr. Tingting Shi, Dr. Xing Lu and Prof. Weirong Zhong for useful discussions and supports in numerical calculations. This work was partially supported by the National Natural Science Foundation of China under Grants No. 12075008 and No. 1200050749. This research was also supported by the Israel Science Foundation (grant No. 499/19) and the Bar- Ilan University Data Science Institute grant for COVID- 19 related research.  
+
+## Author contribution  
+
+X.Z. developed the concept. X.Z., B.B. and S.B. designed the framework. X.Z. and Z.R. performed the numerical simulations. All authors jointly analyzed the results and developed the analytical framework. X.Z., B.B. and S.B. wrote the paper.  
+
+## Code availability  
+
+All code to study and reproduce the results shown here will be made freely available online upon publication.
+
+<--- Page Split --->
+
+## References  
+
+[1] Hufnagel, L., Brockmann, D. and Geisel, T., Forecast and control of epidemics in a globalized world, Proc. Natl. Acad. Sci. U. S. A. 101 15124- 15129 (2004).  
+
+[2] Colizza, V., Barrat, A., Barthélemy, M. and Vespignani, A., The role of the airline transportation network in the prediction and predictability of global epidemics, Proc. Natl. Acad. Sci. U. S. A. 103, 2015- 2020 (2006).  
+
+[3] Balcan, D. et al. Multiscale mobility networks and the spatial spreading of infectious diseases, Proc. Natl. Acad. Sci. U. S. A. 106, 21484- 21489 (2009).  
+
+[4] Brockmann, D. and Helbing, D., The Hidden Geometry of Complex, Network- Driven Contagion Phenomena, Science, 342, 1337- 1342 (2013).  
+
+[5] Liu, Q. et al. Measurability of the epidemic reproduction number in data- driven contact networks, Proc. Natl. Acad. Sci. U. S. A. 115, 12680- 12685 (2018).  
+
+[6] Pastor- Satorras, R., Castellano, C., Van Mieghem, P. and Vespignani, A. Epidemic processes in complex networks, Rev. Mod. Phys. 87, 925- 979 (2015).  
+
+[7] Zhang, J. et al. Changes in contact patterns shape the dynamics of the COVID- 19 outbreak in China, Science, 368, 1481- 1486 (2020).  
+
+[8] Maier, B. F. and Brockmann, D., Effective containment explains subexponential growth in recent confirmed COVID- 19 cases in China, Science, 368, 742- 746 (2020).  
+
+[9] Schlosser, F. et al. COVID- 19 lockdown induces disease- mitigating structural changes in mobility networks, Proc. Natl. Acad. Sci. U. S. A. 117, 32883- 32890 (2020).  
+
+[10] Aleta, A. et al. Modelling the impact of testing, contact tracing and household quarantine on second waves of COVID- 19, Nature Human Behaviour, 4, 964971 (2020).  
+
+[11] Vespignani, A. et al. Modelling COVID- 19, Nature Reviews Physics, 2, 279281 (2020).  
+
+[12] te Vrugt, M., Bickmann, J. and Wittkowski, R., Effects of social distancing and isolation on epidemic spreading modeled via dynamical density functional theory, Nature Communications 11, 5576 (2020).  
+
+[13] Kogan, N. E. et al. An early warning approach to monitor COVID- 19 activity with multiple digital traces in near real time, Science Advances 7, eabd6989 (2021).  
+
+[14] Yao, H. et al. Patient- derived mutations impact pathogenicity of SARS- CoV- 2, medRxiv 2020.04.14.20060160 (2020).  
+
+[15] Konno, Y. et al. SARS- CoV- 2 ORF3b is a potent interferon antagonist whose activity is further increased by a naturally occurring elongation variant, bioRxiv 2020.05.11.088179 (2020).  
+
+[16] Hu, J. et al. The D614G mutation of SARS- CoV- 2 spike protein enhances viral infectivity and decreases neutralization sensitivity to individual convalescent sera, bioRxiv 2020.06.20.161323 (2020).
+
+<--- Page Split --->
+
+[17] Korber, B. er al. on behalf of the Sheffield COVID- 19 Genomics Group, Tracking changes in SARS- CoV- 2 Spike: evidence that D614G increases infectivity of the COVID- 19 virus, Cell (2020).  
+
+[18] Grubaugh, N.D., Hanage, W.P., and Rasmussen, A.L., Making sense of mutation: what D614G means for the COVID- 19 pandemic remains unclear, Cell (2020).  
+
+[19] Ojosnegros, S. and Beerenwinkel, N., Models of RNA virus evolution and their roles in vaccine design, Immunome Research 6(Suppl 2):S5 (2010).  
+
+[20] Nougairède, A., Charrel, R. N. and Raoult, D., Models cannot predict future outbreaks: A/H1N1 virus, the paradigm, Eur J Epidemiol 26:183, 186 (2011).  
+
+[21] Bedford, T., et al. Integrating influenza antigenic dynamics with molecular evolution, eLife 3:e01914 (2014).  
+
+[22] Nelson, M. I. and Holmes, E. C., The evolution of epidemic influenza, Nature Reviews Genetics 8, pages196,205 (2007).  
+
+[23] Ghedin, E. et al. Large- scale sequencing of human influenza reveals the dynamic nature of viral genome evolution, Nature 437, pages 1162, 1166 (2005).  
+
+[24] Earn, D. J.D., Dushoff, J. and Levin, S. A., Ecology and evolution of the flu, TRENDS in Ecology & Evolution 17, Pages 334- 340 (2002).  
+
+[25] Lopman, B., Zambon, M. and Brown, D. W., The Evolution of Norovirus, the Gastric Flu, PLoS Medicine 5(2): e42. (2008).  
+
+[26] Uwimana, A. et al., Emergence and clonal expansion of in vitro artemisinin- resistant Plasmodium falciparum kelch13 R561H mutant parasites in Rwanda, Nature Medicine (2020).  
+
+[27] Birnbaum, J. et al., A Kelch13- defined endocytosis pathway mediates artemisinin resistance in malaria parasites, Science 367, 51- 59 (2020).  
+
+[28] Alexander, H. and Day, T., Risk factors for the evolutionary emergence of pathogens. J. R. Soc. Interface 7, 1455- 1474 (2010).  
+
+[29] Uekermann, F. and Sneppen, K., Spreading of multiple epidemics with cross immunization. Phys. Rev. E, 86, 036108 (2012).  
+
+[30] Hébert- Dufresne, L., Patterson- Lomba, O., Goerg, G. M. and Althouse, B. M., Pathogen Mutation Modeled by Competition Between Site and Bond Percolation, Phys. Rev. Lett. 110, 108103 (2013).  
+
+[31] Alexandrou, C., Harmandaris, V., Irakleous, A., Koutsou, G. and Savva, N., Modeling the evolution of COVID- 19, arXiv:2008.03165 (2020).  
+
+[32] Eletreby, R. et al., The effects of evolutionary adaptations on spreading processes in complex networks, Proc. Natl. Acad. Sci. U. S. A. 117, 5664 (2020).  
+
+[33] Anderson, R. M., Heesterbeek, H., Klinkenberg, S. and Hollingsworth, T.D., How will country- based mitigation measures influence the course of the COVID- 19 epidemic? The Lancet, 395, 10228 (2020).
+
+<--- Page Split --->
+
+[34] Shen, C. and Bar- Yam, Y, COVID- 19: How to win (2020).  
+
+[35] Meidan, D. et al. Alternating quarantine for sustainable epidemic mitigation, Nature Communications 12,220 (2021).  
+
+[36] Böttcher, L., Woolley- Meza, O., Araújo, N. A. M., Herrmann, H. J. and Helbing, D., Disease- induced resource constraints can trigger explosive epidemics, Scientific Reports 5, 16571 (2015).  
+
+[37] Boccaletti, S. et al. Explosive transitions in complex networks? structure and dynamics: Percolation and synchronization, Physics Reports 660, 1- 94 (2016).  
+
+[38] Liu, Q., Wang, W., Tang, M., Zhou, T. and Lai, Y., Explosive spreading on complex networks: The role of synergy, Phys. Rev. E 95, 042320 (2017).  
+
+[39] Pires, M. A., Oestereich, A. L. and Crokidakis, N., Sudden transitions in coupled opinion and epidemic dynamics with vaccination, J. Stat. Mech. 053407 (2018).  
+
+[40] Mieghem, P. V., Explosive phase transition in susceptible- infected- susceptible epidemics with arbitrary small but nonzero self- infection rate, Phys. Rev. E, 101, 032303 (2020).  
+
+[41] Zhang, X., Ruan, Z., Zheng, M., Barzel, B. and Boccaletti, S., Epidemic spreading under infection- reduced- recovery, Chaos, Solitons and Fractals 140, 110130 (2020).  
+
+[42] Pastor- Satorras, R., Castellano, C., Mieghem, P. V. and Vespignani, A., Epidemic processes in complex networks, Rev. Mod. Phys. 87, 925 (2015).  
+
+[43] Yang, H., Gu, C., Tang, M., Cai, S. and Lai, Y., Suppression of epidemic spreading in time- varying multiplex networks, Applied Mathematical Modelling, 75, 806- 818 (2019).  
+
+[44] Zhai, Z., Long, Y., Tang, M., Liu, Z. and Lai, Y., Optimal inference of the start of COVID- 19, Phys. Rev. Research 3, 013155 (2021).  
+
+[45] Gómez- Gardeñes, J., Lotero, L., Taraskin, S. N. and Pérez- Reche, F. J., Explosive contagion in networks, Sci. Rep. 6, 19767 (2016).  
+
+[46] Wu, J., Zheng, M., Xu, K. and Gu, C., Effects of two channels on explosive information spreading, Nonlinear Dynamics, 99, 2387- 2397, (2020).  
+
+[47] Hébert- Dufresne, L., Scarpino, S. V. and Young, J., Macroscopic patterns of interacting contagions are indistinguishable from social reinforcement, Nature Physics 16, 426 (2020).  
+
+[48] Gross, T., Lima, C. J. D. and Blasius, B., Epidemic dynamics on an adaptive network, Phys. Rev. Lett. 96, 208701 (2006).  
+
+[49] Böttcher, L., Woolley- Meza, O., Goles, E., Helbing, D. and Herrmann, H. J., Connectivity disruption sparks explosive epidemic spreading, Phys. Rev. E 93, 042315 (2016).  
+
+[50] Epstein, J.M., Modelling to contain pandemics. Nature 460,687 (2009).  
+
+[51] Harush, U. and Barzel, B., Dynamic patterns of information flow in complex net- works, Nat. Comm. 8, 2181 (2017).  
+
+[52] Hens, C., Harush, U., Cohen, R. and Barzel, B., Spatiotemporal signal propagation in complex networks, Nat. Phys. 15, 403 (2019).
+
+<--- Page Split --->
+
+[53] Hacohen, A., Cohen, R., Efroni, S., Barzel, B. and Bachelet, I., Digitizable therapeutics for decentralized mitigation of global pandemics, Scientific Reports 9, 14345 (2019).  
+
+[54] Ferguson, N. M. et al. Impact of Non- Pharmaceutical Interventions (npis) to Reduce COVID- 19 Mortality and Health- Care Demand. (Imperial College COVID- 19 Response Team, 2020).  
+
+[55] Li, Q. et al. Early transmission dynamics in Wuhan, China, of novel Coronavirus- infected pneumonia. N. Engl. J. Med. 382, 1199 (2020).  
+
+[56] WHO. Coronavirus Disease (COVID- 2019) Situation Report 30. (2020).  
+
+[57] Ferretti, L. et al. Quantifying SARS- CoV- 2 transmission suggests epidemic control with digital contact tracing. Science 368, 6936 (2020).  
+
+[58] Backer, J. A., Klinkenberg, D. and Wallinga, J. Incubation period of 2019 novel Coronavirus (2019- ncov) infections among travelers from Wuhan, China, 2028 January 2020. Eurosurveillance 25, 2000062 (2020).  
+
+[59] Linton, N. M., et al. Incubation period and other epidemiological characteristics of 2019 novel Coronavirus infections with right truncation: a statistical analysis of publicly available case data. J. Clin. Med. 9, 538 (2020)  
+
+[60] Tao, Y. et al. High incidence of asymptomatic SARS- CoV- 2 infection, Chongqing, China. medRxiv https://doi.org/10.1101/2020.03.16.20037259 (2020).  
+
+[61] Mizumoto, K., Kagaya, K., Zarebski, A. and Chowell, G. Estimating the asymptomatic proportion of Coronavirus disease 2019 (COVID- 19) cases on board the Diamond Princess cruise ship, Yokohama, Japan, 2020. Eurosurveillance 25, 2000180 (2020).  
+
+[62] Pan, X. et al. Asymptomatic cases in a family cluster with SARS- CoV- 2 infection. Lancet Infect. Dis. 20, 410 (2020).  
+
+[63] Lu, X. et al. SARS- CoV- 2 infection in children. N. Engl. J. Med. 382, 16631665 (2020).  
+
+[64] Al- Tawfiq, J. A. Asymptomatic coronavirus infection: MERS- CoV and SARS- CoV- 2 (COVID- 19). Travel Med. Infect. Dis. 35, 101608 (2020).  
+
+[65] Colson, P. et al. Children account for a small proportion of diagnoses of SARS- CoV- 2 infection and do not exhibit greater viral loads than adults. Eur. J. Clin. Microbiol. Infect. Dis. 26, 15 (2020).  
+
+[66] Song, H. et al. A considerable proportion of individuals with asymptomatic SARS- CoV- 2 infection in Tibetan population. medRxiv https://doi.org/10.1101/2020.03.27.20043836 (2020).  
+
+[67] Dong, Y. et al. Epidemiology of COVID- 19 among children in China. Pediatrics 145, e20200702 (2020).  
+
+[68] Dong, E., Du, H. and Gardner, L., An interactive web- based dashboard to track COVID- 19 in real time. Lancet Infect. Dis. 20, 533534 (2020).  
+
+[69] Bar- On, Y. M., Flamholz, A., Phillips, R. and Milo, R., SARS- CoV- 2 (COVID- 19) by the numbers. eLife 9, e57309 (2020).
+
+<--- Page Split --->
+
+[70] Leyva, I. et al., Explosive first- order transition to synchrony in networked chaotic oscillators, Phys. Rev. Lett. 108 (16), 168702 (2012).  
+
+[71] Zhang, X., Hu, X., Kurths, J. and Liu, Z., Explosive synchronization in a general complex network, Phys. Rev. E 88, 010802(R) (2013).  
+
+[72] Zhang, X., Boccaletti, S., Guan, S. and Liu, Z., Explosive synchronization in adaptive and multilayer networks, Phys. Rev. Lett. 114, 038701 (2015).
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1: The phases of a pandemic under pathogen mutation. (a) Pandemic phase. For \(R_{0} > 1\) we observe the classic pandemic phase. The prevalence \(\rho (t)\) vs. \(t\) (top) grows continuously as the fitness \(\bar{F} (t)\) (bottom) increases due to mutation and natural selection. (b) Lethargic phase. For \(R_{0}< 1\) we have \(\rho (t)\) exponentially decaying to zero. The mutation rate \(\sigma = 0.01\) is too slow, \(\bar{F} (t)\) remains almost constant (bottom), and the pathogen fails to reach critical fitness \(F_{c}\) (grey dashed line) on time. (c) Mutated phase. We now remain in the sub-pandemic regime \(R_{0}< 1\) , but increased the mutation rate to \(\sigma = 1\) . For small \(t\) we observe \(\rho (t)\) rapidly decaying (top). However, thanks to the rapid mutations \(\bar{F} (t)\) reaches critical fitness (grey dashed line) within a short time. Following this point the disease reemerges and \(\rho (t)\) changes course, turning pandemic. This is observed in the snapshots at bottom through the appearance of sporadic instances of high fitness pathogens (middle, dark red nodes), which then spread to infect the majority of the population. (d) \(\sigma ,R_{0}\) phase diagram. To systematically observe the different phases we varied \(R_{0}\in (0,1.5)\) and \(\sigma \in (10^{-3},10)\) , capturing a total of 1,050 epidemiological scenarios, with different \(\mu ,\beta\) and \(\sigma\) . For each scenario we ran 50 stochastic realizations and measured the probability \(P\) to have \(\rho (t\to \infty) > 0\) , i.e. pandemic. We observe three phases with sharp boundaries between them. First, the pandemic phase (red) for \(R_{0} > 1\) , independent of \(\sigma\) , as predicted by the classic SIS model. In addition to that the sub-pandemic regime \(R_{0}< 1\) is split into two phases: Under small \(\sigma\) , \(P\) tends to zero (blue) and the pathogen fails to spread, giving rise to the lethargic phase. For large \(\sigma\) , the spreading probability becomes almost certain, as \(p\sim 1\) (green), and we observe a mutation driven contagion. The gap between these phases (grey) indicates an abrupt transition from \(P\to 0\) to \(P\to 1\) , a dramatic shift occurring within a narrow range of \(R_{0},\sigma\) values. This grey range is well-approximated by our theoretical prediction (solid black line) as appears in Eq. (11). All simulations, here and throughout, were done on a random network of \(N = 5,000\) nodes and \(\bar{k} = 15\) . The disease parameters were set to \(\mu = 0.1\) and the infection rate was set variably to \(\beta = \mu R_{0} / \bar{k}\) , to obtain the different values of \(R_{0}\) . The mutation rate \(\sigma\) is specified in each figure. In each scenario we set the initial condition to \(\rho (t) = 0.2\) . </center>
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2: The transition to the mutated phase. To observe a mutated phase a critical mutation must arise before the pathogen is eliminated, namely before \(\rho (t)\) crosses \(1 / N\) (grey dashed lines), capturing the unit line in which there is a single infected individual among the \(N\) node population. (a) \(\rho (t)\) vs. \(t\) (grey solid line) as obtained from Eq. (8) in the lethargic phase ( \(R_{0} = 0.25\) , \(\sigma = 0.01\) ). The critical mutation occurs at the minimum point \((t_{c})\) , which is below the unit line. Therefore the epidemic decays prior to the appearance of the critical mutation. Indeed, the stochastic simulation (blue solid line) approaches zero prevalence, never reaching the positive branch of \(\rho (t)\) . (b) Setting \(\sigma = 0.16\) the system is at criticality. \(\rho (t_{c})\) is adjacent to the unit line, and hence we observe critical behavior: some realizations decay (blue), whereas others successfully mutate (green). (c) Under \(\sigma = 0.5\) , the system is in the mutated phase, \(\rho (t_{c})\) is sufficiently above the unit line and the critical mutation is reached with probability \(P \to 1\) . (d) The lethargic-mutated phase boundary in Eq. (11) depends on the initial size of the infected population \(\mathcal{I}_{0}\) . Here we show this boundary for \(\mathcal{I}_{0} = 10^{2}, \ldots , 10^{8}\) (grey solid lines). (e) The long term prevalence \(\rho = \rho (t \to \infty)\) vs. \(R_{0}\) under \(\sigma = 0.1\) (yellow dashed path in panel (d)). Approaching from small \(R_{0}\) (left to right) we begin with an initial infection of \(\mathcal{I}_{0} = 10^{2}\) and observe a transition to the mutated phase at \(R_{0} = R_{\mathrm{High}}\) . In the opposite direction, however, as we begin with large \(R_{0}\) we approach the transition from an already pandemic state with \(\mathcal{I}_{0} \sim 10^{4}\) . Now the phase boundary traverses through \(R_{0} = R_{\mathrm{Low}}\) . Both transitions are also marked by circles in panel (d). We, therefore arrive at a hysteresis phenomenon, in which the critical transition point depends on the current state of the spread. Consequently, preemptive mitigation, done when the spread is still at its embryonic stage (\(\mathcal{I}_{0}\) small), is more effective than reactive mitigation, applied when \(\mathcal{I}_{0}\) is already large. </center>
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3: The volatile phase. (a) The \(\sigma ,R_{0}\) phase diagram under the bounded fitness of Eq. (12). We now observe a volatile phase, in which \(\rho \rightarrow 0\) (blue), when \(\sigma\) is too large. Hence, the mutated phase (green) now only appears in the Goldilocks zone in which the mutation rate in not too high nor too low. The theoretical prediction of (13) is also shown (black solid line on right). (b) \(\rho\) vs. \(R_{0}\) under \(\sigma = 3\) (yellow path in panel (a)). As opposed to the lethargic-mutated phase transition, the shift from volatile to mutated follows a continuous second order transition. (c) \(\rho (t)\) vs. \(t\) in the volatile phase decaying, as predicted, to the healthy state \(\rho = 0\) . (d) \(\overline{F} (t)\) changes rapidly thanks to the large \(\sigma\) , and crosses the critical \(F_{c}\) (grey dashed line) early on. However the rapid mutations prevent the slower natural selection from securing a steady increase in \(\overline{F} (t)\) . Hence, the achieved fitness cannot be stably sustained for the pathogen to continually spread. (e) Indeed, we observe multiple instances of critical fitness (dark red) that fail to reproduce and dominate the pathogen population. </center>
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4: Vaccination under the threat of mutation. (a) The SARS-CoV-2 disease cycle. Upon exposure (yellow) individuals enter a pre-symptomatic phase (purple), from which they later develop mild \((I_{M})\) , severe \((I_{S})\) or critical symptoms \((I_{C})\) , determining the duration of their infected phase and their probability to recover (green) or decease (grey). (b) Vaccine resistance is risky under a coexistence of both infected \((\rho)\) and vaccinated \((V)\) individuals (center). When \(\rho\) is small, the probability of mutation is marginal (right); when \((V)\) is small the selection pressure for resistance is weak (left). (c) Under slow vaccination \(\rho (t)\) increases (red). As a result, when vaccines gain coverage we enter the risky zone (shaded), and become potentially vulnerable to resistance mutation. Indeed, when such mutation occurs (orange line), the trend is reversed, \(\rho (t)\) increases and the vaccine coverage \(V(t)\) plummets (blue). (d) We present several snapshots to track the state of the spread. In snapshot 2 we observe a premature mutation (orange node) that fails to spread, since \(V(t)\) at that point is still small (blue nodes). Later (snapshot 4), with the system in the risky zone of high \(\rho (t)\) and \(V(t)\) , such mutations rapidly take over, as seen by the coverage of the orange nodes in snapshot 5. (e) - (f) Rapid vaccination in and of itself may be insufficient. The system quickly enters the risky zone (shaded) and with \(R_{0} > 1\) , a single resistance mutation eventually outruns our vaccination efforts. (g) - (h) Successful eradication of the disease is achieved under a combination of rapid vaccination (blue) and suppression of \(R_{0}\) , e.g., through social distancing. Pushing \(R_{0}\) down suppresses \(\rho (t)\) , and hence avoids the risky zone by locating the system in the right hand side of panel (b). (i) The probability \(P\) to observe a pandemic state as a function of the vaccination rate \(\eta\) for different values of \(R_{0}\) . To alleviate the risk of vaccine resistant spread we must remain in the blue zone, in which we not only invest in the vaccine roll-out \((\eta)\) , but also in suppressing the spread (reducing \(R_{0}\) ). (j) Similar, albeit less dramatic results are also observed under our low risk scenario \(\mathcal{P} = 0.25\) . </center>
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1 </center>  
+
+The phases of a pandemic under pathogen mutation. (see Manuscript file for full figure caption)
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2 </center>  
+
+The transition to the mutated phase. (see Manuscript file for full figure caption)
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3 </center>  
+
+The volatile phase. (see Manuscript file for full figure caption)
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4 </center>  
+
+Vaccination under the threat of mutation. (see Manuscript file for full figure caption)  
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.
+
+<--- Page Split --->
+
+SupplementaryInformation.pdf
+
+<--- Page Split --->

preprint/preprint__b3c85f4e7eda79d6b312b4b17f3b683b2b0d1fb0bb454f9d6b0b7786fc1b5793/preprint__b3c85f4e7eda79d6b312b4b17f3b683b2b0d1fb0bb454f9d6b0b7786fc1b5793_det.mmd

@@ -0,0 +1,619 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 848, 144]]<|/det|>
+# Epidemic spreading under pathogen evolution  
+
+<|ref|>text<|/ref|><|det|>[[44, 163, 560, 203]]<|/det|>
+Xiyun Zhang Jinan University https://orcid.org/0000- 0002- 7694- 234X  
+
+<|ref|>text<|/ref|><|det|>[[44, 209, 730, 250]]<|/det|>
+Ruan Zhongyuan Institute of Cyberspace Security, Zhejiang University of Technology, Hangzhou  
+
+<|ref|>text<|/ref|><|det|>[[44, 255, 670, 296]]<|/det|>
+Muhua Zheng East China Normal University https://orcid.org/0000- 0002- 9811- 529X  
+
+<|ref|>text<|/ref|><|det|>[[44, 302, 311, 342]]<|/det|>
+Jie Zhou East China Normal University  
+
+<|ref|>text<|/ref|><|det|>[[44, 348, 208, 386]]<|/det|>
+Boccaletti Stefano CNR  
+
+<|ref|>text<|/ref|><|det|>[[44, 393, 580, 435]]<|/det|>
+Baruch Barzel ( baruchbarzel@gmail.com) Bar- llan University https://orcid.org/0000- 0001- 8862- 4384  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 476, 102, 494]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 514, 612, 534]]<|/det|>
+Keywords: pathogen evolution, epidemic, COVID- 19, mutant strain  
+
+<|ref|>text<|/ref|><|det|>[[44, 552, 296, 571]]<|/det|>
+Posted Date: April 13th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 590, 463, 609]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 373402/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 627, 909, 670]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 706, 936, 749]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on October 20th, 2022. See the published version at https://doi.org/10.1038/s41467- 022- 34027- 9.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[168, 78, 826, 103]]<|/det|>
+# Epidemic spreading under pathogen evolution  
+
+<|ref|>text<|/ref|><|det|>[[243, 112, 752, 131]]<|/det|>
+Xiyun Zhang, \(^{1,*}\) Zhongyuan Ruan, \(^{2}\) Muhua Zheng, \(^{3}\) Jie Zhou, \(^{4}\)  
+
+<|ref|>text<|/ref|><|det|>[[308, 136, 686, 153]]<|/det|>
+Stefano Boccaletti \(^{5,6,7,8,\dagger}\) & Baruch Barzel \(^{9,10,\dagger}\)  
+
+<|ref|>text<|/ref|><|det|>[[140, 159, 881, 374]]<|/det|>
+1. Department of Physics, Jinan University, Guangzhou, Guangdong 510632, China  
+2. Institute of Cyberspace Security, Zhejiang University of Technology, Hangzhou, Zhejiang 310023, China  
+3. School of Physics and Electronic Engineering, Jiangsu University, Zhenjiang, Jiangsu, 212013, China  
+4. School of Physics and Electronic Science, East China Normal University, Shanghai 200241, China  
+5. CNR - Institute of Complex Systems, Via Madonna del Piano 10, I-50019 Sesto Fiorentino, Italy  
+6. Unmanned Systems Research Institute, Northwestern Polytechnical University, Xi'an 710072, China  
+7. Moscow Institute of Physics and Technology (National Research University), 9 Institutskiy per., Dolgoprudny, Moscow Region, 141701, Russian Federation  
+8. Universidad Rey Juan Carlos, Calle Tulipán s/n, 28933 Móstoles, Madrid, Spain  
+9. Department of Mathematics, Bar-Ilan University, Ramat-Gan, 5290002, Israel  
+10. Gonda Multidisciplinary Brain Research Center, Bar-Ilan University, Ramat-Gan, 5290002, Israel  
+
+<|ref|>text<|/ref|><|det|>[[144, 377, 458, 391]]<|/det|>
+\* Correspondence: xiyunzhang@jnu.edu.cn  
+
+<|ref|>text<|/ref|><|det|>[[145, 398, 583, 413]]<|/det|>
+↑ These Authors equally contributed to the manuscript  
+
+<|ref|>text<|/ref|><|det|>[[115, 429, 881, 748]]<|/det|>
+Battling a widespread pandemic is an arms race between our mitigation efforts, e.g., social distancing or vaccination, and the pathogen's evolving persistence. This is being observed firsthand during the current COVID- 19 crisis, as novel mutations are constantly challenging our global vaccination race. To address this, we introduce here a general framework for epidemic spreading under pathogen evolution, which shows that mutations can fundamentally alter the projection of the spread. Specifically, we detect a new pandemic phase - the mutated phase - in which, despite the fact that the pathogen is initially non- pandemic ( \(R_{0} < 1\) ), it may still spread due to the emergence of a critical mutation. The boundaries of this phase portray a balance between the epidemic and the evolutionary time- scales. If the mutation rate is too low, the pathogen prevalence decays prior to the appearance of a critical mutation. On the other hand, if mutations are too rapid, the pathogen evolution becomes volatile and, once again, it fails to spread. Between these two extremes, however, a broad range of conditions exists in which an initially sub- pandemic pathogen will eventually gain prevalence. This is especially relevant during vaccination, which creates, as it progresses, increasing selection pressure towards vaccine- resistance. To overcome this, we show that vaccination campaigns must be accompanied by fierce mitigation efforts, to suppress the potential rise of a resistant mutant strain.  
+
+<|ref|>text<|/ref|><|det|>[[115, 754, 881, 858]]<|/det|>
+Evolutionary time- scales are often considered to be vast, occurring gradually over the course of millions of years. However, if prevalent enough, a species may undergo even rare mutations at relatively short time- scales. This is especially relevant during the course of a widespread and prolonged pandemic. \(^{1 - 6}\) The global spread ensures a sufficiently large pool of pathogens for mutations to occur, and on top of that, the long duration of the pandemic \(^{7 - 13}\) affords the pathogens sufficient time to evolve.  
+
+<|ref|>text<|/ref|><|det|>[[115, 865, 881, 917]]<|/det|>
+Such troubling scenario is currently unfolding in the case of COVID- 19, where novel mutations of the SARS- CoV- 2 virus continue to challenge our mitigation efforts. \(^{14 - 18}\) They are, however, equally relevant in other infections, such as influenza A, forcing us to distribute a dedicated
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 82, 881, 152]]<|/det|>
+vaccine in each yearly cycle. \(^{19 - 24}\) Another notable example is norovirus, whose enhanced transmission, likely due to mutation, led to an observable spike in gastric flu patients in England and Wales from 1991 to 2006, \(^{25}\) and finally, beyond viruses, artemisinin-resistance, a parasite mutation, rendered void the common treatment of malaria in Africa. \(^{26,27}\)  
+
+<|ref|>text<|/ref|><|det|>[[116, 160, 882, 300]]<|/det|>
+The common approach for tracking the spread of evolving pathogens is to introduce several competing strains and extract their interacting contagion process. \(^{28 - 32}\) This captures the patterns of spread of already evolved pathogens, overlooking the dynamics, and most importantly, the time- scales, of the evolution itself. Indeed, in an ongoing pandemic, mutations represent a gradual random process, in which an originally unfit pathogen mutates step- by- step via a series of small changes, until reaching a critical mutation that allows it to efficiently spread. Such process may take a significant amount of time, and, in some cases, the disease may taper off before such critical mutation has the opportunity to take over.  
+
+<|ref|>text<|/ref|><|det|>[[116, 308, 882, 430]]<|/det|>
+Another crucial aspect, absent when considering pre- mutated strains, is the fact that pathogen evolution is responsive. As we tighten our mitigation, either through prophylactic measures \(^{33 - 35}\) or via pharmaceutical interventions, we induce a selective pressure for mitigation resistance. For example, if one enforces social distancing to push the reproduction rate \(R_{0}\) below the pandemic threshold, the pathogen becomes naturally pressured towards higher transmissibility. Similarly, if one employs therapeutic treatment to expedite recovery, natural selection will push the pathogen to higher drug- persistence.  
+
+<|ref|>text<|/ref|><|det|>[[116, 436, 882, 560]]<|/det|>
+To address this, we introduce here an evolving pathogen model, which encompasses the delicate interplay between the pathogen's spread and its developing fitness. The evolution, a random walk in fitness space, is driven by the pathogen's mutation rate. At the same time the natural selection, in which the fitter strains proliferate, is pushed by the epidemiological parameters, characterizing how fast a mutated strain propagates. Together, we identify a rather broad set of conditions - the mutated phase - in which a non- pandemic pathogen will eventually reach an evolved pandemic state.  
+
+<|ref|>text<|/ref|><|det|>[[116, 567, 882, 832]]<|/det|>
+We find that besides the classic epidemiological parameters, i.e. infection/recovery rates, two additional components factor in - the mutation rate governing the evolutionary time- scales, and the number of infected individuals, which determines the likelihood of a critical mutation to occur within the relevant time- frame. Therefore, as opposed to classic pandemic transitions, which depend solely on the epidemiological parameters, \(^{1 - 5}\) here the current prevalence \(\rho (t)\) of the pathogen has direct impact on its anticipated spread. This has significant implications pertaining to our two main mitigation strategies \(\bullet\) Social distancing suppresses the reproduction number \(R_{0}\) to below the pandemic threshold. \(^{7 - 13}\) However, if many have already been infected, i.e. \(\rho (t)\) is large, then a stricter suppression may be required to avoid the emergence of a critical mutation. This indicates that the projection of the spread, and hence also its mitigation, depends on its present state \(\rho (t)\) - a hysteresis phenomenon, unobserved in the classic modeling frameworks \(^{36 - 41}\) \(\bullet\) Vaccination campaigns create strong evolutionary pressure towards a vaccine resistant mutation, whose risk, once again, is directly related to the current pathogen prevalence. Hence, to succeed, we show that vaccine roll- out must be coupled with fierce suppression via social distancing.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 80, 420, 101]]<|/det|>
+## Evolving pathogen model  
+
+<|ref|>text<|/ref|><|det|>[[115, 105, 881, 247]]<|/det|>
+Consider a social random network of \(N\) individuals linked through the adjacency matrix \(A\equiv\) \(\{A_{i j}\}\) and with average degree \(\overline{k}\) . At \(t = 0\) the network experiences an outbreak, which then spreads via the susceptible- infected- susceptible42 (SIS) dynamics. In the classic SIS formulation, the projected spread is driven by two time- independent parameters: the recovery rate \(\mu\) and the infection rate \(\beta\) , whose ratio \(R_{0} = \overline{k}\beta /\mu\) , the reproduction number, determines the state of the system - pandemic ( \(R_{0}\geq 1\) ) or healthy ( \(R_{0}< 1\) ). Here, however, the pathogen is allowed to evolve, therefore these parameter may change over the course of the spread. This is captured by the individual recovery rate  
+
+<|ref|>equation<|/ref|><|det|>[[434, 263, 878, 296]]<|/det|>
+\[\mu_{i}(t) = \frac{1}{F_{i}(t)}\mu , \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 306, 881, 395]]<|/det|>
+where the fitness \(F_{i}(t)\) stands for the level of mutation of the pathogen carried by individual \(i\) at time \(t\) , hence the unmutated pathogen has \(F_{i}(0) = 1\) . The above equation models the fact that (i) each individual \(i\) may carry a distinct version of the virus; (ii) this version may gradually change in time \(t\) due to mutations. The smaller is \(\mu_{i}(t)\) , the higher is the transmissibility of the pathogen, as described by the evolving reproduction number  
+
+<|ref|>equation<|/ref|><|det|>[[398, 410, 878, 449]]<|/det|>
+\[R_{i}(t) = \frac{\overline{k}\beta}{\mu_{i}(t)} = R_{0}F_{i}(t). \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 457, 881, 528]]<|/det|>
+Indeed, a low rate of recovery \(\mu_{i}(t)\) extends the duration of the infectious state, providing individual \(i\) with more opportunities to infect their peers. Hence, as the r.h.s. of (2) indicates, pathogens with increased \(F_{i}(t)\) exhibit higher reproduction, and therefore spread more efficiently than their lower fitness competitors.  
+
+<|ref|>text<|/ref|><|det|>[[115, 535, 881, 641]]<|/det|>
+Mutation may also impact the transmissibility of the pathogen directly by altering the value of the infection rate \(\beta\) , e.g., by evolving a more infectious strain. However, in the SIS framework, the relevant parameter is not \(\mu\) nor \(\beta\) , but their ratio, as provided by \(R_{i}(t)\) .8,11,12,43,44 Therefore, for simplicity, in (1) we only track the pathogen evolution through \(\mu_{i}(t)\) , and its subsequent \(R_{i}(t)\) , setting \(\beta\) stationary. To complement this analysis, in Supplementary Section 1 we examine the case of \(\beta\) - mutations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 648, 881, 682]]<|/det|>
+The spread is driven by the infection, recovery and mutation processes. The process of infection between a pair of individuals \(i\) and \(j\) is modeled by  
+
+<|ref|>equation<|/ref|><|det|>[[404, 700, 878, 754]]<|/det|>
+\[\begin{array}{r l r}{{S_{i}+I_{j}}}&{{\xrightarrow{A_{i j}\beta}}I_{i}+I_{j}}\\ {{}}&{{}}&{{}}\\ {{F_{i}(t)}}&{{=}}&{{F_{j}(t),}}\end{array} \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 763, 881, 850]]<|/det|>
+in which a susceptible \((S)\) individual \(i\) interacts with their infected \((I)\) neighbor \(j\) ( \(A_{i j} = 1\) ) at rate \(\beta\) . This leads to both individuals becoming infected. The newly infected individual \(i\) inherits \(j\) 's pathogen, and hence in (4) we set \(i\) 's fitness at the time of infection equal to that of \(j\) . Both fitness parameters, \(F_{i}(t)\) and \(F_{j}(t)\) may later change via mutation. Next, we consider the process of recovery  
+
+<|ref|>equation<|/ref|><|det|>[[452, 869, 878, 893]]<|/det|>
+\[I_{i}\xrightarrow{\mu_{i}(t)}S_{i}, \quad (5)\]
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 880, 117]]<|/det|>
+in which an infected individual \(I_{i}\) transitions to \(S_{i}\) at the evolved recovery rate \(\mu_{i}(t)\) of (1). Finally, the process of mutation follows  
+
+<|ref|>equation<|/ref|><|det|>[[337, 135, 878, 192]]<|/det|>
+\[\left\{ \begin{array}{l l}{F_{i}(0) = 1}\\ {F_{i}(t + 1) = \max \left(F_{i}(t) + \delta_{i}(t)~,~0\right)} \end{array} \right., \quad (6)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 881, 309]]<|/det|>
+capturing a random walk with variable step size \(\delta_{i}(t)\) , i.e. a sequence of random shifts in fitness, caused by small discrepancies in the pathogen's reproduction. Note that \(F_{i}(t)\) is prohibited from becoming negative, as, indeed, a below zero fitness in (2) is meaningless. The case where \(F_{i}(t)\) does approach zero corresponds to \(\mu_{i}(t) \to \infty\) in (1), a limit in which recovery is instantaneous, and hence the pathogen is unfit for reproduction. Such strains will be rapidly eliminated from the pathogen pool.  
+
+<|ref|>text<|/ref|><|det|>[[115, 315, 881, 421]]<|/det|>
+The magnitude of each mutation step is extracted from a zero- mean normal distribution, namely \(\delta_{i}(t) \sim \mathcal{N}(0, \sigma^{2})\) . Consequently, in the limit where \(\sigma = 0\) , we have \(\delta_{i}(t) = 0\) at all times, mutations are suppressed, and Eqs. (3) - (6) converge to the classic SIS model, with \(R_{i}(t) = R_{0}\) , a constant reproduction number. In contrast, as \(\sigma\) is increased, significant mutations become more frequent and the pathogens rapidly evolve. We therefore vary \(\sigma\) to control the mutation rate of the pathogens.  
+
+<|ref|>text<|/ref|><|det|>[[115, 427, 881, 533]]<|/det|>
+Taken together, our modeling framework accounts for the dynamics of infection and recovery (SIS) under the effect of pathogen mutation. As the spread progresses, pathogens evolve via Eq. (6), blindly altering their epidemiological parameters at random. Natural selection, however, will favor the positive mutations, in which \(\delta_{i}(t) > 0\) . Indeed, such mutations lead to higher fitness, reducing the recovery rate \(\mu_{i}(t)\) , and consequently increasing \(R_{i}(t)\) . Such pathogens, with increased \(R_{i}(t)\) , will proliferate more rapidly, and will eventually dominate the population.  
+
+<|ref|>text<|/ref|><|det|>[[115, 539, 881, 645]]<|/det|>
+Critical mutation. Consider an outbreak of a pathogen with \(R_{0} < 1\) , i.e. below the epidemic threshold. This can be either due to the pathogen's initial sub- pandemic parameters, or a result of mitigation, e.g., social distancing to reduce \(\beta\) . In the classic SIS formulation, such pathogen with fail to penetrate the network. However, in the presence of mutations ( \(\sigma > 0\) ) the pathogen may potentially undergo selection, reach \(R_{i}(t) > 1\) , and from that time onward begin to proliferate. This represents a critical mutation, which, using (2), translates to  
+
+<|ref|>equation<|/ref|><|det|>[[459, 661, 877, 694]]<|/det|>
+\[F_{c} = \frac{1}{R_{0}}, \quad (7)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 703, 880, 755]]<|/det|>
+the critical fitness that, once crossed, may lead an initially non- pandemic pathogen to become pandemic. The smaller is \(R_{0}\) the higher is \(F_{c}\) , as, indeed, weakly transmissible pathogens require a longer evolutionary path to reach pandemic spread.  
+
+<|ref|>text<|/ref|><|det|>[[115, 761, 880, 795]]<|/det|>
+Next, we analyze the spreading patterns of our evolving pathogens, seeking the conditions for the appearance of the critical mutation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 809, 560, 830]]<|/det|>
+## Phase-diagram of evolving pathogens  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 881, 907]]<|/det|>
+To examine the behavior of (1) - (6) we constructed an Erdős- Rényi (ER) network with \(N = 5,000\) nodes and \(\overline{k} = 15\) , providing a testing ground upon which we incorporate a series of epidemic scenarios (Fig. 1). Each scenario is characterized by a different selection of our model's three epidemiological parameters: \(\mu\) and \(\beta\) , which determine the pathogen's unmutated repro
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 881, 136]]<|/det|>
+duction \(R_{0}\) , and \(\sigma\) , which controls the rate of mutation. We then follow the spread by measuring the prevalence \(\rho (t)\) , which monitors the fraction of infected individuals vs. time. We also track the pathogen's evolution via the population averaged fitness \(\overline{F} (t) = (1 / N)\sum_{i = 1}^{N}F_{i}(t)\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 141, 881, 230]]<|/det|>
+Pandemic (Fig. 1a, red). In our first scenario we set \(\mu = 0.1\) , \(\beta = 8\times 10^{- 3}\) and \(\sigma = 10^{- 2}\) . This captures a pandemic pathogen, which, using \(\overline{k} = 15\) , has \(R_{0} = 1.2 > 1\) , namely it can spread even without mutation. Indeed \(\rho (t)\) rapidly climbs to gain macroscopic coverage, congruent with the prediction of the classic SIS model, but this time constantly growing, due to the gradual, but continuous, increase in fitness \(\overline{F} (t)\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 237, 881, 377]]<|/det|>
+Mutated (Fig. 1c, green). Next we reduce the infection rate to \(\beta = 1.67\times 10^{- 3}\) , an initial reproduction of \(R_{0} = 0.25< 1\) . This describes a pathogen whose transmissibility is significantly below the epidemic threshold, and therefore, following the initial outbreak we observe a decline in \(\rho (t)\) , which by \(t\sim 50\) almost approaches zero, as the disease seems to be tapering off. In this scenario, however, we set a faster mutation rate \(\sigma = 1\) . As a result, despite the initial remission, at around \(t\sim 15\) , the pathogen undergoes a critical mutation as \(\overline{F} (t)\) crosses the critical \(F_{c} =\) \(1 / R_{0} = 4\) (grey dashed line) and transitions into the pandemic regime. Consequently, \(\rho (t)\) changes course, the disease reemerges and the mutated pathogens successfully spreads.  
+
+<|ref|>text<|/ref|><|det|>[[115, 383, 881, 471]]<|/det|>
+Lethargic (Fig. 1b, blue). We now remain in the sub- pandemic regime, with \(R_{0} = 0.25\) , but with a much slower mutation rate, set again to \(\sigma = 10^{- 2}\) . As above, \(\rho (t)\) declines, however the pathogen evolution is now too slow - it is lethargic, and cannot reach critical fitness on time. Therefore, the disease fails to penetrate the network, lacking the opportunity for the critical mutation to occur.  
+
+<|ref|>text<|/ref|><|det|>[[115, 478, 881, 566]]<|/det|>
+Taken together, the dynamics of the spread are driven by three parameters: the initial epidemiological characteristics of the pathogen, \(\mu\) and \(\beta\) , which determine \(R_{0}\) , and the mutation rate \(\sigma\) , which governs the time- scale for the appearance of the critical mutation. Therefore, to determine the conditions for a mutation- driven contagion, as observed in Fig. 1c, we investigate the balance between the decay in \(\rho (t)\) vs. the gradual increase in \(\overline{F} (t)\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 574, 297, 589]]<|/det|>
+## The mutated phase  
+
+<|ref|>text<|/ref|><|det|>[[115, 596, 880, 630]]<|/det|>
+To understand the dynamics of the evolving pathogen model, we show in Supplementary Section 2 that at the initial stages of the spread, the prevalence \(\rho (t)\) follows  
+
+<|ref|>equation<|/ref|><|det|>[[433, 653, 878, 672]]<|/det|>
+\[\rho (t) = \rho (0)e^{\xi (t)}. \quad (8)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 685, 881, 718]]<|/det|>
+The time- dependent exponential rate \(\xi (t)\) is determined by the epidemiological/mutation rates via  
+
+<|ref|>equation<|/ref|><|det|>[[361, 737, 878, 768]]<|/det|>
+\[\xi (t) = -\mu (1 - R_{0})t + \frac{1}{2}\sigma^{2}\mu^{2}R_{0}^{2}t^{3}, \quad (9)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 776, 881, 917]]<|/det|>
+whose two terms characterize the pre- mutated vs. post- mutated spread of the pathogen. The first term, linear in \(t\) , represents the initial patterns of spread, which are determined by the original pathogen parameters, \(\mu ,R_{0}\) . For \(R_{0}< 1\) this describes an exponential decay, a la SIS dynamics in the sub- pandemic regime. At later times, however, as \(t^{3}\) becomes large, the second term begins to dominate, and the exponential decay is replaced by a rapid proliferation, now driven by the mutation rate \(\sigma\) . The transition between these two behaviors - decay vs. proliferation - occurs at \(\tau_{c} = \sqrt{2(1 - R_{0}) / 3\mu\sigma^{2}R_{0}^{2}}\) , which provides the anticipated time- scale for the appearance of the critical mutation \(F_{c}\) in (7).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 881, 152]]<|/det|>
+This analysis portrays the mutated phase as a balance between two competing time- scales: on the one hand the exponential decay of the sub- pandemic pathogen, and on the other hand the evolutionary time- scale \(\tau_{c}\) for the appearance of the critical mutation. For the evolution to win this race the pathogen must not vanish before \(t = \tau_{c}\) . This imposes the condition (Fig. 2a- c)  
+
+<|ref|>equation<|/ref|><|det|>[[451, 170, 878, 201]]<|/det|>
+\[\rho (\tau_{c})\geq \frac{1}{N}, \quad (10)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 210, 881, 280]]<|/det|>
+ensuring that at \(\tau_{c}\) there are still one or more individuals hosting the pathogen. Indeed, \(\rho (\tau_{c})< 1 / N\) indicates that on average, at \(t = \tau_{c}\) less than a single individual is left in the infected pool. Under this condition, the critical mutation is too late, the spread has already tapered off, and the exponential growth driven by the positive term in (9) is averted.  
+
+<|ref|>text<|/ref|><|det|>[[115, 286, 880, 321]]<|/det|>
+Taking \(\rho (\tau_{c})\) from (8), we can now use (10) to express the boundary of the mutated phase, predicting the critical mutation rate as (Supplementary Section 2)  
+
+<|ref|>equation<|/ref|><|det|>[[376, 339, 878, 380]]<|/det|>
+\[\sigma_{c}\sim \left(\frac{\sqrt{\mu(1 - R_{0})^{3}}}{2R_{0}}\right)\frac{1}{\ln(\mathcal{I}_{0})}, \quad (11)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 390, 881, 496]]<|/det|>
+where \(\mathcal{I}_{0} = N\rho (t = 0)\) is the number of individuals infected at \(t = 0\) . Equation (11) describes the minimal mutation rate required for the pathogen to evolve a pandemic strain. For \(R_{0} = 1\) it predicts \(\sigma_{c} = 0\) , as such pathogen can indeed spread even without mutation. However, as \(R_{0}\) is decreased, for example under mitigation, the pathogen prevalence rapidly declines, and hence it must evolve at an accelerated rate to reach critical fitness. This is expressed in (11) by an increased \(\sigma_{c}\) , which approaches infinity as \(R_{0}\to 0\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 503, 881, 696]]<|/det|>
+To test our predicted phase transition we simulate in Fig. 1d an array of 1,050 realizations of Eqs. (1) - (6), representing different epidemiological scenarios. We varied \(R_{0}\) from 0 to 1.5, i.e. from non- transmissible to highly contagious, and scanned a spectrum of mutation rates from \(\sigma = 10^{- 3}\) to \(\sigma = 10\) , spanning four orders of magnitude. Simulating each scenario 50 times we observe the probability \(P\) for the disease to spread. This is done by tracking the pathogen's long- term prevalence \(\rho = \rho (t\to \infty)\) and counting the realizations in which \(\rho \to 0\) vs. those where \(\rho >0\) . As predicted, we find that the pandemic state, classically observed only at \(R_{0}\geq 1\) , now extends to lower \(R_{0}\) in the presence of sufficiently rapid mutations. This gives rise to the mutated phase (green), in which an initially decaying contagion suddenly turns pandemic. The transition between the lethargic and the mutated states (grey zone) is well- approximated by our theoretical prediction of Eq. (11), as depicted by the black solid line.  
+
+<|ref|>text<|/ref|><|det|>[[115, 704, 881, 915]]<|/det|>
+Equation (11) shows that \(\sigma_{c}\) depends not only on the epidemiological characteristics of the pathogen \((\mu ,R_{0})\) , but also on the initial condition, here captured by the number of infected individuals \(\mathcal{I}_{0} = \rho (t = 0)N\) . If \(\mathcal{I}_{0}\) is large the critical rate \(\sigma_{c}\) becomes lower, in effect expanding the bounds of the mutated phase. To understand this consider the evolutionary paths followed by the pathogens as they reproduce. These paths represent random trajectories in fitness space, each starting from \(F_{i}(0) = 1\) , and with a small probability crossing the critical fitness \(F_{c}\) . The more such attempts are made, the higher the chances that at least one of these paths will be successful. Therefore, a higher initial prevalence \(\mathcal{I}_{0}\) of the pathogen increases the probability for the appearance of a critical mutation, enabling a mutated phase even with low \(\sigma\) . In simple words, even rare mutations may occur if the initial pathogen pool \((\mathcal{I}_{0})\) is large enough. Indeed, in Fig. 2d we find that the phase boundary shifts towards lower \(\sigma_{c}\) as the initial prevalence is increased (grey shaded lines). Hence, a greater \(\mathcal{I}_{0}\) , indeed, expands the mutated phase.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 882, 241]]<|/det|>
+Hysteresis. This dependence on \(\mathcal{I}_{0}\) indicates that the transition of Eq. (11) behaves differently if we approach it from the pandemic state or from the healthy state. To observe this let us fix the mutation rate at \(\sigma = 0.1\) and gradually increase \(R_{0}\) , seeking the critical point where the system shifts to the mutated phase. This is mapped to a vertical trajectory in the \(\sigma ,R_{0}\) plane (Fig. 2d, yellow dashed line). At each value of \(R_{0}\) we instigate an outbreak with \(\rho (0) = 0.2\) , and observe its long- term prevalence \(\rho\) . For small \(R_{0}\) this outbreak decays and the system reverts to the healthy state \(\rho = 0\) . However, as we transition into the mutated phase, here predicted at \(R_{0} = R_{\mathrm{High}}\approx 0.6\) , the pathogen turns pandemic and its prevalence abruptly changes to \(\rho \approx 0.85\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 248, 882, 372]]<|/det|>
+To reverse this transition the naive approach is to push \(R_{0}\) slightly below this critical point, for instance, by practicing social distancing to reduce transmission. The challenge is that now, moving in the opposite direction - from large to small \(R_{0}\) - our initial condition is pandemic, with prevalence of order unity ( \(\sim 85\%\) ), and hence \(\mathcal{I}_{0}\sim N\) . Under these conditions, Eq. (11) predicts that, for our fixed \(\sigma\) , the critical \(R_{0}\) is now lower, at \(R_{\mathrm{Low}} = 0.35\) . This results in a hysteresis phenomenon, in which criticality occurs at different points depending on the state from which we approach the transition (Fig. 2e).  
+
+<|ref|>text<|/ref|><|det|>[[115, 377, 882, 501]]<|/det|>
+We find, therefore, that pathogen evolution fundamentally changes the phase space of epidemic spreading. First it predicts a broad range of conditions - the mutated phase - in which a sub- pandemic pathogen can gain prevalence. On top of that, it also predicts that this phase exhibits a discontinuous transition, characterized by hysteresis, a phenomenon unobserved in the classic SIS dynamics, yet congruent with other models \(^{38,41,45 - 49}\) that incorporate feedback between a pathogen's prevalence ( \(\rho (t)\) ) and its potency ( \(R_{i}(t)\) ). These two observations have direct implications on mitigation:  
+
+<|ref|>text<|/ref|><|det|>[[144, 508, 882, 666]]<|/det|>
+- Soft mitigation is risky. Most mitigation strategies seek a minimal approach, aiming to drive \(R_{0}\) just below unity. This is understandable as (i) major restrictions on social interactions are costly and difficult to sustain \(^{50}\) for extended periods; (ii) having \(R_{0}< 1\) , even by a small margin, is assumed to naturally suppress the spread, as it leads \(\rho (t)\) to decay exponentially towards zero. Our analysis, however, shows that this is insufficient. For \(R_{0}\lesssim 1\) we have \(\sigma_{c}\to 0\) , indicating that even a relatively stable pathogen, with a low mutation rate, may eventually break through. Using Eq. (11) we can predict for a given \(\sigma\) , the level of tolerable \(R_{0}\) that is sufficient to mitigate the mutated phase risk, providing guidelines for effective mitigation.  
+
+<|ref|>text<|/ref|><|det|>[[144, 673, 882, 831]]<|/det|>
+- The sooner the better. Another common assumption, driven by the classic epidemic phase-diagram, is that the projected state \(\rho (t\to \infty)\) depends only on \(R_{0}\) , i.e. the epidemiological parameters. The current state of the spread \(\rho (0)\) at the time we implement our mitigation, plays no role. The observed hysteresis, however, shows that successful mitigation strongly depends on the prevalence at the time of instigation. If the pathogen has already gained sufficient ground, we will need to suppress the reproduction number below \(R_{\mathrm{Low}}\) , namely the lower phase-boundary in Fig. 2e. It is, therefore, crucial to respond early, and initiate our mitigation when \(\rho (t)\) is still small, eradicating the pandemic before mutations may determine a risk for its reemergence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 838, 881, 907]]<|/det|>
+Bounded fitness. Our mutation process in Eq. (6) allows the pathogen an unbounded random walk in fitness space. In reality, however, there are practical restrictions on fitness, as \(R_{i}(t)\) cannot grow ad infinitum. Therefore, we now consider our evolving pathogen model, substituting the mutation in (6) with
+
+<--- Page Split --->
+<|ref|>equation<|/ref|><|det|>[[305, 95, 878, 122]]<|/det|>
+\[F_{i}(t + 1) = \min \left(F_{\mathrm{max}},\max \left(F_{i}(t) + \delta_{i}(t),0\right)\right), \quad (12)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 130, 882, 237]]<|/det|>
+in which the pathogen fitness is bounded from above by \(F_{\mathrm{max}}\) and from below by zero. Setting \(F_{\mathrm{max}} = 20\) we now revisit our phase- diagram (Fig. 3a). For small \(\sigma\) , mutations are slow, and the evolution path is unaffected by the upper bound on \(F_{i}(t)\) . Therefore, we continue to observe the same transition as in the unbounded model of Fig. 1d. As we increase \(\sigma\) , however, we witness a second transition, this time back to the healthy state, indicating that now, mutations are too rapid. This captures the final phase of our evolving pathogen model - the volatile phase:  
+
+<|ref|>text<|/ref|><|det|>[[115, 242, 881, 313]]<|/det|>
+Volatile (Fig. 3, blue). When the mutation rate is too high the pathogen fitness becomes unstable. On the one hand it can rapidly reach critical fitness, yet, on the other hand, due to the random nature of its frequent mutations, it fails to sustain this fitness - resulting in an irregular \(\overline{F} (t)\) , that fluctuates above and below the critical \(F_{c}\) (Fig. 3c).  
+
+<|ref|>text<|/ref|><|det|>[[115, 318, 881, 444]]<|/det|>
+To gain deeper insight into the volatile phase, consider the natural selection process, here driven by the reproduction benefit of the fitter strains. This process is not instantaneous, and requires several reproduction instances, i.e. generations, to gain a sufficient spreading advantage. With \(\sigma\) too high, natural selection is confounded, the pathogen shown no consistent gain in fitness and, as Fig. 3c indicates, \(\rho (t)\) decays exponentially to zero. In Supplementary Section 3 we use a time- scale analysis, similar to the one leading to Eq. (11), to show that the volatile phase occurs when \(\sigma\) exceeds  
+
+<|ref|>equation<|/ref|><|det|>[[394, 460, 878, 500]]<|/det|>
+\[\sigma_{c}\sim \sqrt{\frac{\mu}{3}}\frac{(F_{\mathrm{max}}R_{0} - 1)^{\frac{3}{2}}}{R_{0}}. \quad (13)\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 507, 880, 543]]<|/det|>
+This prediction is, indeed, confirmed by our simulated phase diagram in Fig. 3a (black solid line).  
+
+<|ref|>text<|/ref|><|det|>[[122, 553, 875, 696]]<|/det|>
+Our phase- diagram illustrates the different forces governing the spread of pathogens in the presence of mutations. While spread is prohibited classically for \(R_{0}< 1\) , here we observe a new, previously undocumented pandemic phase, in which the disease can successfully permeate despite having an initially low reproduction rate. The conditions for this phase require a balance between three separate time- scales: (i) The time for the initial outbreak \(\rho (0)\) to reach near zero prevalence \(\tau_{r}\) ; (ii) The time for the pathogen to evolve beyond critical fitness \(\tau_{c}\) ; (iii) The time for the natural selection to lock- in the fitter mutations \(\tau_{s}\) . Pathogens with small \(R_{0}\) , we find, can still spread provided that  
+
+<|ref|>equation<|/ref|><|det|>[[446, 720, 871, 737]]<|/det|>
+\[\tau_{r} > \tau_{c} > \tau_{s}. \quad (14)\]  
+
+<|ref|>text<|/ref|><|det|>[[122, 748, 875, 873]]<|/det|>
+The l.h.s. of (14) ensures that the pathogen can reach critical fitness before reaching zero prevalence. This gives rise to the first transition of Eq. (11), between the lethargic and the mutated phases. The r.h.s. of (14) is responsible for the second transition, from mutated to volatile. It ensures that fitter pathogens do not undergo additional mutation before they have time to proliferate via natural selection. Therefore, we observe a Goldilocks zone, in which the mutation rate \(\sigma\) is just right: on the one hand, enabling unfit pathogens to cross the Rubicon towards pandemicity, but on the other hand, avoiding aimless capricious mutations.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 80, 846, 100]]<|/det|>
+## Mutation risk in vaccine distribution - the case of COVID-19  
+
+<|ref|>text<|/ref|><|det|>[[115, 106, 882, 266]]<|/det|>
+Vaccination during an ongoing pandemic is, by nature, a competition between the rate of the vaccine roll- out and the spread of the pathogen. \(^{51 - 53}\) Therefore, naively, to win this race all one has to do is disseminate the vaccine as efficiently as possible, aiming to reach the majority of the population before the pathogen does. This, however, ignores the role of mutations, which may gravely impact even the most efficient vaccination campaign. Such mutations may, generally, be less fit epidemiologically, i.e. have a lower \(F\) and consequently a lower \(R_{i}(t)\) . Therefore, absent a vaccine, they will be rapidly overcome by the faster spreading pathogen strains. However, once the vaccine becomes widespread, resistance, even if less contagious, becomes a highly desirable trait, and a resistant mutation, if occurs, will take over the population.  
+
+<|ref|>text<|/ref|><|det|>[[115, 272, 882, 430]]<|/det|>
+To examine this in a realistic setting we consider the spread of SARS- CoV- 2, currently battled by a global vaccination effort. To model the disease dynamics we collected data on the COVID- 19 infection cycle (Fig. 4a), which includes a well- documented and elaborate set of transitions. \(^{54 - 64}\) Upon infection, individuals enter a pre- symptomatic state, which lasts, on average 5 days. During this period, typically within \(2 - 4\) days they begin to shed the virus and infect their network contacts (PS, purple). This continues until the onset of mild \((I_{M})\) , severe \((I_{S})\) or critical \((I_{C})\) symptoms, at which point they enter isolation and cease to spread the virus. A fraction \((\sim 30\%)\) of infected individuals never go on to develop noticeable symptoms (AS, top arrow), and hence they continue to spread the virus until their full recovery \((R)\) , typically within \(\sim 7\) days.  
+
+<|ref|>text<|/ref|><|det|>[[115, 436, 881, 524]]<|/det|>
+To evaluate the infection rate \(\beta\) we used empirical data on the observed spread in 12 different countries. \(^{68}\) Focusing on the early stages of the contagion, prior to the instigation of mitigation strategies, we find that \(\beta = 5 \times 10^{- 2}\) best fits the observed spreading dynamics. This corresponds to a reproduction rate of \(R_{0} \approx 2.6\) , congruent with existing valuations of \(R_{0}\) under COVID- 19. \(^{55,69}\) For details on the data analysis see Supplementary Section 4.  
+
+<|ref|>text<|/ref|><|det|>[[115, 532, 654, 548]]<|/det|>
+Here we complement this disease cycle by two additional processes  
+
+<|ref|>text<|/ref|><|det|>[[140, 555, 882, 720]]<|/det|>
+- Vaccination. The population is vaccinated at a rate \(\nu\) , quantifying the percentage of the (susceptible) population that receives the vaccine per unit time (day).- Resistance. At each time-step, the pathogen may undergo a vaccine-resistant mutation with probability \(p\) . This mutation has no bearing on its epidemiological parameters \(\mu , \beta\) , thus providing no additional spreading advantage, other than being resistant to the vaccine. The larger is the infected population \((\rho (t)N)\) the greater is the risk for such mutation, hence we quantify the mutation risk via \(\mathcal{P} = pN\) , and examine two scenarios: high risk with \(\mathcal{P} = 2.5\) and low risk, setting \(\mathcal{P} = 0.25\) . For a population of \(N \sim 10^{9}\) both cases capture a very rare mutation with \(p \sim 10^{-9}\) and \(10^{-10}\) , respectively.  
+
+<|ref|>text<|/ref|><|det|>[[115, 726, 882, 850]]<|/det|>
+Two factors drive the level of risk in this process. The prevalence \(\rho (t)\) determines the size of the pathogen pool, which must be large for the rare mutation to be realized. The vaccine coverage \(V(t)\) determines the selective advantage of the resistant strain, which becomes marginal if only a small fraction of the population is inoculated. Therefore the highest risk occurs under the coexistence of both infected and vaccinated individuals. This enables the interaction between these two populations paving the way for both mutation (large \(\rho (t)\) ) and selection (large \(V(t)\) ), and hence potentially driving the system towards vaccine resistance (Fig. 4b).  
+
+<|ref|>text<|/ref|><|det|>[[115, 856, 882, 909]]<|/det|>
+To observe this we simulated three vaccination strategies, under the high risk \(\mathcal{P} = 2.5\) scenario- Slow (Fig. 4c,d). First we assume a slow vaccination rate of \(\eta = 10^{- 3}\) , a \(0.1\%\) daily coverage. Such slow vaccination is insufficient to suppress \(\rho (t)\) , allowing us, after some time to enter the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 882, 330]]<|/det|>
+risk zone in which \(\rho (t)\) coexists with \(V(t)\) (shaded). Mutations occurring within this window (orange) are likely to proliferate. Indeed, we find that in the long term, vaccination fails, and the resistant strain gains coverage. \(\bullet\) Rapid (Fig. 4e,f). To overcome this we simulate a rapid vaccine roll- out with \(\eta = 10^{- 2}\) , capturing an optimistic scenario, in which \(1\%\) of the population is inoculated per day. Despite these favorable conditions we continue to enter the risk zone, as the pathogen is allowed to spread freely in parallel to our vaccination efforts. The result is, as before, an increased likelihood of a resistant mutation, which, once again, regardless of our efficient dissemination, renders our vaccination void. \(\bullet\) Combined effort (Fig. 4g,h). The only way to avoid the risk zone is to minimize the potential interaction between infected and vaccinated individuals. Since \(V(t)\) will inevitably grow - indeed, this is the goal of vaccine distribution - we must contain \(\rho (t)\) , namely aim for the right- most branch of the risk curve in Fig. 4b. This requires a combined effort of both rapid vaccination ( \(\eta = 10^{- 2}\) ) and fierce mitigation to suppress \(R_{0}\) . The result is a successful elimination of the pathogen with \(V(t) \to 1\) and \(\rho (t) \to 0\) .  
+
+<|ref|>text<|/ref|><|det|>[[115, 336, 882, 442]]<|/det|>
+In Fig. 4i,j we systematically plot the spreading probability \(P\) in function of \(\eta , R_{0}\) under our high/low risk scenarios. We find that for COVID- 19, having \(R_{0} \approx 2.6\) (black solid line) the risk of vaccine resistance is significant, even under large \(\eta\) . Reducing \(R_{0}\) via social distancing helps alleviate this risk. For example, for \(\mathcal{P} = 2.5\) , even is we assume a rapid roll- out (large \(\eta\) ), we must reach \(R_{0} \lesssim 2\) to remain within a low mutation risk (blue). Under \(\mathcal{P} = 0.25\) , it is sufficient to aim for \(R_{0} \lesssim 3\) , roughly the natural state of SARS- CoV- 2.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 456, 393, 476]]<|/det|>
+## Discussion and outlook  
+
+<|ref|>text<|/ref|><|det|>[[115, 483, 882, 623]]<|/det|>
+The phase diagram of epidemic spreading is a crucial tool for forecasting and mitigating pandemic risks. First, it identifies the relevant control parameters, such as \(\mu , \beta\) and \(\bar{k}\) in our SIS framework, or additional parameters in more complex contagion processes, whose value determines \(R_{0}\) and hence the expected patterns of spread. The phase boundaries, then, help us assess the state of the system - healthy or pandemic - and provide guidelines for our response. For example, social distancing to reduce \(\bar{k}\) , therapeutic treatment to increase \(\mu\) or mask wearing to suppress \(\beta\) , all aimed to navigate the system's location along the pandemic phase- diagram towards the desired healthy state.  
+
+<|ref|>text<|/ref|><|det|>[[115, 630, 882, 841]]<|/det|>
+The common thread binding all of these strategies is the assumption that the epidemiological control parameters themselves are constant in time, and hence our intervention must just push them beyond the static phase- boundary, from which point on the epidemic will decay towards \(\rho \to 0\) spontaneously. This is, indeed, relevant if the temporal evolution of \(\mu , \beta\) is slow compared to the epidemic spreading dynamics - as observed in the case of our lethargic phase. However, once the epidemiological parameters can change at a sufficiently high rate, it fundamentally changes the rules of the game. This is because now, not only are the parameters dynamic, but, thanks to natural selection, they also become responsive. If, for instance, we develop drug- based treatment to increase the recovery rate \(\mu\) , we inevitably also generate selection pressure towards drug persistence. Similarly, if we vaccinate or practice distancing to reduce \(\bar{k}, \beta\) , we initiate an evolutionary race towards higher transmissibility or vaccine resistance. This was clearly observed in our analysis of the COVID- 19 vaccine dissemination.  
+
+<|ref|>text<|/ref|><|det|>[[115, 848, 882, 918]]<|/det|>
+The result is a complex interplay between the spreading dynamics ( \(R_{0}\) ), the instantaneous prevalence of the pathogen ( \(\rho (t)\) ), and the dynamic evolution of its parameters ( \(\sigma\) ), which reshapes the pandemic phase diagram. It not only expands the pandemic risk to a range of \(R_{0} < 1\) , but also predicts an explosive transition pattern, i.e. the hysteresis of Fig. 2e, that is
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 882, 294]]<|/det|>
+not observed in standard epidemiological transitions. This altered phase diagram, and its abrupt first- order like transition, we have shown, has crucial implications pertaining to our mitigation strategies. Yet, more broadly, as a physical phenomenon, it offers an interesting mechanism for explosive transitions. Most often, such abrupt phase- shifts are caused by internal suppression rules, that hold back the transition until it breaks through in an explosive fashion. \(^{37,70 - 72}\) In contrast, here what holds back the transition is the waiting time for the critical mutation. Until its appearance the system behaves in one way ( \(R_{0} < 1\) ), but once it occurs, the system suddenly enters the pandemic regime ( \(R_{0} > 1\) ). The explosiveness is therefore traced to a local event, whose probability depends on the system's initial parameters ( \(R_{0}, \mathcal{I}_{0}, \sigma\) ). This local event then changes fundamentally the state of the system - capturing a feedback between the system's phase and its intrinsic control parameters. We believe this describes a unique mechanism, inherent to the basic ingredients of our biological system, reproduction, mutation and selection.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 308, 344, 328]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[116, 335, 882, 422]]<|/det|>
+X.Z. thanks Dr. Xiaobo Chen, Dr. Tingting Shi, Dr. Xing Lu and Prof. Weirong Zhong for useful discussions and supports in numerical calculations. This work was partially supported by the National Natural Science Foundation of China under Grants No. 12075008 and No. 1200050749. This research was also supported by the Israel Science Foundation (grant No. 499/19) and the Bar- Ilan University Data Science Institute grant for COVID- 19 related research.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 436, 360, 455]]<|/det|>
+## Author contribution  
+
+<|ref|>text<|/ref|><|det|>[[116, 462, 881, 515]]<|/det|>
+X.Z. developed the concept. X.Z., B.B. and S.B. designed the framework. X.Z. and Z.R. performed the numerical simulations. All authors jointly analyzed the results and developed the analytical framework. X.Z., B.B. and S.B. wrote the paper.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 529, 317, 549]]<|/det|>
+## Code availability  
+
+<|ref|>text<|/ref|><|det|>[[115, 556, 880, 590]]<|/det|>
+All code to study and reproduce the results shown here will be made freely available online upon publication.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 80, 245, 100]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[117, 110, 884, 147]]<|/det|>
+[1] Hufnagel, L., Brockmann, D. and Geisel, T., Forecast and control of epidemics in a globalized world, Proc. Natl. Acad. Sci. U. S. A. 101 15124- 15129 (2004).  
+
+<|ref|>text<|/ref|><|det|>[[125, 158, 881, 211]]<|/det|>
+[2] Colizza, V., Barrat, A., Barthélemy, M. and Vespignani, A., The role of the airline transportation network in the prediction and predictability of global epidemics, Proc. Natl. Acad. Sci. U. S. A. 103, 2015- 2020 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[125, 223, 880, 258]]<|/det|>
+[3] Balcan, D. et al. Multiscale mobility networks and the spatial spreading of infectious diseases, Proc. Natl. Acad. Sci. U. S. A. 106, 21484- 21489 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[125, 269, 880, 304]]<|/det|>
+[4] Brockmann, D. and Helbing, D., The Hidden Geometry of Complex, Network- Driven Contagion Phenomena, Science, 342, 1337- 1342 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[125, 315, 880, 350]]<|/det|>
+[5] Liu, Q. et al. Measurability of the epidemic reproduction number in data- driven contact networks, Proc. Natl. Acad. Sci. U. S. A. 115, 12680- 12685 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[125, 361, 880, 396]]<|/det|>
+[6] Pastor- Satorras, R., Castellano, C., Van Mieghem, P. and Vespignani, A. Epidemic processes in complex networks, Rev. Mod. Phys. 87, 925- 979 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[125, 407, 880, 442]]<|/det|>
+[7] Zhang, J. et al. Changes in contact patterns shape the dynamics of the COVID- 19 outbreak in China, Science, 368, 1481- 1486 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[125, 453, 880, 489]]<|/det|>
+[8] Maier, B. F. and Brockmann, D., Effective containment explains subexponential growth in recent confirmed COVID- 19 cases in China, Science, 368, 742- 746 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[125, 499, 880, 535]]<|/det|>
+[9] Schlosser, F. et al. COVID- 19 lockdown induces disease- mitigating structural changes in mobility networks, Proc. Natl. Acad. Sci. U. S. A. 117, 32883- 32890 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 545, 880, 581]]<|/det|>
+[10] Aleta, A. et al. Modelling the impact of testing, contact tracing and household quarantine on second waves of COVID- 19, Nature Human Behaviour, 4, 964971 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 591, 848, 609]]<|/det|>
+[11] Vespignani, A. et al. Modelling COVID- 19, Nature Reviews Physics, 2, 279281 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 619, 880, 672]]<|/det|>
+[12] te Vrugt, M., Bickmann, J. and Wittkowski, R., Effects of social distancing and isolation on epidemic spreading modeled via dynamical density functional theory, Nature Communications 11, 5576 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 683, 880, 718]]<|/det|>
+[13] Kogan, N. E. et al. An early warning approach to monitor COVID- 19 activity with multiple digital traces in near real time, Science Advances 7, eabd6989 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 880, 764]]<|/det|>
+[14] Yao, H. et al. Patient- derived mutations impact pathogenicity of SARS- CoV- 2, medRxiv 2020.04.14.20060160 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 880, 828]]<|/det|>
+[15] Konno, Y. et al. SARS- CoV- 2 ORF3b is a potent interferon antagonist whose activity is further increased by a naturally occurring elongation variant, bioRxiv 2020.05.11.088179 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 840, 880, 892]]<|/det|>
+[16] Hu, J. et al. The D614G mutation of SARS- CoV- 2 spike protein enhances viral infectivity and decreases neutralization sensitivity to individual convalescent sera, bioRxiv 2020.06.20.161323 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 883, 135]]<|/det|>
+[17] Korber, B. er al. on behalf of the Sheffield COVID- 19 Genomics Group, Tracking changes in SARS- CoV- 2 Spike: evidence that D614G increases infectivity of the COVID- 19 virus, Cell (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 147, 881, 181]]<|/det|>
+[18] Grubaugh, N.D., Hanage, W.P., and Rasmussen, A.L., Making sense of mutation: what D614G means for the COVID- 19 pandemic remains unclear, Cell (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 193, 881, 227]]<|/det|>
+[19] Ojosnegros, S. and Beerenwinkel, N., Models of RNA virus evolution and their roles in vaccine design, Immunome Research 6(Suppl 2):S5 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 881, 273]]<|/det|>
+[20] Nougairède, A., Charrel, R. N. and Raoult, D., Models cannot predict future outbreaks: A/H1N1 virus, the paradigm, Eur J Epidemiol 26:183, 186 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[115, 285, 881, 319]]<|/det|>
+[21] Bedford, T., et al. Integrating influenza antigenic dynamics with molecular evolution, eLife 3:e01914 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[115, 331, 881, 365]]<|/det|>
+[22] Nelson, M. I. and Holmes, E. C., The evolution of epidemic influenza, Nature Reviews Genetics 8, pages196,205 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[115, 377, 881, 411]]<|/det|>
+[23] Ghedin, E. et al. Large- scale sequencing of human influenza reveals the dynamic nature of viral genome evolution, Nature 437, pages 1162, 1166 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[115, 422, 881, 457]]<|/det|>
+[24] Earn, D. J.D., Dushoff, J. and Levin, S. A., Ecology and evolution of the flu, TRENDS in Ecology & Evolution 17, Pages 334- 340 (2002).  
+
+<|ref|>text<|/ref|><|det|>[[115, 469, 881, 503]]<|/det|>
+[25] Lopman, B., Zambon, M. and Brown, D. W., The Evolution of Norovirus, the Gastric Flu, PLoS Medicine 5(2): e42. (2008).  
+
+<|ref|>text<|/ref|><|det|>[[115, 515, 881, 549]]<|/det|>
+[26] Uwimana, A. et al., Emergence and clonal expansion of in vitro artemisinin- resistant Plasmodium falciparum kelch13 R561H mutant parasites in Rwanda, Nature Medicine (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 561, 881, 595]]<|/det|>
+[27] Birnbaum, J. et al., A Kelch13- defined endocytosis pathway mediates artemisinin resistance in malaria parasites, Science 367, 51- 59 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 607, 881, 641]]<|/det|>
+[28] Alexander, H. and Day, T., Risk factors for the evolutionary emergence of pathogens. J. R. Soc. Interface 7, 1455- 1474 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[115, 653, 881, 687]]<|/det|>
+[29] Uekermann, F. and Sneppen, K., Spreading of multiple epidemics with cross immunization. Phys. Rev. E, 86, 036108 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[115, 699, 881, 750]]<|/det|>
+[30] Hébert- Dufresne, L., Patterson- Lomba, O., Goerg, G. M. and Althouse, B. M., Pathogen Mutation Modeled by Competition Between Site and Bond Percolation, Phys. Rev. Lett. 110, 108103 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[115, 763, 881, 797]]<|/det|>
+[31] Alexandrou, C., Harmandaris, V., Irakleous, A., Koutsou, G. and Savva, N., Modeling the evolution of COVID- 19, arXiv:2008.03165 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 809, 881, 843]]<|/det|>
+[32] Eletreby, R. et al., The effects of evolutionary adaptations on spreading processes in complex networks, Proc. Natl. Acad. Sci. U. S. A. 117, 5664 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 855, 881, 907]]<|/det|>
+[33] Anderson, R. M., Heesterbeek, H., Klinkenberg, S. and Hollingsworth, T.D., How will country- based mitigation measures influence the course of the COVID- 19 epidemic? The Lancet, 395, 10228 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 630, 100]]<|/det|>
+[34] Shen, C. and Bar- Yam, Y, COVID- 19: How to win (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 110, 881, 145]]<|/det|>
+[35] Meidan, D. et al. Alternating quarantine for sustainable epidemic mitigation, Nature Communications 12,220 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 156, 881, 208]]<|/det|>
+[36] Böttcher, L., Woolley- Meza, O., Araújo, N. A. M., Herrmann, H. J. and Helbing, D., Disease- induced resource constraints can trigger explosive epidemics, Scientific Reports 5, 16571 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[115, 219, 881, 254]]<|/det|>
+[37] Boccaletti, S. et al. Explosive transitions in complex networks? structure and dynamics: Percolation and synchronization, Physics Reports 660, 1- 94 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 264, 881, 300]]<|/det|>
+[38] Liu, Q., Wang, W., Tang, M., Zhou, T. and Lai, Y., Explosive spreading on complex networks: The role of synergy, Phys. Rev. E 95, 042320 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[115, 310, 881, 345]]<|/det|>
+[39] Pires, M. A., Oestereich, A. L. and Crokidakis, N., Sudden transitions in coupled opinion and epidemic dynamics with vaccination, J. Stat. Mech. 053407 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[115, 355, 881, 391]]<|/det|>
+[40] Mieghem, P. V., Explosive phase transition in susceptible- infected- susceptible epidemics with arbitrary small but nonzero self- infection rate, Phys. Rev. E, 101, 032303 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 400, 881, 436]]<|/det|>
+[41] Zhang, X., Ruan, Z., Zheng, M., Barzel, B. and Boccaletti, S., Epidemic spreading under infection- reduced- recovery, Chaos, Solitons and Fractals 140, 110130 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 446, 881, 481]]<|/det|>
+[42] Pastor- Satorras, R., Castellano, C., Mieghem, P. V. and Vespignani, A., Epidemic processes in complex networks, Rev. Mod. Phys. 87, 925 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[115, 492, 881, 527]]<|/det|>
+[43] Yang, H., Gu, C., Tang, M., Cai, S. and Lai, Y., Suppression of epidemic spreading in time- varying multiplex networks, Applied Mathematical Modelling, 75, 806- 818 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 537, 881, 572]]<|/det|>
+[44] Zhai, Z., Long, Y., Tang, M., Liu, Z. and Lai, Y., Optimal inference of the start of COVID- 19, Phys. Rev. Research 3, 013155 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 583, 881, 618]]<|/det|>
+[45] Gómez- Gardeñes, J., Lotero, L., Taraskin, S. N. and Pérez- Reche, F. J., Explosive contagion in networks, Sci. Rep. 6, 19767 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 629, 881, 664]]<|/det|>
+[46] Wu, J., Zheng, M., Xu, K. and Gu, C., Effects of two channels on explosive information spreading, Nonlinear Dynamics, 99, 2387- 2397, (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 674, 881, 709]]<|/det|>
+[47] Hébert- Dufresne, L., Scarpino, S. V. and Young, J., Macroscopic patterns of interacting contagions are indistinguishable from social reinforcement, Nature Physics 16, 426 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 719, 881, 754]]<|/det|>
+[48] Gross, T., Lima, C. J. D. and Blasius, B., Epidemic dynamics on an adaptive network, Phys. Rev. Lett. 96, 208701 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[115, 764, 881, 800]]<|/det|>
+[49] Böttcher, L., Woolley- Meza, O., Goles, E., Helbing, D. and Herrmann, H. J., Connectivity disruption sparks explosive epidemic spreading, Phys. Rev. E 93, 042315 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 810, 725, 827]]<|/det|>
+[50] Epstein, J.M., Modelling to contain pandemics. Nature 460,687 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[115, 838, 881, 872]]<|/det|>
+[51] Harush, U. and Barzel, B., Dynamic patterns of information flow in complex net- works, Nat. Comm. 8, 2181 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[115, 883, 881, 918]]<|/det|>
+[52] Hens, C., Harush, U., Cohen, R. and Barzel, B., Spatiotemporal signal propagation in complex networks, Nat. Phys. 15, 403 (2019).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 881, 118]]<|/det|>
+[53] Hacohen, A., Cohen, R., Efroni, S., Barzel, B. and Bachelet, I., Digitizable therapeutics for decentralized mitigation of global pandemics, Scientific Reports 9, 14345 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 128, 881, 180]]<|/det|>
+[54] Ferguson, N. M. et al. Impact of Non- Pharmaceutical Interventions (npis) to Reduce COVID- 19 Mortality and Health- Care Demand. (Imperial College COVID- 19 Response Team, 2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 192, 881, 226]]<|/det|>
+[55] Li, Q. et al. Early transmission dynamics in Wuhan, China, of novel Coronavirus- infected pneumonia. N. Engl. J. Med. 382, 1199 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 237, 733, 254]]<|/det|>
+[56] WHO. Coronavirus Disease (COVID- 2019) Situation Report 30. (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 265, 881, 300]]<|/det|>
+[57] Ferretti, L. et al. Quantifying SARS- CoV- 2 transmission suggests epidemic control with digital contact tracing. Science 368, 6936 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 311, 881, 363]]<|/det|>
+[58] Backer, J. A., Klinkenberg, D. and Wallinga, J. Incubation period of 2019 novel Coronavirus (2019- ncov) infections among travelers from Wuhan, China, 2028 January 2020. Eurosurveillance 25, 2000062 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 374, 881, 426]]<|/det|>
+[59] Linton, N. M., et al. Incubation period and other epidemiological characteristics of 2019 novel Coronavirus infections with right truncation: a statistical analysis of publicly available case data. J. Clin. Med. 9, 538 (2020)  
+
+<|ref|>text<|/ref|><|det|>[[115, 437, 880, 472]]<|/det|>
+[60] Tao, Y. et al. High incidence of asymptomatic SARS- CoV- 2 infection, Chongqing, China. medRxiv https://doi.org/10.1101/2020.03.16.20037259 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 483, 881, 535]]<|/det|>
+[61] Mizumoto, K., Kagaya, K., Zarebski, A. and Chowell, G. Estimating the asymptomatic proportion of Coronavirus disease 2019 (COVID- 19) cases on board the Diamond Princess cruise ship, Yokohama, Japan, 2020. Eurosurveillance 25, 2000180 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 547, 881, 580]]<|/det|>
+[62] Pan, X. et al. Asymptomatic cases in a family cluster with SARS- CoV- 2 infection. Lancet Infect. Dis. 20, 410 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 592, 857, 609]]<|/det|>
+[63] Lu, X. et al. SARS- CoV- 2 infection in children. N. Engl. J. Med. 382, 16631665 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 620, 881, 654]]<|/det|>
+[64] Al- Tawfiq, J. A. Asymptomatic coronavirus infection: MERS- CoV and SARS- CoV- 2 (COVID- 19). Travel Med. Infect. Dis. 35, 101608 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 666, 881, 717]]<|/det|>
+[65] Colson, P. et al. Children account for a small proportion of diagnoses of SARS- CoV- 2 infection and do not exhibit greater viral loads than adults. Eur. J. Clin. Microbiol. Infect. Dis. 26, 15 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 881, 780]]<|/det|>
+[66] Song, H. et al. A considerable proportion of individuals with asymptomatic SARS- CoV- 2 infection in Tibetan population. medRxiv https://doi.org/10.1101/2020.03.27.20043836 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 793, 881, 825]]<|/det|>
+[67] Dong, Y. et al. Epidemiology of COVID- 19 among children in China. Pediatrics 145, e20200702 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 837, 881, 871]]<|/det|>
+[68] Dong, E., Du, H. and Gardner, L., An interactive web- based dashboard to track COVID- 19 in real time. Lancet Infect. Dis. 20, 533534 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 883, 881, 917]]<|/det|>
+[69] Bar- On, Y. M., Flamholz, A., Phillips, R. and Milo, R., SARS- CoV- 2 (COVID- 19) by the numbers. eLife 9, e57309 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 881, 116]]<|/det|>
+[70] Leyva, I. et al., Explosive first- order transition to synchrony in networked chaotic oscillators, Phys. Rev. Lett. 108 (16), 168702 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[115, 127, 881, 162]]<|/det|>
+[71] Zhang, X., Hu, X., Kurths, J. and Liu, Z., Explosive synchronization in a general complex network, Phys. Rev. E 88, 010802(R) (2013).  
+
+<|ref|>text<|/ref|><|det|>[[115, 174, 881, 209]]<|/det|>
+[72] Zhang, X., Boccaletti, S., Guan, S. and Liu, Z., Explosive synchronization in adaptive and multilayer networks, Phys. Rev. Lett. 114, 038701 (2015).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[156, 85, 836, 592]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 606, 881, 907]]<|/det|>
+<center>Figure 1: The phases of a pandemic under pathogen mutation. (a) Pandemic phase. For \(R_{0} > 1\) we observe the classic pandemic phase. The prevalence \(\rho (t)\) vs. \(t\) (top) grows continuously as the fitness \(\bar{F} (t)\) (bottom) increases due to mutation and natural selection. (b) Lethargic phase. For \(R_{0}< 1\) we have \(\rho (t)\) exponentially decaying to zero. The mutation rate \(\sigma = 0.01\) is too slow, \(\bar{F} (t)\) remains almost constant (bottom), and the pathogen fails to reach critical fitness \(F_{c}\) (grey dashed line) on time. (c) Mutated phase. We now remain in the sub-pandemic regime \(R_{0}< 1\) , but increased the mutation rate to \(\sigma = 1\) . For small \(t\) we observe \(\rho (t)\) rapidly decaying (top). However, thanks to the rapid mutations \(\bar{F} (t)\) reaches critical fitness (grey dashed line) within a short time. Following this point the disease reemerges and \(\rho (t)\) changes course, turning pandemic. This is observed in the snapshots at bottom through the appearance of sporadic instances of high fitness pathogens (middle, dark red nodes), which then spread to infect the majority of the population. (d) \(\sigma ,R_{0}\) phase diagram. To systematically observe the different phases we varied \(R_{0}\in (0,1.5)\) and \(\sigma \in (10^{-3},10)\) , capturing a total of 1,050 epidemiological scenarios, with different \(\mu ,\beta\) and \(\sigma\) . For each scenario we ran 50 stochastic realizations and measured the probability \(P\) to have \(\rho (t\to \infty) > 0\) , i.e. pandemic. We observe three phases with sharp boundaries between them. First, the pandemic phase (red) for \(R_{0} > 1\) , independent of \(\sigma\) , as predicted by the classic SIS model. In addition to that the sub-pandemic regime \(R_{0}< 1\) is split into two phases: Under small \(\sigma\) , \(P\) tends to zero (blue) and the pathogen fails to spread, giving rise to the lethargic phase. For large \(\sigma\) , the spreading probability becomes almost certain, as \(p\sim 1\) (green), and we observe a mutation driven contagion. The gap between these phases (grey) indicates an abrupt transition from \(P\to 0\) to \(P\to 1\) , a dramatic shift occurring within a narrow range of \(R_{0},\sigma\) values. This grey range is well-approximated by our theoretical prediction (solid black line) as appears in Eq. (11). All simulations, here and throughout, were done on a random network of \(N = 5,000\) nodes and \(\bar{k} = 15\) . The disease parameters were set to \(\mu = 0.1\) and the infection rate was set variably to \(\beta = \mu R_{0} / \bar{k}\) , to obtain the different values of \(R_{0}\) . The mutation rate \(\sigma\) is specified in each figure. In each scenario we set the initial condition to \(\rho (t) = 0.2\) . </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[118, 133, 880, 600]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 613, 881, 860]]<|/det|>
+<center>Figure 2: The transition to the mutated phase. To observe a mutated phase a critical mutation must arise before the pathogen is eliminated, namely before \(\rho (t)\) crosses \(1 / N\) (grey dashed lines), capturing the unit line in which there is a single infected individual among the \(N\) node population. (a) \(\rho (t)\) vs. \(t\) (grey solid line) as obtained from Eq. (8) in the lethargic phase ( \(R_{0} = 0.25\) , \(\sigma = 0.01\) ). The critical mutation occurs at the minimum point \((t_{c})\) , which is below the unit line. Therefore the epidemic decays prior to the appearance of the critical mutation. Indeed, the stochastic simulation (blue solid line) approaches zero prevalence, never reaching the positive branch of \(\rho (t)\) . (b) Setting \(\sigma = 0.16\) the system is at criticality. \(\rho (t_{c})\) is adjacent to the unit line, and hence we observe critical behavior: some realizations decay (blue), whereas others successfully mutate (green). (c) Under \(\sigma = 0.5\) , the system is in the mutated phase, \(\rho (t_{c})\) is sufficiently above the unit line and the critical mutation is reached with probability \(P \to 1\) . (d) The lethargic-mutated phase boundary in Eq. (11) depends on the initial size of the infected population \(\mathcal{I}_{0}\) . Here we show this boundary for \(\mathcal{I}_{0} = 10^{2}, \ldots , 10^{8}\) (grey solid lines). (e) The long term prevalence \(\rho = \rho (t \to \infty)\) vs. \(R_{0}\) under \(\sigma = 0.1\) (yellow dashed path in panel (d)). Approaching from small \(R_{0}\) (left to right) we begin with an initial infection of \(\mathcal{I}_{0} = 10^{2}\) and observe a transition to the mutated phase at \(R_{0} = R_{\mathrm{High}}\) . In the opposite direction, however, as we begin with large \(R_{0}\) we approach the transition from an already pandemic state with \(\mathcal{I}_{0} \sim 10^{4}\) . Now the phase boundary traverses through \(R_{0} = R_{\mathrm{Low}}\) . Both transitions are also marked by circles in panel (d). We, therefore arrive at a hysteresis phenomenon, in which the critical transition point depends on the current state of the spread. Consequently, preemptive mitigation, done when the spread is still at its embryonic stage (\(\mathcal{I}_{0}\) small), is more effective than reactive mitigation, applied when \(\mathcal{I}_{0}\) is already large. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 230, 880, 623]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 634, 881, 768]]<|/det|>
+<center>Figure 3: The volatile phase. (a) The \(\sigma ,R_{0}\) phase diagram under the bounded fitness of Eq. (12). We now observe a volatile phase, in which \(\rho \rightarrow 0\) (blue), when \(\sigma\) is too large. Hence, the mutated phase (green) now only appears in the Goldilocks zone in which the mutation rate in not too high nor too low. The theoretical prediction of (13) is also shown (black solid line on right). (b) \(\rho\) vs. \(R_{0}\) under \(\sigma = 3\) (yellow path in panel (a)). As opposed to the lethargic-mutated phase transition, the shift from volatile to mutated follows a continuous second order transition. (c) \(\rho (t)\) vs. \(t\) in the volatile phase decaying, as predicted, to the healthy state \(\rho = 0\) . (d) \(\overline{F} (t)\) changes rapidly thanks to the large \(\sigma\) , and crosses the critical \(F_{c}\) (grey dashed line) early on. However the rapid mutations prevent the slower natural selection from securing a steady increase in \(\overline{F} (t)\) . Hence, the achieved fitness cannot be stably sustained for the pathogen to continually spread. (e) Indeed, we observe multiple instances of critical fitness (dark red) that fail to reproduce and dominate the pathogen population. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[154, 80, 840, 655]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 662, 881, 923]]<|/det|>
+<center>Figure 4: Vaccination under the threat of mutation. (a) The SARS-CoV-2 disease cycle. Upon exposure (yellow) individuals enter a pre-symptomatic phase (purple), from which they later develop mild \((I_{M})\) , severe \((I_{S})\) or critical symptoms \((I_{C})\) , determining the duration of their infected phase and their probability to recover (green) or decease (grey). (b) Vaccine resistance is risky under a coexistence of both infected \((\rho)\) and vaccinated \((V)\) individuals (center). When \(\rho\) is small, the probability of mutation is marginal (right); when \((V)\) is small the selection pressure for resistance is weak (left). (c) Under slow vaccination \(\rho (t)\) increases (red). As a result, when vaccines gain coverage we enter the risky zone (shaded), and become potentially vulnerable to resistance mutation. Indeed, when such mutation occurs (orange line), the trend is reversed, \(\rho (t)\) increases and the vaccine coverage \(V(t)\) plummets (blue). (d) We present several snapshots to track the state of the spread. In snapshot 2 we observe a premature mutation (orange node) that fails to spread, since \(V(t)\) at that point is still small (blue nodes). Later (snapshot 4), with the system in the risky zone of high \(\rho (t)\) and \(V(t)\) , such mutations rapidly take over, as seen by the coverage of the orange nodes in snapshot 5. (e) - (f) Rapid vaccination in and of itself may be insufficient. The system quickly enters the risky zone (shaded) and with \(R_{0} > 1\) , a single resistance mutation eventually outruns our vaccination efforts. (g) - (h) Successful eradication of the disease is achieved under a combination of rapid vaccination (blue) and suppression of \(R_{0}\) , e.g., through social distancing. Pushing \(R_{0}\) down suppresses \(\rho (t)\) , and hence avoids the risky zone by locating the system in the right hand side of panel (b). (i) The probability \(P\) to observe a pandemic state as a function of the vaccination rate \(\eta\) for different values of \(R_{0}\) . To alleviate the risk of vaccine resistant spread we must remain in the blue zone, in which we not only invest in the vaccine roll-out \((\eta)\) , but also in suppressing the spread (reducing \(R_{0}\) ). (j) Similar, albeit less dramatic results are also observed under our low risk scenario \(\mathcal{P} = 0.25\) . </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[80, 110, 888, 787]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 807, 115, 826]]<|/det|>
+<center>Figure 1 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 850, 872, 871]]<|/det|>
+The phases of a pandemic under pathogen mutation. (see Manuscript file for full figure caption)
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[68, 60, 920, 635]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 658, 118, 678]]<|/det|>
+<center>Figure 2 </center>  
+
+<|ref|>text<|/ref|><|det|>[[44, 700, 728, 721]]<|/det|>
+The transition to the mutated phase. (see Manuscript file for full figure caption)
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[79, 60, 916, 525]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 555, 118, 575]]<|/det|>
+<center>Figure 3 </center>  
+
+<|ref|>text<|/ref|><|det|>[[44, 597, 576, 618]]<|/det|>
+The volatile phase. (see Manuscript file for full figure caption)
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[84, 65, 840, 770]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 789, 117, 808]]<|/det|>
+<center>Figure 4 </center>  
+
+<|ref|>text<|/ref|><|det|>[[44, 830, 768, 850]]<|/det|>
+Vaccination under the threat of mutation. (see Manuscript file for full figure caption)  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 874, 310, 901]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 924, 765, 944]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[61, 46, 353, 65]]<|/det|>
+SupplementaryInformation.pdf
+
+<--- Page Split --->

preprint/preprint__b3cdb805334a0d61d0b77d7ab9ffa2231861aba3b0f051ea183438476e3179db/images_list.json

@@ -0,0 +1,62 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "FIG. 1. Phase-sensitive NV magnetometry (a) Continuous sampling induced phase reviving signals, known as the quantum heterodyne (Q-dyne) detection. The phase reviving frequency changes with the external field and cannot be used for phase-sensitive detection. The signal responses \\(G(s_{\\phi},k)\\) should be small to ensure measurement linearity. \\(R\\) is the detected photon rate, and \\(C\\) is the detected signal contrast. (b) Unlike the Q-dyne detection, the quantum phase-sensitive detection (QPSD) is based on the rotating frame modulation induced by the evolving phase difference of the two driving MW fields. Two frequency-offset MWs acquire a phase difference of \\(\\alpha - \\beta = 2\\pi \\delta f\\Delta t\\) after the sampling time interval \\(\\Delta t\\) . In the Bloch sphere picture, it can be understand as the MW2 defined rotating frame \\(x_{2}y_{2}z\\) rotates with rate of \\(\\delta f\\) referring to the MW1 defined rotating frame \\(x_{1}y_{1}z\\) . The acquired quantum phase is \\(\\theta = 2\\pi \\delta f\\Delta t - \\phi\\) at sample 2. Through the quantum phase modulation, the acquired readout representing the Bloch vector projections is as shown in (c), where we present the measurements of the quantum phase \\(\\phi = 0\\) and \\(\\phi \\neq 0\\) . By demodulating the acquired signal with a lock-in amplifier, we can get the phase values. The dashed box shows the measurement sequences we applied in experiments. Except for the last \\(\\pi /2\\) pulse applied with MW2, all the other driving pulses are generated by MW1. The fluorescence signal is demodulated at the frequency of \\(1 / (2T_{seq})\\) to get a fluorescence intensity readout for a sample. The QPSD readout is acquired with the demodulation at \\(\\delta f\\) . (d) Schematic of the experiment. NV centers ensemble in diamond is used to perform the QPSD readout.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        92,
+        874,
+        423
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "FIG. 2. Sensing performance of the QPSD. (a) Spectra and linearity comparison of the normal Ramsey readout and the QPSD readout. We apply \\(T_{\\phi} = 6.25\\mu s\\) in both measurements. The applied peak-to-peak field and the readout are plotted showing the linearity of the measurements. (b) Spectra and linearity comparison of normal Hahn-echo readout and the QPSD readout. The applied phase accumulation time \\(T_{\\phi} = 12.5\\mu s\\) . Thus, the Hahn-echo measurements performs a higher sensitivity but smaller dynamic range than the Ramsey measurements. (c) Signal response to different sampling frequencies. The measurements use the same calibration field, and the readouts are normalized to be plotted in the same vertical axis. (d) Measurement bandwidth. Ramsey and Hahn-echo sequences are applied to measure test fields at different frequencies with the same magnitude. The heterodyne frequency responses are limited in bandwidth by the cut-off frequency of the LIA.",
+    "footnote": [],
+    "bbox": [
+      [
+        125,
+        92,
+        872,
+        558
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "FIG. 3. Frequency Offset Heterodyne readout. (a) Hahn-echo sequence is used for this demonstration. The detected phase of the ac signal is locked by the sequence when the frequency is \\(f_{ac} = f_{\\phi}\\) . Otherwise, a heterodyne signal of \\(|f_{ac} - f_{\\phi}|\\) can be measured. The colored regions mark where the quantum phase is accumulated, while phase accumulations at the other areas are canceled in the spin evolution. The figure shows identical heterodyne signals due to \\(f_{\\phi} - f_{1} = f_{2} - f_{\\phi}\\) . (b) We apply ac fields at different frequencies with an offset of \\(5 \\mathrm{~Hz}\\) to the sensor so that \\(5 \\mathrm{~Hz}\\) peaks can be detected as the signal response. The signal frequency response of the Hahn-echo sequence and CPMG-2 sequence are plotted after normalization, respectively. In both measurements \\(T_{\\phi} = 50 \\mu \\mathrm{s}\\) , and the sampling frequency is \\(2 \\mathrm{kHz}\\) according to the applied sequence length. The red lines are the filter functions in theory. (c) Signal frequency response of Hahn-echo measurements with \\(1 / T_{seq} = 10 \\mathrm{kHz}\\) . The dash line indicates the filter introduced by the lock-in amplifier. (d) Sequence dependency of the frequency resolution. In means. 1, a \\(20.005 \\mathrm{kHz}\\) filed is applied and measured by sequences with \\(T_{\\phi} = 50 \\mu \\mathrm{s} \\pm 4 \\mathrm{ns}\\) , and \\(T_{seq} = 20 T_{\\phi}\\) . In means. 2, we keep \\(T_{\\phi}\\) unchanged, and offset \\(T_{seq}\\) with \\(\\pm 4 \\mathrm{ns}\\) . In means. 3, the frequency of the applied field is changed to \\(16.005 \\mathrm{kHz}\\) while the other parameters are the same as means. 2.",
+    "footnote": [],
+    "bbox": [
+      [
+        128,
+        88,
+        863,
+        465
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "FIG. 4. Detection of arbitrary audio signals. (a) Phase response of the QPSD measurement. A 20.08 kHz signal with sequential phase changing is applied to the sensor. The bars show the phases at different time. Stars mark the readout of the sensor, and the curve is the simulated readout. (b) An ac field is applied with the frequency, amplitude and phase switched every \\(100\\mathrm{ms}\\) . The light blue areas corresponding to the right \\(y\\) -axis shows the applied field of around \\(10\\mathrm{kHz}\\) , and the red curve shows the QPSD readout. (c) Spectral comparison of the applied signal and the detected magnetic field in a narrow bandwidth. The applied signal is a sum of 20 different sine signals within \\(400\\mathrm{Hz}\\) bandwidth. (d) A signal with wide bandwidth between 10 to \\(15\\mathrm{kHz}\\) is applied and detected by varying the sequence. We set an \\(800\\mathrm{Hz}\\) bandwidth for the measurement of each sequence and use 6 measurements to cover the entire bandwidth. The red dash line shows the spectrum of the output of the AWG, and the solid black line is the spectrum of the detected magnetic field signal. The inset figure is the phase-noise power spectrum density plotted within 1 Hz bandwidth cut.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        293,
+        863,
+        670
+      ]
+    ],
+    "page_idx": 15
+  }
+]

preprint/preprint__b3cdb805334a0d61d0b77d7ab9ffa2231861aba3b0f051ea183438476e3179db/preprint__b3cdb805334a0d61d0b77d7ab9ffa2231861aba3b0f051ea183438476e3179db.mmd

@@ -0,0 +1,325 @@
+
+# Quantum-assisted distortion-free audio signal sensing  
+
+Chen Zhang (chen.zhang@pi3. uni- stuttgart.de)  
+
+University of Stuttgart  
+
+Durga Dasari University of Stuttgart  
+
+Matthias Widmann University of Stuttgart  
+
+Jonas Meinel University of Stuttgart https://orcid.org/0000- 0003- 4040- 8361  
+
+Vadim Vorobyov University of Stuttgart  
+
+Polina Kapitanova ITMO University  
+
+Elizaveta Nenasheva Giricond Research Institute, Ceramics Co., Ltd.  
+
+Kazuo Nakamura Tokyo Gas Co., Ltd. https://orcid.org/0000- 0002- 3412- 834X  
+
+Hitoshi Sumiya Sumitomo Electric Industries  
+
+Shinobu Onoda National Institutes for Quantum Science and Technology https://orcid.org/0000- 0003- 1425- 0708  
+
+Junichi Isoya University of Tsukuba https://orcid.org/0000- 0002- 9598- 625X  
+
+Jörg Wrachtrup University of Stuttgart  
+
+## Article  
+
+Keywords: Quantum sensors, metrology, audio signal sensing  
+
+Posted Date: November 19th, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 1068484/v1
+
+<--- Page Split --->
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Version of Record: A version of this preprint was published at Nature Communications on August 8th, 2022. See the published version at https://doi.org/10.1038/s41467-022-32150-1.
+
+<--- Page Split --->
+
+# Quantum-assisted distortion-free audio signal sensing  
+
+Chen Zhang, \(^{1, *}\) Durga Dasari, \(^{1, \dagger}\) Matthias Widmann, \(^{1}\) Jonas Meinel, \(^{1}\) Vadim  
+
+Vorobyov, \(^{1}\) Polina Kapitanova, \(^{2}\) Elizaveta Nenasheva, \(^{3}\) Kazuo Nakamura, \(^{4}\)  
+
+Hitoshi Sumiya, \(^{5}\) Shinobu Onoda, \(^{6}\) Junichi Isoya, \(^{7}\) and Jörg Wrachtrup \(^{1}\)  
+
+\(^{1}\) 3rd Institute of Physics, University of Stuttgart,  
+
+Pfaffenwaldring 57, Stuttgart 70569, Germany  
+
+\(^{2}\) Department of Physics and Engineering,  
+
+ITMO University, Saint Petersburg 197101, Russia  
+
+\(^{3}\) Giricond Research Institute, Ceramics Co. Ltd., Saint Petersburg 194223, Russia  
+
+\(^{4}\) Hydrogen and Carbon Management Technology Section,  
+
+Hydrogen and Carbon Management Technology Strategy Department,  
+
+Tokyo Gas Co. Ltd., Yokohama 230- 0045, Japan  
+
+\(^{5}\) Advanced Materials Labotatory, Sumitomo Electric Industries Ltd., Itami 664- 0016, Japan  
+
+\(^{6}\) Takasaki Advanced Radiation Research Institute,  
+
+National Institutes for Quantum Science and Technology, Takasaki 370- 1292, Japan  
+
+\(^{7}\) Faculty of Pure and Applied Sciences,  
+
+University of Tsukuba, Tsukuba 305- 8573, Japan
+
+<--- Page Split --->
+
+## Abstract  
+
+Quantum sensors are keeping the cutting- edge sensitivities in metrology. However, for high- sensitive measurements of arbitrary signals, limitations in linear dynamic range could introduce distortions when sensing the frequency, magnitude and phase of unknown signals. Here, we overcome these limitations with advanced sensing protocol that combines quantum phase- sensitive detection with the heterodyne readout. We present theoretical and experimental investigations using nitrogen- vacancy centers in diamond, showing the ability to sense audio signals with a 98 dB linear dynamic range, a 31 pT/Hz \(^{1 / 2}\) sensitivity, and arbitrary frequency resolution. Further, we perform the quantum assisted distortion free audio signal (melody piece, speech) sensing with high fidelity. The methods developed here could broaden the horizon for quantum sensors towards applications in telecommunication, where high- fidelity and low- distortion at multiple frequency bands within small sensing volumes are required.  
+
+## 18 I. INTRODUCTION  
+
+Quantum sensors are setting new frontiers of sensing techniques with their extraordinary performances in sensitivity and stability [1- 5]. These techniques rely on either measuring the line- shift of spin or atomic transition frequencies or reading out the relative populations of the occupied energy levels using interferometric methods [6, 7]. In most cases, there are trade- off relations between the sensitivity and other features in metrology [8]. For example, a high- sensitive measurement acquired by detecting the transition line shift requires a narrow linewidth, which, on the other hand, will limit the dynamic range. Interferometric measurements detect a sinusoidal response, and linearity is only achieved when the phase signal is in a small dynamic range. It sets a massive limitation on the sensitivity when sensing an unknown signal that gets measured beyond this linear regime, for example, when the working point of the sensor is at the maxima or minima of the sinusoidal signal response. Thus, it becomes a bottleneck for high sensitivity measurements that are required in many cutting- edge applications. Operating within the linear dynamic range (LDR) can be crucial for reconstructing unknown signals. One way to directly extract the phase factor, which is
+
+<--- Page Split --->
+
+linear to the physical quantity to be detected, is to use phase- sensitive detection known as the classical lock- in technique. In this work, using a modified sensing scheme that introduces an external readout phase modulation, we acquire the target quantum phase signal after demodulation. Therefore, the LDR is no longer limited to the small- angle approximation. Hence our method combines large dynamic range with maximum sensitivity.  
+
+Nitrogen- vacancy (NV) centers in diamond have been at the forefront in performing high- sensitive measurements of various physical quantities, viz., magnetic and electric field, temperature, and strain distributions internal and external to diamond [9–13]. The NV magnetometry has been performed under bias fields ranging from zero- field to a few Tesla, and for sensing signals with frequencies ranging from DC to a few GHz [14–17]. While dynamical- decoupling techniques are usually employed for high sensitivity [9, 18, 19], arbitrary frequency resolution can be achieved with the quantum heterodyne (Q- dyne) detection [20, 21]. However, both methods suffer from a limited LDR when they are applied to measure arbitrary signals.  
+
+For high dynamic range measurements, a closed- loop frequency- locking scheme together with optically detected magnetic resonance (ODMR) can be used to track resonance frequency shifts [22]. However, this scheme cannot be used for ac field measurements in combination with interferometric methods, if the signal frequency is higher than the readout sampling frequency. Phase- estimation algorithms can effectively improve the LDR in Ramsey measurements by varying the sequence with adaptive feedback schemes [23, 24]. However, for the case of ac sensing schemes e.g. Hahn- echo, as varying the sequence itself will change the sensor response to the ac signals, such methods become less applicable. Therefore, a technique is still missing, that addresses the LDR while maintaining high sensitivity and frequency resolution, for example, in sensing arbitrary radio- frequency fields within a broad bandwidth.  
+
+Sensing radio- frequency signals by electric- field sensors, either conventional electronic receivers or the Rydberg atom sensors, need antennas to collect and guide the electric signals towards the sensors [25–28]. Although the receivers can be highly integrated, the dimension of antennas can scale to meters due to the signal wavelength. The size becomes critical when there is limited space for the sensor, for example, in a satellite. In this regard, quantum magnetometers can be very attractive due to their small sensing volume and high sensitivity [29]. A flux concentrator can be used as a substitute to conventional antennas
+
+<--- Page Split --->
+
+65 for obtaining high signal gain. Independent of the signal wavelength, the dimensions of such flux concentrators can be as small as a few centimeters [30, 31].  
+
+67 In this paper, we demonstrate quantum- assisted distortion- free audio signal sensing with 68 NV center ensembles in diamond using the quantum- phase- sensitive detection (QPSD) technique combined with heterodyne readout. Firstly, we introduce the QPSD technique, which 70 can provide an extended LDR in interferometry measurements by using two synchronized 71 driving fields. Then, we present the heterodyne readout, which can interpret e.g. ac signals 72 to get frequency information. Taking advantage of the bandwidth of the Hahn- echo sequence 73 and the frequency comb induced by the continuous sampling, we demonstrate measurements 74 of audio signals around 10 kHz, beyond the coherence limit without losing sensitivity. Fi- 75 nally, we present arbitrary audio signal measurements with a LDR of 98 dB at a sensitivity 76 of 31 pT/Hz1/2. A piece of melody and a speech are encoded as magnetic field signals and 77 measured by the NV magnetometer. By using the sensor as a quantum radio, we demon- 78 strate the application potentials for areas such as quantum- assisted telecommunication and 79 unknown signal exploration.  
+
+## 80 II. RESULTS  
+
+## 81 A. Quantum Phase Sensitive Detection  
+
+82 In interferometric measurements, the quantum phase to be detected is usually converted 83 to a quantum state population difference, resulting in a sinusoid readout [9]. A way to 84 extract the phase factor from the sinusoidal readout is to modulate the phase with a specific 85 frequency and perform phase- sensitive modulation. Such a quantum phase modulation can 86 be introduced by using the difference between the quantum phase of the sensor to an ex- 87 ternal oscillator. The Q- dyne method uses such a strategy for resolving frequency of signals 88 better than the relaxation time of the sensor, as shown in Fig. 1a [20, 21]. However, it 89 cannot be used for phase- sensitive detection because the Q- dyne frequency is also what to 90 be resolved and an extra modulation is still needed [21]. Another way to introduce such a 91 phase modulation is to use the frequency offset between two different driving fields of the 92 sensor [32, 33]. These driving fields define two rotating frames, and the evolution of the spin 93 as seen from one rotating frame will lead to a quantum phase modulation due to the relative
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>FIG. 1. Phase-sensitive NV magnetometry (a) Continuous sampling induced phase reviving signals, known as the quantum heterodyne (Q-dyne) detection. The phase reviving frequency changes with the external field and cannot be used for phase-sensitive detection. The signal responses \(G(s_{\phi},k)\) should be small to ensure measurement linearity. \(R\) is the detected photon rate, and \(C\) is the detected signal contrast. (b) Unlike the Q-dyne detection, the quantum phase-sensitive detection (QPSD) is based on the rotating frame modulation induced by the evolving phase difference of the two driving MW fields. Two frequency-offset MWs acquire a phase difference of \(\alpha - \beta = 2\pi \delta f\Delta t\) after the sampling time interval \(\Delta t\) . In the Bloch sphere picture, it can be understand as the MW2 defined rotating frame \(x_{2}y_{2}z\) rotates with rate of \(\delta f\) referring to the MW1 defined rotating frame \(x_{1}y_{1}z\) . The acquired quantum phase is \(\theta = 2\pi \delta f\Delta t - \phi\) at sample 2. Through the quantum phase modulation, the acquired readout representing the Bloch vector projections is as shown in (c), where we present the measurements of the quantum phase \(\phi = 0\) and \(\phi \neq 0\) . By demodulating the acquired signal with a lock-in amplifier, we can get the phase values. The dashed box shows the measurement sequences we applied in experiments. Except for the last \(\pi /2\) pulse applied with MW2, all the other driving pulses are generated by MW1. The fluorescence signal is demodulated at the frequency of \(1 / (2T_{seq})\) to get a fluorescence intensity readout for a sample. The QPSD readout is acquired with the demodulation at \(\delta f\) . (d) Schematic of the experiment. NV centers ensemble in diamond is used to perform the QPSD readout. </center>  
+
+94 rotation of the two frames, i.e., rotating frame modulation. The modulation frequency only 95 depends on the frequency difference of the two driving fields, as shown in Fig. 1b and c. 96 By performing multiple measurements within a modulation cycle and upon using lock- in 97 detection, we will achieve phase- sensitive detection. Below we mathematically describe this 98 relative evolution of the sensor under such interferometric measurement with two- frequency
+
+<--- Page Split --->
+
+99 driving fields.  
+
+100 Aligning an external field \(B_{0}\) along the NV axis, we use the two- level subspace of the NV ground triplet in the derivation. Thus, the Hamiltonian of the system can be written as:  
+
+\[\mathcal{H} = \omega_{0}S_{z} + \gamma_{e}B_{1}\cos \left(2\pi f_{1}t + \alpha\right)S_{x}, \quad (1)\]  
+
+102 where \(\omega_{0}\) is the transition frequency of the two- level subspace, \(B_{1}\) is the oscillating magnetic field perpendicular to the NV axis, \(f_{1}\) and \(\alpha\) are the frequency and phase of the driving field, and \(\gamma_{e}\) is the electron gyromagnetic ratio. In the rotating frame defined by the resonance frequency, the time- dependent Hamiltonian is  
+
+\[\mathcal{H}_{1}^{\prime} = \Omega_{1}\cos \left(\delta \omega_{1} + \alpha\right)S_{x} + \Omega_{1}\sin \left(\delta \omega_{1}t + \alpha\right)S_{y}, \quad (2)\]  
+
+106 where \(\delta \omega_{1} = 2\pi (f_{0} - f_{1})\) , and \(\Omega_{1} = \gamma_{e}B_{1} / 2\) is the Rabi frequency introduced by MW1. In 107 interferometry measurements, a \(\pi /2\) pulse prepares the spin state from the polarized state to 108 an equalized population, and another \(\pi /2\) pulse projects the quantum phase as a population 109 difference after the sensing procedure. We use the second driving field, MW2, to offset the 110 frequency of the second \(\pi /2\) pulse. \(\delta \omega_{2},\beta\) and \(\Omega_{2}\) are used to denote the frequency offset, 111 Rabi frequency, and phase of MW2. After this, the measured spin- expectation value is  
+
+\[\langle S_{z}\rangle = \sin \left[\phi +\frac{\pi}{2}\left(\frac{\delta\omega_{1}}{\Omega_{1}} -\frac{\delta\omega_{2}}{\Omega_{2}}\right) + \alpha -\beta \right], \quad (3)\]  
+
+112 where \(\phi\) is the acquired quantum phase which contains the information we want to measure, 113 both of the MWs are near- resonant with \(\delta \omega_{1}\ll \Omega_{1}\) , \(\delta \omega_{2}\ll \Omega_{2}\) . Therefore, the off- resonant 114 term can be neglected, and the phase difference term \(\alpha - \beta\) will evolve with time so that 115 there is  
+
+\[\langle S_{z}\rangle \approx \sin \left(\phi +2\pi \delta f\cdot t\right), \quad (4)\]  
+
+116 where \(\delta f\) is the frequency difference of the two MWs.  
+
+117 The above result can be seen as a modulation of the rotating frame itself. As schematically 118 shown in Fig. 1b (left Bloch sphere), we can assume that the two driving fields have the same 119 phase at the duration of the second \(\pi /2\) pulse, and this defines an instantaneous rotating 120 frame with coordinates \(x_{1}y_{1}z\) . Thus, the readout is similar to that of the regular Ramsey
+
+<--- Page Split --->
+
+interferometry using a single driving field. After an interval of \(\Delta t\) , MW2 develops a phase difference of \(2\pi \delta f\Delta t\) . Since the quantum phase is finally measured by MW2, the Bloch vector rotates in the new instantaneous rotating frame with coordinates \(x_{2}y_{2}z\) , as shown in Fig. 1b (the right Bloch sphere). The rotating frame defined by MW2 rotates continuously around the \(z\) - axis with the frequency of \(\delta f\) . Due to this, the fluorescence readout also modulates in a sinusoidal fashion, as shown in Fig. 1c. While the readout signal frequency depends on \(\delta f\) and its amplitude depends on the signal contrast, the initial phase, \(\phi\) , is linear to the field to be measured. Through the external modulation induced by the MWs, the working point of the sensor evolves in the entire phase range, which ensures the LDR of the initial phase measurement covering \([-\pi ,\pi ]\) . By fitting or demodulating the fluorescence signal, we can resolve the changing of the phase factor \(\phi\) between each modulation cycle and find measurement linearity for the external field.  
+
+The measurement sequence we applied in the experiment is depicted in Fig. 1c, in which \(T_{\phi}\) is the field sensing time, \(T_{seq}\) is the sequence length of one measurement, and we use a second measurement with the final pulse changed to \((\pi /2)_{- x}\) . As a result, the fluorescence signal is modulated with a frequency of \(f_{s} = 1 / (2T_{seq})\) , which is also the sampling frequency of the fluorescence readout. The demodulation of the fluorescence signal, denoted by Demod. 1 in Fig. 1d, has a readout bandwidth \(f_{s} / 2\) set by the Shannon sampling theorem. The readout is further demodulated by another demodulator of the lock- in amplifier (LIA), denoted as Demod. 2. Upon measuring \(N\) samples of the fluorescence readout, the bandwidth of the phase readout is narrowed down to \(f_{s} / (2N)\) . These measurements are schematically shown in Fig. 1d.  
+
+The sensitivity of such measurements can be derived based on the fitting of the \(N\) samples in the measurement time of \(N \cdot 2T_{seq}\) . The minimum detectable phase is derived as  
+
+\[\delta \phi = \frac{2}{\sqrt{N}} \frac{1}{C \sqrt{N}}, \quad (5)\]  
+
+where \(C\) is the fluorescence signal contrast, \(\mathcal{N}\) is the detected photon counts in each measurement. The sensitivity to external magnetic field, however, is still subject to the applied MW sequence, can be derived as  
+
+\[\eta = \frac{2}{\gamma_{e}|G(\omega)|C}\sqrt{\frac{2T_{seq}}{\mathcal{N}}}, \quad (6)\]
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>FIG. 2. Sensing performance of the QPSD. (a) Spectra and linearity comparison of the normal Ramsey readout and the QPSD readout. We apply \(T_{\phi} = 6.25\mu s\) in both measurements. The applied peak-to-peak field and the readout are plotted showing the linearity of the measurements. (b) Spectra and linearity comparison of normal Hahn-echo readout and the QPSD readout. The applied phase accumulation time \(T_{\phi} = 12.5\mu s\) . Thus, the Hahn-echo measurements performs a higher sensitivity but smaller dynamic range than the Ramsey measurements. (c) Signal response to different sampling frequencies. The measurements use the same calibration field, and the readouts are normalized to be plotted in the same vertical axis. (d) Measurement bandwidth. Ramsey and Hahn-echo sequences are applied to measure test fields at different frequencies with the same magnitude. The heterodyne frequency responses are limited in bandwidth by the cut-off frequency of the LIA. </center>  
+
+148 where \(|G(\omega)|\) is the MW filter function which is usually used to describe the transfer function 149 of such a sensor from magnetic field to quantum phase. In comparison to the conventional 150 fluorescence readout, the sensitivity of QPSD readout deteriorates by a factor of \(\sqrt{2}\) . Details 151 about the sensitivity derivation can be seen in the Supplementary Materials (see Supple- 152 mentary Note 3).
+
+<--- Page Split --->
+
+In Fig. 2a and b, we compare the regular interferometry (single driving field) and with the measurements obtained from the QPSD readout described above. The strength of the applied external ac fields ranges from 0 to 3 \(\mu \mathrm{T}\) . For Ramsey measurements, the applied fields are at a frequency of 46 Hz, and we use a field sensing time \(T_{\phi ,R a m s e y} = 6.25\mu \mathrm{s}\) . For Hahn- echo measurements, we use external fields at 80 kHz+46 Hz and the field sensing time \(T_{\phi ,H a h n} = 2T_{\phi ,R a m s e y}\) . The test fields are sent to the diamond by a calibrated loop antenna. The signal readout of the regular interferometry measurements is proportional to \(\sin (\phi)\) , where \(\phi \propto \gamma_{e}B\) is the accumulated quantum phase. The response is linear only when \(\phi\) is small, limiting the dynamic range. Thus, the regular Ramsey and Hahn- echo readout quickly saturate due to this limited LDR. We plot the Fourier transform of the readout signals also in the figures. The harmonics of the 46 Hz signal rise significantly in the fluorescence readout spectral due to the saturation induced by the limited LDR, compared to the QPSD readout which shows the linearity over the entire field range. The high- order harmonics of the signal detected by the QPSD readout are small and mainly arise from the function generator. In the measurements, one could see the linewidth broadening induced by the increasing signal power. The peak at 100 Hz, which is consistently seen in both the Ramsey and Hahn- echo measurements, comes from the electronics instrumentation. Other side peaks seen near the original signal frequency in the QPSD readout spectra are due to the mixing of the 100 Hz power line harmonics and the 92 Hz signal harmonics in the LIA.  
+
+Besides the LDR, the method also demonstrates measurement robustness to changing of \(T_{seq}\) . The motivation of using different \(T_{seq}\) is to get different sampling frequencies as well as measurement bandwidth. Signal responses to different sampling frequencies, i.e., different \(1 / 2T_{seq}\) , are plotted in Fig. 2c. Characterized by the same test field, fluorescence readout shows varying signal responses over the sampling frequency range, while the QPSD readout almost stays at the same level because the measured phase factor only changes with the external field and the sensing time \(T_{\phi}\) . It also indicates that the QPSD readout does not change with varying of fluorescence signal contrast, which is affected by the low spin polarization rate when \(T_{seq}\) is short in the regime of low excitation laser power.  
+
+The measurement bandwidth of the QPSD readout is shown in Fig. 2d, where the signal responses to different test field frequencies are plotted. The plotted values are the magnitudes at the corresponding frequencies in the Fourier transform of the QPSD readout. For the measurements based on the Hahn- echo sequence, we detected the heterodyne signal
+
+<--- Page Split --->
+
+185 for the ac fields. The applied sequence length, \(T_{seq} = 100\mu \mathrm{s}\) , gives the referencing frequency \(f_{s} = 5\mathrm{kHz}\) for Demod. 1. We apply the second driving field with the frequency offset at \(\delta f = 500\mathrm{Hz}\) to have \(N = 10\) samples in a modulation cycle. Due to this, there will be flexibility in deciding the single measurement bandwidth by setting the time constant of Demod. 2. We choose different settings corresponding to the cut- off frequency of the filter at \(100\mathrm{Hz}\) and \(200\mathrm{Hz}\) . Finally, one can conclude that the rotating frame modulation provides QPSD readout magnetometry that has enhanced LDR and robustness in a flexible bandwidth. As we show below, this can be used for measurement of arbitrary fields with low distortion.  
+
+## B. Frequency Offset Heterodyne Readout  
+
+Heterodyne readout has been used to improve the frequency resolution remarkably in nuclear magnetic resonance spectroscopy. It is also a way to achieve high precision microwave sensing [34- 36]. High- order dynamical decoupling sequences are used to narrow the spectral linewidth by decoupling the sensor response from unwanted signal frequencies [20, 21]. Here arises a trade- off between the measurable signal bandwidth and fidelity. High- order dynamical decoupling can ensure a high sensitivity but only allows to measure signals within the narrow bandwidth defined by the sequence. On the other hand, the lower limit on the detectable signal frequency is set by the decoherence time of the sensor. Here, we will use the Hahn- echo sequence in combination with the QPSD readout to measure signals at frequencies that go beyond the coherence time of the sensor.  
+
+In Qdyne, the sampling time usually satisfies \(T_{seq} \neq mT_{\phi}\) so as to get the heterodyne signal [21]. The frequency of this heterodyne signal depends on the timing offset. Here, we choose the measurement sampling time \(T_{seq} = mT_{\phi}\) to obtain the heterodyne readout depending on the signal frequency offset from \(1 / T_{\phi}\) . As a result, the detected phase of signals at frequencies of \(n / T_{\phi}\) is locked by the sequence, where \(n\) can be a random integer. On the other hand, the frequency offset of signals can also introduce phase revivals, i.e. frequency offset heterodyne signal, as shown in Fig. 3a. The detected heterodyne frequency would be the exact offset of the signal frequency to \(1 / T_{\phi}\) .  
+
+The frequency offset heterodyne readout is modeled based on the MW sequence filter [37, 38]. Sampling happens in each time interval of \([NmT_{\phi}, (Nm + 1)T_{\phi}]\) , where \(N \in \mathbb{Z}\) . For
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>FIG. 3. Frequency Offset Heterodyne readout. (a) Hahn-echo sequence is used for this demonstration. The detected phase of the ac signal is locked by the sequence when the frequency is \(f_{ac} = f_{\phi}\) . Otherwise, a heterodyne signal of \(|f_{ac} - f_{\phi}|\) can be measured. The colored regions mark where the quantum phase is accumulated, while phase accumulations at the other areas are canceled in the spin evolution. The figure shows identical heterodyne signals due to \(f_{\phi} - f_{1} = f_{2} - f_{\phi}\) . (b) We apply ac fields at different frequencies with an offset of \(5 \mathrm{~Hz}\) to the sensor so that \(5 \mathrm{~Hz}\) peaks can be detected as the signal response. The signal frequency response of the Hahn-echo sequence and CPMG-2 sequence are plotted after normalization, respectively. In both measurements \(T_{\phi} = 50 \mu \mathrm{s}\) , and the sampling frequency is \(2 \mathrm{kHz}\) according to the applied sequence length. The red lines are the filter functions in theory. (c) Signal frequency response of Hahn-echo measurements with \(1 / T_{seq} = 10 \mathrm{kHz}\) . The dash line indicates the filter introduced by the lock-in amplifier. (d) Sequence dependency of the frequency resolution. In means. 1, a \(20.005 \mathrm{kHz}\) filed is applied and measured by sequences with \(T_{\phi} = 50 \mu \mathrm{s} \pm 4 \mathrm{ns}\) , and \(T_{seq} = 20 T_{\phi}\) . In means. 2, we keep \(T_{\phi}\) unchanged, and offset \(T_{seq}\) with \(\pm 4 \mathrm{ns}\) . In means. 3, the frequency of the applied field is changed to \(16.005 \mathrm{kHz}\) while the other parameters are the same as means. 2. </center>  
+
+215 a random ac signal component \(B_{ac}(t) = B(\omega)e^{- i[\omega t + \phi (\omega)]}\) and a measurement with the MW 216 \(\pi\) - pulse number of \(n\) , we can get the accumulated quantum phase as (see Supplementary 217 Note 2)  
+
+\[\phi_{r}(N) = |G_{n}(\omega)|e^{i\left(-\frac{\omega T_{\phi}}{2} -\frac{P}{2}\pi\right)}\gamma_{e}B(\omega)e^{-i\phi (\omega)}e^{-i\omega NmT_{\phi}}, \quad (7)\]  
+
+218 where \(N\) denotes the sampling timestamp, \(G_{n}(\omega) = |G_{n}(\omega)|e^{i\left(-\frac{\omega T_{\phi}}{2} -\frac{P}{2}\pi\right)}\) is the MW filter
+
+<--- Page Split --->
+
+219 function, \(P = 1\) when the \(\pi\) - pulse number \(n\) is odd and \(P = 2\) when \(n\) is even. Particularly, when \(n = 1\) i.e. Hahn- echo sequence is applied, the filter function satisfies \(|G_{1}(2\pi /T_{\phi})| =\) \(|G_{1}(\pi /T_{\phi})|\) . In principle, measurements of signals at a wide frequency range is feasible by choosing the appropriate \(T_{\phi}\) in Hahn- echo measurements. For example, by using \(T_{\phi}< 1\mu \mathrm{s}\) , one can achieve detection of signals at frequencies higher than 1 MHz. It is more challenging to measure a signal at a lower frequency, such as a signal at 10 kHz, for the reason that a longer \(T_{2}\) is required. With the property described above, it is feasible to use \(T_{\phi} = 50\mu \mathrm{s}\) rather than \(T_{\phi} = 100\mu \mathrm{s}\) to achieve the measurement with a better sensitivity due to the higher signal contrast when \(T_{\phi}\) is smaller. For diamonds which have NV center ensembles with \(T_{2}< 100\mu \mathrm{s}\) , the property makes it feasible to measure signals at the frequencies lower than \(1 / T_{2}\) beyond the coherence limit.  
+
+Given a reference frequency \(\omega_{r e f}~ = ~k\omega_{s},k~\in \mathbb{N}\) , where \(\omega_{s}~ = ~2\pi /(m T_{\phi})\) and \(\omega \in\) \((\omega_{r e f} - \omega_{s} / 2,\omega_{r e f} + \omega_{s} / 2)\) , the evolving phase factor can be rewritten as \(e^{- i\omega N m T_{\phi}}~ =\) \(e^{- i\omega H t}\delta (t - N T_{s})\) , where \(\omega_{H}~ = ~\omega - \omega_{r e f}\) is the heterodyne frequency, \(\delta (t)\) is the Dirac function, and \(T_{s} = m T_{\phi}\) is the sampling period. Thus, the readout signal turns to be  
+
+\[\phi_{r}(t) = G(\omega)\sum_{N = -\infty}^{\infty}\gamma_{e}B_{H}(t)\delta (t - N T_{s}), \quad (8)\]  
+
+where \(B_{H}(t) = B(\omega)e^{- i(\omega_{H}t + \phi)}\) contains all the information from the origin signal to be measured. As discussed in previous section, the quantum phase readout bandwidth is limited by the cut- off frequency \(f_{c}\) of the filter of LIA. For any signal with a frequency range of \([(k - 1)f_{s} + f_{c},(k + 1)f_{s} - f_{c}]\) , aliasing can be filtered. Although a smaller \(f_{c}\) makes the measurement bandwidth narrower, it ensures signals that in a larger frequency range can be detected without aliasing. By changing \(T_{\phi}\) together with \(T_{seq}\) , we can resolve a spectrum in multiple frequency bands with a series of sequences.  
+
+We present two specific examples of the measured frequency responses by using the Hahn- echo and CPMG- 2 sequence. We plot both the theoretical MW filter function and the experimentally measured signal responses together in Fig. 3b. The field sensing time for both experiment and theory calculations is set to be \(T_{\phi} = 50\mu \mathrm{s}\) . In the experiments, we measured the amplitudes of the frequency offset heterodyne signals with \(T_{seq} = 250\mu \mathrm{s}\) , i.e., the magnetic field sampling rate is 4 kHz. Due to this reason, the measured MW filters are combed with a frequency distance of 4 kHz. Aliasing signals exist between the main lobes
+
+<--- Page Split --->
+
+at a distance of 2 kHz, because the readout sampling frequency is \(f_{s} = 2 \mathrm{kHz}\) .  
+
+In order to measure signals that distribute in larger bandwidth, we can increase the sampling frequency, for example, to \(f_{s} = 5 \mathrm{kHz}\) . The spectrum is plotted in Fig. 3c in decibel, from which one can see that magnitudes are the same at 10 kHz and 20 kHz, i.e., \(1 / (2T_{\phi})\) and \(1 / T_{\phi}\) as discussed in the derivation. The insets of Fig. 3c depict the signals that the quantum sensor detects during \(T_{\phi}\) at the two frequencies. In this measurement, the bandwidth limited by the filter of the LIA is at 200 Hz, i.e., the single measurement bandwidth is 400 Hz, and the detectable signal frequency range is 9600 Hz.  
+
+We notice that a single measurement cannot tell if the ac field frequency offset is positive or negative from the heterodyne readout. Additional measurement is needed to distinguish the direction of the frequency offset. By adding a difference to the phase accumulation time \(T_{\phi}\) as well as the sequence time, we can change the reference frequency \(\omega_{ref}\) to get a different heterodyne frequency. By seeing if the heterodyne frequency increases or decreases, we can determine if the signal frequency is larger or smaller than the reference frequency. As the measurements presented in Fig. 3d that \(T_{\phi} = 50 \mu \mathrm{s}\) is offset by a difference of 4 ns and \(T_{seq} = 10 T_{\phi}\) changes accordingly, the detected heterodyne frequency of the signal shift in two different directions. We further investigated the dependency of the heterodyne frequency on the parameters by performing measurements that vary (i) \(T_{seq}\) , (ii) \(T_{seq}\) and \(\omega_{ref}\) . When \(T_{\phi}\) keeps unchanged, the heterodyne frequency shifts by  
+
+\[\Delta \omega_{H} = \omega_{ref}\Delta T_{seq} / T_{seq}. \quad (9)\]  
+
+Using the equation, we can estimate the frequency fidelity of the given sequence. For example, with a timing error \(< 3 \mathrm{ps}\) , the frequency fidelity of a signal around 10 kHz could be only 0.06 mHz. The frequency resolution can be arbitrarily high with a long \(T_{seq}\) at the cost of bandwidth.  
+
+## C. Sensing of Arbitrary Audio Signals  
+
+We demonstrate measurements of arbitrary audio signals by combining the QPSD readout and the frequency offset heterodyne detection. We first generate a signal at 20.08 kHz with its phase varying with time (see Fig. 4a). The MW filter is set by the Hahn- echo sequence
+
+<--- Page Split --->
+
+275 with \(T_{\phi} = 50\mu \mathrm{s}\) . With the reference frequency is at \(20\mathrm{kHz}\) , the heterodyne readout is at \(80\mathrm{Hz}\) , as seen from the simulated curve. The phase of the external field is switched with a cycle of \(80\mathrm{Hz}\) and \(40\mathrm{Hz}\) so that the experimental readout displays the phase change, as shown in Fig. 4a.  
+
+279 Next, we apply a field with its frequency, amplitude and phase all arbitrarily changing. 280 The signal frequency is around \(10\mathrm{kHz}\) and the signal bandwidth is within \(400\mathrm{Hz}\) . Using  
+
+![](images/Figure_4.jpg)
+
+<center>FIG. 4. Detection of arbitrary audio signals. (a) Phase response of the QPSD measurement. A 20.08 kHz signal with sequential phase changing is applied to the sensor. The bars show the phases at different time. Stars mark the readout of the sensor, and the curve is the simulated readout. (b) An ac field is applied with the frequency, amplitude and phase switched every \(100\mathrm{ms}\) . The light blue areas corresponding to the right \(y\) -axis shows the applied field of around \(10\mathrm{kHz}\) , and the red curve shows the QPSD readout. (c) Spectral comparison of the applied signal and the detected magnetic field in a narrow bandwidth. The applied signal is a sum of 20 different sine signals within \(400\mathrm{Hz}\) bandwidth. (d) A signal with wide bandwidth between 10 to \(15\mathrm{kHz}\) is applied and detected by varying the sequence. We set an \(800\mathrm{Hz}\) bandwidth for the measurement of each sequence and use 6 measurements to cover the entire bandwidth. The red dash line shows the spectrum of the output of the AWG, and the solid black line is the spectrum of the detected magnetic field signal. The inset figure is the phase-noise power spectrum density plotted within 1 Hz bandwidth cut. </center>
+
+<--- Page Split --->
+
+\(1 / T_{seq} = 10\mathrm{kHz}\) , we can measure the signals close to \(10\mathrm{kHz}\) with the same sensitivity as the \(20\mathrm{kHz}\) signal. The signal length is one second and consists of ten \(100\mathrm{ms}\) parts. In Fig. 4b, both the applied field waveform and the QPSD readout are plotted. The heterodyne frequencies well resolve the frequency differences in the original waveform. The amplitudes of the readout also correspond to the applied field strength.  
+
+As discussed previously, the measurement bandwidth used in the experiment is \(400\mathrm{Hz}\) . For this, we perform a spectrum analysis as shown in Fig. 4c. The signal to be measured is a sum of 20 tones with random frequencies, amplitudes and phases. In order to distinguish the sign of frequency offsets for each component, we measure the signal using an alternative sequence with \(T_{\phi} = 50\mu \mathrm{s} + 2\mathrm{ns}\) . The sharp peaks observed in the spectrum should shift according to the changes of the measurement sequence, else we exclude them as noise signals generated from our electronics. As shown in Fig. 4c, the applied frequencies are properly resolved. Additionally, we find a \(9.93\mathrm{kHz}\) noise spike from the environment. The errors in magnitude of the measured signal could be induced by the LIA filter, as shown earlier in Fig. 2d. The errors could also be caused by an insufficient sampling number for demodulating the rotating frame modulation. In the measurements, we apply sequences with their lengths corresponding to a sampling frequency \(f_{s} = 5\mathrm{kHz}\) . The frequency difference of the two MWs is \(\delta f = 500\mathrm{Hz}\) and \(N = 10\) for reading out a phase sample. To increase the measurement precision, if we use a smaller \(\delta f\) , it will decrease the bandwidth.  
+
+Though the bandwidth of each measurement sequence is limited, we can still measure a signal with a wider bandwidth by merging several measurements. The condition is that the signal bandwidth should not be larger than the sampling frequency to avoid frequency aliasing. In Fig. 4d, we perform a spectrum analysis for a signal within a bandwidth from \(10\mathrm{kHz}\) to \(15\mathrm{kHz}\) . The signal to be detected is a sum of 10 components with their frequencies randomly distributed in the bandwidth. The signal is generated by an arbitrary signal generator (AWG) and sent to the test field coil. The dotted curve in the figure displays the spectrum of the electrical signal from the AWG. There are some harmonics near each main component due to the limited AWG internal clock and signal length. The components at different frequencies are measured by varying \(T_{\phi}\) to get different referencing frequencies for heterodyne detection. The inset figure shows the power spectrum of the QPSD readout noise within \(1\mathrm{Hz}\) bandwidth, from which we calculate the square root of the standard deviation \(\sigma_{phase} = 0.0022^{\circ}\) . The magnetic field sensitivity depends on the
+
+<--- Page Split --->
+
+applied sequence and the corresponding frequency response. In the case of the Hahn- echo sequence with \(T_{\phi} = 50\mu \mathrm{s}\) , we have a calibrated scalar factor of \(k_{sf} = 0.071^{\circ} / \mathrm{nT}\) , and the magnetic field sensitivity is \(\eta = 31\mathrm{pT} / \sqrt{\mathrm{Hz}}\) . Taking the scalar factor into the calculation of LDR \([-\pi ,\pi ]\) , we can get the dynamic range in decibels as \(20\log [\pi /(k_{sf}\eta)] = 98\mathrm{dB}\) . The sensitivity can be further optimized by, e.g., using higher laser power, applying high- order dynamical decoupling sequences, and implementing flux concentration.  
+
+Finally, we demonstrate the detection and demodulation of audio signals. Although 10 kHz is within the audio frequency band, most of the daily audio sounds have frequencies ranging from hundreds of Hz to a few kHz. Therefore, signals need to be modulated to a detectable frequency range. For this, we have used (i) a melody piece composed of 3 tones and (ii) a small part from Dr. Martin Luther King Jr.'s famous speech "I have a dream", to test the waveform reconstruction by the diamond quantum sensor. The tones of the melody have frequencies distributed between 500 Hz and 700 Hz. Therefore, we mix it with a 9.5 kHz reference to get the signal modulated around 10 kHz and broadcast the mixed signal to the diamond. For case (ii), we have to compress the signal bandwidth into 200 Hz and then modulate it with a 10 kHz reference. The audio reconstructed from the diamond sensor can be heard and compared with the original audio (see Supplementary Audio S1- S4).  
+
+## III. DISCUSSIONS  
+
+In this work, we overcome the LDR limitation of the conventional interferometric readout through a new technique that includes the QPSD scheme and the frequency offset heterodyne readout. The technique allows one to measure unknown signals with maximal sensitivity independent of their dynamic range. It improves the feasibility for quantum sensors to perform high- sensitive measurements of different physical quantities using interferometric methods, beyond magnetometry.  
+
+Theoretically, the extended LDR comes from the multiple measurements that have the quantum phase evolving through the entire phase range \([-\pi ,\pi ]\) so that the initial phase factor that contains the external field information can be resolved. Such an extended phase range affects the measurement bandwidth as well as the sensitivity. In theory, the sensitivity does not deteriorate a lot from the conventional fluorescence readout except for a factor of \(\sqrt{2}\) . While in the experiment, we suffer from a low contrast \(C = 0.19\%\) due to
+
+<--- Page Split --->
+
+343 the low excitation laser power (80 mW). The contrast and the fluorescence photon count 344 can significantly increase when the laser reaches saturation power [39]. Different dynamical 345 decoupling sequences can also improve the magnetic field sensitivity through the filter func- 346 tion \(G(\omega)\) . Flux concentration could further improve the signal- noise ratio [31, 40]. The flux 347 concentrator can be very small compared to conventional dipole antennas because the gain 348 no longer depends on the signal wavelength. With the millimeter size diamond dimension, 349 the flux concentration factor can easily reach a factor of hundreds when using a concentrator 350 in centimeters.  
+
+351 The QPSD readout can also enhance the capability of vector magnetometry. Convention- 352 ally, fluorescence emitted from NV centers in multiple orientations is measured sequentially 353 to acquire the vector components. Similar to the methods developed here, one could also 354 modulate the signal on each orientation with different modulation frequencies [41]. Per- 355 forming measurements on different NV orientations with appropriate synchronization can 356 suppress the phase errors in vector reconstruction.  
+
+357 In conclusion, we demonstrated high- sensitive distortion- free quantum- assisted detection 358 of audio signals, including melody and speech, using the QPSD scheme in combination with 359 the heterodyne readout. A further improvement in sensitivity can be achieved by using 360 flux concentrators. One could also generalize the current methods to achieve vector magne- 361 tometry with extended LDR. We envisage that the techniques developed here will have the 362 potential to develop low- distortion, small- volume quantum sensors for various applications 363 in science and technology.  
+
+## IV. METHODS  
+
+## A. Experimental setup  
+
+366 The diamond used in the experiment is a (111)- oriented (0.5 mm) cube obtained from a 367 single crystal grown by the temperature gradient method at high- pressure high- temperature 368 (HPHT) conditions. The diamond is 99.97% \(^{12}\mathrm{C}\) enriched, and has an initial nitrogen concentration of 1.4 ppm. The final NV concentration is 0.4 ppm after electron irradiation 370 and annealing. Dephasing time of the NV ensemble is obtained as \(T_{2}^{*} = 8.5\mu \mathrm{s}\) by Ramsey 371 sequence, and a decoherence time \(T_{2} = 200\mu \mathrm{s}\) is measured by Hahn- echo sequence. The dia
+
+<--- Page Split --->
+
+372 mond is positioned at center of a home- built three dimension coils system, and is illuminated 373 by a 532 nm laser (Lighthous Sprout- G) at around 80 mW. Microwave signals are generated 374 from two sources (Rohde&Schwarz, SMIQ03B) and are individually cut by two switches 375 before the combination. Measurement sequences are generated by a data timing generator 376 (Tekreoxin, DTG5274). After the combination and amplification of the MW signals, MW 377 pulses are feed to the diamond through a dielectric resonator antenna [42]. The detected 378 fluorescence signal is demodulated by a LIA (Zurich Instruments, HF2LI) which has two 379 independent differential input channels and demodulators. To generate arbitrary magnetic 380 fields, we write signals to an AWG (Tektronix, AWG520) with \(10^{5}\) samples per second out- 381 put sampling rate. The test signals are continuously repeated and sent to a copper loop 382 near the diamond.  
+
+## B. Spectrum analyzing  
+
+384 The spectrum to be analyzed is divided into several sections with the bandwidth set 385 by the LIA for data acquisition. In each section, the center frequency determines \(T_{\phi}\) of 386 the measurement sequence. Usually, the center frequency satisfies \(f_{c} = 1 / T_{\phi} + \epsilon /T_{seq}\) 387 where \(\epsilon = 0,\pm 1\) . A time trace is recorded after running the sequence, and a spectrum is 388 acquired from the Fourier transform of the time trace. However, the spectrum is a fold of 389 the two sidebands with repect to the center frequency. The sequence with \(T_{\phi}^{\prime} = T_{\phi} + t_{clk}\) 390 and \(T_{seq}^{\prime} = mT_{\phi}^{\prime}\) is applied to get an alternate spectrum with analyzed frequencies shift by 391 \(\Delta f = \pm \left|1 / T_{\phi} - 1 / T_{\phi}^{\prime}\right|\) . The direction of the frequency shift shows which side the signal 392 component belongs to. In the algorithm, we set a threshold to separate signal spikes from 393 noise, and use the known sequences induced spectrum frequency shift to distinguish the signs 394 of the signal offset frequency to the center frequency. The signal spikes that do not shift 395 according \(\Delta f\) are recognized as systematic noise spikes. Then, the spectrum of the selected 396 section can be replotted as the example shown in Fig. 4c. After measuring the spectra of 397 all the sections, we can get the final spectrum by merging them together.
+
+<--- Page Split --->
+
+We acknowledge financial support by European Union's Horizon 2020 research and innovation program ASTERIQS under grant No. 820394, European Research Council advanced grant No. 742610, SMeI, Federal Ministry of Education and Research (BMBF) project MiLiQuant and Quamapolis, German Research Foundation grant GRK 2198 and 2642, and Japan Society for the Promotion of Science (JSPS) KAKENHI No. 17H02751.
+
+<--- Page Split --->
+
+[12] Chen, E. H. et al. High- sensitivity spin- based electrometry with an ensemble of nitrogen- vacancy centers in diamond. Physical Review A 95, 053417 (2017).[13] Tisler, J. et al. Single defect center scanning near- field optical microscopy on graphene. Nano Lett. 13, 3152–3156 (2013).[14] Zheng, H. J. et al. Zero- field magnetometry based on nitrogen- vacancy ensembles in diamond. Phys. Rev. Appl. 11 (2019).[15] Wickenbrock, A. et al. Microwave- free magnetometry with nitrogen- vacancy centers in diamond. Appl. Phys. Lett. 109 (2016).[16] Barry, J. F. et al. Optical magnetic detection of single- neuron action potentials using quantum defects in diamond. Proc. Nat. Acad. Sci. 201601513 (2016).[17] Wang, P. et al. High- resolution vector microwave magnetometry based on solid- state spins in diamond. Nat. Commun. 6, 6631 (2015).[18] Farfurnik, D. et al. Optimizing a dynamical decoupling protocol for solid- state electronic spin ensembles in diamond. Phys. Rev. B 92 (2015).[19] de Lange, G., Riste, D., Dobrovitski, V. V. & Hanson, R. Single- spin magnetometry with multipulse sensing sequences. Phys. Rev. Lett. 106 (2011).[20] Schmitt, S. et al. Submillihertz magnetic spectroscopy performed with a nanoscale quantum sensor. Science 356, 832–836 (2017).[21] Boss, J. M., Cujia, K. S., Zopes, J. & Degen, C. L. Quantum sensing with arbitrary frequency resolution. Science 356, 837–840 (2017).[22] Clevenson, H. et al. Robust high- dynamic- range vector magnetometry with nitrogen- vacancy centers in diamond. Appl. Phys. Lett. 112.[23] Nusran, N. M., Momeen, M. U. & Dutt, M. V. G. High- dynamic- range magnetometry with a single electronic spin in diamond. Nat. Nano. 7, 109–113 (2012).[24] Joas, T. et al. Online adaptive quantum characterization of a nuclear spin. Npj Quantum Inf. 7, 1–8 (2021).[25] Gurses, B. V., Whitmore, K. T. & Cohen, M. B. Ultra- sensitive broadband ”awesome” electric field receiver for nanovolt low- frequency signals. Rev. Sci. Instrum. 92 (2021).[26] Cohen, M. B. et al. Broadband longwave radio remote sensing instrumentation. Rev. Sci. Instrum. 89 (2018).
+
+<--- Page Split --->
+
+[27] Sedlacek, J. A. et al. Microwave electrometry with rydberg atoms in a vapour cell using bright atomic resonances. Nat. Phys. 8, 819- 824 (2012).  
+
+[28] Meyer, D. H., Kunz, P. D. & Cox, K. C. Waveguide- coupled rydberg spectrum analyzer from 0 to 20 ghz. Phys. Rev. Appl. 15 (2021).  
+
+[29] Boto, E. et al. Moving magnetoencephalography towards real- world applications with a wearable system. Nature 555, 657 (2018).  
+
+[30] Gerginov, V., da Silva, F. C. S. & Howe, D. Prospects for magnetic field communications and location using quantum sensors. Rev. Sci. Instrum. 88 (2017).  
+
+[31] Fescenko, I. et al. Diamond magnetometer enhanced by ferrite flux concentrators. Phys. Rev. Res. 2 (2020).  
+
+[32] Seidel, D. & Muga, J. G. Two- frequency ramsey interferometry. Phys. Rev. A 75 (2007).  
+
+[33] Vutha, A. C. & Hessels, E. A. Frequency- offset separated oscillatory fields. Phys. Rev. A 92 (2015).  
+
+[34] Meinel, J. et al. Heterodyne sensing of microwaves with a quantum sensor. Nat. Commun. 12, 2737 (2021).  
+
+[35] Chu, Y. M. et al. Precise spectroscopy of high- frequency oscillating fields with a single- qubit sensor. Phys. Rev. Appl. 15 (2021).  
+
+[36] Staudenmaier, N., Schmitt, S., McGuinness, L. P. & Jelezko, F. Phase- sensitive quantum spectroscopy with high- frequency resolution. Phys. Rev. A 104 (2021).  
+
+[37] Staudacher, T. et al. Nuclear magnetic resonance spectroscopy on a (5- nanometer)(3) sample volume. Science 339, 561- 563 (2013).  
+
+[38] Bar- Gill, N., Pham, L. M., Jarmola, A., Budker, D. & Walsworth, R. L. Solid- state electronic spin coherence time approaching one second. Nat. Commun. 4, 1743 (2013).  
+
+[39] Zhang, C. et al. Diamond magnetometry and gradiometry towards subpicotesla dc field measurement. Phys. Rev. Appl. 15 (2021).  
+
+[40] Griffith, W. C., Jimenez- Martinez, R., Shah, V., Knappe, S. & Kitching, J. Miniature atomic magnetometer integrated with flux concentrators. Appl. Phys. Lett. 94, 023502 (2009).  
+
+[41] Schloss, J. M., Barry, J. F., Turner, M. J. & Walsworth, R. L. Simultaneous broadband vector magnetometry using solid- state spins. Phys. Rev. Appl. 10 (2018).  
+
+[42] Kapitanova, P. et al. 3d uniform manipulation of nv centers in diamond using a dielectric resonator antenna. Jetp Lett. 108, 588- 595 (2018).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+SupplementaryInformation.docx AudioS1melody.wav AudioS2detmelody.wav AudioS3voice.wav AudioS4detvoice.wav
+
+<--- Page Split --->

preprint/preprint__b3cdb805334a0d61d0b77d7ab9ffa2231861aba3b0f051ea183438476e3179db/preprint__b3cdb805334a0d61d0b77d7ab9ffa2231861aba3b0f051ea183438476e3179db_det.mmd

@@ -0,0 +1,436 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 848, 177]]<|/det|>
+# Quantum-assisted distortion-free audio signal sensing  
+
+<|ref|>text<|/ref|><|det|>[[44, 196, 490, 216]]<|/det|>
+Chen Zhang (chen.zhang@pi3. uni- stuttgart.de)  
+
+<|ref|>text<|/ref|><|det|>[[50, 219, 248, 236]]<|/det|>
+University of Stuttgart  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 248, 281]]<|/det|>
+Durga Dasari University of Stuttgart  
+
+<|ref|>text<|/ref|><|det|>[[44, 289, 248, 327]]<|/det|>
+Matthias Widmann University of Stuttgart  
+
+<|ref|>text<|/ref|><|det|>[[44, 335, 600, 375]]<|/det|>
+Jonas Meinel University of Stuttgart https://orcid.org/0000- 0003- 4040- 8361  
+
+<|ref|>text<|/ref|><|det|>[[44, 380, 248, 419]]<|/det|>
+Vadim Vorobyov University of Stuttgart  
+
+<|ref|>text<|/ref|><|det|>[[44, 426, 207, 465]]<|/det|>
+Polina Kapitanova ITMO University  
+
+<|ref|>text<|/ref|><|det|>[[44, 472, 456, 512]]<|/det|>
+Elizaveta Nenasheva Giricond Research Institute, Ceramics Co., Ltd.  
+
+<|ref|>text<|/ref|><|det|>[[44, 519, 585, 559]]<|/det|>
+Kazuo Nakamura Tokyo Gas Co., Ltd. https://orcid.org/0000- 0002- 3412- 834X  
+
+<|ref|>text<|/ref|><|det|>[[44, 565, 308, 604]]<|/det|>
+Hitoshi Sumiya Sumitomo Electric Industries  
+
+<|ref|>text<|/ref|><|det|>[[44, 611, 905, 653]]<|/det|>
+Shinobu Onoda National Institutes for Quantum Science and Technology https://orcid.org/0000- 0003- 1425- 0708  
+
+<|ref|>text<|/ref|><|det|>[[44, 658, 605, 698]]<|/det|>
+Junichi Isoya University of Tsukuba https://orcid.org/0000- 0002- 9598- 625X  
+
+<|ref|>text<|/ref|><|det|>[[44, 704, 248, 743]]<|/det|>
+Jörg Wrachtrup University of Stuttgart  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 789, 101, 806]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 826, 576, 846]]<|/det|>
+Keywords: Quantum sensors, metrology, audio signal sensing  
+
+<|ref|>text<|/ref|><|det|>[[44, 864, 345, 883]]<|/det|>
+Posted Date: November 19th, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 902, 473, 920]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 1068484/v1
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 44, 911, 87]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 123, 920, 167]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on August 8th, 2022. See the published version at https://doi.org/10.1038/s41467-022-32150-1.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[186, 85, 808, 109]]<|/det|>
+# Quantum-assisted distortion-free audio signal sensing  
+
+<|ref|>text<|/ref|><|det|>[[170, 127, 828, 150]]<|/det|>
+Chen Zhang, \(^{1, *}\) Durga Dasari, \(^{1, \dagger}\) Matthias Widmann, \(^{1}\) Jonas Meinel, \(^{1}\) Vadim  
+
+<|ref|>text<|/ref|><|det|>[[180, 157, 812, 179]]<|/det|>
+Vorobyov, \(^{1}\) Polina Kapitanova, \(^{2}\) Elizaveta Nenasheva, \(^{3}\) Kazuo Nakamura, \(^{4}\)  
+
+<|ref|>text<|/ref|><|det|>[[186, 185, 805, 207]]<|/det|>
+Hitoshi Sumiya, \(^{5}\) Shinobu Onoda, \(^{6}\) Junichi Isoya, \(^{7}\) and Jörg Wrachtrup \(^{1}\)  
+
+<|ref|>text<|/ref|><|det|>[[285, 216, 710, 238]]<|/det|>
+\(^{1}\) 3rd Institute of Physics, University of Stuttgart,  
+
+<|ref|>text<|/ref|><|det|>[[295, 245, 700, 264]]<|/det|>
+Pfaffenwaldring 57, Stuttgart 70569, Germany  
+
+<|ref|>text<|/ref|><|det|>[[315, 270, 677, 290]]<|/det|>
+\(^{2}\) Department of Physics and Engineering,  
+
+<|ref|>text<|/ref|><|det|>[[275, 297, 720, 316]]<|/det|>
+ITMO University, Saint Petersburg 197101, Russia  
+
+<|ref|>text<|/ref|><|det|>[[145, 323, 848, 344]]<|/det|>
+\(^{3}\) Giricond Research Institute, Ceramics Co. Ltd., Saint Petersburg 194223, Russia  
+
+<|ref|>text<|/ref|><|det|>[[252, 350, 739, 370]]<|/det|>
+\(^{4}\) Hydrogen and Carbon Management Technology Section,  
+
+<|ref|>text<|/ref|><|det|>[[204, 376, 789, 396]]<|/det|>
+Hydrogen and Carbon Management Technology Strategy Department,  
+
+<|ref|>text<|/ref|><|det|>[[275, 403, 714, 422]]<|/det|>
+Tokyo Gas Co. Ltd., Yokohama 230- 0045, Japan  
+
+<|ref|>text<|/ref|><|det|>[[115, 428, 880, 449]]<|/det|>
+\(^{5}\) Advanced Materials Labotatory, Sumitomo Electric Industries Ltd., Itami 664- 0016, Japan  
+
+<|ref|>text<|/ref|><|det|>[[115, 455, 710, 475]]<|/det|>
+\(^{6}\) Takasaki Advanced Radiation Research Institute,  
+
+<|ref|>text<|/ref|><|det|>[[144, 481, 853, 502]]<|/det|>
+National Institutes for Quantum Science and Technology, Takasaki 370- 1292, Japan  
+
+<|ref|>text<|/ref|><|det|>[[325, 508, 666, 528]]<|/det|>
+\(^{7}\) Faculty of Pure and Applied Sciences,  
+
+<|ref|>text<|/ref|><|det|>[[285, 535, 710, 555]]<|/det|>
+University of Tsukuba, Tsukuba 305- 8573, Japan
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[452, 87, 541, 106]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[113, 115, 883, 409]]<|/det|>
+Quantum sensors are keeping the cutting- edge sensitivities in metrology. However, for high- sensitive measurements of arbitrary signals, limitations in linear dynamic range could introduce distortions when sensing the frequency, magnitude and phase of unknown signals. Here, we overcome these limitations with advanced sensing protocol that combines quantum phase- sensitive detection with the heterodyne readout. We present theoretical and experimental investigations using nitrogen- vacancy centers in diamond, showing the ability to sense audio signals with a 98 dB linear dynamic range, a 31 pT/Hz \(^{1 / 2}\) sensitivity, and arbitrary frequency resolution. Further, we perform the quantum assisted distortion free audio signal (melody piece, speech) sensing with high fidelity. The methods developed here could broaden the horizon for quantum sensors towards applications in telecommunication, where high- fidelity and low- distortion at multiple frequency bands within small sensing volumes are required.  
+
+<|ref|>sub_title<|/ref|><|det|>[[98, 466, 319, 484]]<|/det|>
+## 18 I. INTRODUCTION  
+
+<|ref|>text<|/ref|><|det|>[[95, 508, 883, 872]]<|/det|>
+Quantum sensors are setting new frontiers of sensing techniques with their extraordinary performances in sensitivity and stability [1- 5]. These techniques rely on either measuring the line- shift of spin or atomic transition frequencies or reading out the relative populations of the occupied energy levels using interferometric methods [6, 7]. In most cases, there are trade- off relations between the sensitivity and other features in metrology [8]. For example, a high- sensitive measurement acquired by detecting the transition line shift requires a narrow linewidth, which, on the other hand, will limit the dynamic range. Interferometric measurements detect a sinusoidal response, and linearity is only achieved when the phase signal is in a small dynamic range. It sets a massive limitation on the sensitivity when sensing an unknown signal that gets measured beyond this linear regime, for example, when the working point of the sensor is at the maxima or minima of the sinusoidal signal response. Thus, it becomes a bottleneck for high sensitivity measurements that are required in many cutting- edge applications. Operating within the linear dynamic range (LDR) can be crucial for reconstructing unknown signals. One way to directly extract the phase factor, which is
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[95, 87, 882, 212]]<|/det|>
+linear to the physical quantity to be detected, is to use phase- sensitive detection known as the classical lock- in technique. In this work, using a modified sensing scheme that introduces an external readout phase modulation, we acquire the target quantum phase signal after demodulation. Therefore, the LDR is no longer limited to the small- angle approximation. Hence our method combines large dynamic range with maximum sensitivity.  
+
+<|ref|>text<|/ref|><|det|>[[95, 218, 882, 450]]<|/det|>
+Nitrogen- vacancy (NV) centers in diamond have been at the forefront in performing high- sensitive measurements of various physical quantities, viz., magnetic and electric field, temperature, and strain distributions internal and external to diamond [9–13]. The NV magnetometry has been performed under bias fields ranging from zero- field to a few Tesla, and for sensing signals with frequencies ranging from DC to a few GHz [14–17]. While dynamical- decoupling techniques are usually employed for high sensitivity [9, 18, 19], arbitrary frequency resolution can be achieved with the quantum heterodyne (Q- dyne) detection [20, 21]. However, both methods suffer from a limited LDR when they are applied to measure arbitrary signals.  
+
+<|ref|>text<|/ref|><|det|>[[95, 457, 882, 740]]<|/det|>
+For high dynamic range measurements, a closed- loop frequency- locking scheme together with optically detected magnetic resonance (ODMR) can be used to track resonance frequency shifts [22]. However, this scheme cannot be used for ac field measurements in combination with interferometric methods, if the signal frequency is higher than the readout sampling frequency. Phase- estimation algorithms can effectively improve the LDR in Ramsey measurements by varying the sequence with adaptive feedback schemes [23, 24]. However, for the case of ac sensing schemes e.g. Hahn- echo, as varying the sequence itself will change the sensor response to the ac signals, such methods become less applicable. Therefore, a technique is still missing, that addresses the LDR while maintaining high sensitivity and frequency resolution, for example, in sensing arbitrary radio- frequency fields within a broad bandwidth.  
+
+<|ref|>text<|/ref|><|det|>[[95, 747, 882, 925]]<|/det|>
+Sensing radio- frequency signals by electric- field sensors, either conventional electronic receivers or the Rydberg atom sensors, need antennas to collect and guide the electric signals towards the sensors [25–28]. Although the receivers can be highly integrated, the dimension of antennas can scale to meters due to the signal wavelength. The size becomes critical when there is limited space for the sensor, for example, in a satellite. In this regard, quantum magnetometers can be very attractive due to their small sensing volume and high sensitivity [29]. A flux concentrator can be used as a substitute to conventional antennas
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[95, 87, 880, 133]]<|/det|>
+65 for obtaining high signal gain. Independent of the signal wavelength, the dimensions of such flux concentrators can be as small as a few centimeters [30, 31].  
+
+<|ref|>text<|/ref|><|det|>[[95, 140, 882, 477]]<|/det|>
+67 In this paper, we demonstrate quantum- assisted distortion- free audio signal sensing with 68 NV center ensembles in diamond using the quantum- phase- sensitive detection (QPSD) technique combined with heterodyne readout. Firstly, we introduce the QPSD technique, which 70 can provide an extended LDR in interferometry measurements by using two synchronized 71 driving fields. Then, we present the heterodyne readout, which can interpret e.g. ac signals 72 to get frequency information. Taking advantage of the bandwidth of the Hahn- echo sequence 73 and the frequency comb induced by the continuous sampling, we demonstrate measurements 74 of audio signals around 10 kHz, beyond the coherence limit without losing sensitivity. Fi- 75 nally, we present arbitrary audio signal measurements with a LDR of 98 dB at a sensitivity 76 of 31 pT/Hz1/2. A piece of melody and a speech are encoded as magnetic field signals and 77 measured by the NV magnetometer. By using the sensor as a quantum radio, we demon- 78 strate the application potentials for areas such as quantum- assisted telecommunication and 79 unknown signal exploration.  
+
+<|ref|>sub_title<|/ref|><|det|>[[97, 533, 252, 550]]<|/det|>
+## 80 II. RESULTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[97, 578, 508, 596]]<|/det|>
+## 81 A. Quantum Phase Sensitive Detection  
+
+<|ref|>text<|/ref|><|det|>[[95, 620, 882, 930]]<|/det|>
+82 In interferometric measurements, the quantum phase to be detected is usually converted 83 to a quantum state population difference, resulting in a sinusoid readout [9]. A way to 84 extract the phase factor from the sinusoidal readout is to modulate the phase with a specific 85 frequency and perform phase- sensitive modulation. Such a quantum phase modulation can 86 be introduced by using the difference between the quantum phase of the sensor to an ex- 87 ternal oscillator. The Q- dyne method uses such a strategy for resolving frequency of signals 88 better than the relaxation time of the sensor, as shown in Fig. 1a [20, 21]. However, it 89 cannot be used for phase- sensitive detection because the Q- dyne frequency is also what to 90 be resolved and an extra modulation is still needed [21]. Another way to introduce such a 91 phase modulation is to use the frequency offset between two different driving fields of the 92 sensor [32, 33]. These driving fields define two rotating frames, and the evolution of the spin 93 as seen from one rotating frame will lead to a quantum phase modulation due to the relative
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 92, 874, 423]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 445, 881, 775]]<|/det|>
+<center>FIG. 1. Phase-sensitive NV magnetometry (a) Continuous sampling induced phase reviving signals, known as the quantum heterodyne (Q-dyne) detection. The phase reviving frequency changes with the external field and cannot be used for phase-sensitive detection. The signal responses \(G(s_{\phi},k)\) should be small to ensure measurement linearity. \(R\) is the detected photon rate, and \(C\) is the detected signal contrast. (b) Unlike the Q-dyne detection, the quantum phase-sensitive detection (QPSD) is based on the rotating frame modulation induced by the evolving phase difference of the two driving MW fields. Two frequency-offset MWs acquire a phase difference of \(\alpha - \beta = 2\pi \delta f\Delta t\) after the sampling time interval \(\Delta t\) . In the Bloch sphere picture, it can be understand as the MW2 defined rotating frame \(x_{2}y_{2}z\) rotates with rate of \(\delta f\) referring to the MW1 defined rotating frame \(x_{1}y_{1}z\) . The acquired quantum phase is \(\theta = 2\pi \delta f\Delta t - \phi\) at sample 2. Through the quantum phase modulation, the acquired readout representing the Bloch vector projections is as shown in (c), where we present the measurements of the quantum phase \(\phi = 0\) and \(\phi \neq 0\) . By demodulating the acquired signal with a lock-in amplifier, we can get the phase values. The dashed box shows the measurement sequences we applied in experiments. Except for the last \(\pi /2\) pulse applied with MW2, all the other driving pulses are generated by MW1. The fluorescence signal is demodulated at the frequency of \(1 / (2T_{seq})\) to get a fluorescence intensity readout for a sample. The QPSD readout is acquired with the demodulation at \(\delta f\) . (d) Schematic of the experiment. NV centers ensemble in diamond is used to perform the QPSD readout. </center>  
+
+<|ref|>text<|/ref|><|det|>[[98, 805, 880, 930]]<|/det|>
+94 rotation of the two frames, i.e., rotating frame modulation. The modulation frequency only 95 depends on the frequency difference of the two driving fields, as shown in Fig. 1b and c. 96 By performing multiple measurements within a modulation cycle and upon using lock- in 97 detection, we will achieve phase- sensitive detection. Below we mathematically describe this 98 relative evolution of the sensor under such interferometric measurement with two- frequency
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[97, 88, 230, 106]]<|/det|>
+99 driving fields.  
+
+<|ref|>text<|/ref|><|det|>[[92, 112, 880, 161]]<|/det|>
+100 Aligning an external field \(B_{0}\) along the NV axis, we use the two- level subspace of the NV ground triplet in the derivation. Thus, the Hamiltonian of the system can be written as:  
+
+<|ref|>equation<|/ref|><|det|>[[344, 188, 877, 209]]<|/det|>
+\[\mathcal{H} = \omega_{0}S_{z} + \gamma_{e}B_{1}\cos \left(2\pi f_{1}t + \alpha\right)S_{x}, \quad (1)\]  
+
+<|ref|>text<|/ref|><|det|>[[92, 236, 881, 338]]<|/det|>
+102 where \(\omega_{0}\) is the transition frequency of the two- level subspace, \(B_{1}\) is the oscillating magnetic field perpendicular to the NV axis, \(f_{1}\) and \(\alpha\) are the frequency and phase of the driving field, and \(\gamma_{e}\) is the electron gyromagnetic ratio. In the rotating frame defined by the resonance frequency, the time- dependent Hamiltonian is  
+
+<|ref|>equation<|/ref|><|det|>[[296, 365, 877, 387]]<|/det|>
+\[\mathcal{H}_{1}^{\prime} = \Omega_{1}\cos \left(\delta \omega_{1} + \alpha\right)S_{x} + \Omega_{1}\sin \left(\delta \omega_{1}t + \alpha\right)S_{y}, \quad (2)\]  
+
+<|ref|>text<|/ref|><|det|>[[92, 412, 881, 567]]<|/det|>
+106 where \(\delta \omega_{1} = 2\pi (f_{0} - f_{1})\) , and \(\Omega_{1} = \gamma_{e}B_{1} / 2\) is the Rabi frequency introduced by MW1. In 107 interferometry measurements, a \(\pi /2\) pulse prepares the spin state from the polarized state to 108 an equalized population, and another \(\pi /2\) pulse projects the quantum phase as a population 109 difference after the sensing procedure. We use the second driving field, MW2, to offset the 110 frequency of the second \(\pi /2\) pulse. \(\delta \omega_{2},\beta\) and \(\Omega_{2}\) are used to denote the frequency offset, 111 Rabi frequency, and phase of MW2. After this, the measured spin- expectation value is  
+
+<|ref|>equation<|/ref|><|det|>[[314, 584, 877, 627]]<|/det|>
+\[\langle S_{z}\rangle = \sin \left[\phi +\frac{\pi}{2}\left(\frac{\delta\omega_{1}}{\Omega_{1}} -\frac{\delta\omega_{2}}{\Omega_{2}}\right) + \alpha -\beta \right], \quad (3)\]  
+
+<|ref|>text<|/ref|><|det|>[[92, 644, 880, 743]]<|/det|>
+112 where \(\phi\) is the acquired quantum phase which contains the information we want to measure, 113 both of the MWs are near- resonant with \(\delta \omega_{1}\ll \Omega_{1}\) , \(\delta \omega_{2}\ll \Omega_{2}\) . Therefore, the off- resonant 114 term can be neglected, and the phase difference term \(\alpha - \beta\) will evolve with time so that 115 there is  
+
+<|ref|>equation<|/ref|><|det|>[[390, 750, 877, 771]]<|/det|>
+\[\langle S_{z}\rangle \approx \sin \left(\phi +2\pi \delta f\cdot t\right), \quad (4)\]  
+
+<|ref|>text<|/ref|><|det|>[[92, 789, 565, 809]]<|/det|>
+116 where \(\delta f\) is the frequency difference of the two MWs.  
+
+<|ref|>text<|/ref|><|det|>[[92, 815, 880, 915]]<|/det|>
+117 The above result can be seen as a modulation of the rotating frame itself. As schematically 118 shown in Fig. 1b (left Bloch sphere), we can assume that the two driving fields have the same 119 phase at the duration of the second \(\pi /2\) pulse, and this defines an instantaneous rotating 120 frame with coordinates \(x_{1}y_{1}z\) . Thus, the readout is similar to that of the regular Ramsey
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 85, 883, 397]]<|/det|>
+interferometry using a single driving field. After an interval of \(\Delta t\) , MW2 develops a phase difference of \(2\pi \delta f\Delta t\) . Since the quantum phase is finally measured by MW2, the Bloch vector rotates in the new instantaneous rotating frame with coordinates \(x_{2}y_{2}z\) , as shown in Fig. 1b (the right Bloch sphere). The rotating frame defined by MW2 rotates continuously around the \(z\) - axis with the frequency of \(\delta f\) . Due to this, the fluorescence readout also modulates in a sinusoidal fashion, as shown in Fig. 1c. While the readout signal frequency depends on \(\delta f\) and its amplitude depends on the signal contrast, the initial phase, \(\phi\) , is linear to the field to be measured. Through the external modulation induced by the MWs, the working point of the sensor evolves in the entire phase range, which ensures the LDR of the initial phase measurement covering \([-\pi ,\pi ]\) . By fitting or demodulating the fluorescence signal, we can resolve the changing of the phase factor \(\phi\) between each modulation cycle and find measurement linearity for the external field.  
+
+<|ref|>text<|/ref|><|det|>[[90, 403, 883, 662]]<|/det|>
+The measurement sequence we applied in the experiment is depicted in Fig. 1c, in which \(T_{\phi}\) is the field sensing time, \(T_{seq}\) is the sequence length of one measurement, and we use a second measurement with the final pulse changed to \((\pi /2)_{- x}\) . As a result, the fluorescence signal is modulated with a frequency of \(f_{s} = 1 / (2T_{seq})\) , which is also the sampling frequency of the fluorescence readout. The demodulation of the fluorescence signal, denoted by Demod. 1 in Fig. 1d, has a readout bandwidth \(f_{s} / 2\) set by the Shannon sampling theorem. The readout is further demodulated by another demodulator of the lock- in amplifier (LIA), denoted as Demod. 2. Upon measuring \(N\) samples of the fluorescence readout, the bandwidth of the phase readout is narrowed down to \(f_{s} / (2N)\) . These measurements are schematically shown in Fig. 1d.  
+
+<|ref|>text<|/ref|><|det|>[[90, 667, 881, 714]]<|/det|>
+The sensitivity of such measurements can be derived based on the fitting of the \(N\) samples in the measurement time of \(N \cdot 2T_{seq}\) . The minimum detectable phase is derived as  
+
+<|ref|>equation<|/ref|><|det|>[[423, 732, 877, 774]]<|/det|>
+\[\delta \phi = \frac{2}{\sqrt{N}} \frac{1}{C \sqrt{N}}, \quad (5)\]  
+
+<|ref|>text<|/ref|><|det|>[[90, 792, 881, 864]]<|/det|>
+where \(C\) is the fluorescence signal contrast, \(\mathcal{N}\) is the detected photon counts in each measurement. The sensitivity to external magnetic field, however, is still subject to the applied MW sequence, can be derived as  
+
+<|ref|>equation<|/ref|><|det|>[[395, 881, 877, 927]]<|/det|>
+\[\eta = \frac{2}{\gamma_{e}|G(\omega)|C}\sqrt{\frac{2T_{seq}}{\mathcal{N}}}, \quad (6)\]
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[125, 92, 872, 558]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 576, 881, 777]]<|/det|>
+<center>FIG. 2. Sensing performance of the QPSD. (a) Spectra and linearity comparison of the normal Ramsey readout and the QPSD readout. We apply \(T_{\phi} = 6.25\mu s\) in both measurements. The applied peak-to-peak field and the readout are plotted showing the linearity of the measurements. (b) Spectra and linearity comparison of normal Hahn-echo readout and the QPSD readout. The applied phase accumulation time \(T_{\phi} = 12.5\mu s\) . Thus, the Hahn-echo measurements performs a higher sensitivity but smaller dynamic range than the Ramsey measurements. (c) Signal response to different sampling frequencies. The measurements use the same calibration field, and the readouts are normalized to be plotted in the same vertical axis. (d) Measurement bandwidth. Ramsey and Hahn-echo sequences are applied to measure test fields at different frequencies with the same magnitude. The heterodyne frequency responses are limited in bandwidth by the cut-off frequency of the LIA. </center>  
+
+<|ref|>text<|/ref|><|det|>[[92, 805, 881, 932]]<|/det|>
+148 where \(|G(\omega)|\) is the MW filter function which is usually used to describe the transfer function 149 of such a sensor from magnetic field to quantum phase. In comparison to the conventional 150 fluorescence readout, the sensitivity of QPSD readout deteriorates by a factor of \(\sqrt{2}\) . Details 151 about the sensitivity derivation can be seen in the Supplementary Materials (see Supple- 152 mentary Note 3).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 80, 883, 585]]<|/det|>
+In Fig. 2a and b, we compare the regular interferometry (single driving field) and with the measurements obtained from the QPSD readout described above. The strength of the applied external ac fields ranges from 0 to 3 \(\mu \mathrm{T}\) . For Ramsey measurements, the applied fields are at a frequency of 46 Hz, and we use a field sensing time \(T_{\phi ,R a m s e y} = 6.25\mu \mathrm{s}\) . For Hahn- echo measurements, we use external fields at 80 kHz+46 Hz and the field sensing time \(T_{\phi ,H a h n} = 2T_{\phi ,R a m s e y}\) . The test fields are sent to the diamond by a calibrated loop antenna. The signal readout of the regular interferometry measurements is proportional to \(\sin (\phi)\) , where \(\phi \propto \gamma_{e}B\) is the accumulated quantum phase. The response is linear only when \(\phi\) is small, limiting the dynamic range. Thus, the regular Ramsey and Hahn- echo readout quickly saturate due to this limited LDR. We plot the Fourier transform of the readout signals also in the figures. The harmonics of the 46 Hz signal rise significantly in the fluorescence readout spectral due to the saturation induced by the limited LDR, compared to the QPSD readout which shows the linearity over the entire field range. The high- order harmonics of the signal detected by the QPSD readout are small and mainly arise from the function generator. In the measurements, one could see the linewidth broadening induced by the increasing signal power. The peak at 100 Hz, which is consistently seen in both the Ramsey and Hahn- echo measurements, comes from the electronics instrumentation. Other side peaks seen near the original signal frequency in the QPSD readout spectra are due to the mixing of the 100 Hz power line harmonics and the 92 Hz signal harmonics in the LIA.  
+
+<|ref|>text<|/ref|><|det|>[[90, 588, 883, 821]]<|/det|>
+Besides the LDR, the method also demonstrates measurement robustness to changing of \(T_{seq}\) . The motivation of using different \(T_{seq}\) is to get different sampling frequencies as well as measurement bandwidth. Signal responses to different sampling frequencies, i.e., different \(1 / 2T_{seq}\) , are plotted in Fig. 2c. Characterized by the same test field, fluorescence readout shows varying signal responses over the sampling frequency range, while the QPSD readout almost stays at the same level because the measured phase factor only changes with the external field and the sensing time \(T_{\phi}\) . It also indicates that the QPSD readout does not change with varying of fluorescence signal contrast, which is affected by the low spin polarization rate when \(T_{seq}\) is short in the regime of low excitation laser power.  
+
+<|ref|>text<|/ref|><|det|>[[90, 826, 882, 925]]<|/det|>
+The measurement bandwidth of the QPSD readout is shown in Fig. 2d, where the signal responses to different test field frequencies are plotted. The plotted values are the magnitudes at the corresponding frequencies in the Fourier transform of the QPSD readout. For the measurements based on the Hahn- echo sequence, we detected the heterodyne signal
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 85, 881, 319]]<|/det|>
+185 for the ac fields. The applied sequence length, \(T_{seq} = 100\mu \mathrm{s}\) , gives the referencing frequency \(f_{s} = 5\mathrm{kHz}\) for Demod. 1. We apply the second driving field with the frequency offset at \(\delta f = 500\mathrm{Hz}\) to have \(N = 10\) samples in a modulation cycle. Due to this, there will be flexibility in deciding the single measurement bandwidth by setting the time constant of Demod. 2. We choose different settings corresponding to the cut- off frequency of the filter at \(100\mathrm{Hz}\) and \(200\mathrm{Hz}\) . Finally, one can conclude that the rotating frame modulation provides QPSD readout magnetometry that has enhanced LDR and robustness in a flexible bandwidth. As we show below, this can be used for measurement of arbitrary fields with low distortion.  
+
+<|ref|>sub_title<|/ref|><|det|>[[137, 366, 529, 385]]<|/det|>
+## B. Frequency Offset Heterodyne Readout  
+
+<|ref|>text<|/ref|><|det|>[[90, 409, 881, 666]]<|/det|>
+Heterodyne readout has been used to improve the frequency resolution remarkably in nuclear magnetic resonance spectroscopy. It is also a way to achieve high precision microwave sensing [34- 36]. High- order dynamical decoupling sequences are used to narrow the spectral linewidth by decoupling the sensor response from unwanted signal frequencies [20, 21]. Here arises a trade- off between the measurable signal bandwidth and fidelity. High- order dynamical decoupling can ensure a high sensitivity but only allows to measure signals within the narrow bandwidth defined by the sequence. On the other hand, the lower limit on the detectable signal frequency is set by the decoherence time of the sensor. Here, we will use the Hahn- echo sequence in combination with the QPSD readout to measure signals at frequencies that go beyond the coherence time of the sensor.  
+
+<|ref|>text<|/ref|><|det|>[[90, 672, 881, 880]]<|/det|>
+In Qdyne, the sampling time usually satisfies \(T_{seq} \neq mT_{\phi}\) so as to get the heterodyne signal [21]. The frequency of this heterodyne signal depends on the timing offset. Here, we choose the measurement sampling time \(T_{seq} = mT_{\phi}\) to obtain the heterodyne readout depending on the signal frequency offset from \(1 / T_{\phi}\) . As a result, the detected phase of signals at frequencies of \(n / T_{\phi}\) is locked by the sequence, where \(n\) can be a random integer. On the other hand, the frequency offset of signals can also introduce phase revivals, i.e. frequency offset heterodyne signal, as shown in Fig. 3a. The detected heterodyne frequency would be the exact offset of the signal frequency to \(1 / T_{\phi}\) .  
+
+<|ref|>text<|/ref|><|det|>[[91, 885, 880, 932]]<|/det|>
+The frequency offset heterodyne readout is modeled based on the MW sequence filter [37, 38]. Sampling happens in each time interval of \([NmT_{\phi}, (Nm + 1)T_{\phi}]\) , where \(N \in \mathbb{Z}\) . For
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[128, 88, 863, 465]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 483, 881, 757]]<|/det|>
+<center>FIG. 3. Frequency Offset Heterodyne readout. (a) Hahn-echo sequence is used for this demonstration. The detected phase of the ac signal is locked by the sequence when the frequency is \(f_{ac} = f_{\phi}\) . Otherwise, a heterodyne signal of \(|f_{ac} - f_{\phi}|\) can be measured. The colored regions mark where the quantum phase is accumulated, while phase accumulations at the other areas are canceled in the spin evolution. The figure shows identical heterodyne signals due to \(f_{\phi} - f_{1} = f_{2} - f_{\phi}\) . (b) We apply ac fields at different frequencies with an offset of \(5 \mathrm{~Hz}\) to the sensor so that \(5 \mathrm{~Hz}\) peaks can be detected as the signal response. The signal frequency response of the Hahn-echo sequence and CPMG-2 sequence are plotted after normalization, respectively. In both measurements \(T_{\phi} = 50 \mu \mathrm{s}\) , and the sampling frequency is \(2 \mathrm{kHz}\) according to the applied sequence length. The red lines are the filter functions in theory. (c) Signal frequency response of Hahn-echo measurements with \(1 / T_{seq} = 10 \mathrm{kHz}\) . The dash line indicates the filter introduced by the lock-in amplifier. (d) Sequence dependency of the frequency resolution. In means. 1, a \(20.005 \mathrm{kHz}\) filed is applied and measured by sequences with \(T_{\phi} = 50 \mu \mathrm{s} \pm 4 \mathrm{ns}\) , and \(T_{seq} = 20 T_{\phi}\) . In means. 2, we keep \(T_{\phi}\) unchanged, and offset \(T_{seq}\) with \(\pm 4 \mathrm{ns}\) . In means. 3, the frequency of the applied field is changed to \(16.005 \mathrm{kHz}\) while the other parameters are the same as means. 2. </center>  
+
+<|ref|>text<|/ref|><|det|>[[92, 792, 880, 864]]<|/det|>
+215 a random ac signal component \(B_{ac}(t) = B(\omega)e^{- i[\omega t + \phi (\omega)]}\) and a measurement with the MW 216 \(\pi\) - pulse number of \(n\) , we can get the accumulated quantum phase as (see Supplementary 217 Note 2)  
+
+<|ref|>equation<|/ref|><|det|>[[275, 865, 877, 896]]<|/det|>
+\[\phi_{r}(N) = |G_{n}(\omega)|e^{i\left(-\frac{\omega T_{\phi}}{2} -\frac{P}{2}\pi\right)}\gamma_{e}B(\omega)e^{-i\phi (\omega)}e^{-i\omega NmT_{\phi}}, \quad (7)\]  
+
+<|ref|>text<|/ref|><|det|>[[92, 907, 880, 933]]<|/det|>
+218 where \(N\) denotes the sampling timestamp, \(G_{n}(\omega) = |G_{n}(\omega)|e^{i\left(-\frac{\omega T_{\phi}}{2} -\frac{P}{2}\pi\right)}\) is the MW filter
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 85, 882, 371]]<|/det|>
+219 function, \(P = 1\) when the \(\pi\) - pulse number \(n\) is odd and \(P = 2\) when \(n\) is even. Particularly, when \(n = 1\) i.e. Hahn- echo sequence is applied, the filter function satisfies \(|G_{1}(2\pi /T_{\phi})| =\) \(|G_{1}(\pi /T_{\phi})|\) . In principle, measurements of signals at a wide frequency range is feasible by choosing the appropriate \(T_{\phi}\) in Hahn- echo measurements. For example, by using \(T_{\phi}< 1\mu \mathrm{s}\) , one can achieve detection of signals at frequencies higher than 1 MHz. It is more challenging to measure a signal at a lower frequency, such as a signal at 10 kHz, for the reason that a longer \(T_{2}\) is required. With the property described above, it is feasible to use \(T_{\phi} = 50\mu \mathrm{s}\) rather than \(T_{\phi} = 100\mu \mathrm{s}\) to achieve the measurement with a better sensitivity due to the higher signal contrast when \(T_{\phi}\) is smaller. For diamonds which have NV center ensembles with \(T_{2}< 100\mu \mathrm{s}\) , the property makes it feasible to measure signals at the frequencies lower than \(1 / T_{2}\) beyond the coherence limit.  
+
+<|ref|>text<|/ref|><|det|>[[90, 377, 881, 479]]<|/det|>
+Given a reference frequency \(\omega_{r e f}~ = ~k\omega_{s},k~\in \mathbb{N}\) , where \(\omega_{s}~ = ~2\pi /(m T_{\phi})\) and \(\omega \in\) \((\omega_{r e f} - \omega_{s} / 2,\omega_{r e f} + \omega_{s} / 2)\) , the evolving phase factor can be rewritten as \(e^{- i\omega N m T_{\phi}}~ =\) \(e^{- i\omega H t}\delta (t - N T_{s})\) , where \(\omega_{H}~ = ~\omega - \omega_{r e f}\) is the heterodyne frequency, \(\delta (t)\) is the Dirac function, and \(T_{s} = m T_{\phi}\) is the sampling period. Thus, the readout signal turns to be  
+
+<|ref|>equation<|/ref|><|det|>[[331, 500, 877, 546]]<|/det|>
+\[\phi_{r}(t) = G(\omega)\sum_{N = -\infty}^{\infty}\gamma_{e}B_{H}(t)\delta (t - N T_{s}), \quad (8)\]  
+
+<|ref|>text<|/ref|><|det|>[[90, 565, 882, 745]]<|/det|>
+where \(B_{H}(t) = B(\omega)e^{- i(\omega_{H}t + \phi)}\) contains all the information from the origin signal to be measured. As discussed in previous section, the quantum phase readout bandwidth is limited by the cut- off frequency \(f_{c}\) of the filter of LIA. For any signal with a frequency range of \([(k - 1)f_{s} + f_{c},(k + 1)f_{s} - f_{c}]\) , aliasing can be filtered. Although a smaller \(f_{c}\) makes the measurement bandwidth narrower, it ensures signals that in a larger frequency range can be detected without aliasing. By changing \(T_{\phi}\) together with \(T_{seq}\) , we can resolve a spectrum in multiple frequency bands with a series of sequences.  
+
+<|ref|>text<|/ref|><|det|>[[90, 752, 882, 930]]<|/det|>
+We present two specific examples of the measured frequency responses by using the Hahn- echo and CPMG- 2 sequence. We plot both the theoretical MW filter function and the experimentally measured signal responses together in Fig. 3b. The field sensing time for both experiment and theory calculations is set to be \(T_{\phi} = 50\mu \mathrm{s}\) . In the experiments, we measured the amplitudes of the frequency offset heterodyne signals with \(T_{seq} = 250\mu \mathrm{s}\) , i.e., the magnetic field sampling rate is 4 kHz. Due to this reason, the measured MW filters are combed with a frequency distance of 4 kHz. Aliasing signals exist between the main lobes
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 87, 770, 108]]<|/det|>
+at a distance of 2 kHz, because the readout sampling frequency is \(f_{s} = 2 \mathrm{kHz}\) .  
+
+<|ref|>text<|/ref|><|det|>[[90, 113, 882, 293]]<|/det|>
+In order to measure signals that distribute in larger bandwidth, we can increase the sampling frequency, for example, to \(f_{s} = 5 \mathrm{kHz}\) . The spectrum is plotted in Fig. 3c in decibel, from which one can see that magnitudes are the same at 10 kHz and 20 kHz, i.e., \(1 / (2T_{\phi})\) and \(1 / T_{\phi}\) as discussed in the derivation. The insets of Fig. 3c depict the signals that the quantum sensor detects during \(T_{\phi}\) at the two frequencies. In this measurement, the bandwidth limited by the filter of the LIA is at 200 Hz, i.e., the single measurement bandwidth is 400 Hz, and the detectable signal frequency range is 9600 Hz.  
+
+<|ref|>text<|/ref|><|det|>[[90, 299, 882, 584]]<|/det|>
+We notice that a single measurement cannot tell if the ac field frequency offset is positive or negative from the heterodyne readout. Additional measurement is needed to distinguish the direction of the frequency offset. By adding a difference to the phase accumulation time \(T_{\phi}\) as well as the sequence time, we can change the reference frequency \(\omega_{ref}\) to get a different heterodyne frequency. By seeing if the heterodyne frequency increases or decreases, we can determine if the signal frequency is larger or smaller than the reference frequency. As the measurements presented in Fig. 3d that \(T_{\phi} = 50 \mu \mathrm{s}\) is offset by a difference of 4 ns and \(T_{seq} = 10 T_{\phi}\) changes accordingly, the detected heterodyne frequency of the signal shift in two different directions. We further investigated the dependency of the heterodyne frequency on the parameters by performing measurements that vary (i) \(T_{seq}\) , (ii) \(T_{seq}\) and \(\omega_{ref}\) . When \(T_{\phi}\) keeps unchanged, the heterodyne frequency shifts by  
+
+<|ref|>equation<|/ref|><|det|>[[397, 612, 877, 632]]<|/det|>
+\[\Delta \omega_{H} = \omega_{ref}\Delta T_{seq} / T_{seq}. \quad (9)\]  
+
+<|ref|>text<|/ref|><|det|>[[90, 660, 880, 760]]<|/det|>
+Using the equation, we can estimate the frequency fidelity of the given sequence. For example, with a timing error \(< 3 \mathrm{ps}\) , the frequency fidelity of a signal around 10 kHz could be only 0.06 mHz. The frequency resolution can be arbitrarily high with a long \(T_{seq}\) at the cost of bandwidth.  
+
+<|ref|>sub_title<|/ref|><|det|>[[89, 815, 498, 835]]<|/det|>
+## C. Sensing of Arbitrary Audio Signals  
+
+<|ref|>text<|/ref|><|det|>[[90, 858, 880, 932]]<|/det|>
+We demonstrate measurements of arbitrary audio signals by combining the QPSD readout and the frequency offset heterodyne detection. We first generate a signal at 20.08 kHz with its phase varying with time (see Fig. 4a). The MW filter is set by the Hahn- echo sequence
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 87, 881, 186]]<|/det|>
+275 with \(T_{\phi} = 50\mu \mathrm{s}\) . With the reference frequency is at \(20\mathrm{kHz}\) , the heterodyne readout is at \(80\mathrm{Hz}\) , as seen from the simulated curve. The phase of the external field is switched with a cycle of \(80\mathrm{Hz}\) and \(40\mathrm{Hz}\) so that the experimental readout displays the phase change, as shown in Fig. 4a.  
+
+<|ref|>text<|/ref|><|det|>[[90, 192, 880, 240]]<|/det|>
+279 Next, we apply a field with its frequency, amplitude and phase all arbitrarily changing. 280 The signal frequency is around \(10\mathrm{kHz}\) and the signal bandwidth is within \(400\mathrm{Hz}\) . Using  
+
+<|ref|>image<|/ref|><|det|>[[123, 293, 863, 670]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[113, 690, 881, 928]]<|/det|>
+<center>FIG. 4. Detection of arbitrary audio signals. (a) Phase response of the QPSD measurement. A 20.08 kHz signal with sequential phase changing is applied to the sensor. The bars show the phases at different time. Stars mark the readout of the sensor, and the curve is the simulated readout. (b) An ac field is applied with the frequency, amplitude and phase switched every \(100\mathrm{ms}\) . The light blue areas corresponding to the right \(y\) -axis shows the applied field of around \(10\mathrm{kHz}\) , and the red curve shows the QPSD readout. (c) Spectral comparison of the applied signal and the detected magnetic field in a narrow bandwidth. The applied signal is a sum of 20 different sine signals within \(400\mathrm{Hz}\) bandwidth. (d) A signal with wide bandwidth between 10 to \(15\mathrm{kHz}\) is applied and detected by varying the sequence. We set an \(800\mathrm{Hz}\) bandwidth for the measurement of each sequence and use 6 measurements to cover the entire bandwidth. The red dash line shows the spectrum of the output of the AWG, and the solid black line is the spectrum of the detected magnetic field signal. The inset figure is the phase-noise power spectrum density plotted within 1 Hz bandwidth cut. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 87, 881, 212]]<|/det|>
+\(1 / T_{seq} = 10\mathrm{kHz}\) , we can measure the signals close to \(10\mathrm{kHz}\) with the same sensitivity as the \(20\mathrm{kHz}\) signal. The signal length is one second and consists of ten \(100\mathrm{ms}\) parts. In Fig. 4b, both the applied field waveform and the QPSD readout are plotted. The heterodyne frequencies well resolve the frequency differences in the original waveform. The amplitudes of the readout also correspond to the applied field strength.  
+
+<|ref|>text<|/ref|><|det|>[[90, 217, 881, 584]]<|/det|>
+As discussed previously, the measurement bandwidth used in the experiment is \(400\mathrm{Hz}\) . For this, we perform a spectrum analysis as shown in Fig. 4c. The signal to be measured is a sum of 20 tones with random frequencies, amplitudes and phases. In order to distinguish the sign of frequency offsets for each component, we measure the signal using an alternative sequence with \(T_{\phi} = 50\mu \mathrm{s} + 2\mathrm{ns}\) . The sharp peaks observed in the spectrum should shift according to the changes of the measurement sequence, else we exclude them as noise signals generated from our electronics. As shown in Fig. 4c, the applied frequencies are properly resolved. Additionally, we find a \(9.93\mathrm{kHz}\) noise spike from the environment. The errors in magnitude of the measured signal could be induced by the LIA filter, as shown earlier in Fig. 2d. The errors could also be caused by an insufficient sampling number for demodulating the rotating frame modulation. In the measurements, we apply sequences with their lengths corresponding to a sampling frequency \(f_{s} = 5\mathrm{kHz}\) . The frequency difference of the two MWs is \(\delta f = 500\mathrm{Hz}\) and \(N = 10\) for reading out a phase sample. To increase the measurement precision, if we use a smaller \(\delta f\) , it will decrease the bandwidth.  
+
+<|ref|>text<|/ref|><|det|>[[90, 588, 881, 926]]<|/det|>
+Though the bandwidth of each measurement sequence is limited, we can still measure a signal with a wider bandwidth by merging several measurements. The condition is that the signal bandwidth should not be larger than the sampling frequency to avoid frequency aliasing. In Fig. 4d, we perform a spectrum analysis for a signal within a bandwidth from \(10\mathrm{kHz}\) to \(15\mathrm{kHz}\) . The signal to be detected is a sum of 10 components with their frequencies randomly distributed in the bandwidth. The signal is generated by an arbitrary signal generator (AWG) and sent to the test field coil. The dotted curve in the figure displays the spectrum of the electrical signal from the AWG. There are some harmonics near each main component due to the limited AWG internal clock and signal length. The components at different frequencies are measured by varying \(T_{\phi}\) to get different referencing frequencies for heterodyne detection. The inset figure shows the power spectrum of the QPSD readout noise within \(1\mathrm{Hz}\) bandwidth, from which we calculate the square root of the standard deviation \(\sigma_{phase} = 0.0022^{\circ}\) . The magnetic field sensitivity depends on the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 85, 882, 240]]<|/det|>
+applied sequence and the corresponding frequency response. In the case of the Hahn- echo sequence with \(T_{\phi} = 50\mu \mathrm{s}\) , we have a calibrated scalar factor of \(k_{sf} = 0.071^{\circ} / \mathrm{nT}\) , and the magnetic field sensitivity is \(\eta = 31\mathrm{pT} / \sqrt{\mathrm{Hz}}\) . Taking the scalar factor into the calculation of LDR \([-\pi ,\pi ]\) , we can get the dynamic range in decibels as \(20\log [\pi /(k_{sf}\eta)] = 98\mathrm{dB}\) . The sensitivity can be further optimized by, e.g., using higher laser power, applying high- order dynamical decoupling sequences, and implementing flux concentration.  
+
+<|ref|>text<|/ref|><|det|>[[90, 245, 882, 531]]<|/det|>
+Finally, we demonstrate the detection and demodulation of audio signals. Although 10 kHz is within the audio frequency band, most of the daily audio sounds have frequencies ranging from hundreds of Hz to a few kHz. Therefore, signals need to be modulated to a detectable frequency range. For this, we have used (i) a melody piece composed of 3 tones and (ii) a small part from Dr. Martin Luther King Jr.'s famous speech "I have a dream", to test the waveform reconstruction by the diamond quantum sensor. The tones of the melody have frequencies distributed between 500 Hz and 700 Hz. Therefore, we mix it with a 9.5 kHz reference to get the signal modulated around 10 kHz and broadcast the mixed signal to the diamond. For case (ii), we have to compress the signal bandwidth into 200 Hz and then modulate it with a 10 kHz reference. The audio reconstructed from the diamond sensor can be heard and compared with the original audio (see Supplementary Audio S1- S4).  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 577, 306, 595]]<|/det|>
+## III. DISCUSSIONS  
+
+<|ref|>text<|/ref|><|det|>[[90, 620, 882, 774]]<|/det|>
+In this work, we overcome the LDR limitation of the conventional interferometric readout through a new technique that includes the QPSD scheme and the frequency offset heterodyne readout. The technique allows one to measure unknown signals with maximal sensitivity independent of their dynamic range. It improves the feasibility for quantum sensors to perform high- sensitive measurements of different physical quantities using interferometric methods, beyond magnetometry.  
+
+<|ref|>text<|/ref|><|det|>[[90, 780, 882, 932]]<|/det|>
+Theoretically, the extended LDR comes from the multiple measurements that have the quantum phase evolving through the entire phase range \([-\pi ,\pi ]\) so that the initial phase factor that contains the external field information can be resolved. Such an extended phase range affects the measurement bandwidth as well as the sensitivity. In theory, the sensitivity does not deteriorate a lot from the conventional fluorescence readout except for a factor of \(\sqrt{2}\) . While in the experiment, we suffer from a low contrast \(C = 0.19\%\) due to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 85, 881, 291]]<|/det|>
+343 the low excitation laser power (80 mW). The contrast and the fluorescence photon count 344 can significantly increase when the laser reaches saturation power [39]. Different dynamical 345 decoupling sequences can also improve the magnetic field sensitivity through the filter func- 346 tion \(G(\omega)\) . Flux concentration could further improve the signal- noise ratio [31, 40]. The flux 347 concentrator can be very small compared to conventional dipole antennas because the gain 348 no longer depends on the signal wavelength. With the millimeter size diamond dimension, 349 the flux concentration factor can easily reach a factor of hundreds when using a concentrator 350 in centimeters.  
+
+<|ref|>text<|/ref|><|det|>[[90, 299, 881, 450]]<|/det|>
+351 The QPSD readout can also enhance the capability of vector magnetometry. Convention- 352 ally, fluorescence emitted from NV centers in multiple orientations is measured sequentially 353 to acquire the vector components. Similar to the methods developed here, one could also 354 modulate the signal on each orientation with different modulation frequencies [41]. Per- 355 forming measurements on different NV orientations with appropriate synchronization can 356 suppress the phase errors in vector reconstruction.  
+
+<|ref|>text<|/ref|><|det|>[[90, 457, 881, 635]]<|/det|>
+357 In conclusion, we demonstrated high- sensitive distortion- free quantum- assisted detection 358 of audio signals, including melody and speech, using the QPSD scheme in combination with 359 the heterodyne readout. A further improvement in sensitivity can be achieved by using 360 flux concentrators. One could also generalize the current methods to achieve vector magne- 361 tometry with extended LDR. We envisage that the techniques developed here will have the 362 potential to develop low- distortion, small- volume quantum sensors for various applications 363 in science and technology.  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 690, 273, 707]]<|/det|>
+## IV. METHODS  
+
+<|ref|>sub_title<|/ref|><|det|>[[92, 736, 360, 754]]<|/det|>
+## A. Experimental setup  
+
+<|ref|>text<|/ref|><|det|>[[90, 778, 881, 931]]<|/det|>
+366 The diamond used in the experiment is a (111)- oriented (0.5 mm) cube obtained from a 367 single crystal grown by the temperature gradient method at high- pressure high- temperature 368 (HPHT) conditions. The diamond is 99.97% \(^{12}\mathrm{C}\) enriched, and has an initial nitrogen concentration of 1.4 ppm. The final NV concentration is 0.4 ppm after electron irradiation 370 and annealing. Dephasing time of the NV ensemble is obtained as \(T_{2}^{*} = 8.5\mu \mathrm{s}\) by Ramsey 371 sequence, and a decoherence time \(T_{2} = 200\mu \mathrm{s}\) is measured by Hahn- echo sequence. The dia
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 86, 882, 371]]<|/det|>
+372 mond is positioned at center of a home- built three dimension coils system, and is illuminated 373 by a 532 nm laser (Lighthous Sprout- G) at around 80 mW. Microwave signals are generated 374 from two sources (Rohde&Schwarz, SMIQ03B) and are individually cut by two switches 375 before the combination. Measurement sequences are generated by a data timing generator 376 (Tekreoxin, DTG5274). After the combination and amplification of the MW signals, MW 377 pulses are feed to the diamond through a dielectric resonator antenna [42]. The detected 378 fluorescence signal is demodulated by a LIA (Zurich Instruments, HF2LI) which has two 379 independent differential input channels and demodulators. To generate arbitrary magnetic 380 fields, we write signals to an AWG (Tektronix, AWG520) with \(10^{5}\) samples per second out- 381 put sampling rate. The test signals are continuously repeated and sent to a copper loop 382 near the diamond.  
+
+<|ref|>sub_title<|/ref|><|det|>[[90, 525, 361, 544]]<|/det|>
+## B. Spectrum analyzing  
+
+<|ref|>text<|/ref|><|det|>[[90, 567, 883, 931]]<|/det|>
+384 The spectrum to be analyzed is divided into several sections with the bandwidth set 385 by the LIA for data acquisition. In each section, the center frequency determines \(T_{\phi}\) of 386 the measurement sequence. Usually, the center frequency satisfies \(f_{c} = 1 / T_{\phi} + \epsilon /T_{seq}\) 387 where \(\epsilon = 0,\pm 1\) . A time trace is recorded after running the sequence, and a spectrum is 388 acquired from the Fourier transform of the time trace. However, the spectrum is a fold of 389 the two sidebands with repect to the center frequency. The sequence with \(T_{\phi}^{\prime} = T_{\phi} + t_{clk}\) 390 and \(T_{seq}^{\prime} = mT_{\phi}^{\prime}\) is applied to get an alternate spectrum with analyzed frequencies shift by 391 \(\Delta f = \pm \left|1 / T_{\phi} - 1 / T_{\phi}^{\prime}\right|\) . The direction of the frequency shift shows which side the signal 392 component belongs to. In the algorithm, we set a threshold to separate signal spikes from 393 noise, and use the known sequences induced spectrum frequency shift to distinguish the signs 394 of the signal offset frequency to the center frequency. The signal spikes that do not shift 395 according \(\Delta f\) are recognized as systematic noise spikes. Then, the spectrum of the selected 396 section can be replotted as the example shown in Fig. 4c. After measuring the spectra of 397 all the sections, we can get the final spectrum by merging them together.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[91, 130, 882, 258]]<|/det|>
+We acknowledge financial support by European Union's Horizon 2020 research and innovation program ASTERIQS under grant No. 820394, European Research Council advanced grant No. 742610, SMeI, Federal Ministry of Education and Research (BMBF) project MiLiQuant and Quamapolis, German Research Foundation grant GRK 2198 and 2642, and Japan Society for the Promotion of Science (JSPS) KAKENHI No. 17H02751.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 78, 884, 905]]<|/det|>
+[12] Chen, E. H. et al. High- sensitivity spin- based electrometry with an ensemble of nitrogen- vacancy centers in diamond. Physical Review A 95, 053417 (2017).[13] Tisler, J. et al. Single defect center scanning near- field optical microscopy on graphene. Nano Lett. 13, 3152–3156 (2013).[14] Zheng, H. J. et al. Zero- field magnetometry based on nitrogen- vacancy ensembles in diamond. Phys. Rev. Appl. 11 (2019).[15] Wickenbrock, A. et al. Microwave- free magnetometry with nitrogen- vacancy centers in diamond. Appl. Phys. Lett. 109 (2016).[16] Barry, J. F. et al. Optical magnetic detection of single- neuron action potentials using quantum defects in diamond. Proc. Nat. Acad. Sci. 201601513 (2016).[17] Wang, P. et al. High- resolution vector microwave magnetometry based on solid- state spins in diamond. Nat. Commun. 6, 6631 (2015).[18] Farfurnik, D. et al. Optimizing a dynamical decoupling protocol for solid- state electronic spin ensembles in diamond. Phys. Rev. B 92 (2015).[19] de Lange, G., Riste, D., Dobrovitski, V. V. & Hanson, R. Single- spin magnetometry with multipulse sensing sequences. Phys. Rev. Lett. 106 (2011).[20] Schmitt, S. et al. Submillihertz magnetic spectroscopy performed with a nanoscale quantum sensor. Science 356, 832–836 (2017).[21] Boss, J. M., Cujia, K. S., Zopes, J. & Degen, C. L. Quantum sensing with arbitrary frequency resolution. Science 356, 837–840 (2017).[22] Clevenson, H. et al. Robust high- dynamic- range vector magnetometry with nitrogen- vacancy centers in diamond. Appl. Phys. Lett. 112.[23] Nusran, N. M., Momeen, M. U. & Dutt, M. V. G. High- dynamic- range magnetometry with a single electronic spin in diamond. Nat. Nano. 7, 109–113 (2012).[24] Joas, T. et al. Online adaptive quantum characterization of a nuclear spin. Npj Quantum Inf. 7, 1–8 (2021).[25] Gurses, B. V., Whitmore, K. T. & Cohen, M. B. Ultra- sensitive broadband ”awesome” electric field receiver for nanovolt low- frequency signals. Rev. Sci. Instrum. 92 (2021).[26] Cohen, M. B. et al. Broadband longwave radio remote sensing instrumentation. Rev. Sci. Instrum. 89 (2018).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[90, 90, 883, 135]]<|/det|>
+[27] Sedlacek, J. A. et al. Microwave electrometry with rydberg atoms in a vapour cell using bright atomic resonances. Nat. Phys. 8, 819- 824 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[90, 143, 881, 189]]<|/det|>
+[28] Meyer, D. H., Kunz, P. D. & Cox, K. C. Waveguide- coupled rydberg spectrum analyzer from 0 to 20 ghz. Phys. Rev. Appl. 15 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[90, 198, 881, 243]]<|/det|>
+[29] Boto, E. et al. Moving magnetoencephalography towards real- world applications with a wearable system. Nature 555, 657 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[90, 251, 881, 297]]<|/det|>
+[30] Gerginov, V., da Silva, F. C. S. & Howe, D. Prospects for magnetic field communications and location using quantum sensors. Rev. Sci. Instrum. 88 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[90, 306, 881, 351]]<|/det|>
+[31] Fescenko, I. et al. Diamond magnetometer enhanced by ferrite flux concentrators. Phys. Rev. Res. 2 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[90, 360, 853, 380]]<|/det|>
+[32] Seidel, D. & Muga, J. G. Two- frequency ramsey interferometry. Phys. Rev. A 75 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[90, 388, 881, 434]]<|/det|>
+[33] Vutha, A. C. & Hessels, E. A. Frequency- offset separated oscillatory fields. Phys. Rev. A 92 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[90, 443, 880, 488]]<|/det|>
+[34] Meinel, J. et al. Heterodyne sensing of microwaves with a quantum sensor. Nat. Commun. 12, 2737 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[90, 497, 881, 543]]<|/det|>
+[35] Chu, Y. M. et al. Precise spectroscopy of high- frequency oscillating fields with a single- qubit sensor. Phys. Rev. Appl. 15 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[90, 552, 881, 598]]<|/det|>
+[36] Staudenmaier, N., Schmitt, S., McGuinness, L. P. & Jelezko, F. Phase- sensitive quantum spectroscopy with high- frequency resolution. Phys. Rev. A 104 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[90, 606, 881, 652]]<|/det|>
+[37] Staudacher, T. et al. Nuclear magnetic resonance spectroscopy on a (5- nanometer)(3) sample volume. Science 339, 561- 563 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[90, 661, 881, 707]]<|/det|>
+[38] Bar- Gill, N., Pham, L. M., Jarmola, A., Budker, D. & Walsworth, R. L. Solid- state electronic spin coherence time approaching one second. Nat. Commun. 4, 1743 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[90, 715, 880, 761]]<|/det|>
+[39] Zhang, C. et al. Diamond magnetometry and gradiometry towards subpicotesla dc field measurement. Phys. Rev. Appl. 15 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[90, 770, 881, 815]]<|/det|>
+[40] Griffith, W. C., Jimenez- Martinez, R., Shah, V., Knappe, S. & Kitching, J. Miniature atomic magnetometer integrated with flux concentrators. Appl. Phys. Lett. 94, 023502 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[90, 824, 881, 870]]<|/det|>
+[41] Schloss, J. M., Barry, J. F., Turner, M. J. & Walsworth, R. L. Simultaneous broadband vector magnetometry using solid- state spins. Phys. Rev. Appl. 10 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[90, 879, 881, 925]]<|/det|>
+[42] Kapitanova, P. et al. 3d uniform manipulation of nv centers in diamond using a dielectric resonator antenna. Jetp Lett. 108, 588- 595 (2018).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 43, 312, 71]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 131, 366, 257]]<|/det|>
+SupplementaryInformation.docx AudioS1melody.wav AudioS2detmelody.wav AudioS3voice.wav AudioS4detvoice.wav
+
+<--- Page Split --->

preprint/preprint__b3f261e76766734f08b87ae1a92ea1300136deecc3c800639ccde5acbd7fbac9/images_list.json

@@ -0,0 +1,70 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        85,
+        880,
+        808
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        80,
+        880,
+        850
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        118,
+        80,
+        880,
+        732
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        80,
+        884,
+        855
+      ]
+    ],
+    "page_idx": 27
+  }
+]

preprint/preprint__b3f261e76766734f08b87ae1a92ea1300136deecc3c800639ccde5acbd7fbac9/preprint__b3f261e76766734f08b87ae1a92ea1300136deecc3c800639ccde5acbd7fbac9.mmd

@@ -0,0 +1,263 @@
+
+# Single cell analysis of early metastasis identifies targetable tumor subpopulation and mechanisms of immune evasion in squamous cell cancers  
+
+Hong Sheng Quah National Cancer Centre Singapore https://orcid.org/0000- 0001- 8430- 8528  
+
+Elaine Yiqun Cao Duke- NUS Medical School  
+
+Lida Suteja National Cancer Centre Singapore  
+
+Hui Leong National Cancer Centre Singapore  
+
+Fui Chong National Cancer Centre Singapore  
+
+Constance Li National Cancer Centre Singapore  
+
+Shilpi Gupta Singapore Immunology Network  
+
+Camille Arcinas National Cancer Centre https://orcid.org/0000- 0001- 5374- 9232  
+
+John Ouyang Duke- NUS Medical School https://orcid.org/0000- 0002- 1239- 1577  
+
+Vivian Ang Singapore Immunology Network  
+
+Daniel Tan Division of Medical Oncology, National Cancer Centre Singapore https://orcid.org/0000- 0002- 6514- 6786  
+
+Subhra BISWAS Human Innate Immunity Lab  
+
+Owen Rackham University of Bristol https://orcid.org/0000- 0002- 4390- 0872  
+
+N. Gopalakrishna lyer ( \(\square\) gopaliyer@singhealth.com.sg) National Cancer Centre Singapore https://orcid.org/0000- 0002- 8812- 6219
+
+<--- Page Split --->
+
+Keywords: pre-metastatic, single-cell genomics, targeted therapy, t-cell receptor, cytotoxic T-lymphocytes, EMT  
+
+Posted Date: October 22nd, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 960593/v1  
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+<--- Page Split --->
+
+1 Single cell analysis of early metastasis identifies targetable tumor subpopulation and mechanisms of immune evasion in squamous cell cancers  
+
+3 Authors: Hong Sheng Quah \(^{1,2*}\) , Elaine Yiqun Cao \(^{3*}\) , Lida Suteja \(^{1}\) , Hui Sun Leong \(^{1}\) , Fui Teen Chong \(^{1}\) , Constance H Li \(^{1}\) , Shilpi Gupta \(^{4}\) , Camille Arcinas \(^{1,2}\) , John F Ouyang \(^{3}\) , Vivian Ang \(^{4}\) , Daniel SW Tan \(^{1,2,5}\) , Subhra K Biswas \(^{4}\) , Owen JL Rackham \(^{3}\) and N Gopalakrishna Iyer \(^{1,2,6,7\#}\)  
+
+## 6 Affiliation:  
+
+7 1 Cancer Therapeutics Research Laboratory, National Cancer Centre Singapore, Singapore  8 2 Academic Clinical Program in Oncology, Duke- NUS Medical School, Singapore  9 3 Program in Cardiovascular and Metabolic Disorders, Duke- NUS Medical School, Singapore  10 4 Singapore Immunology Network, Singapore  11 5 Division of Medical Oncology, National Cancer Centre Singapore, Singapore  12 6 Department of Head and Neck Surgery, National Cancer Centre Singapore, Singapore  13 7 Division of Medical Sciences, National Cancer Centre Singapore, Singapore  14 15 1 These authors contributed equally to this study  16 1 Corresponding author  
+
+## 17 Correspondence to:  
+
+18 N Gopalakrishna Iyer, 11 Hospital Crescent, National Cancer Centre Singapore, Singapore 169610.  19 Email: gopaliyer@singhealth.com.sg, Tel: +65- 64368000, Fax: +65- 62257559  20 Keywords: pre- metastatic, single- cell genomics, targeted therapy, t- cell receptor, cytotoxic T- lymphocytes, EMT
+
+<--- Page Split --->
+
+Profiling tumors at single- cell resolution provides an opportunity to understand complexities underpinning lymph- node metastases in head and neck squamous- cell carcinoma. Single- cell RNAseq (scRNAseq) analysis of cancer- cell trajectories identified a sub- population of pre- metastatic cells, driven by actionable pathways including AXL and AURK. Blocking these two proteins blunted tumor invasion in patient- derived cultures. Furthermore, scRNAseq analyses of tumor- infiltrating CD8+ T- lymphocytes showed two distinct trajectories to T- cell dysfunction, corroborated by their clonal architecture based on single- cell T- cell receptor sequencing. By determining key modulators of these trajectories, followed by validation using external datasets and functional experiments, we uncovered a novel role for SOX4 in mediating T- cell exhaustion. Finally, interactome- analyses between pre- metastatic tumor- cells and CD8+ T- lymphocytes uncovered a putative role for the Midkine pathway in immune- modulation; this was confirmed by scRNAseq of tumors from humanized mice. Aside from specific findings, this study demonstrates the importance of tumor heterogeneity analyses in identifying key vulnerabilities during early metastasis.
+
+<--- Page Split --->
+
+## Introduction  
+
+In most solid tumors development of lymph node metastasis portends poor outcomes, pre- dating distant metastasis \(^{1 - 3}\) . In head and neck squamous cell cancers (HNSCC), these patients are treated with curative intent by surgery and radiation therapy with the prime objective of eradicating existing and future disease by depleting clones with a metastatic potential \(^{4,5}\) . Metastasis is a continuum of phenotypes ranging from pre- metastatic features (eg lympho- vascular invasion), circulating tumor cells/emboli, microscopic lymph node deposits, gross nodal involvement and adjacent soft- tissue invasion, oligo- metastasis and finally, full blown distant metastasis \(^{6}\) . Most studies focus on the terminal event, highlighting the role of definitive epithelial- mesenchymal transition (EMT); however bulk analyses in HNSCC suggests that EMT does not appear to be a pre- requisite for lymph node dissemination \(^{7 - 11}\) . Recent studies have also highlighted that EMT itself exists as a spectrum, and tumor cells exhibit a significant amount of plasticity which may account for the range of clinical manifestations observed \(^{12,13}\) . Single- cell analyses have the ability to resolve both issues: identification of rare clones with true metastatic potential and identifying pathways and vulnerabilities that can be exploited in the clinical setting to prevent further dissemination of these.  
+
+The role of the immune system during the metastatic cascade is gaining clinical relevance with current advancements in checkpoint blockade therapies \(^{14}\) . This is especially pertinent in the context of lymph node metastasis, as lymph nodes are believed to be the main organ for T- cell priming, expansion and trafficking \(^{15}\) . Understanding the mechanisms by which tumors evade immune- based killing within lymph nodes is critical to target early metastases \(^{16 - 19}\) . Again, this can be addressed by single- cell analyses by defining the immune landscape, and in- depth dissection of interactions involved during immune evasion at the primary and nodal sites.  
+
+Here, we profiled primary and early (nodal) metastatic HNSCC tumors using single- cell RNAseq (scRNAseq) and TCRseq (scTCRseq) with two major objectives: to identify metastatic tumor subpopulations and identification of targetable vulnerabilities, and to determine the evolutionary trajectory of tumor- targeting T- cells as well as dissecting pathways employed by tumors to evade immune destruction during nodal dissemination.
+
+<--- Page Split --->
+
+## Single-cell transcriptional states of primary and lymph node metastasis in HNSCC  
+
+To delineate 'whole- tumor' single- cell landscapes in primary tumors and lymph node metastases, we developed a protocol to rapidly process freshly resected tissue for single- cell RNA sequencing (scRNAseq) and establishing primary cultures (Figure 1A) \(^{20,21}\) . Tumors were harvested from fourteen treatment- naive patients with locally advanced, HPV- negative HNSCC from primary and cervical lymph nodes (Supplementary Table S1 and S2). Seven pairs were processed for scRNAseq and single- cell T- cell receptor sequencing (scTCRseq), while primary cultures were successfully established for seven.  
+
+scRNAseq data for fresh tumors describes 53,459 cells (3,553- 11,308 per patient) and 23,148 genes, with a median of 776 genes per cell (details on quality controls steps in Methods and Supplementary Figure 1A- B). Using Seurat v3.0, the data was normalized, pooled, and clustered (Figure 1B). Canonical markers were used to broadly annotate these populations into: epithelial (KRT7, KRT17), salivary (STATH), fibroblasts (COL1A2), endothelial (PECAM) and immune (PTPRC) cells (Figure 1B and Supplementary Figure 1C). Fibroblasts were further subdivided into cancer associated fibroblasts (CAFs; MMP2) and myofibroblasts (ACTA2), while immune cells were organized into T- (CD3E, NKG7), NK- (NKG7, XCL2), B- (CD79A), plasma- (IGHG1), mast- (TPSAB1), conventional (LAMP3) and plasmacytoid (LILR4) dendritic cells, as well as macrophages/monocytes (CD163). These were well- distributed across samples from all patients, apart from salivary cells, which were only observed in one patient, likely due to harvest of adjacent parotid gland tissue (HN263). However, there were differences in composition between primary and metastatic sites (Figure 1C), with higher proportions of CAFs and TAMs in the primary tumor, versus more B- cells, plasma cells and dendritic cells at the metastatic sites, typical of a lymph node. These were similar to cellular composition proportions derived from bulk data from TCGA (Supplementary Figure 1D). Inferred copy number variant analyses on the epithelial population showed that aneuploidy was evident in \(>95\%\) of cells validating that this population comprised cancer cells (Figure 1D and Supplementary Figure 1E). Copy number alterations (CNAs) were further analyzed using the CopyKat algorithm \(^{22}\) , and identified those frequently observed in HNSCC \(^{23}\) , including gains across chromosomes 7 and 8q and loss of 3p and 5q (Supplementary Figure 1F). Significant overlap of CNAs was also noted between the primary and metastatic sites in each patient (Supplementary Figure 1G).  
+
+## Tumor cells demonstrate varying degree of epithelial-mesenchymal transition during metastasis  
+
+We next focused on tumor cells (total of 6,115 cells & 17,784 genes) by extracting only the epithelial population with aneuploidy. Using Seurat 3.0, we pooled and re- analysed this subset, visualized as distinct clusters for each individual patient, with varying degree of overlap across cells from primary and nodal sites (Figure 2A and Supplementary 2A). Tumor cell data can be accessed and interrogated as an interactive web application via the following Shiny app (http://hnc.ddnetbio.com/). Tumors from patients HN242, HN257 and HN272 show
+
+<--- Page Split --->
+
+significant overlap in tumor cells derived from both sites, while patients HN251 and HN279 show distinct site- specific sub- clusters. When comparing EMT gene markers in primary vs nodal metastases populations, nodal tumor cells had higher EMT scores compared to the primary in all patients except HN257 (Figure 2B).  
+
+To identify the pre- nodal metastases subpopulation in primary tumors, we built trajectories using Monocle 2.0, and labelled the origin and direction based on the ground truth of site (ie primary tumor presumed to pre- date nodal disease), incorporating EMT- scores, and CytoTRACE (see Methods). The latter is a tool to determine degrees of differentiation, assuming de- differentiation co- occurs with the metastatic phenotype<sup>24,25</sup>. This approach was effective in identifying pre- nodal cells in patients HN242, HN251, HN272 and HN279 (Figure 2C- D and Supplementary Figure 2C, 2G and 2H). For patients HN251 and HN279, pseudo- time ordering demonstrated an ordered, progressive, step- wise transition from primary to nodal disease. Nodal tumor cells largely dominate the end of the trajectory with higher CytoTrace scores. Major pathways over- represented across pseudotime include epithelial de- differentiation, oxidative phosphorylation and EMT (Figure 2E). Even in more complex trajectories such as HN272, the same approach was used to determine the likely trajectory to lymph node metastases, and identify sub- populations of primary cells (pre- nodal cells) that are similar to and likely gave rise to the metastatic phenotype (Supplementary Figure 2C). We next applied GeneSwitches<sup>26</sup> to identify actionable genes associated with the trajectory from primary to pre- nodal cells; these identified AXL, Aurora kinase, TYMS and STAT2 at potentially critical genes in this process (Figure 2F and Supplementary Figure 2D- F). This approach was validated on an external dataset comprising scRNASeq data from 5 tumors from primary and nodal sites available for analyses (2076 cells) (Supplementary Figure 2H- P)<sup>12</sup>. In three of these (p25, p26 and p28), EMT was higher in nodal tumor cells compared to the primary, hence could be resolved using the method described to identify a pre- nodal subpopulation (Supplementary Figure 2P- R). Several actionable genes identified through GeneSwitches appear to be implicated in this dataset as well: AXL (p25, p26, P28), STAT2 (p25, p26) and AURK (p26, p28) (Supplementary Figure 2U).  
+
+In contrast, analyses of patient HN257 was more complicated as the primary tumor had higher EMT scores than nodal tumor cells, and tumor trajectories were haphazard with no directionality (Supplementary Figure 2H). Cytotrace showed a distinct de- differentiated sub- population in the primary tumor that had high EMT scores and expression of SNAI2 (Figure 2G and Supplementary Figure 2I- J). We hypothesized that this was an aggressive, rapidly evolving tumor subpopulation. Differential expression analyses identified a panel of 132 up- regulated and 45 down- regulated genes in this subpopulation involved in oxidative phosphorylation and tumor metabolism, and immune evasion, respectively (Figure 2H, Supplementary Table S3). Based on these gene sets, tumors in TCGA with the same signature (based on RNASeq data) had significantly poorer outcomes (Figure 2I and Supplementary Figure 2K). In the validation scRNASeq dataset above, two of the tumors (p5 and p20) also showed a similar trend, with specific subpopulations in the primary tumor with high EMT scores (Supplementary Figure 2S- T). Therefore, we postulate that in these tumors, distinct sub- populations in the primary tumor showed a more aggressive phenotype, that likely evolved after nodal dissemination had occurred.
+
+<--- Page Split --->
+
+## Identifying vulnerabilities to target pre-metastatic tumor cells  
+
+We then proceed to test whether targets identified in this manner presented an opportunity for therapeutic intervention. scRNAseq using the C1 platform was performed on patient- derived cultures (PDCs) from primary and nodal metastatic sites \((n = 7\) pairs). The data was processed using Seurat 3.0 and PAGODA (pathway and gene set overdispersion analysis) (Figure 3A and Supplementary Figure 3A- B). We derived scRNAseq data for a total of 1,317 cells and 55,216 genes. Similar to above, tumor- cell clusters were based on individual patients. However, PDCs demonstrated distinct separation between primary and metastatic cells, with EMT as one of the major differentiating principal component pathways (Figure 3B and Supplementary Figure 3A- B). Here, pre- nodal cells in HN137, HN159 and HN220 were identified as small primary subpopulations that clustered with metastatic cells.  
+
+Differential expression analyses for these pre- nodal populations identified AXL (in HN137) and AURKB (in HN159 and HN220) as putative actionable targets (Figure 3C and Supplementary Table S4- 6). Expression of these genes was validated using immunohistochemistry or immunofluorescence in both PDCs and respective tumor tissue, and this was recapitulated on flow cytometry for AXL (HN137) and AURK (HN159 and HN220), respectively (Figure 3D and Supplementary Figure 3C- D). In HN137, expression of protein and transcript AXL was detected in a majority of metastatic cells compared with only a small sub- population of primary cells. Similarly, for HN159 and HN220, AURKB expression was significantly lower in metastatic cells, compared to primary cells. We focused on AXL and AURKB because both have specific inhibitors: BGB324 targeting cells with high AXL expression, and barasertib (pan- AURK inhibitor) targeting cells with limiting AURKA/AURKB levels. There were no differences in clonogenicity between primary and metastatic cultures from patient HN137 treated with BGB324, nor HN159 and HN220 treated with barasertib (Supplementary Figure 3E- G). In contrast, all three metastatic lines HN137, HN159 and HN220 (treated with their respective drugs) demonstrated lower cell migration/invasion compared to untreated cultures, measured by scratch and Boyden chamber invasion assays (Figure 3E- G): AXL- inhibition significantly reduced invasive potential of both primary and metastatic cells of HN137 (Figure 3E) while AURK- inhibition significantly reduced the invasive potential of only metastatic cells of HN159 and HN220 (Figure 3F and G). As AXL is a surface membrane protein, primary cells were sorted into AXL low-, medium- and high- expressing cells. As predicted, BGB324 specifically inhibited invasion only in the AXL- high primary subpopulation compared to AXL- low cells (Figure 3H). These data indicate AXL and AURKB play major roles in invasion and provide an opportunity for specific anti- metastatic therapy.  
+
+## Evolution of CD8+ T-cells derived from analysis of primary tumor and lymph node metastasis  
+
+CD3+ T- cells form one of the major subpopulations sequenced at both primary and nodal sites. Data from 10,168 cells (covering 13,729 genes) were pooled, analyzed using Seurat, and visualized as ten distinct T- cell clusters (Figure 4A). The identity of each cluster was delineated based on differential gene expression of known T- cell markers (Figure 4B and Supplementary Figure 4A- B). Some were distinct for CD4+ cells (Tregs and Tfh) and CD8+ cells (Pre- dysfunctional, Dysfunctional, Proliferative), while others comprise both CD4+ and CD8+ lineages
+
+<--- Page Split --->
+
+(Naive- like and Transitional). Majority of naive- like cells were derived from nodal tissue while the remaining clusters appear to have equal representation from the primary and nodal metastatic sites (Figure 4B and Supplementary Figure 4C).  
+
+CD8+ T- cells (total of 3,387 cells, 11,847 genes) were extracted from this pooled T- cell dataset and re- analyzed after regression for cell cycle- driven artefacts to identify lineage- based clusters. CD8+ T- cell data can be accessed and interrogated as an interactive web application using the following Shiny app (http://hnc.ddnetbio.com/). Six distinct clusters were labelled as naive, transitional, tissue- resident memory, pre- dysfunctional, proliferative and late dysfunctional based on canonical markers (Figure 4C- D). Using Slingshot, we performed trajectory analyses on the CD8+ T- cells using the CXCL13- high, LAYN- high exhausted/senescent population as the endpoint \(^{27}\) , and this identified two convergent trajectories (Figure 4E). Expression plots across Trajectory 1 showed a progressive loss of naive markers, gradual gain of dysfunctional (and senescent) markers and an intervening proliferative 'burst', that likely reflects expanding clones of tumor targeting CD8+ cells (Figure 4F). Specifically, this lineage suggests a scenario where naive CD8+ T- cells from lymph nodes or circulation were trafficking into the primary tumor with loss of circulating markers KLF2, SELL and CCR7, gain of tissue resident marker CD103/ITGAE, progressive decline in the expression of naive genes TCF7, IL7R, CCR7, and gradual gain of dysfunctional markers (TIM3, CTLA4, TIGIT, CXCL13, LAYN) with an intermediary proliferative burst with high levels of MK167, TOP2A, TYMS (Figure 4B, 4E- F). This is also reflected by progressive increase from GZMK to GZMB, PRF1, and IFNG in pre- dysfunctional to dysfunctional cells. In contrast, the trajectory of tissue- resident memory (TRM) to dysfunctional cells (Trajectory 2) shows fewer genes being activated as the expression level of many of the tissue resident (ITGAE), dysfunctional (CTLA4) and granzymes (GZMs) genes were already upregulated (Figure 4B). The Geneswitches algorithm was applied to trajectory 1 (naive- to- dysfunction) to predict key gene expression changes across pseudotime and identify factors that could account for these (Figure 4G) \(^{26}\) . Our results indicate the major nodes appear to be an early loss of KLF2, intermediate increase in NKG7 and late increase in SOX4, DUSP4 and RBPJ (Figure 4G- H).  
+
+## Modulating genes driving tumor-targeting cells dysfunction/exhaustion  
+
+Based on the data above, expression of SOX4, DUSP4 and RBPJ appears to coincide with the transition between dysfunction and exhaustion, but whether these genes modulate the process remains untested. We attempted to validate these findings in two separate datasets. Re- analysis of data from Puram et al (scRNAseq from 542 CD8+ T cells) showed that expression levels of SOX4 and RBPJ were higher in dysfunctional CD8 cell populations, while DUSP4 expression was more generalized (Figure 5A and Supplementary Figure 5A- C) \(^{12}\) . The second scRNAseq dataset comprised T- cells obtained from cutaneous squamous- cell carcinoma patients before and after treatment with PD1- blockade (Supplementary Figure 5D) \(^{28}\) . Here, all three genes showed higher expression in the exhausted CD8 subpopulation in this dataset (Figure 5B and Supplementary Figure 5E). However, only levels of SOX4 and DUSP4 were reduced after PD1 blockade, where there is expected re- activation of tumor- targeting clones and reduction in the exhaustion phenotype (Figure 5C). Combining these
+
+<--- Page Split --->
+
+results, SOX4 appears to be the most likely gene associated during the transition between pre- dysfunction to dysfunction/exhaustion. To test whether SOX4 plays a causative role in T- cell dysfunction, we performed RNAi- based knock- down on activated PBMCs. Cells were transfected with Accell pooled siRNA against SOX4, DUSP4 or non- targeting siRNA as controls, activated with anti- CD3/CD28 microbeads and harvested for flow cytometry. Remarkably, SOX4 knockdown resulted in a reduction in senescent CD57+ and dysfunctional PD1+ and CD39+ populations, compared to DUSP4 and control siRNAs (Figure 5D and Supplementary Figure 5F- G). Taken together, these data provide functional validation for our CD8+ T- cell trajectory mapping and implicates SOX4 as an important driver of T- cell dysfunction/exhaustion.  
+
+## Establishing clonal architecture in \(\mathsf{CD8 + }\) T-cells using single-cell T-cell receptor sequencing  
+
+Clonal identifiers obtained by TCR analysis allows for elucidation of CDR3 sequences as well as providing a unique dataset to infer the lineage structure of T- cells. Specifically, our current dataset can be used to model clonal selection and amplification across the \(\mathsf{CD8 + }\) T- cell subpopulations and trajectories. We recovered productive TCR- alpha and TCR- beta sequences from 1,461 and 1,948 cells, respectively, and identified 1,590 unique TCR sequences. No shared clones were found between patients, with unique TCRs for each patient. Clonal expansion was seen in \(17.39\%\) of \(\mathsf{CD8 + }\) cells, and clone size ranged from 2 to 60 cells per clone (Figure 5E, Supplementary Figure 5H and 5I). Clonal overlap between the two different sites for each tumor (primary and lymph node) was demonstrated in patients HN257 and HN272 (Figure 5F). There was a progressive increase in clonality across the dysfunctional gradient, with evidence of single naive or TRM- derived clones subsequently expanding to give rise to multiple dysfunctional clones that span these trajectories (Figure 5F and 5G).  
+
+There appeared to be patient- specific biases for one trajectory over the other. For example, there are \(\mathsf{CD8 + }\) T- cell clones in patient HN272 that followed a naive- dysfunction trajectory (Trajectory 1), with expansion of lymph node derived naive clonotypes, migrating to the primary site and captured there along a dysfunctional gradient (pre- dysfunctional, proliferative and then late- dysfunction) (Figure 5F). This supports a previous observation which suggests that circulation is one of the major sources of tumor- targeting dysfunctional cells, which in this case is the regional lymphatics draining nodal tissue<sup>28</sup>. In contrast, in patient HN263 and selected \(\mathsf{CD8 + }\) T- cell clones in patient HN272, the dysfunctional gradient appears to comprise of tissue resident memory (TRM) cells derived from the primary tumor, which amplified into putative tumor- targeting clonotypes (Figure 5F). This is consistent with a model of ongoing differentiation and proliferation of dysfunctional T- cells at the tumor site itself<sup>29</sup>. It is likely that both mechanisms contribute to the dysfunction gradient, sometimes even within the same patient. For example, lineage tracing in HN257 and HN272 demonstrates extensive trafficking and interplay between the primary site and lymph node: there is evidence of lymph node- derived naive cells expanded in the primary site as expected, but also surprisingly TRM cells expanding and subsequently migrating to the lymph node (Figure 5F and 5G). This scTCR data adds intriguing complexity to concepts of clonal expansion and lineage structure in a treatment naive setting.
+
+<--- Page Split --->
+
+## Pre-nodal cells and immune micro-environment  
+
+Our analyses identified a pre- nodal sub- population in primary tumors with intrinsic properties of invasion and migration. However, metastasis also requires acquisition of an immune evasion phenotype. To test whether the pre- nodal cells identified above demonstrated specific immune- modulatory phenotypes, we subjected three tumors (from our study) and two tumors (from the Puram dataset) each with a minimum RNAseq dataset to interactome analyses using Cellchat. To do this, we divided primary tumor cells into two subpopulations (primary and pre- nodal) and analyzed the interactions of these two tumor subpopulations with \(\mathsf{CD8 + }\) \(\mathsf{CD4 + }\) and T- reg lymphocytes and TAMs. For HN251, HN272 and HN279, the analysis showed similar trends in primary to pre- nodal malignant cells, with increasing interactions between the pre- nodal subpopulation and T- lymphocytes, specifically with \(\mathsf{CD8 + }\) cells (Figure 6A). The analyses implicated a number of pathways that were differentially modulated by primary versus pre- nodal populations on T- lymphocytes (Supplementary Figure 6A- C). In particular, the interaction between Midkine (MDK, secreted by tumor cells) and a number of MDK- receptors (ITGA4, ITGA6, ITGB1, NCL, LRP1) on \(\mathsf{CD8 + }\) T- cells appears to be a recurrent immunosuppressive pathway seen across all three patients (Figure 6B). Applying the same approach to the external dataset also implicated the MDK pathway as being differentially activated by the pre- nodal population in one (p17) out of two tumors tested (Figure 6B and Supplementary Figure 6D- E).  
+
+Recent published data suggest that MDK- driven modulation is important for immune evasion in melanomas with activation of NFKB and its downstream pathways<sup>30</sup>. To test whether MDK- driven immune suppression dampens the effect of immune checkpoint blockade (ICB) therapy, we developed a humanized mouse model engrafted with pre- nodal cells from the tumor of patient HN279, and treated these with PD1- blockade. As expected, the majority of cancer- cells expressed MDK (Figure 6C- 6D and Supplementary Figure 6F- G), together with a number of genes associated with the pre- nodal phenotype (eg SNAI2, AXL, STAT2) that were unaffected by ICB (Figure 6E and Supplementary Figure 6H). In contrast, expression of AURKB and TOP2A (cell cycle genes) in cancer cells was significantly downregulated after pembrolizumab treatment (Figure 6E), indicating a reduction in cancer cell proliferation.  
+
+Analyses of the \(\mathsf{CD8 + }\) T- cell fraction revealed naive, TRM, transitional, proliferative and dysfunctional/exhausted subpopulations, with an additional cytotoxic populations (likely bystander) (Figure 6F and Supplementary Figure 6I). \(\mathsf{CD8 + }\) cells from mice treated with pembrolizumab showed reduction in naive, dysfunctional and memory with concomitant increase in proliferative, cytotoxic/bystander, tissue resident subpopulations compared to untreated mice (Figure 6G). These changes suggest a re- invigoration and re- activation of dysfunctional and memory, respectively, into tumor- targeting cells<sup>29</sup>. Remarkably, analyses of MDK receptor- expressing CD8 cells (ITGA4, ITGB1, NCL) showed the opposite trend, with an increase in dysfunctional and reduction in the proliferative (tumor- targeting) populations (Figure 6H and Supplementary Figure 6J). These findings suggest MDK- signaling promotes immune- suppression, that abrogates re- invigoration by PD1- blockade. Indeed, these changes were also associated with NFKB1 activation which is significantly higher in the dysfunctional CD8 population after pembrolizumab treatment (Figure 6I). Moreover, plotting the expression levels of several MDK- receptors (ITGA4, ITGB1, NCL) with NFKB1 show a good correlation in gene expression in \(\mathsf{CD8 + }\) T- cells where the
+
+<--- Page Split --->
+
+276 RNA could be quantified (Figure 6J). Taken together, these results implicate MDK- signaling as a pathway through which pre- nodal cells evade CD8- mediated immune- editing.
+
+<--- Page Split --->
+
+Currently available algorithms analyzing single- cell data have the ability to construct evolutionary trajectories, which are especially powerful in studying specific events in space (eg relationships between different tumor sites, eg primary vs lymph node metastasis) and time (eg pre- and post- treatment analysis)12,28. Here, we applied these to explore early lymph node metastasis across tumor and immune sub- compartments within the tumor. Analysis of tumor cells shows that nodal metastasis is an early event, where canonical epithelial- to- mesenchymal transition is less apparent than postulated. Our findings support previous studies that suggest EMT is not an all- or- none phenomenon, but instead occurs in graded levels 31,32. This is in contrast to in vitro systems (including our own) where cultured tumor cells from lymph nodes display more canonical features of EMT33. Despite overlap between tumor cells derived from primary and nodal sites, trajectory mapping could define evolutionary pathways at individual tumor levels, although this process require a combination of trajectory algorithms, scoring for aggressiveness (based on EMT and stemness) and knowledge of the ground truth. These have expanded the results of previous studies in the identification of a pre- nodal or metastatic population 12, and importantly identified actionable drivers of that could be targeted for anti- metastatic therapy, in this case AXL and AURK. Targeting AXL would not only prevent pathways involved in dissemination, but presumably reduce tumor heterogeneity by targeting the specific clones34. The role of aurora kinases is less clear; rather than impacting the metastatic process, it is possible that this vulnerability reflects a generalized reduction in cell cycling that occurs during EMT with a concomitant sensitivity to all cell cycle inhibitors. We recently demonstrated the same phenomenon during drug resistance: reduction in cell proliferation, limited AURK expression and sensitivity to inhibitors of AURK and other cell cycle targets35. Nevertheless, the ability to profile tumors and identify vulnerabilities in metastasis- inducing clones is an attractive notion, with increasing interest in low- dose, long term anti- metastatic therapy.  
+
+Alignment of CD8+ T lymphocyte populations is driven by existing knowledge on T- cell maturation. The fact that we could pool data across different patients increased the number of cells available and in itself was a form of validation. The alignment was further supported by single- cell VDJ sequencing, which reinforced trajectories from naive or memory populations, towards clonally expanded, dysfunctional and potentially tumor targeting CD8+ subpopulations These supported the notion that both naive CD8+ cells from adjacent lymph nodes and tissue resident CD8+ T- cells were sources of expanded tumor- infiltrating CD8+ T cells, and trafficking was bidirectional. Strikingly, this trajectory could be used to identify novel modulators of T- cell dysfunction by studying gene expression changes along pseudotime, and was used to identify SOX4 as novel driver of dysfunction in CD8 cells. Interactome analyses performed to identify signaling networks within CD8+ T cells during early metastasis converged onto the MDK pathway. Remarkably, in a humanized mouse model, MDK signaling was associated with a reduced ability to reinvigorate exhausted T- cells. This is supported by a recent publication which identified that the MDK pathway could abrogate immune reactivation by ICB therapy in melanoma, and this could be reversed using MDK- specific inhibitors30. In a similar context, MDK- inhibition could be explored in the prevention and treatment of tumor metastasis in HNSCC and add synergy to PD1- blockade which is the current standard of care in metastatic HNSCC.
+
+<--- Page Split --->
+
+In conclusion, we applied single- cell genomics to uncover pathways and mechanisms that mediate early nodal metastasis in HNSCC. The data presented here shows that early metastasis is a much more nuanced process than previously presumed. Collectively these indicate the discovery potential of single cell studies and existing computational tools, when applied to specific clinical contexts and questions. Future studies will focus on more specific tumor subpopulations including CD8+ cells and the impact of treatment on tumor recurrence and metastasis.
+
+<--- Page Split --->
+
+## Methods  
+
+Tumor collection and processing  
+
+Patient tumors were harvested in the operating room and transported to the lab for processing within 30 minutes. Tumors were a priori confirmed histologically to be HNSCC and patients were consented prior to surgery. This study is approved by SingHealth Centralized Institutional Review Board (CIRB: 2014/2093, 2018/2512, 2016/2757). All tumors were dissociated using the gentleMACS™ Octo system (Miltenyi Biotech, Bergisch Gladbach, Germany) as described in manufacturer's protocol. These were subjected to filtration, washing and magnetic bead separation, where required, prior to single cell capturing (details in Supplementary methods).  
+
+## Patient-derived cell cultures  
+
+Cultures were established as previously described<sup>20,21</sup>. Cells were maintained in complete RPMI (C/RPMI) containing \(10\%\) FBS, \(1\%\) pen- strep, \(1\%\) anti- mycotic and a humidified incubator at \(37^{\circ}C\) with \(5\%\) CO2. All lines were tested and confirmed to be free of mycoplasma using an EZ- PCR Mycoplasma Detection Kit (Biological Industries, Kibbutz Beit Haemek, Israel) at the time of experiments. Cells were processed for scRNAseq and for immunostaining as described in Supplementary methods. Invasion assays cells were treated with or without \(0.25 \mu M\) of bementinib (BGB324) or \(0.25 \mu M\) of barasetrib (both from Selleck Chem, Houston, TX), then seeded on an \(8 \mu m\) filter membrane within a 24- well transwell insert (Corning, New York City, NY), with C/RPMI at bottom of wells of 24- well Falcon TC Companion Plate (Corning, New York City, NY). After 72hrs, the bottom of each inserts was fixed and stained for quantification of invaded cells. Cell invasion area was determined by quantifying the area with crystal violet staining using the ImageJ software.  
+
+## Humanized mouse model  
+
+Sixteen NOG- EXL (hGM- CSF/hIL- 3 NOG) mice (hNOG- EXL), pre- engrafted with human CD34+ hematopoietic stem cells, were procured from CIEA- SlgN. At 16 weeks post- engraftment, mice were injected subcutaneously with cells from HN279, and treated intraperitoneally with \(12.5mg / kg\) of pembrolizumab or with phosphate buffered saline (PBS) on day 17, 19, 21 and 24. Mice were euthanized on day 25 and tumors harvested for dissociation and preparation for scRNAseq as described.  
+
+## Small-interfering RNA knock-down of SOX4 and DUSP4  
+
+Peripheral blood mononuclear cells (PBMCs) from healthy donors were cultured at a density of \(0.5 - 1 \times 10^{6}\) cells/ml in 24- well plate (Corning, New York City, NY), containing TexMACS Medium and T- Cell TransAct (both from Miltenyi Biotech, Bergisch Gladbach, Germany) at 1:200 dilution. A final concentration of \(1 \mu M\) of Accell pooled small- interfering RNA (siRNA) targeting human SOX4 (Gene ID 6659) or DUSP4 (Gene ID 1846), or non
+
+<--- Page Split --->
+
+targeting siRNA (all from Dharmacon, Lafayette, CO) was added into respective wells. After 5 days of incubation, cells were harvested for flow cytometry.  
+
+Flow cytometry  
+
+For AXL surface staining, trypsinized cells were stained with fluorochrome- conjugated antibody recognizing AXL (#108724; R&D systems, Minneapolis, MN) or with mouse IgG1 isotype antibody (MOPC- 21; BD Biosciences, Franklin Lakes, NJ). For intracellular AURKB staining, trypsinized cells were fixed and permeabilized with a Foxp3/Transcription Factor Staining Buffer Set (eBioscience, San Diego, CA) according to the manufacturer protocol. After fixation, cells were stained with primary antibody recognizing AURKB (clone RM278; Invitrogen, Carlsbad, CA) or rabbit IgG1 isotype antibody (DA1E; R&D systems, Minneapolis, MN), and subsequently with goat anti- rabbit IgG secondary antibody conjugated to Alex Fluor 647 (#A32733; Waltham, MA). For siRNA knock- down PBMC experiments, harvested cells were stained with fluorochrome- conjugated antibodies recognizing CD57 (HNK- 1), LAG3 (11C3C65), CD39 (A1) and CD4 (OKT4) all from Biolegend, San Diego, CA; PD1 (J105) and CD8 (SK1) from eBioscience, San Diego, CA; and CD4 (SK3) from BD Biosciences, Franklin Lakes, NJ. These cells were stained for 30mins on ice in the dark with 2% BSA in PBS. Live/dead cells were distinguished using a Fixable Live Dead Blue Dead Cell Stain Kit (Thermofisher, Waltham, MA). Cells were acquired and analyzed using a BD FACSCanto II instrument and FlowJo v10.5.3 software (both from BD Biosciences, Franklin Lakes, NJ) respectively.  
+
+Generation of single cell gene expression and TCR libraries by droplet- based (10x system) and microfluidic- based technologies  
+
+The 5' gene expression (GEX) and TCR single cell RNA libraries from tumors were prepared using the 10x Chromium Single Cell V(D)J Reagent Kits (10x Genomics, Pleasanton, CA), as described in the manufacturer's protocol. Briefly, freshly dissociated tumor cells were sorted into CD45+ and CD45- fractions, mixed at a 1:1 ratio and loaded into the Single Cell A Chip for gel bead- in- emulsion (GEM) generation and barcoding, targeting for a cell recovery of 4000- 7000 cells per sample. Reverse transcription, cDNA amplification, GEX and TCR library construction were performed as described. For C1, single cell suspensions were loaded and captured using medium- sized (10- 17um) Fluidigm Integrated Fluidic Circuit (IFC) and a Fluidigm C1 instrument (Fluidigm, South San Francisco, CA), according to the manufacturer's protocol. cDNA product was harvested from the IFC, barcoded for individual cell identity and pooled. Sequencing was performed by an Illumina Hiseq 4000 (Illumina, San Diego, CA) with 151- bp single- ended or pair- ended reads.
+
+<--- Page Split --->
+
+394 Data processing of single- cell RNA- seq libraries and clustering 395 scRNAseq reads were aligned to the GRCh38 reference genome and quantified using Cellranger count (10x 396 Genomics, version 2.2.0). Downstream analyses were performed using Seurat (version 3.1.5). For malignant- cell 397 analysis, we isolated subsets of cells identified as malignant cells based on broad clustering and reprocessed 398 using Seurat without patient alignment, since tumor cells tend to be patient specific. For T- cell clustering, we 399 isolated subsets of cells identified as T- cells based on broad clustering. Cells were then re- clustered using Seurat 400 alignment across patients similar as with previous analysis. For CD8+ T- cell clustering, CD8+ T- cells were 401 extracted from the T- cell clustering based on the following two criteria: 1) in Pre- dysfunctional, Dysfunctional 402 and Proliferative clusters, and with zero CD4 expression, 2) in Naive- like, Memory and Transitional clusters, with 403 zero CD4 and positive CD8 (either CD8A or CD8B) expression. TCR reads were mapped to 404 vdj_GRCh38_alts_ensembl- 3.1.0- 3.1.0 reference genome and quantified using cellranger count (10x Genomics, 405 version 3.1.0). Further details on Seurat analysis, UMAP visualization and use of the following algorithms: 406 InferCNV, Monocle, PAGODA, Slingshot, functional annotation, Geneswitches, Cytotrace and other analyses 407 tools, are described in detail in Supplementary methods.  
+
+## Statistical analysis  
+
+Statistical analysis was performed using GraphPad Prism software (GraphPad Software, Inc., San Diego, CA), or otherwise indicated in the figure legends and Supplementary methods.
+
+<--- Page Split --->
+
+## 413 Acknowledgements  
+
+We would like to thank all patients and families who contributed to this project. Additional technical support and access for 10x experiments were obtained from the Laboratory of Cell Therapy and Cancer Vaccine, National Cancer Centre Singapore (under Dr Han- Chong Toh), C1 experiments from the Laboratory of Cancer Epigenome (under Professor Bin- Tean Teh and Mr Cedric Ng), and humanized mouse experiments from the HuNIT IAF- PP program (Teja Celhar, Yunqian Zhao, Hui Chen Tay, Takeshi Takahashi and Jerry Chan, Singapore Immunology Network- A\*STAR). This project was funded through the following grants awarded to the respective investigators, for which we are extremely grateful: Khoo Postdoctoral Fellowship Award to QHS, National Medical Research Council (Singapore) Clinician Scientist Awards to NGI (NMRC/CSA/001/2016, MOH- 000325- 00), the Peter Fu Head and Neck Cancer Program (under the Oncology Academic Clinical Program, National Cancer Centre Singapore) to NGI, core funding by Singapore Immunology Network (A\*STAR) to SKB and a Singapore National Research Foundation grant [NRF- CRP20- 2017- 0002] to OR and JO.  
+
+425  
+
+Competing interest: NGI has/had a consulting or advisory role in PairX Therapeutics and Invitrocue PLC, and received honoraria from Kalbe Biotech and Agilent, all of which are outside this submitted work. DSWT received honoraria from Bristol- Myers Squibb, Takeda Pharmaceuticals, Novartis, Roche, and Pfizer; and has consulting or advisory role in Novartis, Merck, Loxo Oncology, AstraZeneca, Roche, and Pfizer. DSWT also received research funding from Novartis (Inst), GlaxoSmithKline (Inst), and AstraZeneca (Inst), outside this submitted work. None of the remaining authors have any other conflicts to report.
+
+<--- Page Split --->
+
+433 1 Steeg, P. S. Targeting metastasis. Nat Rev Cancer 16, 201- 218, doi:10.1038/nrc.2016.25 (2016). 434 2 Chaffer, C. L. & Weinberg, R. A. A perspective on cancer cell metastasis. Science 331, 1559- 1564, doi:10.1126/science.1203543 (2011). 436 3 Gupta, G. P. & Massague, J. Cancer metastasis: building a framework. Cell 127, 679- 695, doi:10.1016/j.cell.2006.11.001 (2006). 438 4 Chow, L. Q. M. Head and Neck Cancer. N Engl J Med 382, 60- 72, doi:10.1056/NEJMra1715715 (2020). 439 5 Iyer, N. G. et al. Randomized trial comparing surgery and adjuvant radiotherapy versus concurrent chemoradiotherapy in patients with advanced, nonmetastatic squamous cell carcinoma of the head and neck: 10- year update and subset analysis. Cancer 121, 1599- 1607, doi:10.1002/cncr.29251 (2015). 443 6 Mehlen, P. & Puisieux, A. Metastasis: a question of life or death. Nat Rev Cancer 6, 449- 458, doi:10.1038/nrc1886 (2006). 445 7 Heerboth, S. et al. EMT and tumor metastasis. Clin Transl Med 4, 6, doi:10.1186/s40169- 015- 0048- 3 (2015). 447 8 Brabletz, T., Kalluri, R., Nieto, M. A. & Weinberg, R. A. EMT in cancer. Nat Rev Cancer 18, 128- 134, doi:10.1038/nrc.2017.118 (2018). 449 9 Williams, E. D., Gao, D., Redfern, A. & Thompson, E. W. Controversies around epithelial- mesenchymal plasticity in cancer metastasis. Nat Rev Cancer 19, 716- 732, doi:10.1038/s41568- 019- 0213- x (2019). 451 10 Vettore, A. L. et al. Mutational landscapes of tongue carcinoma reveal recurrent mutations in genes of therapeutic and prognostic relevance. Genome Med 7, 98, doi:10.1186/s13073- 015- 0219- 2 (2015). 453 11 Tan, D. S. et al. Tongue carcinoma infrequently harbor common actionable genetic alterations. BMC Cancer 14, 679, doi:10.1186/1471- 2407- 14- 679 (2014). 455 12 Puram, S. V. et al. Single- Cell Transcriptomic Analysis of Primary and Metastatic Tumor Ecosystems in Head and Neck Cancer. Cell 171, 1611- 1624 e1624, doi:10.1016/j.cell.2017.10.044 (2017). 457 13 McFaline- Figueroa, J. L. et al. A pooled single- cell genetic screen identifies regulatory checkpoints in the continuum of the epithelial- to- mesenchymal transition. Nat Genet 51, 1389- 1398, doi:10.1038/s41588- 019- 0489- 5 (2019). 460 14 Wei, S. C., Duffy, C. R. & Allison, J. P. Fundamental Mechanisms of Immune Checkpoint Blockade Therapy. Cancer Discov 8, 1069- 1086, doi:10.1158/2159- 8290. CD- 18- 0367 (2018). 462 15 Mempel, T. R., Henrickson, S. E. & Von Andrian, U. H. T- cell priming by dendritic cells in lymph nodes occurs in three distinct phases. Nature 427, 154- 159, doi:10.1038/nature02238 (2004). 463 16 van der Leun, A. M., Thommen, D. S. & Schumacher, T. N. CD8(+) T cell states in human cancer: insights from single- cell analysis. Nat Rev Cancer, doi:10.1038/s41568- 019- 0235- 4 (2020). 466 17 Dunn, G. P., Old, L. J. & Schreiber, R. D. The three Es of cancer immunoediting. Annu Rev Immunol 22, 329- 360, doi:10.1146/annurev.immunol.22.012703.104803 (2004). 468 18 Dunn, G. P., Bruce, A. T., Ikeda, H., Old, L. J. & Schreiber, R. D. Cancer immunoediting: from immunosuppressive to tumor escape. Nat Immunol 3, 991- 998, doi:10.1038/nij1102- 991 (2002). 469 19 Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646- 674, doi:10.1016/j.cell.2011.02.013 (2011). 471 20 Chia, S. et al. Phenotype- driven precision oncology as a guide for clinical decisions one patient at a time. Nat Commun 8, 435, doi:10.1038/s41467- 017- 00451- 5 (2017). 472 21 Leong, H. S. et al. Targeting cancer stem cell plasticity through modulation of epidermal growth factor and insulin- like growth factor receptor signaling in head and neck squamous cell cancer. Stem Cells Transl Med 3, 1055- 1065, doi:10.5966/sctm.2013- 0214 (2014). 473 22 Gao, R. et al. Delineating copy number and clonal substructure in human tumors from single- cell transcriptomes. Nat Biotechnol 39, 599- 608, doi:10.1038/s41587- 020- 00795- 2 (2021). 474 23 Yang, J., Chen, Y., Luo, H. & Cai, H. The Landscape of Somatic Copy Number Alterations in Head and Neck Squamous Cell Carcinoma. Front Oncol 10, 321, doi:10.3389/fonc.2020.00321 (2020). 485 24 Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat Biotechnol 32, 381- 386, doi:10.1038/nbt.2859 (2014). 483 25 Gulati, G. S. et al. Single- cell transcriptional diversity is a hallmark of developmental potential. Science 367, 405- 411, doi:10.1126/science.aax0249 (2020).
+
+<--- Page Split --->
+
+485 26 Cao, E. Y., Ouyang, J. F. & Rackham, O. J. L. GeneSwitches: ordering gene expression and functional 486 events in single-cell experiments. Bioinformatics 36, 3273- 3275, doi:10.1093/bioinformatics/btaa099 487 (2020). 488 27 Street, K. et al. Slingshot: cell lineage and pseudotime inference for single-cell transcriptomics. BMC 489 Genomics 19, 477, doi:10.1186/s12864- 018- 4772- 0 (2018). 490 28 Yost, K. E. et al. Clonal replacement of tumor- specific T cells following PD- 1 blockade. Nat Med 25, 491 1251- 1259, doi:10.1038/s41591- 019- 0522- 3 (2019). 492 29 Li, H. et al. Dysfunctional CD8 T Cells Form a Proliferative, Dynamically Regulated Compartment within 493 Human Melanoma. Cell, doi:10.1016/j.cell.2018.11.043 (2018). 494 30 Cerezo- Wallis, D. et al. Midkine rewires the melanoma microenvironment toward a tolerogenic and 495 immune- resistant state. Nat Med 26, 1865- 1877, doi:10.1038/s41591- 020- 1073- 3 (2020). 496 31 Revenco, T. et al. Context Dependency of Epithelial- to- Mesenchymal Transition for Metastasis. Cell 497 Rep 29, 1458- 1468 e1453, doi:10.1016/j.celrep.2019.09.081 (2019). 498 32 Chen, Y. et al. Dual reporter genetic mouse models of pancreatic cancer identify an epithelial- to- 499 mesenchymal transition- independent metastasis program. EMBO Mol Med 10, 500 doi:10.15252/emmm.201809085 (2018). 501 33 Sharma, A. et al. Longitudinal single- cell RNA sequencing of patient- derived primary cells reveals 502 drug- induced infidelity in stem cell hierarchy. Nat Commun 9, 4931, doi:10.1038/s41467- 018- 07261- 3 503 (2018). 504 34 Antony, J. & Huang, R. Y. AXL- Driven EMT State as a Targetable Conduit in Cancer. Cancer Res 77, 505 3725- 3732, doi:10.1158/0008- 5472.CAN- 17- 0392 (2017). 506 35 Low, J. L. et al. A chemical genetic screen identifies Aurora kinases as a therapeutic target in EGFR 507 T790M negative, gefitinib- resistant head and neck squamous cell carcinoma (HNSCC). EBioMedicine 508 64, 103220, doi:10.1016/j.ebiom.2021.103220 (2021).
+
+<--- Page Split --->
+
+Figure 1. Tumor samples for single cell RNAseq (A) Workflow of sample acquisition, processing, and analyses for single cell transcriptome and TCR clonality of tumors (and patient-derived cultures) from primary and metastatic lymph nodes of HNSCC patients. Diagram was created with BioRender.com. (B) Uniform manifold approximation and projection (UMAP) of scRNAseq data from all cells within primary tumors and metastatic lymph nodes from 7 patients. Clusters are denoted by colors and labelled according to inferred cell types. Violin plots show the expression of selected genes used to define the inferred cell types. (C) Distribution of different cell types (color) for each patient sample (top) and comparing primary and metastatic samples (bottom) as indicated on the y-axis. (D) Chromosomal gains and losses prediction for malignant epithelial cells by inferCNV using non- malignant cells from respective samples as controls. Cyan indicates primary malignant epithelial; yellow indicates lymph node malignant epithelial; sample identities on the y-axis, chromosome numbers on the x-axis.  
+
+Figure 2. scRNAseq analysis of malignant epithelial cells and identification of pre-metastatic sub- population. (A) UMAP of malignant epithelial cells only, clustered by Seurat clusters (left), patients (middle), and tissue origin (primary/metastatic) (right). (B) Boxplot showing epithelial-mesenchymal transition (EMT) scores across patients and tissue origin (primary versus metastasis). Line represents mean scores, while box represents 2 standard deviations. (C) and (D) Monocle plots demonstrating the derivation of pre-metastatic populations in HN251 (C) and HN279 (D) based on (from left to right) tissue origin, monocle clusters, EMT scores, CytoTRACE scores to derive trajectory. (E) Gene ontology pathways that are significantly altered across pseudotime derived in C and D. (F) Potentially actionable genes identified to be increased in pre-metastatic population. (G) t-SNE plot of tumor cells in HN257 showing a highly aggressive sub-population in the primary tumor with high CytoTRACE scores and expression of SNAI2. (H) Gene set enrichment analysis (GSEA) showing normalized enrichment scores and (I) Kaplan-Meier plot of TCGA data showing overall survival in patients with high versus low scores based on genes expressed by the specific subpopulation in (G). Shaded area shows 95% confidence interval and p-value as indicated based on log-rank test.  
+
+Figure 3. Functional analysis of actionable genes enriched in pre-metastatic population in patient- derived cultures (PDCs). (A) Dimension reduction plots based on PAGODA for PDCs derived from matched primary and metastatic lymph nodes (nodal metastatic; M). Clusters are denoted by patient identity and site of origin (left), and Seurat clusters (right). (B) Heatmap of differentially expressed pathways (rows) across samples and tumor origin (columns), showing selected Hallmark and Gene Ontology (GO) gene sets. Bars on the top of the heat map indicate the site of sample origins, clusters and patient samples corresponding to those of (A). (C) Boxplot showing the gene expression level of AXL (left) and AURKB (right) of malignant cells from primary and metastatic PDCs for the indicated patients. Line represents mean expression, while box represents 2 standard deviations; colors and cluster numbers of the bars correspond to (A). (D) Immunocytochemistry of AXL in HN137 and AURKB in HN159 and HN220 of primary and metastatic PDCs. Scale bar indicates 100 μm. (E) Representative micrographs from Boyden chamber assays of invaded cells (purple) (top), and quantification of invaded cells (bottom) in barplots from primary and metastatic cell cultures treated with or without BGB324 or barasetrib.
+
+<--- Page Split --->
+
+\(^{**}p < 0.01\) , \(^{***}p < 0.001\) , \(^{****}p < 0.0001\) (significant difference) using student t-test compared with untreated at corresponding site of origin. Error bars represent one standard deviation. (H) Flow cytometry dot plots representing anti- AXL (left) and mouse IgG1 isotype control (right) staining of primary and metastatic PDCs of HN137. (I) Gating used for identification and isolation of \(\mathsf{AXL}^{\mathsf{h}\mathsf{i}}\) , \(\mathsf{AXL}^{\mathsf{mid}}\) and \(\mathsf{AXL}^{\mathsf{neg / low}}\) from HN137 primary PDC by FACS sorting (left). Micrographs representing isolated AXL- based subpopulations treated with or without BGB324 and their respective invasive potential in Boyden chamber assays (right).  
+
+Figure 4. scRNAseq analysis of tumor infiltrating T- cells and establishing a trajectory for tumor- targeting CD8+ lymphocytes. (A) UMAP of tumor infiltrating T- cells from primary and metastatic tumors with clusters denoted by colors and labelled with inferred cell identities. (B) Heatmap of differentially expressed genes (rows) between cells classified into inferred T- cell subsets. Bars on the top of the heatmap indicate the site of origin and cell type corresponding to those of (A) with selected genes indicated. (C) UMAP of all CD8 T- cells from primary and metastatic tumors. Clusters are denoted by colours and labelled with inferred cell identities based on (D) expression of selected genes used for CD8 T- cell subset annotation for. (E) Slingshot analysis of CD8 T- cells showing two potential trajectories giving rise to tumor- targeting CD8+ cells: Trajectory 1 (top)- from naive to dysfunctional and Trajectory 2 (bottom)- memory to dysfunctional. (F) Graphs showing the estimate scores of curated genes related to naive- like (IL7R, TXNIP, SELL, CCR7, TCF7), proliferative (MKI67, HMG82, TYMS), dysfunctional (GZMB, GNYL, CTLA4, LAYN, LAG3, TIG17) populations, and expression of CXCL13 during the development of CD8 T- cell along the naive- proliferation- dysfunction axis in Trajectory 1. (G) Geneswitches output showing ordering of the top switching genes along the naive to dysfunctional (Trajectory 1) CD8 T- cell axis using. Key genes are highlighted with enlarged font size. (H) UMAP projections of expression levels for genes highlighted in (G).  
+
+Figure 5. Functional analysis of genes involved in CD8 dysfunction and T- cell receptor sequencing analysis. Violin plots showing expression of SOX4, DUSP4 and RBPJ in CD8 T- cell subpopulations derived from published cohorts of scRNAseq meta- dataset from (A) HNSCC and (B) skin squamous cell cancer<sup>12,28</sup>. (C) Boxplots showing expression of SOX4, DUSP4 and RBPJ in CD8 T- cell subpopulations from (B), grouped by pre- and post- pembrolizumab treatment. \(^{*}p < 0.05\) and \(^{***}p < 0.0001\) denotes a significant difference compared with pretreatment of corresponding CD8 T- cell subsets by paired t- test. (B- C) X- axis labels: CD8_ mem = CD8 memory; CD8_eff = CD8 effector; CD8_act = CD8 activated; CD8_ex_act = CD8 exhausted/activated; CD8_ex = CD8 exhausted. (D) Bar graph showing percentage of CD8 T- cells expressing CD39, CD57, LAG3 or PD1 from PBMCs that were activated and cultured with siNT, siSOX4 or siDUSP4 for 5 days (n = 4). Black lines and error bars represent mean ± SEM. \(^{*}p < 0.05\) , \(^{**}p < 0.01\) , \(^{****}p < 0.0001\) (significant difference) by paired t- test compared with siNT of respective markers. (E) Barplots of the percentage of TCR clone(s) detected once (n=1), twice (n=2) or more than two times (n>2) across the CD8 T- cell subpopulations of all patients with HNSCC subjected to scRNAseq. (F) UMAP projection of CD8 T- cells from HN272, HN263 and HN257 colored by selected TCR clonotypes. (G) Schematic diagram summarizing the development and trafficking of CD8 T- cell clones between
+
+<--- Page Split --->
+
+primary tumor, lymph node and metastasis, and bloodstream of HN272, HN263 and HN257 based on the clonotype data from (F). Diagram was created with BioRender.com.  
+
+Figure 6. Determining the interaction between pre-metastatic malignant cells and CD8+ T lymphocyte populations. (A) Hierarchical plot derived from Cellchat analyses showing ligand-receptor interactions between tumor cells (primary and pre-nodal subpopulations) with T-lymphocytes (CD8+, CD4+ and Treg cells) and TAMs. Circle sizes are proportional to the number of cells in each cell group available for and edge width represents the communication probability with number of potential ligand-receptor pair as indicated. (B) Dot (bubble) plots showing significant MDK ligand-receptor pairs contributing to the signaling from primary or pre-metastatic cancer cells (epithelial) to Treg, CD4 or CD8 T-cells. The dot color and size represent the calculated communication probability, and p-values determined from one-sided permutation test. (C) UMAP of cells derived from tumors of humanized NOG-EXL mice treated with or without anti-PD1. Clusters are denoted by colors labelled with inferred cell types, with a 2D projection of MDK gene expression (inset). (D) Frequency of MDK+ (blue) and MDK- (orange) malignant cells in control or anti-PD1 treated mice. (E) Expression level of selected genes involved in tumor cell proliferation in malignant cells from control or anti-PD1-treated mice. \(*p < 0.05\) and \(**p < 0.01\) indicate significant difference by unpaired t test when compared to control. (F) UMAP of tumor infiltrating CD8 T-cells only extracted from (C). Clusters are denoted by colors labelled with inferred cell identities. (G) Distribution of CD8 T-cell subpopulations in control vs anti-PD1 treated mice. (H) Delta (Δ) percentage of CD8 T-cells expressing the specific MDK-receptors ITGA4, ITGB1 or NCL showing changes in dysfunctional, transitional and proliferating subpopulations, comparing untreated versus anti-PD1 treated mice. Delta percentage is determined by the percentage of MDK receptor+ CD8+ T-cells from anti-PD1 treated mice minus that of the control mice. (I) Expression of NFKB1 in the three CD8 subpopulations in controls and anti-PD1 treated mice. \(*p < 0.05\) indicates significant difference by unpaired t test when compared to control. (J) Scatterplot showing the correlation of expression between NFKB1 with the following MDK receptor(s): ITGA4, ITGB1 and/or NCL in the dysfunctional CD8 T-cells subpopulation. Each dot represents one dysfunctional CD8 T-cell from control (red) or anti-PD1 (blue) treated mice. The R and p values were determined using Pearson correlation statistical analysis.
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1</center>
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2 </center>
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3</center>
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Figure 5</center>
+
+<--- Page Split --->
+![PLACEHOLDER_28_0]
+
+<center>Figure 6</center>
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+SupplementaryFigure16v3. pdf SupplementaryTables.pdf scSupplementaryMethodsv2. pdf
+
+<--- Page Split --->

preprint/preprint__b3f261e76766734f08b87ae1a92ea1300136deecc3c800639ccde5acbd7fbac9/preprint__b3f261e76766734f08b87ae1a92ea1300136deecc3c800639ccde5acbd7fbac9_det.mmd

@@ -0,0 +1,335 @@
+<|ref|>title<|/ref|><|det|>[[42, 107, 955, 208]]<|/det|>
+# Single cell analysis of early metastasis identifies targetable tumor subpopulation and mechanisms of immune evasion in squamous cell cancers  
+
+<|ref|>text<|/ref|><|det|>[[42, 229, 707, 271]]<|/det|>
+Hong Sheng Quah National Cancer Centre Singapore https://orcid.org/0000- 0001- 8430- 8528  
+
+<|ref|>text<|/ref|><|det|>[[42, 277, 285, 316]]<|/det|>
+Elaine Yiqun Cao Duke- NUS Medical School  
+
+<|ref|>text<|/ref|><|det|>[[42, 323, 352, 364]]<|/det|>
+Lida Suteja National Cancer Centre Singapore  
+
+<|ref|>text<|/ref|><|det|>[[42, 370, 352, 410]]<|/det|>
+Hui Leong National Cancer Centre Singapore  
+
+<|ref|>text<|/ref|><|det|>[[42, 416, 352, 456]]<|/det|>
+Fui Chong National Cancer Centre Singapore  
+
+<|ref|>text<|/ref|><|det|>[[42, 463, 352, 503]]<|/det|>
+Constance Li National Cancer Centre Singapore  
+
+<|ref|>text<|/ref|><|det|>[[42, 509, 338, 549]]<|/det|>
+Shilpi Gupta Singapore Immunology Network  
+
+<|ref|>text<|/ref|><|det|>[[42, 555, 616, 595]]<|/det|>
+Camille Arcinas National Cancer Centre https://orcid.org/0000- 0001- 5374- 9232  
+
+<|ref|>text<|/ref|><|det|>[[42, 601, 641, 641]]<|/det|>
+John Ouyang Duke- NUS Medical School https://orcid.org/0000- 0002- 1239- 1577  
+
+<|ref|>text<|/ref|><|det|>[[42, 648, 338, 688]]<|/det|>
+Vivian Ang Singapore Immunology Network  
+
+<|ref|>text<|/ref|><|det|>[[42, 694, 927, 757]]<|/det|>
+Daniel Tan Division of Medical Oncology, National Cancer Centre Singapore https://orcid.org/0000- 0002- 6514- 6786  
+
+<|ref|>text<|/ref|><|det|>[[42, 763, 307, 803]]<|/det|>
+Subhra BISWAS Human Innate Immunity Lab  
+
+<|ref|>text<|/ref|><|det|>[[42, 809, 707, 850]]<|/det|>
+Owen Rackham University of Bristol https://orcid.org/0000- 0002- 4390- 0872  
+
+<|ref|>text<|/ref|><|det|>[[42, 855, 707, 896]]<|/det|>
+N. Gopalakrishna lyer ( \(\square\) gopaliyer@singhealth.com.sg) National Cancer Centre Singapore https://orcid.org/0000- 0002- 8812- 6219
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 940, 87]]<|/det|>
+Keywords: pre-metastatic, single-cell genomics, targeted therapy, t-cell receptor, cytotoxic T-lymphocytes, EMT  
+
+<|ref|>text<|/ref|><|det|>[[44, 105, 330, 125]]<|/det|>
+Posted Date: October 22nd, 2021  
+
+<|ref|>text<|/ref|><|det|>[[42, 144, 463, 163]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 960593/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 181, 911, 224]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[75, 84, 881, 129]]<|/det|>
+1 Single cell analysis of early metastasis identifies targetable tumor subpopulation and mechanisms of immune evasion in squamous cell cancers  
+
+<|ref|>text<|/ref|><|det|>[[75, 152, 881, 211]]<|/det|>
+3 Authors: Hong Sheng Quah \(^{1,2*}\) , Elaine Yiqun Cao \(^{3*}\) , Lida Suteja \(^{1}\) , Hui Sun Leong \(^{1}\) , Fui Teen Chong \(^{1}\) , Constance H Li \(^{1}\) , Shilpi Gupta \(^{4}\) , Camille Arcinas \(^{1,2}\) , John F Ouyang \(^{3}\) , Vivian Ang \(^{4}\) , Daniel SW Tan \(^{1,2,5}\) , Subhra K Biswas \(^{4}\) , Owen JL Rackham \(^{3}\) and N Gopalakrishna Iyer \(^{1,2,6,7\#}\)  
+
+<|ref|>sub_title<|/ref|><|det|>[[75, 235, 195, 250]]<|/det|>
+## 6 Affiliation:  
+
+<|ref|>text<|/ref|><|det|>[[75, 263, 741, 472]]<|/det|>
+7 1 Cancer Therapeutics Research Laboratory, National Cancer Centre Singapore, Singapore  8 2 Academic Clinical Program in Oncology, Duke- NUS Medical School, Singapore  9 3 Program in Cardiovascular and Metabolic Disorders, Duke- NUS Medical School, Singapore  10 4 Singapore Immunology Network, Singapore  11 5 Division of Medical Oncology, National Cancer Centre Singapore, Singapore  12 6 Department of Head and Neck Surgery, National Cancer Centre Singapore, Singapore  13 7 Division of Medical Sciences, National Cancer Centre Singapore, Singapore  14 15 1 These authors contributed equally to this study  16 1 Corresponding author  
+
+<|ref|>sub_title<|/ref|><|det|>[[75, 497, 258, 511]]<|/det|>
+## 17 Correspondence to:  
+
+<|ref|>text<|/ref|><|det|>[[70, 533, 881, 630]]<|/det|>
+18 N Gopalakrishna Iyer, 11 Hospital Crescent, National Cancer Centre Singapore, Singapore 169610.  19 Email: gopaliyer@singhealth.com.sg, Tel: +65- 64368000, Fax: +65- 62257559  20 Keywords: pre- metastatic, single- cell genomics, targeted therapy, t- cell receptor, cytotoxic T- lymphocytes, EMT
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[66, 120, 884, 380]]<|/det|>
+Profiling tumors at single- cell resolution provides an opportunity to understand complexities underpinning lymph- node metastases in head and neck squamous- cell carcinoma. Single- cell RNAseq (scRNAseq) analysis of cancer- cell trajectories identified a sub- population of pre- metastatic cells, driven by actionable pathways including AXL and AURK. Blocking these two proteins blunted tumor invasion in patient- derived cultures. Furthermore, scRNAseq analyses of tumor- infiltrating CD8+ T- lymphocytes showed two distinct trajectories to T- cell dysfunction, corroborated by their clonal architecture based on single- cell T- cell receptor sequencing. By determining key modulators of these trajectories, followed by validation using external datasets and functional experiments, we uncovered a novel role for SOX4 in mediating T- cell exhaustion. Finally, interactome- analyses between pre- metastatic tumor- cells and CD8+ T- lymphocytes uncovered a putative role for the Midkine pathway in immune- modulation; this was confirmed by scRNAseq of tumors from humanized mice. Aside from specific findings, this study demonstrates the importance of tumor heterogeneity analyses in identifying key vulnerabilities during early metastasis.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 209, 98]]<|/det|>
+## Introduction  
+
+<|ref|>text<|/ref|><|det|>[[115, 105, 882, 384]]<|/det|>
+In most solid tumors development of lymph node metastasis portends poor outcomes, pre- dating distant metastasis \(^{1 - 3}\) . In head and neck squamous cell cancers (HNSCC), these patients are treated with curative intent by surgery and radiation therapy with the prime objective of eradicating existing and future disease by depleting clones with a metastatic potential \(^{4,5}\) . Metastasis is a continuum of phenotypes ranging from pre- metastatic features (eg lympho- vascular invasion), circulating tumor cells/emboli, microscopic lymph node deposits, gross nodal involvement and adjacent soft- tissue invasion, oligo- metastasis and finally, full blown distant metastasis \(^{6}\) . Most studies focus on the terminal event, highlighting the role of definitive epithelial- mesenchymal transition (EMT); however bulk analyses in HNSCC suggests that EMT does not appear to be a pre- requisite for lymph node dissemination \(^{7 - 11}\) . Recent studies have also highlighted that EMT itself exists as a spectrum, and tumor cells exhibit a significant amount of plasticity which may account for the range of clinical manifestations observed \(^{12,13}\) . Single- cell analyses have the ability to resolve both issues: identification of rare clones with true metastatic potential and identifying pathways and vulnerabilities that can be exploited in the clinical setting to prevent further dissemination of these.  
+
+<|ref|>text<|/ref|><|det|>[[115, 388, 882, 534]]<|/det|>
+The role of the immune system during the metastatic cascade is gaining clinical relevance with current advancements in checkpoint blockade therapies \(^{14}\) . This is especially pertinent in the context of lymph node metastasis, as lymph nodes are believed to be the main organ for T- cell priming, expansion and trafficking \(^{15}\) . Understanding the mechanisms by which tumors evade immune- based killing within lymph nodes is critical to target early metastases \(^{16 - 19}\) . Again, this can be addressed by single- cell analyses by defining the immune landscape, and in- depth dissection of interactions involved during immune evasion at the primary and nodal sites.  
+
+<|ref|>text<|/ref|><|det|>[[115, 538, 882, 621]]<|/det|>
+Here, we profiled primary and early (nodal) metastatic HNSCC tumors using single- cell RNAseq (scRNAseq) and TCRseq (scTCRseq) with two major objectives: to identify metastatic tumor subpopulations and identification of targetable vulnerabilities, and to determine the evolutionary trajectory of tumor- targeting T- cells as well as dissecting pathways employed by tumors to evade immune destruction during nodal dissemination.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 122, 684, 138]]<|/det|>
+## Single-cell transcriptional states of primary and lymph node metastasis in HNSCC  
+
+<|ref|>text<|/ref|><|det|>[[115, 159, 882, 286]]<|/det|>
+To delineate 'whole- tumor' single- cell landscapes in primary tumors and lymph node metastases, we developed a protocol to rapidly process freshly resected tissue for single- cell RNA sequencing (scRNAseq) and establishing primary cultures (Figure 1A) \(^{20,21}\) . Tumors were harvested from fourteen treatment- naive patients with locally advanced, HPV- negative HNSCC from primary and cervical lymph nodes (Supplementary Table S1 and S2). Seven pairs were processed for scRNAseq and single- cell T- cell receptor sequencing (scTCRseq), while primary cultures were successfully established for seven.  
+
+<|ref|>text<|/ref|><|det|>[[115, 305, 882, 715]]<|/det|>
+scRNAseq data for fresh tumors describes 53,459 cells (3,553- 11,308 per patient) and 23,148 genes, with a median of 776 genes per cell (details on quality controls steps in Methods and Supplementary Figure 1A- B). Using Seurat v3.0, the data was normalized, pooled, and clustered (Figure 1B). Canonical markers were used to broadly annotate these populations into: epithelial (KRT7, KRT17), salivary (STATH), fibroblasts (COL1A2), endothelial (PECAM) and immune (PTPRC) cells (Figure 1B and Supplementary Figure 1C). Fibroblasts were further subdivided into cancer associated fibroblasts (CAFs; MMP2) and myofibroblasts (ACTA2), while immune cells were organized into T- (CD3E, NKG7), NK- (NKG7, XCL2), B- (CD79A), plasma- (IGHG1), mast- (TPSAB1), conventional (LAMP3) and plasmacytoid (LILR4) dendritic cells, as well as macrophages/monocytes (CD163). These were well- distributed across samples from all patients, apart from salivary cells, which were only observed in one patient, likely due to harvest of adjacent parotid gland tissue (HN263). However, there were differences in composition between primary and metastatic sites (Figure 1C), with higher proportions of CAFs and TAMs in the primary tumor, versus more B- cells, plasma cells and dendritic cells at the metastatic sites, typical of a lymph node. These were similar to cellular composition proportions derived from bulk data from TCGA (Supplementary Figure 1D). Inferred copy number variant analyses on the epithelial population showed that aneuploidy was evident in \(>95\%\) of cells validating that this population comprised cancer cells (Figure 1D and Supplementary Figure 1E). Copy number alterations (CNAs) were further analyzed using the CopyKat algorithm \(^{22}\) , and identified those frequently observed in HNSCC \(^{23}\) , including gains across chromosomes 7 and 8q and loss of 3p and 5q (Supplementary Figure 1F). Significant overlap of CNAs was also noted between the primary and metastatic sites in each patient (Supplementary Figure 1G).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 770, 790, 787]]<|/det|>
+## Tumor cells demonstrate varying degree of epithelial-mesenchymal transition during metastasis  
+
+<|ref|>text<|/ref|><|det|>[[115, 806, 882, 910]]<|/det|>
+We next focused on tumor cells (total of 6,115 cells & 17,784 genes) by extracting only the epithelial population with aneuploidy. Using Seurat 3.0, we pooled and re- analysed this subset, visualized as distinct clusters for each individual patient, with varying degree of overlap across cells from primary and nodal sites (Figure 2A and Supplementary 2A). Tumor cell data can be accessed and interrogated as an interactive web application via the following Shiny app (http://hnc.ddnetbio.com/). Tumors from patients HN242, HN257 and HN272 show
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 881, 144]]<|/det|>
+significant overlap in tumor cells derived from both sites, while patients HN251 and HN279 show distinct site- specific sub- clusters. When comparing EMT gene markers in primary vs nodal metastases populations, nodal tumor cells had higher EMT scores compared to the primary in all patients except HN257 (Figure 2B).  
+
+<|ref|>text<|/ref|><|det|>[[115, 164, 882, 595]]<|/det|>
+To identify the pre- nodal metastases subpopulation in primary tumors, we built trajectories using Monocle 2.0, and labelled the origin and direction based on the ground truth of site (ie primary tumor presumed to pre- date nodal disease), incorporating EMT- scores, and CytoTRACE (see Methods). The latter is a tool to determine degrees of differentiation, assuming de- differentiation co- occurs with the metastatic phenotype<sup>24,25</sup>. This approach was effective in identifying pre- nodal cells in patients HN242, HN251, HN272 and HN279 (Figure 2C- D and Supplementary Figure 2C, 2G and 2H). For patients HN251 and HN279, pseudo- time ordering demonstrated an ordered, progressive, step- wise transition from primary to nodal disease. Nodal tumor cells largely dominate the end of the trajectory with higher CytoTrace scores. Major pathways over- represented across pseudotime include epithelial de- differentiation, oxidative phosphorylation and EMT (Figure 2E). Even in more complex trajectories such as HN272, the same approach was used to determine the likely trajectory to lymph node metastases, and identify sub- populations of primary cells (pre- nodal cells) that are similar to and likely gave rise to the metastatic phenotype (Supplementary Figure 2C). We next applied GeneSwitches<sup>26</sup> to identify actionable genes associated with the trajectory from primary to pre- nodal cells; these identified AXL, Aurora kinase, TYMS and STAT2 at potentially critical genes in this process (Figure 2F and Supplementary Figure 2D- F). This approach was validated on an external dataset comprising scRNASeq data from 5 tumors from primary and nodal sites available for analyses (2076 cells) (Supplementary Figure 2H- P)<sup>12</sup>. In three of these (p25, p26 and p28), EMT was higher in nodal tumor cells compared to the primary, hence could be resolved using the method described to identify a pre- nodal subpopulation (Supplementary Figure 2P- R). Several actionable genes identified through GeneSwitches appear to be implicated in this dataset as well: AXL (p25, p26, P28), STAT2 (p25, p26) and AURK (p26, p28) (Supplementary Figure 2U).  
+
+<|ref|>text<|/ref|><|det|>[[115, 615, 882, 872]]<|/det|>
+In contrast, analyses of patient HN257 was more complicated as the primary tumor had higher EMT scores than nodal tumor cells, and tumor trajectories were haphazard with no directionality (Supplementary Figure 2H). Cytotrace showed a distinct de- differentiated sub- population in the primary tumor that had high EMT scores and expression of SNAI2 (Figure 2G and Supplementary Figure 2I- J). We hypothesized that this was an aggressive, rapidly evolving tumor subpopulation. Differential expression analyses identified a panel of 132 up- regulated and 45 down- regulated genes in this subpopulation involved in oxidative phosphorylation and tumor metabolism, and immune evasion, respectively (Figure 2H, Supplementary Table S3). Based on these gene sets, tumors in TCGA with the same signature (based on RNASeq data) had significantly poorer outcomes (Figure 2I and Supplementary Figure 2K). In the validation scRNASeq dataset above, two of the tumors (p5 and p20) also showed a similar trend, with specific subpopulations in the primary tumor with high EMT scores (Supplementary Figure 2S- T). Therefore, we postulate that in these tumors, distinct sub- populations in the primary tumor showed a more aggressive phenotype, that likely evolved after nodal dissemination had occurred.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 84, 552, 100]]<|/det|>
+## Identifying vulnerabilities to target pre-metastatic tumor cells  
+
+<|ref|>text<|/ref|><|det|>[[115, 111, 883, 303]]<|/det|>
+We then proceed to test whether targets identified in this manner presented an opportunity for therapeutic intervention. scRNAseq using the C1 platform was performed on patient- derived cultures (PDCs) from primary and nodal metastatic sites \((n = 7\) pairs). The data was processed using Seurat 3.0 and PAGODA (pathway and gene set overdispersion analysis) (Figure 3A and Supplementary Figure 3A- B). We derived scRNAseq data for a total of 1,317 cells and 55,216 genes. Similar to above, tumor- cell clusters were based on individual patients. However, PDCs demonstrated distinct separation between primary and metastatic cells, with EMT as one of the major differentiating principal component pathways (Figure 3B and Supplementary Figure 3A- B). Here, pre- nodal cells in HN137, HN159 and HN220 were identified as small primary subpopulations that clustered with metastatic cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 315, 883, 723]]<|/det|>
+Differential expression analyses for these pre- nodal populations identified AXL (in HN137) and AURKB (in HN159 and HN220) as putative actionable targets (Figure 3C and Supplementary Table S4- 6). Expression of these genes was validated using immunohistochemistry or immunofluorescence in both PDCs and respective tumor tissue, and this was recapitulated on flow cytometry for AXL (HN137) and AURK (HN159 and HN220), respectively (Figure 3D and Supplementary Figure 3C- D). In HN137, expression of protein and transcript AXL was detected in a majority of metastatic cells compared with only a small sub- population of primary cells. Similarly, for HN159 and HN220, AURKB expression was significantly lower in metastatic cells, compared to primary cells. We focused on AXL and AURKB because both have specific inhibitors: BGB324 targeting cells with high AXL expression, and barasertib (pan- AURK inhibitor) targeting cells with limiting AURKA/AURKB levels. There were no differences in clonogenicity between primary and metastatic cultures from patient HN137 treated with BGB324, nor HN159 and HN220 treated with barasertib (Supplementary Figure 3E- G). In contrast, all three metastatic lines HN137, HN159 and HN220 (treated with their respective drugs) demonstrated lower cell migration/invasion compared to untreated cultures, measured by scratch and Boyden chamber invasion assays (Figure 3E- G): AXL- inhibition significantly reduced invasive potential of both primary and metastatic cells of HN137 (Figure 3E) while AURK- inhibition significantly reduced the invasive potential of only metastatic cells of HN159 and HN220 (Figure 3F and G). As AXL is a surface membrane protein, primary cells were sorted into AXL low-, medium- and high- expressing cells. As predicted, BGB324 specifically inhibited invasion only in the AXL- high primary subpopulation compared to AXL- low cells (Figure 3H). These data indicate AXL and AURKB play major roles in invasion and provide an opportunity for specific anti- metastatic therapy.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 765, 765, 781]]<|/det|>
+## Evolution of CD8+ T-cells derived from analysis of primary tumor and lymph node metastasis  
+
+<|ref|>text<|/ref|><|det|>[[117, 793, 882, 896]]<|/det|>
+CD3+ T- cells form one of the major subpopulations sequenced at both primary and nodal sites. Data from 10,168 cells (covering 13,729 genes) were pooled, analyzed using Seurat, and visualized as ten distinct T- cell clusters (Figure 4A). The identity of each cluster was delineated based on differential gene expression of known T- cell markers (Figure 4B and Supplementary Figure 4A- B). Some were distinct for CD4+ cells (Tregs and Tfh) and CD8+ cells (Pre- dysfunctional, Dysfunctional, Proliferative), while others comprise both CD4+ and CD8+ lineages
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 881, 145]]<|/det|>
+(Naive- like and Transitional). Majority of naive- like cells were derived from nodal tissue while the remaining clusters appear to have equal representation from the primary and nodal metastatic sites (Figure 4B and Supplementary Figure 4C).  
+
+<|ref|>text<|/ref|><|det|>[[115, 155, 883, 608]]<|/det|>
+CD8+ T- cells (total of 3,387 cells, 11,847 genes) were extracted from this pooled T- cell dataset and re- analyzed after regression for cell cycle- driven artefacts to identify lineage- based clusters. CD8+ T- cell data can be accessed and interrogated as an interactive web application using the following Shiny app (http://hnc.ddnetbio.com/). Six distinct clusters were labelled as naive, transitional, tissue- resident memory, pre- dysfunctional, proliferative and late dysfunctional based on canonical markers (Figure 4C- D). Using Slingshot, we performed trajectory analyses on the CD8+ T- cells using the CXCL13- high, LAYN- high exhausted/senescent population as the endpoint \(^{27}\) , and this identified two convergent trajectories (Figure 4E). Expression plots across Trajectory 1 showed a progressive loss of naive markers, gradual gain of dysfunctional (and senescent) markers and an intervening proliferative 'burst', that likely reflects expanding clones of tumor targeting CD8+ cells (Figure 4F). Specifically, this lineage suggests a scenario where naive CD8+ T- cells from lymph nodes or circulation were trafficking into the primary tumor with loss of circulating markers KLF2, SELL and CCR7, gain of tissue resident marker CD103/ITGAE, progressive decline in the expression of naive genes TCF7, IL7R, CCR7, and gradual gain of dysfunctional markers (TIM3, CTLA4, TIGIT, CXCL13, LAYN) with an intermediary proliferative burst with high levels of MK167, TOP2A, TYMS (Figure 4B, 4E- F). This is also reflected by progressive increase from GZMK to GZMB, PRF1, and IFNG in pre- dysfunctional to dysfunctional cells. In contrast, the trajectory of tissue- resident memory (TRM) to dysfunctional cells (Trajectory 2) shows fewer genes being activated as the expression level of many of the tissue resident (ITGAE), dysfunctional (CTLA4) and granzymes (GZMs) genes were already upregulated (Figure 4B). The Geneswitches algorithm was applied to trajectory 1 (naive- to- dysfunction) to predict key gene expression changes across pseudotime and identify factors that could account for these (Figure 4G) \(^{26}\) . Our results indicate the major nodes appear to be an early loss of KLF2, intermediate increase in NKG7 and late increase in SOX4, DUSP4 and RBPJ (Figure 4G- H).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 648, 619, 665]]<|/det|>
+## Modulating genes driving tumor-targeting cells dysfunction/exhaustion  
+
+<|ref|>text<|/ref|><|det|>[[115, 675, 883, 893]]<|/det|>
+Based on the data above, expression of SOX4, DUSP4 and RBPJ appears to coincide with the transition between dysfunction and exhaustion, but whether these genes modulate the process remains untested. We attempted to validate these findings in two separate datasets. Re- analysis of data from Puram et al (scRNAseq from 542 CD8+ T cells) showed that expression levels of SOX4 and RBPJ were higher in dysfunctional CD8 cell populations, while DUSP4 expression was more generalized (Figure 5A and Supplementary Figure 5A- C) \(^{12}\) . The second scRNAseq dataset comprised T- cells obtained from cutaneous squamous- cell carcinoma patients before and after treatment with PD1- blockade (Supplementary Figure 5D) \(^{28}\) . Here, all three genes showed higher expression in the exhausted CD8 subpopulation in this dataset (Figure 5B and Supplementary Figure 5E). However, only levels of SOX4 and DUSP4 were reduced after PD1 blockade, where there is expected re- activation of tumor- targeting clones and reduction in the exhaustion phenotype (Figure 5C). Combining these
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 882, 253]]<|/det|>
+results, SOX4 appears to be the most likely gene associated during the transition between pre- dysfunction to dysfunction/exhaustion. To test whether SOX4 plays a causative role in T- cell dysfunction, we performed RNAi- based knock- down on activated PBMCs. Cells were transfected with Accell pooled siRNA against SOX4, DUSP4 or non- targeting siRNA as controls, activated with anti- CD3/CD28 microbeads and harvested for flow cytometry. Remarkably, SOX4 knockdown resulted in a reduction in senescent CD57+ and dysfunctional PD1+ and CD39+ populations, compared to DUSP4 and control siRNAs (Figure 5D and Supplementary Figure 5F- G). Taken together, these data provide functional validation for our CD8+ T- cell trajectory mapping and implicates SOX4 as an important driver of T- cell dysfunction/exhaustion.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 293, 744, 310]]<|/det|>
+## Establishing clonal architecture in \(\mathsf{CD8 + }\) T-cells using single-cell T-cell receptor sequencing  
+
+<|ref|>text<|/ref|><|det|>[[115, 321, 882, 536]]<|/det|>
+Clonal identifiers obtained by TCR analysis allows for elucidation of CDR3 sequences as well as providing a unique dataset to infer the lineage structure of T- cells. Specifically, our current dataset can be used to model clonal selection and amplification across the \(\mathsf{CD8 + }\) T- cell subpopulations and trajectories. We recovered productive TCR- alpha and TCR- beta sequences from 1,461 and 1,948 cells, respectively, and identified 1,590 unique TCR sequences. No shared clones were found between patients, with unique TCRs for each patient. Clonal expansion was seen in \(17.39\%\) of \(\mathsf{CD8 + }\) cells, and clone size ranged from 2 to 60 cells per clone (Figure 5E, Supplementary Figure 5H and 5I). Clonal overlap between the two different sites for each tumor (primary and lymph node) was demonstrated in patients HN257 and HN272 (Figure 5F). There was a progressive increase in clonality across the dysfunctional gradient, with evidence of single naive or TRM- derived clones subsequently expanding to give rise to multiple dysfunctional clones that span these trajectories (Figure 5F and 5G).  
+
+<|ref|>text<|/ref|><|det|>[[115, 546, 882, 868]]<|/det|>
+There appeared to be patient- specific biases for one trajectory over the other. For example, there are \(\mathsf{CD8 + }\) T- cell clones in patient HN272 that followed a naive- dysfunction trajectory (Trajectory 1), with expansion of lymph node derived naive clonotypes, migrating to the primary site and captured there along a dysfunctional gradient (pre- dysfunctional, proliferative and then late- dysfunction) (Figure 5F). This supports a previous observation which suggests that circulation is one of the major sources of tumor- targeting dysfunctional cells, which in this case is the regional lymphatics draining nodal tissue<sup>28</sup>. In contrast, in patient HN263 and selected \(\mathsf{CD8 + }\) T- cell clones in patient HN272, the dysfunctional gradient appears to comprise of tissue resident memory (TRM) cells derived from the primary tumor, which amplified into putative tumor- targeting clonotypes (Figure 5F). This is consistent with a model of ongoing differentiation and proliferation of dysfunctional T- cells at the tumor site itself<sup>29</sup>. It is likely that both mechanisms contribute to the dysfunction gradient, sometimes even within the same patient. For example, lineage tracing in HN257 and HN272 demonstrates extensive trafficking and interplay between the primary site and lymph node: there is evidence of lymph node- derived naive cells expanded in the primary site as expected, but also surprisingly TRM cells expanding and subsequently migrating to the lymph node (Figure 5F and 5G). This scTCR data adds intriguing complexity to concepts of clonal expansion and lineage structure in a treatment naive setting.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 456, 100]]<|/det|>
+## Pre-nodal cells and immune micro-environment  
+
+<|ref|>text<|/ref|><|det|>[[115, 111, 883, 435]]<|/det|>
+Our analyses identified a pre- nodal sub- population in primary tumors with intrinsic properties of invasion and migration. However, metastasis also requires acquisition of an immune evasion phenotype. To test whether the pre- nodal cells identified above demonstrated specific immune- modulatory phenotypes, we subjected three tumors (from our study) and two tumors (from the Puram dataset) each with a minimum RNAseq dataset to interactome analyses using Cellchat. To do this, we divided primary tumor cells into two subpopulations (primary and pre- nodal) and analyzed the interactions of these two tumor subpopulations with \(\mathsf{CD8 + }\) \(\mathsf{CD4 + }\) and T- reg lymphocytes and TAMs. For HN251, HN272 and HN279, the analysis showed similar trends in primary to pre- nodal malignant cells, with increasing interactions between the pre- nodal subpopulation and T- lymphocytes, specifically with \(\mathsf{CD8 + }\) cells (Figure 6A). The analyses implicated a number of pathways that were differentially modulated by primary versus pre- nodal populations on T- lymphocytes (Supplementary Figure 6A- C). In particular, the interaction between Midkine (MDK, secreted by tumor cells) and a number of MDK- receptors (ITGA4, ITGA6, ITGB1, NCL, LRP1) on \(\mathsf{CD8 + }\) T- cells appears to be a recurrent immunosuppressive pathway seen across all three patients (Figure 6B). Applying the same approach to the external dataset also implicated the MDK pathway as being differentially activated by the pre- nodal population in one (p17) out of two tumors tested (Figure 6B and Supplementary Figure 6D- E).  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 882, 635]]<|/det|>
+Recent published data suggest that MDK- driven modulation is important for immune evasion in melanomas with activation of NFKB and its downstream pathways<sup>30</sup>. To test whether MDK- driven immune suppression dampens the effect of immune checkpoint blockade (ICB) therapy, we developed a humanized mouse model engrafted with pre- nodal cells from the tumor of patient HN279, and treated these with PD1- blockade. As expected, the majority of cancer- cells expressed MDK (Figure 6C- 6D and Supplementary Figure 6F- G), together with a number of genes associated with the pre- nodal phenotype (eg SNAI2, AXL, STAT2) that were unaffected by ICB (Figure 6E and Supplementary Figure 6H). In contrast, expression of AURKB and TOP2A (cell cycle genes) in cancer cells was significantly downregulated after pembrolizumab treatment (Figure 6E), indicating a reduction in cancer cell proliferation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 647, 882, 904]]<|/det|>
+Analyses of the \(\mathsf{CD8 + }\) T- cell fraction revealed naive, TRM, transitional, proliferative and dysfunctional/exhausted subpopulations, with an additional cytotoxic populations (likely bystander) (Figure 6F and Supplementary Figure 6I). \(\mathsf{CD8 + }\) cells from mice treated with pembrolizumab showed reduction in naive, dysfunctional and memory with concomitant increase in proliferative, cytotoxic/bystander, tissue resident subpopulations compared to untreated mice (Figure 6G). These changes suggest a re- invigoration and re- activation of dysfunctional and memory, respectively, into tumor- targeting cells<sup>29</sup>. Remarkably, analyses of MDK receptor- expressing CD8 cells (ITGA4, ITGB1, NCL) showed the opposite trend, with an increase in dysfunctional and reduction in the proliferative (tumor- targeting) populations (Figure 6H and Supplementary Figure 6J). These findings suggest MDK- signaling promotes immune- suppression, that abrogates re- invigoration by PD1- blockade. Indeed, these changes were also associated with NFKB1 activation which is significantly higher in the dysfunctional CD8 population after pembrolizumab treatment (Figure 6I). Moreover, plotting the expression levels of several MDK- receptors (ITGA4, ITGB1, NCL) with NFKB1 show a good correlation in gene expression in \(\mathsf{CD8 + }\) T- cells where the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 83, 881, 127]]<|/det|>
+276 RNA could be quantified (Figure 6J). Taken together, these results implicate MDK- signaling as a pathway through which pre- nodal cells evade CD8- mediated immune- editing.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 114, 883, 567]]<|/det|>
+Currently available algorithms analyzing single- cell data have the ability to construct evolutionary trajectories, which are especially powerful in studying specific events in space (eg relationships between different tumor sites, eg primary vs lymph node metastasis) and time (eg pre- and post- treatment analysis)12,28. Here, we applied these to explore early lymph node metastasis across tumor and immune sub- compartments within the tumor. Analysis of tumor cells shows that nodal metastasis is an early event, where canonical epithelial- to- mesenchymal transition is less apparent than postulated. Our findings support previous studies that suggest EMT is not an all- or- none phenomenon, but instead occurs in graded levels 31,32. This is in contrast to in vitro systems (including our own) where cultured tumor cells from lymph nodes display more canonical features of EMT33. Despite overlap between tumor cells derived from primary and nodal sites, trajectory mapping could define evolutionary pathways at individual tumor levels, although this process require a combination of trajectory algorithms, scoring for aggressiveness (based on EMT and stemness) and knowledge of the ground truth. These have expanded the results of previous studies in the identification of a pre- nodal or metastatic population 12, and importantly identified actionable drivers of that could be targeted for anti- metastatic therapy, in this case AXL and AURK. Targeting AXL would not only prevent pathways involved in dissemination, but presumably reduce tumor heterogeneity by targeting the specific clones34. The role of aurora kinases is less clear; rather than impacting the metastatic process, it is possible that this vulnerability reflects a generalized reduction in cell cycling that occurs during EMT with a concomitant sensitivity to all cell cycle inhibitors. We recently demonstrated the same phenomenon during drug resistance: reduction in cell proliferation, limited AURK expression and sensitivity to inhibitors of AURK and other cell cycle targets35. Nevertheless, the ability to profile tumors and identify vulnerabilities in metastasis- inducing clones is an attractive notion, with increasing interest in low- dose, long term anti- metastatic therapy.  
+
+<|ref|>text<|/ref|><|det|>[[115, 587, 883, 909]]<|/det|>
+Alignment of CD8+ T lymphocyte populations is driven by existing knowledge on T- cell maturation. The fact that we could pool data across different patients increased the number of cells available and in itself was a form of validation. The alignment was further supported by single- cell VDJ sequencing, which reinforced trajectories from naive or memory populations, towards clonally expanded, dysfunctional and potentially tumor targeting CD8+ subpopulations These supported the notion that both naive CD8+ cells from adjacent lymph nodes and tissue resident CD8+ T- cells were sources of expanded tumor- infiltrating CD8+ T cells, and trafficking was bidirectional. Strikingly, this trajectory could be used to identify novel modulators of T- cell dysfunction by studying gene expression changes along pseudotime, and was used to identify SOX4 as novel driver of dysfunction in CD8 cells. Interactome analyses performed to identify signaling networks within CD8+ T cells during early metastasis converged onto the MDK pathway. Remarkably, in a humanized mouse model, MDK signaling was associated with a reduced ability to reinvigorate exhausted T- cells. This is supported by a recent publication which identified that the MDK pathway could abrogate immune reactivation by ICB therapy in melanoma, and this could be reversed using MDK- specific inhibitors30. In a similar context, MDK- inhibition could be explored in the prevention and treatment of tumor metastasis in HNSCC and add synergy to PD1- blockade which is the current standard of care in metastatic HNSCC.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 883, 210]]<|/det|>
+In conclusion, we applied single- cell genomics to uncover pathways and mechanisms that mediate early nodal metastasis in HNSCC. The data presented here shows that early metastasis is a much more nuanced process than previously presumed. Collectively these indicate the discovery potential of single cell studies and existing computational tools, when applied to specific clinical contexts and questions. Future studies will focus on more specific tumor subpopulations including CD8+ cells and the impact of treatment on tumor recurrence and metastasis.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 85, 185, 99]]<|/det|>
+## Methods  
+
+<|ref|>text<|/ref|><|det|>[[118, 115, 340, 130]]<|/det|>
+Tumor collection and processing  
+
+<|ref|>text<|/ref|><|det|>[[115, 141, 882, 288]]<|/det|>
+Patient tumors were harvested in the operating room and transported to the lab for processing within 30 minutes. Tumors were a priori confirmed histologically to be HNSCC and patients were consented prior to surgery. This study is approved by SingHealth Centralized Institutional Review Board (CIRB: 2014/2093, 2018/2512, 2016/2757). All tumors were dissociated using the gentleMACS™ Octo system (Miltenyi Biotech, Bergisch Gladbach, Germany) as described in manufacturer's protocol. These were subjected to filtration, washing and magnetic bead separation, where required, prior to single cell capturing (details in Supplementary methods).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 331, 310, 345]]<|/det|>
+## Patient-derived cell cultures  
+
+<|ref|>text<|/ref|><|det|>[[115, 357, 882, 570]]<|/det|>
+Cultures were established as previously described<sup>20,21</sup>. Cells were maintained in complete RPMI (C/RPMI) containing \(10\%\) FBS, \(1\%\) pen- strep, \(1\%\) anti- mycotic and a humidified incubator at \(37^{\circ}C\) with \(5\%\) CO2. All lines were tested and confirmed to be free of mycoplasma using an EZ- PCR Mycoplasma Detection Kit (Biological Industries, Kibbutz Beit Haemek, Israel) at the time of experiments. Cells were processed for scRNAseq and for immunostaining as described in Supplementary methods. Invasion assays cells were treated with or without \(0.25 \mu M\) of bementinib (BGB324) or \(0.25 \mu M\) of barasetrib (both from Selleck Chem, Houston, TX), then seeded on an \(8 \mu m\) filter membrane within a 24- well transwell insert (Corning, New York City, NY), with C/RPMI at bottom of wells of 24- well Falcon TC Companion Plate (Corning, New York City, NY). After 72hrs, the bottom of each inserts was fixed and stained for quantification of invaded cells. Cell invasion area was determined by quantifying the area with crystal violet staining using the ImageJ software.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 612, 294, 626]]<|/det|>
+## Humanized mouse model  
+
+<|ref|>text<|/ref|><|det|>[[115, 639, 881, 743]]<|/det|>
+Sixteen NOG- EXL (hGM- CSF/hIL- 3 NOG) mice (hNOG- EXL), pre- engrafted with human CD34+ hematopoietic stem cells, were procured from CIEA- SlgN. At 16 weeks post- engraftment, mice were injected subcutaneously with cells from HN279, and treated intraperitoneally with \(12.5mg / kg\) of pembrolizumab or with phosphate buffered saline (PBS) on day 17, 19, 21 and 24. Mice were euthanized on day 25 and tumors harvested for dissociation and preparation for scRNAseq as described.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 786, 492, 801]]<|/det|>
+## Small-interfering RNA knock-down of SOX4 and DUSP4  
+
+<|ref|>text<|/ref|><|det|>[[115, 821, 881, 903]]<|/det|>
+Peripheral blood mononuclear cells (PBMCs) from healthy donors were cultured at a density of \(0.5 - 1 \times 10^{6}\) cells/ml in 24- well plate (Corning, New York City, NY), containing TexMACS Medium and T- Cell TransAct (both from Miltenyi Biotech, Bergisch Gladbach, Germany) at 1:200 dilution. A final concentration of \(1 \mu M\) of Accell pooled small- interfering RNA (siRNA) targeting human SOX4 (Gene ID 6659) or DUSP4 (Gene ID 1846), or non
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 880, 121]]<|/det|>
+targeting siRNA (all from Dharmacon, Lafayette, CO) was added into respective wells. After 5 days of incubation, cells were harvested for flow cytometry.  
+
+<|ref|>text<|/ref|><|det|>[[115, 165, 226, 180]]<|/det|>
+Flow cytometry  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 882, 502]]<|/det|>
+For AXL surface staining, trypsinized cells were stained with fluorochrome- conjugated antibody recognizing AXL (#108724; R&D systems, Minneapolis, MN) or with mouse IgG1 isotype antibody (MOPC- 21; BD Biosciences, Franklin Lakes, NJ). For intracellular AURKB staining, trypsinized cells were fixed and permeabilized with a Foxp3/Transcription Factor Staining Buffer Set (eBioscience, San Diego, CA) according to the manufacturer protocol. After fixation, cells were stained with primary antibody recognizing AURKB (clone RM278; Invitrogen, Carlsbad, CA) or rabbit IgG1 isotype antibody (DA1E; R&D systems, Minneapolis, MN), and subsequently with goat anti- rabbit IgG secondary antibody conjugated to Alex Fluor 647 (#A32733; Waltham, MA). For siRNA knock- down PBMC experiments, harvested cells were stained with fluorochrome- conjugated antibodies recognizing CD57 (HNK- 1), LAG3 (11C3C65), CD39 (A1) and CD4 (OKT4) all from Biolegend, San Diego, CA; PD1 (J105) and CD8 (SK1) from eBioscience, San Diego, CA; and CD4 (SK3) from BD Biosciences, Franklin Lakes, NJ. These cells were stained for 30mins on ice in the dark with 2% BSA in PBS. Live/dead cells were distinguished using a Fixable Live Dead Blue Dead Cell Stain Kit (Thermofisher, Waltham, MA). Cells were acquired and analyzed using a BD FACSCanto II instrument and FlowJo v10.5.3 software (both from BD Biosciences, Franklin Lakes, NJ) respectively.  
+
+<|ref|>text<|/ref|><|det|>[[115, 550, 880, 587]]<|/det|>
+Generation of single cell gene expression and TCR libraries by droplet- based (10x system) and microfluidic- based technologies  
+
+<|ref|>text<|/ref|><|det|>[[115, 601, 882, 812]]<|/det|>
+The 5' gene expression (GEX) and TCR single cell RNA libraries from tumors were prepared using the 10x Chromium Single Cell V(D)J Reagent Kits (10x Genomics, Pleasanton, CA), as described in the manufacturer's protocol. Briefly, freshly dissociated tumor cells were sorted into CD45+ and CD45- fractions, mixed at a 1:1 ratio and loaded into the Single Cell A Chip for gel bead- in- emulsion (GEM) generation and barcoding, targeting for a cell recovery of 4000- 7000 cells per sample. Reverse transcription, cDNA amplification, GEX and TCR library construction were performed as described. For C1, single cell suspensions were loaded and captured using medium- sized (10- 17um) Fluidigm Integrated Fluidic Circuit (IFC) and a Fluidigm C1 instrument (Fluidigm, South San Francisco, CA), according to the manufacturer's protocol. cDNA product was harvested from the IFC, barcoded for individual cell identity and pooled. Sequencing was performed by an Illumina Hiseq 4000 (Illumina, San Diego, CA) with 151- bp single- ended or pair- ended reads.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[60, 82, 890, 395]]<|/det|>
+394 Data processing of single- cell RNA- seq libraries and clustering 395 scRNAseq reads were aligned to the GRCh38 reference genome and quantified using Cellranger count (10x 396 Genomics, version 2.2.0). Downstream analyses were performed using Seurat (version 3.1.5). For malignant- cell 397 analysis, we isolated subsets of cells identified as malignant cells based on broad clustering and reprocessed 398 using Seurat without patient alignment, since tumor cells tend to be patient specific. For T- cell clustering, we 399 isolated subsets of cells identified as T- cells based on broad clustering. Cells were then re- clustered using Seurat 400 alignment across patients similar as with previous analysis. For CD8+ T- cell clustering, CD8+ T- cells were 401 extracted from the T- cell clustering based on the following two criteria: 1) in Pre- dysfunctional, Dysfunctional 402 and Proliferative clusters, and with zero CD4 expression, 2) in Naive- like, Memory and Transitional clusters, with 403 zero CD4 and positive CD8 (either CD8A or CD8B) expression. TCR reads were mapped to 404 vdj_GRCh38_alts_ensembl- 3.1.0- 3.1.0 reference genome and quantified using cellranger count (10x Genomics, 405 version 3.1.0). Further details on Seurat analysis, UMAP visualization and use of the following algorithms: 406 InferCNV, Monocle, PAGODA, Slingshot, functional annotation, Geneswitches, Cytotrace and other analyses 407 tools, are described in detail in Supplementary methods.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 425, 245, 440]]<|/det|>
+## Statistical analysis  
+
+<|ref|>text<|/ref|><|det|>[[118, 452, 880, 490]]<|/det|>
+Statistical analysis was performed using GraphPad Prism software (GraphPad Software, Inc., San Diego, CA), or otherwise indicated in the figure legends and Supplementary methods.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[68, 84, 258, 99]]<|/det|>
+## 413 Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[66, 111, 883, 348]]<|/det|>
+We would like to thank all patients and families who contributed to this project. Additional technical support and access for 10x experiments were obtained from the Laboratory of Cell Therapy and Cancer Vaccine, National Cancer Centre Singapore (under Dr Han- Chong Toh), C1 experiments from the Laboratory of Cancer Epigenome (under Professor Bin- Tean Teh and Mr Cedric Ng), and humanized mouse experiments from the HuNIT IAF- PP program (Teja Celhar, Yunqian Zhao, Hui Chen Tay, Takeshi Takahashi and Jerry Chan, Singapore Immunology Network- A\*STAR). This project was funded through the following grants awarded to the respective investigators, for which we are extremely grateful: Khoo Postdoctoral Fellowship Award to QHS, National Medical Research Council (Singapore) Clinician Scientist Awards to NGI (NMRC/CSA/001/2016, MOH- 000325- 00), the Peter Fu Head and Neck Cancer Program (under the Oncology Academic Clinical Program, National Cancer Centre Singapore) to NGI, core funding by Singapore Immunology Network (A\*STAR) to SKB and a Singapore National Research Foundation grant [NRF- CRP20- 2017- 0002] to OR and JO.  
+
+<|ref|>text<|/ref|><|det|>[[66, 360, 92, 374]]<|/det|>
+425  
+
+<|ref|>text<|/ref|><|det|>[[66, 395, 883, 520]]<|/det|>
+Competing interest: NGI has/had a consulting or advisory role in PairX Therapeutics and Invitrocue PLC, and received honoraria from Kalbe Biotech and Agilent, all of which are outside this submitted work. DSWT received honoraria from Bristol- Myers Squibb, Takeda Pharmaceuticals, Novartis, Roche, and Pfizer; and has consulting or advisory role in Novartis, Merck, Loxo Oncology, AstraZeneca, Roche, and Pfizer. DSWT also received research funding from Novartis (Inst), GlaxoSmithKline (Inst), and AstraZeneca (Inst), outside this submitted work. None of the remaining authors have any other conflicts to report.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 120, 880, 878]]<|/det|>
+433 1 Steeg, P. S. Targeting metastasis. Nat Rev Cancer 16, 201- 218, doi:10.1038/nrc.2016.25 (2016). 434 2 Chaffer, C. L. & Weinberg, R. A. A perspective on cancer cell metastasis. Science 331, 1559- 1564, doi:10.1126/science.1203543 (2011). 436 3 Gupta, G. P. & Massague, J. Cancer metastasis: building a framework. Cell 127, 679- 695, doi:10.1016/j.cell.2006.11.001 (2006). 438 4 Chow, L. Q. M. Head and Neck Cancer. N Engl J Med 382, 60- 72, doi:10.1056/NEJMra1715715 (2020). 439 5 Iyer, N. G. et al. Randomized trial comparing surgery and adjuvant radiotherapy versus concurrent chemoradiotherapy in patients with advanced, nonmetastatic squamous cell carcinoma of the head and neck: 10- year update and subset analysis. Cancer 121, 1599- 1607, doi:10.1002/cncr.29251 (2015). 443 6 Mehlen, P. & Puisieux, A. Metastasis: a question of life or death. Nat Rev Cancer 6, 449- 458, doi:10.1038/nrc1886 (2006). 445 7 Heerboth, S. et al. EMT and tumor metastasis. Clin Transl Med 4, 6, doi:10.1186/s40169- 015- 0048- 3 (2015). 447 8 Brabletz, T., Kalluri, R., Nieto, M. A. & Weinberg, R. A. EMT in cancer. Nat Rev Cancer 18, 128- 134, doi:10.1038/nrc.2017.118 (2018). 449 9 Williams, E. D., Gao, D., Redfern, A. & Thompson, E. W. Controversies around epithelial- mesenchymal plasticity in cancer metastasis. Nat Rev Cancer 19, 716- 732, doi:10.1038/s41568- 019- 0213- x (2019). 451 10 Vettore, A. L. et al. Mutational landscapes of tongue carcinoma reveal recurrent mutations in genes of therapeutic and prognostic relevance. Genome Med 7, 98, doi:10.1186/s13073- 015- 0219- 2 (2015). 453 11 Tan, D. S. et al. Tongue carcinoma infrequently harbor common actionable genetic alterations. BMC Cancer 14, 679, doi:10.1186/1471- 2407- 14- 679 (2014). 455 12 Puram, S. V. et al. Single- Cell Transcriptomic Analysis of Primary and Metastatic Tumor Ecosystems in Head and Neck Cancer. Cell 171, 1611- 1624 e1624, doi:10.1016/j.cell.2017.10.044 (2017). 457 13 McFaline- Figueroa, J. L. et al. A pooled single- cell genetic screen identifies regulatory checkpoints in the continuum of the epithelial- to- mesenchymal transition. Nat Genet 51, 1389- 1398, doi:10.1038/s41588- 019- 0489- 5 (2019). 460 14 Wei, S. C., Duffy, C. R. & Allison, J. P. Fundamental Mechanisms of Immune Checkpoint Blockade Therapy. Cancer Discov 8, 1069- 1086, doi:10.1158/2159- 8290. CD- 18- 0367 (2018). 462 15 Mempel, T. R., Henrickson, S. E. & Von Andrian, U. H. T- cell priming by dendritic cells in lymph nodes occurs in three distinct phases. Nature 427, 154- 159, doi:10.1038/nature02238 (2004). 463 16 van der Leun, A. M., Thommen, D. S. & Schumacher, T. N. CD8(+) T cell states in human cancer: insights from single- cell analysis. Nat Rev Cancer, doi:10.1038/s41568- 019- 0235- 4 (2020). 466 17 Dunn, G. P., Old, L. J. & Schreiber, R. D. The three Es of cancer immunoediting. Annu Rev Immunol 22, 329- 360, doi:10.1146/annurev.immunol.22.012703.104803 (2004). 468 18 Dunn, G. P., Bruce, A. T., Ikeda, H., Old, L. J. & Schreiber, R. D. Cancer immunoediting: from immunosuppressive to tumor escape. Nat Immunol 3, 991- 998, doi:10.1038/nij1102- 991 (2002). 469 19 Hanahan, D. & Weinberg, R. A. Hallmarks of cancer: the next generation. Cell 144, 646- 674, doi:10.1016/j.cell.2011.02.013 (2011). 471 20 Chia, S. et al. Phenotype- driven precision oncology as a guide for clinical decisions one patient at a time. Nat Commun 8, 435, doi:10.1038/s41467- 017- 00451- 5 (2017). 472 21 Leong, H. S. et al. Targeting cancer stem cell plasticity through modulation of epidermal growth factor and insulin- like growth factor receptor signaling in head and neck squamous cell cancer. Stem Cells Transl Med 3, 1055- 1065, doi:10.5966/sctm.2013- 0214 (2014). 473 22 Gao, R. et al. Delineating copy number and clonal substructure in human tumors from single- cell transcriptomes. Nat Biotechnol 39, 599- 608, doi:10.1038/s41587- 020- 00795- 2 (2021). 474 23 Yang, J., Chen, Y., Luo, H. & Cai, H. The Landscape of Somatic Copy Number Alterations in Head and Neck Squamous Cell Carcinoma. Front Oncol 10, 321, doi:10.3389/fonc.2020.00321 (2020). 485 24 Trapnell, C. et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat Biotechnol 32, 381- 386, doi:10.1038/nbt.2859 (2014). 483 25 Gulati, G. S. et al. Single- cell transcriptional diversity is a hallmark of developmental potential. Science 367, 405- 411, doi:10.1126/science.aax0249 (2020).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 84, 880, 435]]<|/det|>
+485 26 Cao, E. Y., Ouyang, J. F. & Rackham, O. J. L. GeneSwitches: ordering gene expression and functional 486 events in single-cell experiments. Bioinformatics 36, 3273- 3275, doi:10.1093/bioinformatics/btaa099 487 (2020). 488 27 Street, K. et al. Slingshot: cell lineage and pseudotime inference for single-cell transcriptomics. BMC 489 Genomics 19, 477, doi:10.1186/s12864- 018- 4772- 0 (2018). 490 28 Yost, K. E. et al. Clonal replacement of tumor- specific T cells following PD- 1 blockade. Nat Med 25, 491 1251- 1259, doi:10.1038/s41591- 019- 0522- 3 (2019). 492 29 Li, H. et al. Dysfunctional CD8 T Cells Form a Proliferative, Dynamically Regulated Compartment within 493 Human Melanoma. Cell, doi:10.1016/j.cell.2018.11.043 (2018). 494 30 Cerezo- Wallis, D. et al. Midkine rewires the melanoma microenvironment toward a tolerogenic and 495 immune- resistant state. Nat Med 26, 1865- 1877, doi:10.1038/s41591- 020- 1073- 3 (2020). 496 31 Revenco, T. et al. Context Dependency of Epithelial- to- Mesenchymal Transition for Metastasis. Cell 497 Rep 29, 1458- 1468 e1453, doi:10.1016/j.celrep.2019.09.081 (2019). 498 32 Chen, Y. et al. Dual reporter genetic mouse models of pancreatic cancer identify an epithelial- to- 499 mesenchymal transition- independent metastasis program. EMBO Mol Med 10, 500 doi:10.15252/emmm.201809085 (2018). 501 33 Sharma, A. et al. Longitudinal single- cell RNA sequencing of patient- derived primary cells reveals 502 drug- induced infidelity in stem cell hierarchy. Nat Commun 9, 4931, doi:10.1038/s41467- 018- 07261- 3 503 (2018). 504 34 Antony, J. & Huang, R. Y. AXL- Driven EMT State as a Targetable Conduit in Cancer. Cancer Res 77, 505 3725- 3732, doi:10.1158/0008- 5472.CAN- 17- 0392 (2017). 506 35 Low, J. L. et al. A chemical genetic screen identifies Aurora kinases as a therapeutic target in EGFR 507 T790M negative, gefitinib- resistant head and neck squamous cell carcinoma (HNSCC). EBioMedicine 508 64, 103220, doi:10.1016/j.ebiom.2021.103220 (2021).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 83, 882, 318]]<|/det|>
+Figure 1. Tumor samples for single cell RNAseq (A) Workflow of sample acquisition, processing, and analyses for single cell transcriptome and TCR clonality of tumors (and patient-derived cultures) from primary and metastatic lymph nodes of HNSCC patients. Diagram was created with BioRender.com. (B) Uniform manifold approximation and projection (UMAP) of scRNAseq data from all cells within primary tumors and metastatic lymph nodes from 7 patients. Clusters are denoted by colors and labelled according to inferred cell types. Violin plots show the expression of selected genes used to define the inferred cell types. (C) Distribution of different cell types (color) for each patient sample (top) and comparing primary and metastatic samples (bottom) as indicated on the y-axis. (D) Chromosomal gains and losses prediction for malignant epithelial cells by inferCNV using non- malignant cells from respective samples as controls. Cyan indicates primary malignant epithelial; yellow indicates lymph node malignant epithelial; sample identities on the y-axis, chromosome numbers on the x-axis.  
+
+<|ref|>text<|/ref|><|det|>[[112, 343, 882, 622]]<|/det|>
+Figure 2. scRNAseq analysis of malignant epithelial cells and identification of pre-metastatic sub- population. (A) UMAP of malignant epithelial cells only, clustered by Seurat clusters (left), patients (middle), and tissue origin (primary/metastatic) (right). (B) Boxplot showing epithelial-mesenchymal transition (EMT) scores across patients and tissue origin (primary versus metastasis). Line represents mean scores, while box represents 2 standard deviations. (C) and (D) Monocle plots demonstrating the derivation of pre-metastatic populations in HN251 (C) and HN279 (D) based on (from left to right) tissue origin, monocle clusters, EMT scores, CytoTRACE scores to derive trajectory. (E) Gene ontology pathways that are significantly altered across pseudotime derived in C and D. (F) Potentially actionable genes identified to be increased in pre-metastatic population. (G) t-SNE plot of tumor cells in HN257 showing a highly aggressive sub-population in the primary tumor with high CytoTRACE scores and expression of SNAI2. (H) Gene set enrichment analysis (GSEA) showing normalized enrichment scores and (I) Kaplan-Meier plot of TCGA data showing overall survival in patients with high versus low scores based on genes expressed by the specific subpopulation in (G). Shaded area shows 95% confidence interval and p-value as indicated based on log-rank test.  
+
+<|ref|>text<|/ref|><|det|>[[112, 648, 882, 904]]<|/det|>
+Figure 3. Functional analysis of actionable genes enriched in pre-metastatic population in patient- derived cultures (PDCs). (A) Dimension reduction plots based on PAGODA for PDCs derived from matched primary and metastatic lymph nodes (nodal metastatic; M). Clusters are denoted by patient identity and site of origin (left), and Seurat clusters (right). (B) Heatmap of differentially expressed pathways (rows) across samples and tumor origin (columns), showing selected Hallmark and Gene Ontology (GO) gene sets. Bars on the top of the heat map indicate the site of sample origins, clusters and patient samples corresponding to those of (A). (C) Boxplot showing the gene expression level of AXL (left) and AURKB (right) of malignant cells from primary and metastatic PDCs for the indicated patients. Line represents mean expression, while box represents 2 standard deviations; colors and cluster numbers of the bars correspond to (A). (D) Immunocytochemistry of AXL in HN137 and AURKB in HN159 and HN220 of primary and metastatic PDCs. Scale bar indicates 100 μm. (E) Representative micrographs from Boyden chamber assays of invaded cells (purple) (top), and quantification of invaded cells (bottom) in barplots from primary and metastatic cell cultures treated with or without BGB324 or barasetrib.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 84, 881, 209]]<|/det|>
+\(^{**}p < 0.01\) , \(^{***}p < 0.001\) , \(^{****}p < 0.0001\) (significant difference) using student t-test compared with untreated at corresponding site of origin. Error bars represent one standard deviation. (H) Flow cytometry dot plots representing anti- AXL (left) and mouse IgG1 isotype control (right) staining of primary and metastatic PDCs of HN137. (I) Gating used for identification and isolation of \(\mathsf{AXL}^{\mathsf{h}\mathsf{i}}\) , \(\mathsf{AXL}^{\mathsf{mid}}\) and \(\mathsf{AXL}^{\mathsf{neg / low}}\) from HN137 primary PDC by FACS sorting (left). Micrographs representing isolated AXL- based subpopulations treated with or without BGB324 and their respective invasive potential in Boyden chamber assays (right).  
+
+<|ref|>text<|/ref|><|det|>[[115, 235, 882, 556]]<|/det|>
+Figure 4. scRNAseq analysis of tumor infiltrating T- cells and establishing a trajectory for tumor- targeting CD8+ lymphocytes. (A) UMAP of tumor infiltrating T- cells from primary and metastatic tumors with clusters denoted by colors and labelled with inferred cell identities. (B) Heatmap of differentially expressed genes (rows) between cells classified into inferred T- cell subsets. Bars on the top of the heatmap indicate the site of origin and cell type corresponding to those of (A) with selected genes indicated. (C) UMAP of all CD8 T- cells from primary and metastatic tumors. Clusters are denoted by colours and labelled with inferred cell identities based on (D) expression of selected genes used for CD8 T- cell subset annotation for. (E) Slingshot analysis of CD8 T- cells showing two potential trajectories giving rise to tumor- targeting CD8+ cells: Trajectory 1 (top)- from naive to dysfunctional and Trajectory 2 (bottom)- memory to dysfunctional. (F) Graphs showing the estimate scores of curated genes related to naive- like (IL7R, TXNIP, SELL, CCR7, TCF7), proliferative (MKI67, HMG82, TYMS), dysfunctional (GZMB, GNYL, CTLA4, LAYN, LAG3, TIG17) populations, and expression of CXCL13 during the development of CD8 T- cell along the naive- proliferation- dysfunction axis in Trajectory 1. (G) Geneswitches output showing ordering of the top switching genes along the naive to dysfunctional (Trajectory 1) CD8 T- cell axis using. Key genes are highlighted with enlarged font size. (H) UMAP projections of expression levels for genes highlighted in (G).  
+
+<|ref|>text<|/ref|><|det|>[[115, 583, 882, 884]]<|/det|>
+Figure 5. Functional analysis of genes involved in CD8 dysfunction and T- cell receptor sequencing analysis. Violin plots showing expression of SOX4, DUSP4 and RBPJ in CD8 T- cell subpopulations derived from published cohorts of scRNAseq meta- dataset from (A) HNSCC and (B) skin squamous cell cancer<sup>12,28</sup>. (C) Boxplots showing expression of SOX4, DUSP4 and RBPJ in CD8 T- cell subpopulations from (B), grouped by pre- and post- pembrolizumab treatment. \(^{*}p < 0.05\) and \(^{***}p < 0.0001\) denotes a significant difference compared with pretreatment of corresponding CD8 T- cell subsets by paired t- test. (B- C) X- axis labels: CD8_ mem = CD8 memory; CD8_eff = CD8 effector; CD8_act = CD8 activated; CD8_ex_act = CD8 exhausted/activated; CD8_ex = CD8 exhausted. (D) Bar graph showing percentage of CD8 T- cells expressing CD39, CD57, LAG3 or PD1 from PBMCs that were activated and cultured with siNT, siSOX4 or siDUSP4 for 5 days (n = 4). Black lines and error bars represent mean ± SEM. \(^{*}p < 0.05\) , \(^{**}p < 0.01\) , \(^{****}p < 0.0001\) (significant difference) by paired t- test compared with siNT of respective markers. (E) Barplots of the percentage of TCR clone(s) detected once (n=1), twice (n=2) or more than two times (n>2) across the CD8 T- cell subpopulations of all patients with HNSCC subjected to scRNAseq. (F) UMAP projection of CD8 T- cells from HN272, HN263 and HN257 colored by selected TCR clonotypes. (G) Schematic diagram summarizing the development and trafficking of CD8 T- cell clones between
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 880, 123]]<|/det|>
+primary tumor, lymph node and metastasis, and bloodstream of HN272, HN263 and HN257 based on the clonotype data from (F). Diagram was created with BioRender.com.  
+
+<|ref|>text<|/ref|><|det|>[[115, 145, 883, 666]]<|/det|>
+Figure 6. Determining the interaction between pre-metastatic malignant cells and CD8+ T lymphocyte populations. (A) Hierarchical plot derived from Cellchat analyses showing ligand-receptor interactions between tumor cells (primary and pre-nodal subpopulations) with T-lymphocytes (CD8+, CD4+ and Treg cells) and TAMs. Circle sizes are proportional to the number of cells in each cell group available for and edge width represents the communication probability with number of potential ligand-receptor pair as indicated. (B) Dot (bubble) plots showing significant MDK ligand-receptor pairs contributing to the signaling from primary or pre-metastatic cancer cells (epithelial) to Treg, CD4 or CD8 T-cells. The dot color and size represent the calculated communication probability, and p-values determined from one-sided permutation test. (C) UMAP of cells derived from tumors of humanized NOG-EXL mice treated with or without anti-PD1. Clusters are denoted by colors labelled with inferred cell types, with a 2D projection of MDK gene expression (inset). (D) Frequency of MDK+ (blue) and MDK- (orange) malignant cells in control or anti-PD1 treated mice. (E) Expression level of selected genes involved in tumor cell proliferation in malignant cells from control or anti-PD1-treated mice. \(*p < 0.05\) and \(**p < 0.01\) indicate significant difference by unpaired t test when compared to control. (F) UMAP of tumor infiltrating CD8 T-cells only extracted from (C). Clusters are denoted by colors labelled with inferred cell identities. (G) Distribution of CD8 T-cell subpopulations in control vs anti-PD1 treated mice. (H) Delta (Δ) percentage of CD8 T-cells expressing the specific MDK-receptors ITGA4, ITGB1 or NCL showing changes in dysfunctional, transitional and proliferating subpopulations, comparing untreated versus anti-PD1 treated mice. Delta percentage is determined by the percentage of MDK receptor+ CD8+ T-cells from anti-PD1 treated mice minus that of the control mice. (I) Expression of NFKB1 in the three CD8 subpopulations in controls and anti-PD1 treated mice. \(*p < 0.05\) indicates significant difference by unpaired t test when compared to control. (J) Scatterplot showing the correlation of expression between NFKB1 with the following MDK receptor(s): ITGA4, ITGB1 and/or NCL in the dysfunctional CD8 T-cells subpopulation. Each dot represents one dysfunctional CD8 T-cell from control (red) or anti-PD1 (blue) treated mice. The R and p values were determined using Pearson correlation statistical analysis.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 85, 880, 808]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[440, 822, 555, 845]]<|/det|>
+<center>Figure 1</center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 80, 880, 850]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[440, 865, 556, 889]]<|/det|>
+<center>Figure 2 </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[118, 80, 880, 732]]<|/det|>
+
+
+<|ref|>image_caption<|/ref|><|det|>[[440, 746, 556, 770]]<|/det|>
+<center>Figure 3</center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 80, 870, 870]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 120, 863, 870]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[440, 866, 556, 890]]<|/det|>
+<center>Figure 5</center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[115, 80, 884, 855]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[440, 861, 556, 884]]<|/det|>
+<center>Figure 6</center>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 42, 311, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[43, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 368, 203]]<|/det|>
+SupplementaryFigure16v3. pdf SupplementaryTables.pdf scSupplementaryMethodsv2. pdf
+
+<--- Page Split --->

preprint/preprint__b3fb9145f0aa1ea90133bd54a3b4c9195977c2f8dc7fba562e2b6fbf7a6ea38a/images_list.json

@@ -0,0 +1,32 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        45,
+        88,
+        740,
+        840
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        45,
+        40,
+        800,
+        789
+      ]
+    ],
+    "page_idx": 18
+  }
+]

preprint/preprint__b3fb9145f0aa1ea90133bd54a3b4c9195977c2f8dc7fba562e2b6fbf7a6ea38a/preprint__b3fb9145f0aa1ea90133bd54a3b4c9195977c2f8dc7fba562e2b6fbf7a6ea38a.mmd

@@ -0,0 +1,336 @@
+
+# Deciphering spatial domains from spatial multi- omics with SpatialGlue  
+
+Yahui Long Agency for Science, Technology and Research (A\*STAR)  
+
+Kok Siong Ang Singapore Immunology Network  
+
+Sha Liao BGI- Shenzhen  
+
+Raman Sethi Agency for Science, Technology and Research (A\*STAR)  
+
+Yang Heng BGI- Shenzhen, Shenzhen  
+
+Chengwei Zhong Agency for Science, Technology and Research (A\*STAR)  
+
+Hang XU Singapore Immunology Network, A \* STAR https://orcid.org/0000- 0002- 8336- 4445  
+
+Nazihah Husna Singapore Immunology Network, A \* STAR  
+
+Min Jian BGI- Shenzhen, Shenzhen  
+
+Lai Guan Ng Singapore Immunology Network https://orcid.org/0000- 0003- 1905- 3586  
+
+Ao Chen BGI- ShenZhen https://orcid.org/0000- 0002- 9699- 8340  
+
+Nicholas Gascoigne  
+
+Yong Loo Lin School of Medicine, National University of Singapore https://orcid.org/0000- 0001- 9980- 4225  
+
+Xun Xu BGI- Shenzhen https://orcid.org/0000- 0002- 5338- 5173  
+
+Jinmiao Chen  
+
+Chen_Jinmiao@immunol.a- star.edu.sg  
+
+Singapore Immunology Network https://orcid.org/0000- 0001- 7547- 6423
+
+<--- Page Split --->
+
+## Brief Communication  
+
+Keywords: Spatial multi- omics, Cross- omics integration, Deep learning, Graph neural networks, Dual attention  
+
+Posted Date: May 16th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 2921471/v1  
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Additional Declarations: There is NO Competing Interest.  
+
+Version of Record: A version of this preprint was published at Nature Methods on June 21st, 2024. See the published version at https://doi.org/10.1038/s41592-024-02316-4.
+
+<--- Page Split --->
+
+## Deciphering spatial domains from spatial multi-omics with SpatialGlue  
+
+Yahui Long \(^{1}\) , Kok Siong Ang \(^{1}\) , Sha Liao \(^{2,3}\) , Raman Sethi \(^{1}\) , Yang Heng \(^{2,3}\) , Chengwei Zhong \(^{1}\) , Hang Xu \(^{1}\) , Nazihah Husna \(^{1}\) , Min Jian \(^{2,4}\) , Lai Guan Ng \(^{1}\) , Ao Chen \(^{2,3,5}\) , Nicholas RJ Gascoigne \(^{6,7,8}\) , Xun Xu \(^{2}\) , Jinmiao Chen \(^{1,6,7*}\)  
+
+\(^{1}\) Singapore Immunology Network (SlgN), Agency for Science, Technology and Research (A\*STAR), 8A Biomedical Grove, Immunos Building, Level 3, Singapore 138648, Singapore  
+
+\(^{2}\) BGI- Shenzhen, Shenzhen, Guangdong, China  
+
+\(^{3}\) BGI Research- Southwest, BGI, Chongqing 401329, China  
+
+\(^{4}\) BGI Research Asia- Pacific, BGI, Singapore 138567, Singapore  
+
+\(^{5}\) JFL- BGI STOmics Center, Jinfeng Laboratory, Chongqing 401329, China  
+
+\(^{6}\) Immunology Translational Research Programme, Yong Loo Lin School of Medicine, National University of Singapore, 5 Science Drive 2, Singapore 117545, Singapore  
+
+\(^{7}\) Department of Microbiology and Immunology, Yong Loo Lin School of Medicine, National University of Singapore, 5 Science Drive 2, Singapore 117545, Singapore  
+
+\(^{8}\) Cancer Translational Research Programme, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore  
+
+\*Corresponding author. Email: chen_jinmiao@immunol.a- star.edu.sg  
+
+## Abstract  
+
+Integration of multiple data modalities in a spatially informed manner remains an unmet need for exploiting spatial multi- omics data. Here, we introduce SpatialGlue, a novel graph neural network with dual- attention mechanism, to decipher spatial domains by capturing the significance of each modality and neighbor graph in cross- omics and intra- omics integration. We demonstrate that SpatialGlue can accurately aggregate cell types into spatial domains at a higher resolution across different tissue types and technology platforms, as well as gain biological insights into cross- modality spatial correlations.  
+
+Key words: Spatial multi- omics, Cross- omics integration, Deep learning, Graph neural networks, Dual attention  
+
+## Main  
+
+Spatial transcriptomics is the next major development in analyzing biological samples since the advent of single- cell transcriptomics. Currently, spatial technologies are expanding to spatial multi- omics with simultaneous profiling of different omics on a single tissue section. These technologies can be roughly divided into two categories, sequencing- based and imaging- based.
+
+<--- Page Split --->
+
+Sequencing- based techniques include DBiT- seq \(^{1}\) , spatial- CITE- seq \(^{2}\) , spatial- ATAC- RNA- seq and CUT&Tag- RNA- seq \(^{3}\) , SPOTS \(^{4}\) , SM- Omics \(^{5}\) , Stereo- CITE- seq \(^{6}\) , and spatial RNA- TCR- seq \(^{7}\) while imaging- based techniques include DNA seqFISH \(^{8}\) , and DNA- MERFISH based DNA and RNA profiling \(^{9}\) . To fully utilize spatial multi- omics data and construct a coherent picture of the tissue under study, spatially aware integration of heterogeneous data modalities is required. Multi- omics data integration poses a significant challenge as different modalities have feature counts that can vary enormously (e.g., protein vs transcripts) and possess different statistical distributions. This challenge is deepened when integrating spatial information with feature counts within each data modality. To our knowledge, there is no tool designed specifically for spatial multi- omics. Existing tools such as SpaGCN \(^{10}\) and GraphST \(^{11}\) target spatial single omics integrated analysis, while GLUE \(^{12}\) and Seurat WNN \(^{13}\) perform multi- omics data integration without employing spatial information.  
+
+Here we introduce SpatialGlue, a graph neural network (GNN) based deep learning model that performs spatial multi- omics data integration (Figure 1a). SpatialGlue employs attention aggregation to achieve data integration on two levels, within- modality spatial information and measurement feature integration, and between- modality integration. SpatialGlue first learns a low dimension embedding space within each modality using spatial and omics data. Within each modality, SpatialGlue constructs a spatial proximity graph and a feature graph which are used separately to encode the pre- processed expression data into a common low dimension embedding space. Here the spatial proximity graph captures spatial relationships between measurement spots, while the feature graph captures feature similarities between spots that can be spatially distant. These constructed graphs can possess unique semantic information that should be integrated. Therefore, we adopted a within- modality attention aggregation layer to adaptively integrate the spatial and feature graph- specific representations and derive modality- specific representations. Specifically, the model learns graph- specific weights to assign importance to each graph. Similarly, the different data modalities can have distinct and complementary contributions to each spot. Thus, we further designed a between- modality attention aggregation layer that learns modality- specific importance weights and adaptively integrates the modality- specific representations to generate the final cross- modality integrated latent representation. The learned weights illustrate the contribution of each modality to the learned latent representation of each spot and consequently the demarcation of different cell types. We believe this approach enables more accurate integration than the common approaches of summation or concatenation.  
+
+We first demonstrate SpatialGlue's capabilities with murine spleen spatial profiling data consisting of protein and transcript measurements \(^{4}\) . The spleen is an important organ within the lymphatic system with functions including B cell maturation in germinal centers formed within B cell follicles. These are complex structures with an array of immune cells present (Figure 1b). The data was generated using SPOTS with the 10x Visium technology capturing whole transcriptomes and extracellular proteins via polyadenylated antibody- derived tag- conjugated (ADT- conjugated) antibodies. The protein detection panel was designed to detect the surface markers of B cells, T cells, and macrophages which are well represented in the spleen. After preprocessing, we performed clustering of each data modality and plotted the clusters on the tissue slide to examine their correspondence between modalities (Figure 1c). The clusters
+
+<--- Page Split --->
+
+clearly did not align, indicating that each modality possessed different information content. We next integrated both modalities with the spatial context, followed by clustering (Figure 1d and Supplementary Figure S1a). Using the protein markers and DEGs, clusters of spots enriched with B cells, T cells, macrophage subsets, and epithelia were annotated \(^{14 - 16}\) (Figure 1f- h and Supplementary S1b). In particular, we could identify epithelial cells and macrophage subsets that were not annotated in the original study. We also plotted the clusters obtained from the integrated analysis onto the individual modalities' UMAPs to examine their respective contributions. The epithelia enriched spots were better separated in the protein modality but were mixed with marginal zone macrophage (MZMΦ) enriched spots in the RNA modality. This contribution to separation by the protein modality can be seen in the learned modality weights (Figure 1e). Conversely, the red pulp macrophage (RpMΦ) enriched spots were better separated in the RNA UMAP and likewise showed greater contribution in the RNA modality weight. For the remaining cell type enriched spots such as marginal metallophilic macrophages (MMMΦ), the spots were more mixed in the individual modalities and better separated in the integrated analysis. Specifically, the MZMΦ spots were mixed with MMMΦ in the RNA modality and RpMΦ in the protein modality. By leveraging on both modalities, our model was able to separate the different cell types.  
+
+Within the white pulp zone, the T cell spots were concentrated in small clusters known as T cell zones. The B cell enriched spots were mainly found in areas adjacent to the T cell clusters. We also visualized the spatial distribution of the cell types' protein markers (Figure 1g). For the B cells, their signal (CD19) indicated that they were much more widely distributed within the spleen with naïve B cells (IgD<sup>+</sup>) colocalized in clusters surrounding the T cell zones. The RpMΦ markers were unsurprisingly the strongest in the red pulp zone while the MMMΦ marker (CD169/Siglec) was expressed in areas surrounding the B and T cell zones. Expectedly, the RpMΦ and MMMΦ marker expressions were mostly mutually exclusive spatially. Finally, EpCAM (epithelia) expression was only found in only a small portion of the slide at the top right where the epithelia spots were annotated. We also examined the markers, Cd209a (MZMΦ) and Siglec1 (MMMΦ) in the RNA modality for their expression by the macrophage clusters, confirming their identity (Figure 1h).  
+
+From the cluster and marker visualization, we observed cell types whose zones were adjoining. Thus, we quantitated the spatial relationship by computing neighborhood enrichment (Figure 1i) and co- occurrence scores with respect to distance from the T and B cell perspective (Figure 1j). In general, we observed a neighborhood enrichment among the B cells, T cells, and MMMΦ. The co- occurrence scores also showed similar trends with B and T cell spots being most likely to be found together at the closest distance. This was followed by MMMΦ which surrounded T and B cell clusters in the white zone. These reflect the layers of cell types that form the follicles and their surroundings. We also note a neighborhood enrichment of MZMΦ and MMMΦ due to the MZMΦ being positioned within the marginal zone surrounding white pulp which in turn was enriched with MMMΦ. We next compared SpatialGlue's output with Seurat WNN's (Supplementary Figure S1c). Compared to SpatialGlue, Seurat WNN assigned a continuous border zone of RpMΦ (cluster 3). Examining the RpMΦ markers F4/80 and CD163, they did not correspond to a similar continuous zone (Figure 1g). Instead, these markers showed a discontinuous and less concentrated boundary layer. Within the white pulp, the T cell
+
+<--- Page Split --->
+
+zone (red spots) was easily distinguished but the B cell and MMMΦ consecutive layering was less distinct compared to SpatialGlue's clusters. In terms of modality weights, Seurat WNN assigned higher weights for all clusters to the protein modality (Supplementary Figure S1d), in contrast to SpatialGlue.  
+
+We next tested SpatialGlue on a mouse thymus dataset acquired with Stereo- CITE- seq spatial multi- omics \(^{6}\) that measures mRNA and protein at sub- cellular resolution. The thymus is a small gland surrounded by a capsule of fibers and collagen (Figure 2a, b). It is divided into two lobes connected by a connective isthmus with each lobe being broadly divided into a central medulla surrounded by an outer cortex layer. In each data modality, broad outlines of the medulla regions and the surrounding cortex could be seen (Figure 2c). However, clusters in the RNA modality did not capture the cortex layers while the clusters in the protein modality were more fragmented. The integrated results showed much more contiguous clusters which we could easily annotate (Figure 2d and Supplementary Figure S2a). Here the cortex was divided into three regions, inner (3), middle (4), and outer (5), and was separated from the medulla by the corticomedullary junction (CMJ) \(^{17}\) . We then confirmed the annotation with the markers and DEGs of cell types expected in each region (Supplementary Figure S2b, c). For most regions, the RNA modality made the greatest contributions, except for the middle cortex (4) and subcapsular zone (7) (Figure 2e). The protein modality's contribution to distinguishing the middle cortex is visible in Figure 2c with the corresponding cluster 3. We again compared SpatialGlue to Seurat WNN on data integration. For this dataset, the Seurat WNN output captured the major regions in the thymus, but the clusters showed greater fragmentation, especially in the cortex (Supplementary Figure S2d). Like the spleen example, Seurat WNN also assigned higher modality weights to the protein modality for all clusters that it found (Supplementary Figure S2e).  
+
+In our final example, we tested SpatialGlue on a P22 mouse brain coronal section dataset acquired using spatial ATAC- RNA- seq \(^{3}\) to measure mRNA and open chromatin regions. We employed the Allen brain atlas reference to annotate anatomical regions such as the cortex layers (ctx), genu of corpus callosum (ccg), lateral septal nucleus (ls), and nucleus accumbens (acb) (Figure 2f). Analyzing individual modalities, we see that they captured various regions with differing accuracy. While both modalities captured the caudoputamen (cp), lateral ventricle (vl), and olfactory limb of the anterior commissure (aco), the RNA modality captured the ccg but was unable to differentiate the ctx layers or the ls (Figure 2g). Meanwhile, the ATAC modality was able to isolate the ls and even the lateral preoptic area (lpo), as well as portions of the acb and some of the ctx layers. With SpatialGlue, the integrated analysis captured all the aforementioned anatomical regions and produced better defined layers in the ctx and anterior cingulate area (aca) regions (Figure 2h and Supplementary Figure S3a). We also found DEGs in the expected brain regions, such as Cux2 and Olfm1 in the cortex layers (ctx- 3,4), and myelin related genes, Tspan2, Cldn11 and Ugt8a expressed in the post- natal developing corpus callosum (11- ccg/aco) (Supplementary Figure 3b). The contributions by each modality to the integrated result for each cluster were illustrated in the learned modality weights (Figure 2i). As expected from the individual analysis, the ATAC modality was the key contributor towards distinguishing many of the ctx layers (1, 2, 3, 4), acb (12), ls (13), and lpo (15). For the RNA modality, it made a greater contribution to the ccg/aco (11), aca (7), and cp (10) clusters.
+
+<--- Page Split --->
+
+Again, these results demonstrated SpatialGlue's ability to extract the relevant information from each modality and integrate them in a spatially aware manner to capture anatomical regions with superior accuracy. Lastly, we compared SpatialGlue to Seurat WNN. Seurat WNN's output also captured many anatomical regions like the ctx layers, ccg, vl, and cp, but did not differentiate the lpo from the acb (Supplementary Figure S3c). Moreover, the Seurat WNN output was grainy without clear boundaries between regions. Seurat WNN also heavily relied on the ATAC data in the data integration but to a much greater extent than SpatialGlue (Supplementary Figure S3d).  
+
+With these three examples, we demonstrated SpatialGlue's ability to effectively integrate multiple data modalities within their spatial context to reveal histologically relevant structures of tissue samples. Our examples spanned different tissue types and different acquisition technologies, which highlighted its applicability to a wide range of data. SpatialGlue was designed to be computation resource efficient. The largest dataset we tested contained 9,215 spots (spatial- ATAC- RNA- seq mouse brain), and it required about 5 mins of wall- clock time on a server with Intel Core i7- 8665U CPU and NVIDIA RTX A6000 GPU. Therefore, we believe SpatialGlue will be an invaluable analysis tool for present and future spatial multi- omics data. Although the examples so far only include two omics data modalities, the framework is extensible to three or more. As data with greater numbers of modalities become available, we will demonstrate SpatialGlue's applicability. We also plan to extend SpatialGlue's functionality with integration of multi- omics data acquired from adjacent tissue slices.  
+
+## Methods  
+
+## Data  
+
+SPOTS mouse spleen dataset Ben- Chetrit et al. \(^{4}\) processed fresh frozen mouse spleen tissue samples and analyzed them using the 10x Visium system supplemented with DNA- barcoded antibody staining. The antibodies (poly(adenylated) antibody- derived tags (ADTs)) enabled protein measurement alongside the transcriptome profiling by 10x Visium. The panel of 21 ADTs was designed to capture the markers of immune cells found in the spleen, including B cells, T cells, and macrophages. A total of 2,653 spots were captured with 32,285 genes per spot. In this example, we used replicate one. We first filtered out genes expressed in fewer than 10 cells. The filtered gene expression counts were then log- transformed and normalized by library size using the SCANPY package. Finally, the top 3,000 HVGs were selected and used as input for PCA. We used the first 50 principal components as the input of the encoder to ensure a consistent input dimension with the ADT data. For the ADT data, we applied CLR normalization to the raw protein expression counts. PCA was then performed on the normalized data and the top 50 principal components were used as input to the encoder.  
+
+Stereo- CITE- seq mouse thymus dataset Murine thymus tissue sample was investigated with the Stereo- CITE- seq spatial multi- omics by Liao et al. \(^{6}\) . The acquired data consisted of 4,697 spots with 23,622 genes and 51 proteins. For the transcriptomic data, we first filtered out genes expressed in fewer than 10 cells and spots with fewer than 80 gene expressed. The filtered gene expression counts were next log- transformed and normalized by library size via the
+
+<--- Page Split --->
+
+SCANPY package 18. Finally, to reduce the dimensionality of the data, the top 3,000 highly variable genes (HVGs) were selected and used as input for PCA. To ensure a consistent input dimension with the ADT data, the first 22 principal components were retained and used as the input of the encoder. For the ADT data, we first filter out proteins expressed in fewer than 50 cells, resulting 22 proteins retained. The protein expression counts were then normalized using CLR (Centered Log Ratio) across each cell. PCA was then performed on the normalized data, and all 22 principal components were used as the input of the encoder.  
+
+Spatial- ATAC- RNA- seq mouse brain dataset This example utilized a juvenile (P22) murine brain tissue sample analyzed by Zhang et al. with Spatial- ATAC- RNA- seq 3. Microfluidic barcoding was used to capture spatial location and combined with in situ Tn5 transposition chemistry to capture chromatin accessibility. The data captured consisted of 9,215 spots with 2,2914 genes and 1,210 peaks. To preprocess the transcriptomic data, cells expressing fewer than 200 genes and genes expressed fewer than 200 cells were filtered out. Next, the gene expression counts were log- transformed and normalized by library size via the SCANPY package. The top 3,000 highly variable genes (HVGs) were selected and used as input to PCA for dimensionality reduction. For consistency with the chromatin peak data, the first 50 principal components were retained and used as input to the encoder. For the chromatin peak data, we used LSI (latent semantic indexing) to reduce the raw chromatin peak counts data to 50 dimensions.  
+
+## The SpatialGlue framework  
+
+We first consider a spatial multi- omics dataset with two different omics modalities, each with a distinct feature set \(X_{1} \in R^{N \times d_{1}}\) and \(X_{2} \in R^{N \times d_{2}}\) . \(N\) denotes the number of spots in the tissue. \(d_{1}\) and \(d_{2}\) are the numbers of features for two omics modalities, respectively. For example, in spatial- ATAC- RNA- seq, \(X_{1}\) and \(X_{2}\) refer to the sets of genes and chromatin regions respectively, while in Stereo- CITE- seq, \(X_{1}\) and \(X_{2}\) refer to the sets of genes and proteins respectively. The primary objective of spatial multi- omics data integration is to learn a mapping function that can project the original individual modality data into a uniform latent space and then integrate the resulting representations. As shown in Figure 1a, the SpatialGlue framework consists of four major modules: (1) Modality- specific GCN encoder, (2) Within- Modality attention aggregation layer, (3) Between- Modality attention aggregation layer, and (4) Modality- specific GCN decoder. The details of each module are described next. Notably, here we demonstrate the SpatialGlue framework with two modalities. Benefiting from the modular design, SpatialGlue readily extends to spatial multi- omics data with more than two modalities.  
+
+## Construction of neighbor graph  
+
+Assuming spots that are spatially adjacent in a tissue usually have similar cell types or cell states, we convert the spatial information to an undirected neighbor graph \(G_{s} = (V, E)\) with \(V\) denoting the set of \(N\) spots and \(E\) denoting the set of connected edges between spots. Let \(A_{s} \in R^{N \times N}\) be the adjacent matrix of graph \(G_{s}\) , where \(A_{s}(i, j) = 1\) if and only if the Euclidean distance between spots \(i\) and \(j\) is less than specific neighbor number \(l\) , otherwise 0. In our examples, we select the top \(r = 6\) nearest spots as neighbors of a given spot for all datasets.
+
+<--- Page Split --->
+
+In a complex tissue sample, it is possible for spots with the same cell types to be spatially non- adjacent to each other, or even far away. To capture the proximity of such spots in a latent space, we explicitly model the relationship between them using a feature graph. Specifically, we apply the \(k\) - nearest neighbor algorithm (KNN) on the PCA embeddings and construct the feature graph \(G_{f}^{m} = (V^{m}, E^{m})\) , where \(V^{m}\) and \(E^{m}\) denote the sets of \(N\) spots and connected edges between spots in the \(m \in \{1,2\}\) - th modality, respectively. For a given spot, we choose the top \(k\) nearest spots as its neighbors. By default, we set \(k\) to 20 for all datasets. We use \(A_{f}^{m} \in R^{N \times N}\) to denote the adjacency matrix of the feature graph \(G_{f}^{m}\) . If spot \(j \in V^{m}\) is the neighbor of spot \(i \in V^{m}\) , then \(A_{f}^{m}(i,j) = 1\) , otherwise 0.  
+
+## Graph convolutional encoder for individual modality  
+
+Each modality (e.g., mRNA or protein) contains a unique feature distribution. To encode each modality in a low dimension embedding space, we use the graph convolution network (GCN) \(^{19}\) , an unsupervised deep graph network, as the encoder of our framework. The main advantage of GCNs is that it can capture the cell expression patterns and neighborhood microenvironment while preserving the high- level global patterns. For each modality, using the pre- processed features as inputs, we separately implement a GCN- encoder on the spatial adjacency graph \(G_{s}\) and the feature graph \(G_{f}\) to learn graph- specific representations. These two neighbor graphs reflect distinct topological semantic relationships between spots, enabling the encoder to capture different local patterns and dependencies of each spot by iteratively aggregating the representations from its neighbors. Specifically, the \(l\) - th ( \(l \in \{1,2, \ldots , L - 1, L\}\) ) layer representation in the encoder are formulated as follows:  
+
+\[H_{s1}^{l} = \sigma (\tilde{A}_{s}H_{s1}^{l - 1}W_{e1}^{l - 1} + b_{e1}^{l - 1}), (1)\] \[H_{f1}^{l} = \sigma (\tilde{A}_{f}^{1}H_{f1}^{l - 1}W_{e1}^{l - 1} + b_{e1}^{l - 1}), (2)\] \[H_{s2}^{l} = \sigma (\tilde{A}_{s}H_{s2}^{l - 1}W_{e2}^{l - 1} + b_{e2}^{l - 1}), (3)\] \[H_{f2}^{l} = \sigma (\tilde{A}_{f}^{2}H_{f2}^{l - 1}W_{e2}^{l - 1} + b_{e2}^{l - 1}), (4)\]  
+
+where \(\tilde{A} = D^{- \frac{1}{2}} AD^{- \frac{1}{2}}\) represents the normalized adjacency matrix of specific graph and \(D\) is a diagonal matrix with diagonal elements being \(D_{ii} = \sum_{j = 1}^{N}A_{ij}\) . In particular, \(\tilde{A}_{s}, \tilde{A}_{f}^{1}\) , and \(\tilde{A}_{f}^{2}\) are the corresponding normalized adjacency matrices of the spatial graph, the feature graph 1, and the feature graph 2, respectively. \(W_{e}\) , and \(b_{e}\) denote a trainable weight matrix and a bias vector, respectively. \(\sigma (\cdot)\) is a nonlinear activation function such as ReLU (Rectified Linear Unit). \(H^{l}\) denotes the \(l\) - th layer output representation, and \(H_{s1}^{0} = H_{f1}^{0}\) and \(H_{s2}^{0} = H_{f2}^{0}\) are set as the original input PCA embedding \(X_{1}\) and \(X_{2}\) respectively. We also specify \(H^{L} \in R^{d_{3}}\) , the output at the \(L\) - th layer, as the final latent representation of the encoder with \(d_{3}\) as the hidden dimension. \(H_{sk}\) and \(H_{fk}\) represent the latent representations derived from the spatial and feature graphs within modality \(k\) , respectively.  
+
+## Within-Modality attention aggregation layer
+
+<--- Page Split --->
+
+For an individual modality, taking its pre- processed features and two graphs as input, we can derive two graph- specific spot representations via the graph convolutional encoder, such as \(H_{s1}\) and \(H_{f1}\) . To integrate the graph- specific representations, we design a Within- Modality attention aggregation layer following the encoder such that its output representation preserves expression similarity and spatial proximity. Given that different neighbor graphs can provide unique semantic information for each spot, the aggregation layer is designed to integrate graph- specific representations in an adaptive manner by capturing the importance of each graph. As a result, we derive a modality- specific representation for each modality. Specifically, for a given spot \(i\) , we first subject its graph- specific representation \(h_{i}^{t}\) to a linear transformation (i.e., a fully connected neural network), and then evaluate the importance of each graph by the similarity of the transformed representation and a trainable weight vector \(q\) . Formally, the attention coefficient \(e_{i}^{t}\) , representing the importance of graph \(t\) to the spot \(i\) , is calculated by:  
+
+\[e_{i}^{t} = q^{T}\cdot \tanh \left(W_{Wi}h_{i}^{t} + b_{Wi}\right), (5)\]  
+
+where \(W_{Wi}\) and \(b_{Wi}\) are the trainable weight matrix and bias vector, respectively. To reduce the number of parameters in the model, all the trainable parameters are shared by the different graph- specific representations. To make the attention coefficient comparable across different graphs, a softmax function is applied to the attention coefficient to derive attention score \(a_{i}^{t}\) .  
+
+\[a_{i}^{t} = \frac{\exp\left(e_{i}^{t}\right)}{\sum_{t = 1}^{T}\exp\left(e_{i}^{t}\right)}, (6)\]  
+
+where \(T\) denotes the number of neighbor graphs (set to 2). \(a_{i}^{t}\) represents the semantic contribution of the \(t\) - th neighbor graph to the representation of spot \(i\) . A higher value of \(a_{i}^{t}\) means greater importance.  
+
+Subsequently, the final representation \(Y_{i}^{m}\) of spot \(i\) in the \(m\) - th modality can be generated by aggregating graph- specific representations according to their attention scores:  
+
+\[Y_{i}^{t} = \sum_{t = 1}^{T}\alpha_{i}^{t}\cdot h_{i}^{t} (7)\]  
+
+such that \(Y_{i}^{m} \in R^{d_{3}}\) preserves the raw cell expressions, cell expression similarity, and spatial proximity within modality \(m\) .  
+
+## Between-Modality attention aggregation layer  
+
+Each individual omic modality provides a partial view of a complex tissue sample, thus requiring an integrated analysis to obtain a comprehensive picture. These views can contain both complementary and contradictory elements, and thus different importance should be assigned to each modality to achieve coherent data integration. Here we use a Between- Modality attention aggregation layer to adaptively integrate the different data modalities. This attention aggregation layer will focus on the more important omics modality by assigning greater weight values to the corresponding representation. Like the Within- Modality layer, we first learn the importance of modality \(m\) by calculating the following coefficient \(w_{i}^{m}\) :  
+
+\[w_{i}^{m} = v^{T}\cdot \tanh \left(W_{Be}y_{i}^{m} + b_{Be}\right), (8)\]
+
+<--- Page Split --->
+
+where \(w_{i}^{m}\) are attention coefficients that represent the importance of modality \(m\) to the representation of spot \(i\) . \(W_{Be}\) , \(b_{Be}\) , and \(\nu\) are learnable weight and bias variables, respectively. Similarly, we further normalize the attention coefficients using the softmax function:  
+
+\[\beta_{i}^{m} = \frac{\exp{(w_{i}^{m})}}{\sum_{m = 1}^{M}\exp{(w_{i}^{m})}}, (9)\]  
+
+where \(\beta_{i}^{m}\) is the normalized attention score that represents the contribution of modality \(m\) to the representation of spot \(i\) .  
+
+Finally, we derive the final representation \(Z_{i}\) of spot \(i\) by aggregating each modality- specific representation according to their attention score \(\beta_{i}^{m}\) :  
+
+\[Z_{i} = \sum_{m = 1}^{M}\beta_{i}^{m}\cdot y_{i}^{m}. (10)\]  
+
+After model training, the latent representation \(Z_{i} \in R^{d_{3}}\) can be used in various downstream analyses, including clustering, visualization, and DEG detection.  
+
+## Model training  
+
+The resulting model is trained jointly with three different loss functions for reconstruction loss, correspondence loss, and adversarial loss. Each loss function is described as follows.  
+
+Reconstruction loss To enforce the learned latent representation to preserve the expression profiles from different modalities, we design an individual decoder for each modality to reverse the integrated representation \(Z_{i}\) back into the normalized expression space. Specifically, by taking output \(Z_{i}\) from the Between- Modality attention aggregation layer as input, the reconstructed representations \(\hat{H}_{1}^{l}\) and \(\hat{H}_{2}^{l}\) from the decoder at the \(l\) - th \((l \in \{1,2, \ldots , L - 1, L\})\) layer are formulated as follows:  
+
+\[\hat{H}_{1}^{l} = \sigma (\tilde{A}_{s}Z_{1}^{l - 1}W_{d1}^{l - 1} + b_{d1}^{l - 1}), (11)\]  
+
+\[\hat{H}_{2}^{l} = \sigma (\tilde{A}_{s}Z_{1}^{l - 1}W_{d2}^{l - 1} + b_{d2}^{l - 1}), (12)\]  
+
+where \(W_{d1}, W_{d2}, b_{d1},\) and \(b_{d2}\) are trainable weight matrices and bias vectors, respectively. \(\hat{H}_{1}^{l}\) and \(\hat{H}_{2}^{l}\) represent the reconstructed expression matrices for omics modalities 1 and 2, respectively.  
+
+SpatialGlue's objective function to minimize the expression reconstruction loss is as follows:  
+
+\[\mathcal{L}_{recon} = \gamma_{1}\sum_{i = 1}^{N}\left\| x_{i}^{1} - \hat{h}_{i}^{1}\right\|_{F}^{2} + \gamma_{2}\sum_{i = 1}^{N}\left\| x_{i}^{2} - \hat{h}_{i}^{2}\right\|_{F}^{2}, (13)\]  
+
+where \(x^{1}\) and \(x^{2}\) represent the original features of modalities 1 and 2, respectively. \(\gamma_{1}\) and \(\gamma_{2}\) are weight factors that are utilized to balance the contribution of different modalities. Due to the differences of sequencing technologies and molecular types, the feature distributions of different omics assays can vary significantly. As such, the weight factors also vary between different
+
+<--- Page Split --->
+
+spatial multi- omics technologies but are fixed for datasets obtained using the same omics technology.  
+
+Correspondence loss While reconstruction loss can enforce the learned latent representation to simultaneously capture the expression information of different modality data, it does not guarantee that the representation manifolds are fully aligned across modalities. To deal with the issue, we add a correspondence loss function. Correspondence loss aims to force consistency between a modality- specific representation \(Y_{m}\) and its corresponding representation \(\hat{Y}_{m}\) obtained through the decoder- encoder of another modality. Mathematically, the correspondence loss is defined as follows:  
+
+\[\mathcal{L}_{corr} = \gamma_{3}\sum_{i = 1}^{N}\| y_{i}^{1} - \hat{y}_{i}^{1}\|_{F}^{2} + \gamma_{4}\sum_{i = 1}^{N}\| y_{i}^{2} - \hat{y}_{i}^{2}\|_{F}^{2},(14)\]  
+
+\[\hat{Y}_{1}^{l} = \sigma \left(\tilde{A}_{s}\left(\sigma (\tilde{A}_{s}Y_{1}^{l - 1}W_{d2}^{l - 1} + b_{d2}^{l - 1})\right)W_{e2}^{l - 1} + b_{e2}^{l - 1}\right), (15)\]  
+
+\[\hat{Y}_{2}^{l} = \sigma \left(\tilde{A}_{s}\left(\sigma (\tilde{A}_{s}Y_{2}^{l - 1}W_{d1}^{l - 1} + b_{d1}^{l - 1})\right)W_{e1}^{l - 1} + b_{e1}^{l - 1}\right). (16)\]  
+
+where \(\gamma_{3}\) and \(\gamma_{4}\) are hyper- parameters, controlling the influences of different modality data. We set \(\hat{Y}_{1}^{0} = Y_{1}\) and \(\hat{Y}_{2}^{0} = Y_{2}\) . \(\sigma (\cdot)\) is a nonlinear activation function, i.e., ReLU (Rectified Linear Unit).  
+
+Adversarial loss To constrain the representations across different modalities in a uniform latent space, we further add a discriminator, i.e., fully connected neural network, in our model. This discriminator tries to align the representations with \(\delta = 1\) denoting modality 1 and \(\delta = 0\) denoting modality 2. Formally, the adversarial loss of the discriminator is defined as:  
+
+\[\mathcal{L}_{adv} = \frac{1}{N}\left[\sum_{i = 1}^{N}[\delta_{i}log(p_{i}) + (1 - \delta_{i})\log (1 - p_{i})]\right],(17)\]  
+
+where \(p_{i}\) and \((1 - p_{i})\) represents the probability scores of the representations assigned to modalities 1 and 2, respectively.  
+
+Therefore, the overall loss function used for model training is defined as:  
+
+\[\mathcal{L}_{total} = \mathcal{L}_{recon} + \mathcal{L}_{corr} + \mathcal{L}_{adv}. (18)\]  
+
+## Implementation details  
+
+For all datasets, a learning rate of 0.0001 was used. To account for differences in feature distribution across the datasets, a tailored group of weight factors \([y_{1}, y_{2}, y_{3}, y_{4}]\) was assigned to each one. The weight factors were [1, 50, 1, 5] for the SPOTS mouse spleen dataset, [1, 10, 1, 10] for the Stereo- CITE- seq mouse thymus dataset, [1, 2.5, 1, 1] for the spatial- ATAC- RNA- seq mouse brain dataset. The training epochs used for the SPOTS mouse spleen, Stereo- CITE- seq mouse thymus, and spatial- ATAC- RNA- seq mouse brain datasets were 900, 1500, and 1500, respectively.  
+
+## Seurat WNN Analysis
+
+<--- Page Split --->
+
+For the SPOTS mouse spleen and Stereo- CITE- seq mouse thymus datasets, we employed the same data preprocessing steps as SpatialGlue and used the same PCA coefficients as input to analysis with Seurat WNN (FindMultiModalNeighbors function) \(^{13}\) . We then adjusted the resolution input to the Seurat FindCluster function to obtain the same number of clusters as SpatialGlue's output. For the Spatial- ATAC- RNA- seq mouse brain dataset, the output obtained from using the same preprocessing steps as SpatialGlue resulted in an excessive number of clusters (more than 20 at a resolution of 0.1). Therefore, we adjusted our preprocessing by first filtering out cells with less than 200 features. Thereafter, we employed SCTransform for normalization and PCA for dimension reduction of the RNA modality. Dimension reduction of the ATAC peaks was performed with Latent Semantic Indexing using the TFIDF and RunSVD functions from the Signac package \(^{20}\) . Seurat WNN analysis was then performed with the top 10 PCs of the RNA modality and 2- 10PCs of the ATAC modality. We discarded the top ATAC PC as it was correlated with sequencing depth.  
+
+## Data availability  
+
+All datasets used in this study are already published and were obtained from public data repositories. The SPOTS mouse spleen data was obtained from the GEO repository (accession no. GSE198353, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE198353) \(^{4}\) , the Stereo- CITE- seq mouse thymus data from BGI and the spatial- ATAC- RNA- seq mouse brain data from AtlasXplore (https://web.atlasxomics.com/visualization/Fan) \(^{3}\) . The details of all datasets used are available in the Methods section. The raw data used in this study have been uploaded to Zenodo and is freely available at https://zenodo.org/record/7879713#.ZE3aOnZBvUk.  
+
+## Code availability  
+
+An open- source Python implementation of the GraphST toolkit is accessible at https://github.com/JinmiaoChenLab/SpatialGlue.  
+
+## Author contributions  
+
+J.C. conceptualized and supervised the project. Y.L. designed the model. Y.L. developed the SpatialGlue software. Y.L., K.S.A., and J.C. wrote the manuscript. Y.L., J.C., R.S, C.Z, H.X., and K.S.A. performed the data analysis. C.Z. and R.S. ran the Seurat WNN algorithm. Y.L. prepared the figures. J.C., N.R.J.G, L.G.N., and N.H. annotated and interpreted the mouse thymus dataset. S.L., Y.H., M.J., A.C., and X.X generated the mouse thymus dataset.  
+
+## Acknowledgements  
+
+We thank Yingrou Tan for her assistance in mouse thymus data interpretation and Min Wu for his comments on the model. The research was supported by A\*STAR under its BMRC Central Research Fund (CRF, UIBR) Award; AI, Analytics and Informatics (AI3) Horizontal Technology Programme Office (HTPO) seed grant (Spatial transcriptomics ST in conjunction with graph neural networks for cell- cell interaction #C211118015) from A\*STAR, Singapore; Open Fund Individual Research Grant (Mapping hematopoietic lineages of healthy and high- risk acute
+
+<--- Page Split --->
+
+myeloid leukemia patients with FLT3- ITD mutations using single- cell omics #OFIRG18nov- 0103) from Ministry of Health, Singapore; National Research Foundation (NRF), Award no. NRF- CRP26- 2021- 0001; the National Research Foundation, Singapore, and Singapore Ministry of Health's National Medical Research Council under its Open Fund- Large Collaborative Grant ("OF- LCG") (MOH- OFLCG18May- 0003). Singapore National Medical Research Council (#NMRC/OFLCG/003/2018).  
+
+## Ethics declarations  
+
+The authors declare that there are no competing interests.  
+
+## References  
+
+1. Liu, Y. et al. High-Spatial-Resolution Multi-Omics Sequencing via Deterministic Barcoding in Tissue. Cell 183, 1665-1681. e18 (2020).  
+2. Liu, Y. et al. High-plex protein and whole transcriptome co-mapping at cellular resolution with spatial CITE-seq. Nat. Biotechnol. (2023) doi:10.1038/s41587-023-01676-0.  
+3. Zhang, D. et al. Spatial epigenome-transcriptome co-profiling of mammalian tissues. Nature 616, 113-122 (2023).  
+4. Ben-Chetrit, N. et al. Integration of whole transcriptome spatial profiling with protein markers. Nat. Biotechnol. (2023) doi:10.1038/s41587-022-01536-3.  
+5. Vickovic, S. et al. SM-Omics is an automated platform for high-throughput spatial multi-omics. Nat. Commun. 13, 795 (2022).  
+6. Liao, S. et al. Integrated Spatial Transcriptomic and Proteomic Analysis of Fresh Frozen Tissue Based on Stereo-seq. bioRxiv 2023.04.28.538364 (2023) doi:10.1101/2023.04.28.538364.  
+7. Hudson, W. H. & Sudmeier, L. J. Localization of T cell clonotypes using the Visium spatial transcriptomics platform. STAR Protoc. 3, 101391 (2022).  
+8. Takei, Y. et al. Integrated spatial genomics reveals global architecture of single nuclei. Nature 590, 344-350 (2021).  
+9. Su, J.-H., Zheng, P., Kinrot, S. S., Bintu, B. & Zhuang, X. Genome-Scale Imaging of the 3D Organization and Transcriptional Activity of Chromatin. Cell 182, 1641-1659. e26 (2020).  
+10. Hu, J. et al. SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nat. Methods 18, 1342-1351 (2021).  
+11. Long, Y. et al. Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST. Nat. Commun. 14, 1155 (2023).  
+12. Cao, Z.-J. & Gao, G. Multi-omics single-cell data integration and regulatory inference with graph-linked embedding. Nat. Biotechnol. 40, 1458-1466 (2022).
+
+<--- Page Split --->
+
+13. Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573-3587. e29 (2021).14. Alexandre, Y. O. & Mueller, S. N. Splenic stromal niches in homeostasis and immunity. Nat. Rev. Immunol. (2023) doi:10.1038/s41577-023-00857-x.15. Borges da Silva, H. et al. Splenic Macrophage Subsets and Their Function during Blood-Borne Infections. Front. Immunol. 6, 480 (2015).16. Backer, R. et al. Effective collaboration between marginal metallophilic macrophages and CD8+ dendritic cells in the generation of cytotoxic T cells. Proc. Natl. Acad. Sci. U. S. A. 107, 216-221 (2010).17. Blackburn, C. C. & Manley, N. R. Developing a new paradigm for thymus organogenesis. Nat. Rev. Immunol. 4, 278-289 (2004).18. Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).19. Kipf, T. N. & Welling, M. Semi-Supervised Classification with Graph Convolutional Networks. (2016) doi:10.48550/ARXIV.1609.02907.20. Stuart, T., Srivastava, A., Madad, S., Lareau, C. A. & Satija, R. Single-cell chromatin state analysis with Signac. Nat. Methods 18, 1333-1341 (2021).  
+
+## Figure legends  
+
+Figure 1: Interpretable deep dual-attention model enables the identification of fine-grained cell types in mouse spleen data generated using SPOTS. (a) Overview of the SpatialGlue framework. A spatial multi-omics technology simultaneously measures two distinct types of molecules, e.g., RNA and surface protein, while preserving spatial context of the tissue. SpatialGlue first uses the \(K\) -nearest neighbor (KNN) algorithm to construct a spatial neighbor graph using the spatial coordinates and a feature neighbor graph with the normalized expression data for each omics modality. Each modality has an implemented GNN- encoder that takes its normalized expressions and neighbor graph to learn two graph- specific representations by iteratively aggregating representations of neighbors. To capture the importance of different graphs, we designed a Within-Modality attention aggregation layer to adaptively integrate graph-specific representations and obtain a modality- specific representation. Finally, to preserve the importance of different modalities, SpatialGlue uses a Between-Modality attention aggregation layer to adaptively integrated modality- specific representations and output the final integrated representation of spots. (b) H&E image of the mouse spleen. (c) UMAP plots and spatial clustering of the RNA and protein expression data. (d) Comparison of UMAP plots of RNA and protein modalities and SpatialGlue integrated output, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of different modalities to each cluster in the mouse spleen dataset. (f) Heatmap showing the expression levels of differentially expressed ADTs for each cluster. (g) Normalized ADT levels of key surface markers for T cell (CD3, CD4, CD8), B cell (IgD, B220, CD19), RpMΦ (F4_80, CD68, CD163), MMMΦ (CD169), and Epithelia (EpCAM). (h) Violin plots indicating the expression distribution of two marker genes in the MMMΦ, MZMΦ, and RpMΦ clusters. (i) Neighborhood enrichment of
+
+<--- Page Split --->
+
+cell type pairs. (j) Cluster co-occurrence score for each cluster at increasing distances. The full names of the abbreviations RpMΦ, MMMΦ, and MZMΦ are red pulp macro, CD169+ MMM, CD209a+ MZM, respectively.  
+
+Figure 2: SpatialGlue accurately integrates different datasets of mouse thymus and mouse brain. The mouse thymus data of protein and RNA modalities was acquired with Stereo- CITE- seq, and the mouse brain data of RNA and ATAC modalities was acquired with spatial- ATAC- RNA- seq. (a) dsDNA image of the mouse thymus. (b) Total mRNA counts in the mouse thymus dataset. (c) UMAP plots and spatial clustering of the RNA and protein expression data in the mouse thymus dataset. (d) Comparison of UMAP plots of RNA and protein modalities and SpatialGlue integrated output in the mouse thymus dataset, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of each modality to each cluster in the mouse thymus dataset. (f) Annotated reference mouse brain coronal section from Allen Mouse Brain Atlas. (g) UMAP plots and spatial clustering of the RNA expression and ATAC data in the mouse brain dataset. (h) Comparison of UMAP plots of RNA and ATAC modalities and SpatialGlue integrated output in the mouse brain dataset, all colored by annotated clusters obtained from the integrated output. (i) Modality weights explaining the importance of each modality to each cluster in the mouse brain dataset. The full names of abbreviation used in the annotations of (f) and (h) are, ctx: cerebral cortex, cp: caudoputamen, vl: lateral ventricle, lpo: lateral preoptic area, aca: anterior cingulate area, ls: lateral septal nucleus, aco: anterior commissure, olfactory limb, acb: nucleus accumbens, cc: corpus callosum.  
+
+## Supplementary Files  
+
+Supplementary Figure S1: Results on mouse spleen. (a) Spatial plots of all clusters and each cluster identified by SpatialGlue. (b) DEGs of clusters found in tissue samples. (c) UMAP and spatial plots of clusters identified by Seurat WNN. (d) Modality weights from Seurat WNN explaining the importance of each modality to each cluster.  
+
+Supplementary Figure S2: Results on mouse thymus. (a) Spatial plots of all clusters and each cluster identified by SpatialGlue. (b) Expression of marker genes and proteins for each cell type. (c) DEGs of clusters found in tissue samples. (d) UMAP and spatial plots of clusters identified by Seurat WNN. (e) Modality weights from Seurat WNN explaining the importance of each modality to each cluster.  
+
+Supplementary Figure S3: Results on mouse brain. (a) Spatial plots of all clusters and each cluster identified by SpatialGlue. (b). DEGs of clusters found in tissue samples. (c) UMAP and spatial plots of clusters identified by Seurat WNN. (d) Modality weights from Seurat WNN explaining the importance of each modality to each cluster.
+
+<--- Page Split --->
+
+## Figures  
+
+![](images/Figure_1.jpg)
+
+<center>Figure 1 </center>  
+
+Interpretable deep dual- attention model enables the identification of fine- grained cell types in mouse spleen data generated using SPOTS. (a) Overview of the SpatialGlue framework. A spatial multi- omics technology simultaneously measures two distinct types of molecules, e.g., RNA and surface protein,
+
+<--- Page Split --->
+
+while preserving spatial context of the tissue. SpatialGlue first uses the K- nearest neighbor (KNN) algorithm to construct a spatial neighbor graph using the spatial coordinates and a feature neighbor graph with the normalized expression data for each omics modality. Each modality has an implemented GNN- encoder that takes its normalized expressions and neighbor graph to learn two graph- specific representations by iteratively aggregating representations of neighbors. To capture the importance of different graphs, we designed a Within- Modality attention aggregation layer to adaptively integrate graph- specific representations and obtain a modality- specific representation. Finally, to preserve the importance of different modalities, SpatialGlue uses a Between- Modality attention aggregation layer to adaptively integrated modality- specific representations and output the final integrated representation of spots. (b) H&E image of the mouse spleen. (c) UMAP plots and spatial clustering of the RNA and protein expression data. (d) Comparison of UMAP plots of RNA and protein modalities and SpatialGlue integrated output, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of different modalities to each cluster in the mouse spleen dataset. (f) Heatmap showing the expression levels of differentially expressed ADTs for each cluster. (g) Normalized ADT levels of key surface markers for T cell (CD3, CD4, CD8), B cell (IgD, B220, CD19), RpMΦ (F4_80, CD68, CD163), MMMΦ (CD169), and Epithelia (EpCAM). (h) Violin plots indicating the expression distribution of two marker genes in the MMMΦ, MZMΦ, and RpMΦ clusters. (i) Neighborhood enrichment of cell type pairs. (j) Cluster co-occurrence score for each cluster at increasing distances. The full names of the abbreviations RpMΦ, MMMΦ, and MZMΦ are red pulp macro, CD169<sup>+</sup> MMM, CD209a<sup>+</sup> MZM, respectively.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2 </center>  
+
+SpatialGlue accurately integrates different datasets of mouse thymus and mouse brain. The mouse thymus data of protein and RNA modalities was acquired with Stereo- CITE- seq, and the mouse brain data of RNA and ATAC modalities was acquired with spatial- ATAC- RNA- seq. (a) dsDNA image of the mouse thymus. (b) Total mRNA counts in the mouse thymus dataset. (c) UMAP plots and spatial clustering of the RNA and protein expression data in the mouse thymus dataset. (d) Comparison of
+
+<--- Page Split --->
+
+UMAP plots of RNA and protein modalities and SpatialGlue integrated output in the mouse thymus dataset, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of each modality to each cluster in the mouse thymus dataset. (f) Annotated reference mouse brain coronal section from Allen Mouse Brain Atlas. (g) UMAP plots and spatial clustering of the RNA expression and ATAC data in the mouse brain dataset. (h) Comparison of UMAP plots of RNA and ATAC modalities and SpatialGlue integrated output in the mouse brain dataset, all colored by annotated clusters obtained from the integrated output. (i) Modality weights explaining the importance of each modality to each cluster in the mouse brain dataset. The full names of abbreviation used in the annotations of (f) and (h) are, ctx: cerebral cortex, cp: caudoputamen, vl: lateral ventricle, lpo: lateral preoptic area, aca: anterior cingulate area, ls: lateral septal nucleus, aco: anterior commissure, olfactory limb, acb: nucleus accumbens, cc: corpus callosum.  
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+SupplementaryFigureS1. pdf SupplementaryFigureS2. pdf SupplementaryFigureS3. pdf SupplementaryTableS1. xlsx SupplementaryTableS2. xlsx SupplementaryTableS3. xlsx
+
+<--- Page Split --->

preprint/preprint__b3fb9145f0aa1ea90133bd54a3b4c9195977c2f8dc7fba562e2b6fbf7a6ea38a/preprint__b3fb9145f0aa1ea90133bd54a3b4c9195977c2f8dc7fba562e2b6fbf7a6ea38a_det.mmd

@@ -0,0 +1,456 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 870, 175]]<|/det|>
+# Deciphering spatial domains from spatial multi- omics with SpatialGlue  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 544, 238]]<|/det|>
+Yahui Long Agency for Science, Technology and Research (A\*STAR)  
+
+<|ref|>text<|/ref|><|det|>[[44, 243, 339, 284]]<|/det|>
+Kok Siong Ang Singapore Immunology Network  
+
+<|ref|>text<|/ref|><|det|>[[44, 290, 180, 330]]<|/det|>
+Sha Liao BGI- Shenzhen  
+
+<|ref|>text<|/ref|><|det|>[[44, 336, 544, 377]]<|/det|>
+Raman Sethi Agency for Science, Technology and Research (A\*STAR)  
+
+<|ref|>text<|/ref|><|det|>[[44, 382, 275, 423]]<|/det|>
+Yang Heng BGI- Shenzhen, Shenzhen  
+
+<|ref|>text<|/ref|><|det|>[[44, 428, 544, 470]]<|/det|>
+Chengwei Zhong Agency for Science, Technology and Research (A\*STAR)  
+
+<|ref|>text<|/ref|><|det|>[[44, 475, 784, 516]]<|/det|>
+Hang XU Singapore Immunology Network, A \* STAR https://orcid.org/0000- 0002- 8336- 4445  
+
+<|ref|>text<|/ref|><|det|>[[44, 521, 427, 562]]<|/det|>
+Nazihah Husna Singapore Immunology Network, A \* STAR  
+
+<|ref|>text<|/ref|><|det|>[[44, 567, 275, 608]]<|/det|>
+Min Jian BGI- Shenzhen, Shenzhen  
+
+<|ref|>text<|/ref|><|det|>[[44, 614, 697, 655]]<|/det|>
+Lai Guan Ng Singapore Immunology Network https://orcid.org/0000- 0003- 1905- 3586  
+
+<|ref|>text<|/ref|><|det|>[[44, 660, 540, 701]]<|/det|>
+Ao Chen BGI- ShenZhen https://orcid.org/0000- 0002- 9699- 8340  
+
+<|ref|>text<|/ref|><|det|>[[44, 706, 220, 725]]<|/det|>
+Nicholas Gascoigne  
+
+<|ref|>text<|/ref|><|det|>[[44, 728, 945, 768]]<|/det|>
+Yong Loo Lin School of Medicine, National University of Singapore https://orcid.org/0000- 0001- 9980- 4225  
+
+<|ref|>text<|/ref|><|det|>[[44, 775, 540, 816]]<|/det|>
+Xun Xu BGI- Shenzhen https://orcid.org/0000- 0002- 5338- 5173  
+
+<|ref|>text<|/ref|><|det|>[[44, 821, 168, 839]]<|/det|>
+Jinmiao Chen  
+
+<|ref|>text<|/ref|><|det|>[[52, 848, 434, 866]]<|/det|>
+Chen_Jinmiao@immunol.a- star.edu.sg  
+
+<|ref|>text<|/ref|><|det|>[[52, 893, 697, 912]]<|/det|>
+Singapore Immunology Network https://orcid.org/0000- 0001- 7547- 6423
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 45, 230, 64]]<|/det|>
+## Brief Communication  
+
+<|ref|>text<|/ref|><|det|>[[42, 83, 907, 125]]<|/det|>
+Keywords: Spatial multi- omics, Cross- omics integration, Deep learning, Graph neural networks, Dual attention  
+
+<|ref|>text<|/ref|><|det|>[[44, 144, 296, 163]]<|/det|>
+Posted Date: May 16th, 2023  
+
+<|ref|>text<|/ref|><|det|>[[42, 182, 474, 202]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 2921471/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 219, 912, 262]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 280, 534, 300]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 335, 935, 378]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Methods on June 21st, 2024. See the published version at https://doi.org/10.1038/s41592-024-02316-4.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 88, 881, 134]]<|/det|>
+## Deciphering spatial domains from spatial multi-omics with SpatialGlue  
+
+<|ref|>text<|/ref|><|det|>[[115, 146, 883, 202]]<|/det|>
+Yahui Long \(^{1}\) , Kok Siong Ang \(^{1}\) , Sha Liao \(^{2,3}\) , Raman Sethi \(^{1}\) , Yang Heng \(^{2,3}\) , Chengwei Zhong \(^{1}\) , Hang Xu \(^{1}\) , Nazihah Husna \(^{1}\) , Min Jian \(^{2,4}\) , Lai Guan Ng \(^{1}\) , Ao Chen \(^{2,3,5}\) , Nicholas RJ Gascoigne \(^{6,7,8}\) , Xun Xu \(^{2}\) , Jinmiao Chen \(^{1,6,7*}\)  
+
+<|ref|>text<|/ref|><|det|>[[115, 216, 884, 253]]<|/det|>
+\(^{1}\) Singapore Immunology Network (SlgN), Agency for Science, Technology and Research (A\*STAR), 8A Biomedical Grove, Immunos Building, Level 3, Singapore 138648, Singapore  
+
+<|ref|>text<|/ref|><|det|>[[115, 267, 496, 285]]<|/det|>
+\(^{2}\) BGI- Shenzhen, Shenzhen, Guangdong, China  
+
+<|ref|>text<|/ref|><|det|>[[115, 297, 588, 315]]<|/det|>
+\(^{3}\) BGI Research- Southwest, BGI, Chongqing 401329, China  
+
+<|ref|>text<|/ref|><|det|>[[115, 327, 627, 346]]<|/det|>
+\(^{4}\) BGI Research Asia- Pacific, BGI, Singapore 138567, Singapore  
+
+<|ref|>text<|/ref|><|det|>[[115, 358, 707, 377]]<|/det|>
+\(^{5}\) JFL- BGI STOmics Center, Jinfeng Laboratory, Chongqing 401329, China  
+
+<|ref|>text<|/ref|><|det|>[[115, 389, 883, 427]]<|/det|>
+\(^{6}\) Immunology Translational Research Programme, Yong Loo Lin School of Medicine, National University of Singapore, 5 Science Drive 2, Singapore 117545, Singapore  
+
+<|ref|>text<|/ref|><|det|>[[115, 439, 883, 476]]<|/det|>
+\(^{7}\) Department of Microbiology and Immunology, Yong Loo Lin School of Medicine, National University of Singapore, 5 Science Drive 2, Singapore 117545, Singapore  
+
+<|ref|>text<|/ref|><|det|>[[115, 488, 883, 525]]<|/det|>
+\(^{8}\) Cancer Translational Research Programme, Yong Loo Lin School of Medicine, National University of Singapore, Singapore, Singapore  
+
+<|ref|>text<|/ref|><|det|>[[115, 538, 666, 556]]<|/det|>
+\*Corresponding author. Email: chen_jinmiao@immunol.a- star.edu.sg  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 570, 211, 590]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[115, 604, 883, 733]]<|/det|>
+Integration of multiple data modalities in a spatially informed manner remains an unmet need for exploiting spatial multi- omics data. Here, we introduce SpatialGlue, a novel graph neural network with dual- attention mechanism, to decipher spatial domains by capturing the significance of each modality and neighbor graph in cross- omics and intra- omics integration. We demonstrate that SpatialGlue can accurately aggregate cell types into spatial domains at a higher resolution across different tissue types and technology platforms, as well as gain biological insights into cross- modality spatial correlations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 745, 883, 782]]<|/det|>
+Key words: Spatial multi- omics, Cross- omics integration, Deep learning, Graph neural networks, Dual attention  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 798, 170, 817]]<|/det|>
+## Main  
+
+<|ref|>text<|/ref|><|det|>[[115, 832, 883, 905]]<|/det|>
+Spatial transcriptomics is the next major development in analyzing biological samples since the advent of single- cell transcriptomics. Currently, spatial technologies are expanding to spatial multi- omics with simultaneous profiling of different omics on a single tissue section. These technologies can be roughly divided into two categories, sequencing- based and imaging- based.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 309]]<|/det|>
+Sequencing- based techniques include DBiT- seq \(^{1}\) , spatial- CITE- seq \(^{2}\) , spatial- ATAC- RNA- seq and CUT&Tag- RNA- seq \(^{3}\) , SPOTS \(^{4}\) , SM- Omics \(^{5}\) , Stereo- CITE- seq \(^{6}\) , and spatial RNA- TCR- seq \(^{7}\) while imaging- based techniques include DNA seqFISH \(^{8}\) , and DNA- MERFISH based DNA and RNA profiling \(^{9}\) . To fully utilize spatial multi- omics data and construct a coherent picture of the tissue under study, spatially aware integration of heterogeneous data modalities is required. Multi- omics data integration poses a significant challenge as different modalities have feature counts that can vary enormously (e.g., protein vs transcripts) and possess different statistical distributions. This challenge is deepened when integrating spatial information with feature counts within each data modality. To our knowledge, there is no tool designed specifically for spatial multi- omics. Existing tools such as SpaGCN \(^{10}\) and GraphST \(^{11}\) target spatial single omics integrated analysis, while GLUE \(^{12}\) and Seurat WNN \(^{13}\) perform multi- omics data integration without employing spatial information.  
+
+<|ref|>text<|/ref|><|det|>[[114, 321, 883, 707]]<|/det|>
+Here we introduce SpatialGlue, a graph neural network (GNN) based deep learning model that performs spatial multi- omics data integration (Figure 1a). SpatialGlue employs attention aggregation to achieve data integration on two levels, within- modality spatial information and measurement feature integration, and between- modality integration. SpatialGlue first learns a low dimension embedding space within each modality using spatial and omics data. Within each modality, SpatialGlue constructs a spatial proximity graph and a feature graph which are used separately to encode the pre- processed expression data into a common low dimension embedding space. Here the spatial proximity graph captures spatial relationships between measurement spots, while the feature graph captures feature similarities between spots that can be spatially distant. These constructed graphs can possess unique semantic information that should be integrated. Therefore, we adopted a within- modality attention aggregation layer to adaptively integrate the spatial and feature graph- specific representations and derive modality- specific representations. Specifically, the model learns graph- specific weights to assign importance to each graph. Similarly, the different data modalities can have distinct and complementary contributions to each spot. Thus, we further designed a between- modality attention aggregation layer that learns modality- specific importance weights and adaptively integrates the modality- specific representations to generate the final cross- modality integrated latent representation. The learned weights illustrate the contribution of each modality to the learned latent representation of each spot and consequently the demarcation of different cell types. We believe this approach enables more accurate integration than the common approaches of summation or concatenation.  
+
+<|ref|>text<|/ref|><|det|>[[114, 719, 883, 903]]<|/det|>
+We first demonstrate SpatialGlue's capabilities with murine spleen spatial profiling data consisting of protein and transcript measurements \(^{4}\) . The spleen is an important organ within the lymphatic system with functions including B cell maturation in germinal centers formed within B cell follicles. These are complex structures with an array of immune cells present (Figure 1b). The data was generated using SPOTS with the 10x Visium technology capturing whole transcriptomes and extracellular proteins via polyadenylated antibody- derived tag- conjugated (ADT- conjugated) antibodies. The protein detection panel was designed to detect the surface markers of B cells, T cells, and macrophages which are well represented in the spleen. After preprocessing, we performed clustering of each data modality and plotted the clusters on the tissue slide to examine their correspondence between modalities (Figure 1c). The clusters
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 401]]<|/det|>
+clearly did not align, indicating that each modality possessed different information content. We next integrated both modalities with the spatial context, followed by clustering (Figure 1d and Supplementary Figure S1a). Using the protein markers and DEGs, clusters of spots enriched with B cells, T cells, macrophage subsets, and epithelia were annotated \(^{14 - 16}\) (Figure 1f- h and Supplementary S1b). In particular, we could identify epithelial cells and macrophage subsets that were not annotated in the original study. We also plotted the clusters obtained from the integrated analysis onto the individual modalities' UMAPs to examine their respective contributions. The epithelia enriched spots were better separated in the protein modality but were mixed with marginal zone macrophage (MZMΦ) enriched spots in the RNA modality. This contribution to separation by the protein modality can be seen in the learned modality weights (Figure 1e). Conversely, the red pulp macrophage (RpMΦ) enriched spots were better separated in the RNA UMAP and likewise showed greater contribution in the RNA modality weight. For the remaining cell type enriched spots such as marginal metallophilic macrophages (MMMΦ), the spots were more mixed in the individual modalities and better separated in the integrated analysis. Specifically, the MZMΦ spots were mixed with MMMΦ in the RNA modality and RpMΦ in the protein modality. By leveraging on both modalities, our model was able to separate the different cell types.  
+
+<|ref|>text<|/ref|><|det|>[[114, 413, 883, 633]]<|/det|>
+Within the white pulp zone, the T cell spots were concentrated in small clusters known as T cell zones. The B cell enriched spots were mainly found in areas adjacent to the T cell clusters. We also visualized the spatial distribution of the cell types' protein markers (Figure 1g). For the B cells, their signal (CD19) indicated that they were much more widely distributed within the spleen with naïve B cells (IgD<sup>+</sup>) colocalized in clusters surrounding the T cell zones. The RpMΦ markers were unsurprisingly the strongest in the red pulp zone while the MMMΦ marker (CD169/Siglec) was expressed in areas surrounding the B and T cell zones. Expectedly, the RpMΦ and MMMΦ marker expressions were mostly mutually exclusive spatially. Finally, EpCAM (epithelia) expression was only found in only a small portion of the slide at the top right where the epithelia spots were annotated. We also examined the markers, Cd209a (MZMΦ) and Siglec1 (MMMΦ) in the RNA modality for their expression by the macrophage clusters, confirming their identity (Figure 1h).  
+
+<|ref|>text<|/ref|><|det|>[[114, 646, 883, 903]]<|/det|>
+From the cluster and marker visualization, we observed cell types whose zones were adjoining. Thus, we quantitated the spatial relationship by computing neighborhood enrichment (Figure 1i) and co- occurrence scores with respect to distance from the T and B cell perspective (Figure 1j). In general, we observed a neighborhood enrichment among the B cells, T cells, and MMMΦ. The co- occurrence scores also showed similar trends with B and T cell spots being most likely to be found together at the closest distance. This was followed by MMMΦ which surrounded T and B cell clusters in the white zone. These reflect the layers of cell types that form the follicles and their surroundings. We also note a neighborhood enrichment of MZMΦ and MMMΦ due to the MZMΦ being positioned within the marginal zone surrounding white pulp which in turn was enriched with MMMΦ. We next compared SpatialGlue's output with Seurat WNN's (Supplementary Figure S1c). Compared to SpatialGlue, Seurat WNN assigned a continuous border zone of RpMΦ (cluster 3). Examining the RpMΦ markers F4/80 and CD163, they did not correspond to a similar continuous zone (Figure 1g). Instead, these markers showed a discontinuous and less concentrated boundary layer. Within the white pulp, the T cell
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 882, 162]]<|/det|>
+zone (red spots) was easily distinguished but the B cell and MMMΦ consecutive layering was less distinct compared to SpatialGlue's clusters. In terms of modality weights, Seurat WNN assigned higher weights for all clusters to the protein modality (Supplementary Figure S1d), in contrast to SpatialGlue.  
+
+<|ref|>text<|/ref|><|det|>[[114, 175, 883, 543]]<|/det|>
+We next tested SpatialGlue on a mouse thymus dataset acquired with Stereo- CITE- seq spatial multi- omics \(^{6}\) that measures mRNA and protein at sub- cellular resolution. The thymus is a small gland surrounded by a capsule of fibers and collagen (Figure 2a, b). It is divided into two lobes connected by a connective isthmus with each lobe being broadly divided into a central medulla surrounded by an outer cortex layer. In each data modality, broad outlines of the medulla regions and the surrounding cortex could be seen (Figure 2c). However, clusters in the RNA modality did not capture the cortex layers while the clusters in the protein modality were more fragmented. The integrated results showed much more contiguous clusters which we could easily annotate (Figure 2d and Supplementary Figure S2a). Here the cortex was divided into three regions, inner (3), middle (4), and outer (5), and was separated from the medulla by the corticomedullary junction (CMJ) \(^{17}\) . We then confirmed the annotation with the markers and DEGs of cell types expected in each region (Supplementary Figure S2b, c). For most regions, the RNA modality made the greatest contributions, except for the middle cortex (4) and subcapsular zone (7) (Figure 2e). The protein modality's contribution to distinguishing the middle cortex is visible in Figure 2c with the corresponding cluster 3. We again compared SpatialGlue to Seurat WNN on data integration. For this dataset, the Seurat WNN output captured the major regions in the thymus, but the clusters showed greater fragmentation, especially in the cortex (Supplementary Figure S2d). Like the spleen example, Seurat WNN also assigned higher modality weights to the protein modality for all clusters that it found (Supplementary Figure S2e).  
+
+<|ref|>text<|/ref|><|det|>[[114, 555, 883, 904]]<|/det|>
+In our final example, we tested SpatialGlue on a P22 mouse brain coronal section dataset acquired using spatial ATAC- RNA- seq \(^{3}\) to measure mRNA and open chromatin regions. We employed the Allen brain atlas reference to annotate anatomical regions such as the cortex layers (ctx), genu of corpus callosum (ccg), lateral septal nucleus (ls), and nucleus accumbens (acb) (Figure 2f). Analyzing individual modalities, we see that they captured various regions with differing accuracy. While both modalities captured the caudoputamen (cp), lateral ventricle (vl), and olfactory limb of the anterior commissure (aco), the RNA modality captured the ccg but was unable to differentiate the ctx layers or the ls (Figure 2g). Meanwhile, the ATAC modality was able to isolate the ls and even the lateral preoptic area (lpo), as well as portions of the acb and some of the ctx layers. With SpatialGlue, the integrated analysis captured all the aforementioned anatomical regions and produced better defined layers in the ctx and anterior cingulate area (aca) regions (Figure 2h and Supplementary Figure S3a). We also found DEGs in the expected brain regions, such as Cux2 and Olfm1 in the cortex layers (ctx- 3,4), and myelin related genes, Tspan2, Cldn11 and Ugt8a expressed in the post- natal developing corpus callosum (11- ccg/aco) (Supplementary Figure 3b). The contributions by each modality to the integrated result for each cluster were illustrated in the learned modality weights (Figure 2i). As expected from the individual analysis, the ATAC modality was the key contributor towards distinguishing many of the ctx layers (1, 2, 3, 4), acb (12), ls (13), and lpo (15). For the RNA modality, it made a greater contribution to the ccg/aco (11), aca (7), and cp (10) clusters.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 883, 236]]<|/det|>
+Again, these results demonstrated SpatialGlue's ability to extract the relevant information from each modality and integrate them in a spatially aware manner to capture anatomical regions with superior accuracy. Lastly, we compared SpatialGlue to Seurat WNN. Seurat WNN's output also captured many anatomical regions like the ctx layers, ccg, vl, and cp, but did not differentiate the lpo from the acb (Supplementary Figure S3c). Moreover, the Seurat WNN output was grainy without clear boundaries between regions. Seurat WNN also heavily relied on the ATAC data in the data integration but to a much greater extent than SpatialGlue (Supplementary Figure S3d).  
+
+<|ref|>text<|/ref|><|det|>[[115, 248, 883, 468]]<|/det|>
+With these three examples, we demonstrated SpatialGlue's ability to effectively integrate multiple data modalities within their spatial context to reveal histologically relevant structures of tissue samples. Our examples spanned different tissue types and different acquisition technologies, which highlighted its applicability to a wide range of data. SpatialGlue was designed to be computation resource efficient. The largest dataset we tested contained 9,215 spots (spatial- ATAC- RNA- seq mouse brain), and it required about 5 mins of wall- clock time on a server with Intel Core i7- 8665U CPU and NVIDIA RTX A6000 GPU. Therefore, we believe SpatialGlue will be an invaluable analysis tool for present and future spatial multi- omics data. Although the examples so far only include two omics data modalities, the framework is extensible to three or more. As data with greater numbers of modalities become available, we will demonstrate SpatialGlue's applicability. We also plan to extend SpatialGlue's functionality with integration of multi- omics data acquired from adjacent tissue slices.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 482, 211, 502]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 518, 155, 534]]<|/det|>
+## Data  
+
+<|ref|>text<|/ref|><|det|>[[115, 547, 883, 787]]<|/det|>
+SPOTS mouse spleen dataset Ben- Chetrit et al. \(^{4}\) processed fresh frozen mouse spleen tissue samples and analyzed them using the 10x Visium system supplemented with DNA- barcoded antibody staining. The antibodies (poly(adenylated) antibody- derived tags (ADTs)) enabled protein measurement alongside the transcriptome profiling by 10x Visium. The panel of 21 ADTs was designed to capture the markers of immune cells found in the spleen, including B cells, T cells, and macrophages. A total of 2,653 spots were captured with 32,285 genes per spot. In this example, we used replicate one. We first filtered out genes expressed in fewer than 10 cells. The filtered gene expression counts were then log- transformed and normalized by library size using the SCANPY package. Finally, the top 3,000 HVGs were selected and used as input for PCA. We used the first 50 principal components as the input of the encoder to ensure a consistent input dimension with the ADT data. For the ADT data, we applied CLR normalization to the raw protein expression counts. PCA was then performed on the normalized data and the top 50 principal components were used as input to the encoder.  
+
+<|ref|>text<|/ref|><|det|>[[116, 799, 883, 890]]<|/det|>
+Stereo- CITE- seq mouse thymus dataset Murine thymus tissue sample was investigated with the Stereo- CITE- seq spatial multi- omics by Liao et al. \(^{6}\) . The acquired data consisted of 4,697 spots with 23,622 genes and 51 proteins. For the transcriptomic data, we first filtered out genes expressed in fewer than 10 cells and spots with fewer than 80 gene expressed. The filtered gene expression counts were next log- transformed and normalized by library size via the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 882, 216]]<|/det|>
+SCANPY package 18. Finally, to reduce the dimensionality of the data, the top 3,000 highly variable genes (HVGs) were selected and used as input for PCA. To ensure a consistent input dimension with the ADT data, the first 22 principal components were retained and used as the input of the encoder. For the ADT data, we first filter out proteins expressed in fewer than 50 cells, resulting 22 proteins retained. The protein expression counts were then normalized using CLR (Centered Log Ratio) across each cell. PCA was then performed on the normalized data, and all 22 principal components were used as the input of the encoder.  
+
+<|ref|>text<|/ref|><|det|>[[115, 229, 882, 431]]<|/det|>
+Spatial- ATAC- RNA- seq mouse brain dataset This example utilized a juvenile (P22) murine brain tissue sample analyzed by Zhang et al. with Spatial- ATAC- RNA- seq 3. Microfluidic barcoding was used to capture spatial location and combined with in situ Tn5 transposition chemistry to capture chromatin accessibility. The data captured consisted of 9,215 spots with 2,2914 genes and 1,210 peaks. To preprocess the transcriptomic data, cells expressing fewer than 200 genes and genes expressed fewer than 200 cells were filtered out. Next, the gene expression counts were log- transformed and normalized by library size via the SCANPY package. The top 3,000 highly variable genes (HVGs) were selected and used as input to PCA for dimensionality reduction. For consistency with the chromatin peak data, the first 50 principal components were retained and used as input to the encoder. For the chromatin peak data, we used LSI (latent semantic indexing) to reduce the raw chromatin peak counts data to 50 dimensions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 444, 350, 462]]<|/det|>
+## The SpatialGlue framework  
+
+<|ref|>text<|/ref|><|det|>[[115, 475, 882, 715]]<|/det|>
+We first consider a spatial multi- omics dataset with two different omics modalities, each with a distinct feature set \(X_{1} \in R^{N \times d_{1}}\) and \(X_{2} \in R^{N \times d_{2}}\) . \(N\) denotes the number of spots in the tissue. \(d_{1}\) and \(d_{2}\) are the numbers of features for two omics modalities, respectively. For example, in spatial- ATAC- RNA- seq, \(X_{1}\) and \(X_{2}\) refer to the sets of genes and chromatin regions respectively, while in Stereo- CITE- seq, \(X_{1}\) and \(X_{2}\) refer to the sets of genes and proteins respectively. The primary objective of spatial multi- omics data integration is to learn a mapping function that can project the original individual modality data into a uniform latent space and then integrate the resulting representations. As shown in Figure 1a, the SpatialGlue framework consists of four major modules: (1) Modality- specific GCN encoder, (2) Within- Modality attention aggregation layer, (3) Between- Modality attention aggregation layer, and (4) Modality- specific GCN decoder. The details of each module are described next. Notably, here we demonstrate the SpatialGlue framework with two modalities. Benefiting from the modular design, SpatialGlue readily extends to spatial multi- omics data with more than two modalities.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 728, 388, 747]]<|/det|>
+## Construction of neighbor graph  
+
+<|ref|>text<|/ref|><|det|>[[115, 759, 882, 871]]<|/det|>
+Assuming spots that are spatially adjacent in a tissue usually have similar cell types or cell states, we convert the spatial information to an undirected neighbor graph \(G_{s} = (V, E)\) with \(V\) denoting the set of \(N\) spots and \(E\) denoting the set of connected edges between spots. Let \(A_{s} \in R^{N \times N}\) be the adjacent matrix of graph \(G_{s}\) , where \(A_{s}(i, j) = 1\) if and only if the Euclidean distance between spots \(i\) and \(j\) is less than specific neighbor number \(l\) , otherwise 0. In our examples, we select the top \(r = 6\) nearest spots as neighbors of a given spot for all datasets.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 261]]<|/det|>
+In a complex tissue sample, it is possible for spots with the same cell types to be spatially non- adjacent to each other, or even far away. To capture the proximity of such spots in a latent space, we explicitly model the relationship between them using a feature graph. Specifically, we apply the \(k\) - nearest neighbor algorithm (KNN) on the PCA embeddings and construct the feature graph \(G_{f}^{m} = (V^{m}, E^{m})\) , where \(V^{m}\) and \(E^{m}\) denote the sets of \(N\) spots and connected edges between spots in the \(m \in \{1,2\}\) - th modality, respectively. For a given spot, we choose the top \(k\) nearest spots as its neighbors. By default, we set \(k\) to 20 for all datasets. We use \(A_{f}^{m} \in R^{N \times N}\) to denote the adjacency matrix of the feature graph \(G_{f}^{m}\) . If spot \(j \in V^{m}\) is the neighbor of spot \(i \in V^{m}\) , then \(A_{f}^{m}(i,j) = 1\) , otherwise 0.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 274, 564, 293]]<|/det|>
+## Graph convolutional encoder for individual modality  
+
+<|ref|>text<|/ref|><|det|>[[114, 305, 883, 512]]<|/det|>
+Each modality (e.g., mRNA or protein) contains a unique feature distribution. To encode each modality in a low dimension embedding space, we use the graph convolution network (GCN) \(^{19}\) , an unsupervised deep graph network, as the encoder of our framework. The main advantage of GCNs is that it can capture the cell expression patterns and neighborhood microenvironment while preserving the high- level global patterns. For each modality, using the pre- processed features as inputs, we separately implement a GCN- encoder on the spatial adjacency graph \(G_{s}\) and the feature graph \(G_{f}\) to learn graph- specific representations. These two neighbor graphs reflect distinct topological semantic relationships between spots, enabling the encoder to capture different local patterns and dependencies of each spot by iteratively aggregating the representations from its neighbors. Specifically, the \(l\) - th ( \(l \in \{1,2, \ldots , L - 1, L\}\) ) layer representation in the encoder are formulated as follows:  
+
+<|ref|>equation<|/ref|><|det|>[[113, 520, 377, 650]]<|/det|>
+\[H_{s1}^{l} = \sigma (\tilde{A}_{s}H_{s1}^{l - 1}W_{e1}^{l - 1} + b_{e1}^{l - 1}), (1)\] \[H_{f1}^{l} = \sigma (\tilde{A}_{f}^{1}H_{f1}^{l - 1}W_{e1}^{l - 1} + b_{e1}^{l - 1}), (2)\] \[H_{s2}^{l} = \sigma (\tilde{A}_{s}H_{s2}^{l - 1}W_{e2}^{l - 1} + b_{e2}^{l - 1}), (3)\] \[H_{f2}^{l} = \sigma (\tilde{A}_{f}^{2}H_{f2}^{l - 1}W_{e2}^{l - 1} + b_{e2}^{l - 1}), (4)\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 662, 883, 860]]<|/det|>
+where \(\tilde{A} = D^{- \frac{1}{2}} AD^{- \frac{1}{2}}\) represents the normalized adjacency matrix of specific graph and \(D\) is a diagonal matrix with diagonal elements being \(D_{ii} = \sum_{j = 1}^{N}A_{ij}\) . In particular, \(\tilde{A}_{s}, \tilde{A}_{f}^{1}\) , and \(\tilde{A}_{f}^{2}\) are the corresponding normalized adjacency matrices of the spatial graph, the feature graph 1, and the feature graph 2, respectively. \(W_{e}\) , and \(b_{e}\) denote a trainable weight matrix and a bias vector, respectively. \(\sigma (\cdot)\) is a nonlinear activation function such as ReLU (Rectified Linear Unit). \(H^{l}\) denotes the \(l\) - th layer output representation, and \(H_{s1}^{0} = H_{f1}^{0}\) and \(H_{s2}^{0} = H_{f2}^{0}\) are set as the original input PCA embedding \(X_{1}\) and \(X_{2}\) respectively. We also specify \(H^{L} \in R^{d_{3}}\) , the output at the \(L\) - th layer, as the final latent representation of the encoder with \(d_{3}\) as the hidden dimension. \(H_{sk}\) and \(H_{fk}\) represent the latent representations derived from the spatial and feature graphs within modality \(k\) , respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 872, 487, 890]]<|/det|>
+## Within-Modality attention aggregation layer
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 314]]<|/det|>
+For an individual modality, taking its pre- processed features and two graphs as input, we can derive two graph- specific spot representations via the graph convolutional encoder, such as \(H_{s1}\) and \(H_{f1}\) . To integrate the graph- specific representations, we design a Within- Modality attention aggregation layer following the encoder such that its output representation preserves expression similarity and spatial proximity. Given that different neighbor graphs can provide unique semantic information for each spot, the aggregation layer is designed to integrate graph- specific representations in an adaptive manner by capturing the importance of each graph. As a result, we derive a modality- specific representation for each modality. Specifically, for a given spot \(i\) , we first subject its graph- specific representation \(h_{i}^{t}\) to a linear transformation (i.e., a fully connected neural network), and then evaluate the importance of each graph by the similarity of the transformed representation and a trainable weight vector \(q\) . Formally, the attention coefficient \(e_{i}^{t}\) , representing the importance of graph \(t\) to the spot \(i\) , is calculated by:  
+
+<|ref|>equation<|/ref|><|det|>[[114, 326, 375, 347]]<|/det|>
+\[e_{i}^{t} = q^{T}\cdot \tanh \left(W_{Wi}h_{i}^{t} + b_{Wi}\right), (5)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 359, 883, 435]]<|/det|>
+where \(W_{Wi}\) and \(b_{Wi}\) are the trainable weight matrix and bias vector, respectively. To reduce the number of parameters in the model, all the trainable parameters are shared by the different graph- specific representations. To make the attention coefficient comparable across different graphs, a softmax function is applied to the attention coefficient to derive attention score \(a_{i}^{t}\) .  
+
+<|ref|>equation<|/ref|><|det|>[[114, 448, 275, 480]]<|/det|>
+\[a_{i}^{t} = \frac{\exp\left(e_{i}^{t}\right)}{\sum_{t = 1}^{T}\exp\left(e_{i}^{t}\right)}, (6)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 493, 883, 551]]<|/det|>
+where \(T\) denotes the number of neighbor graphs (set to 2). \(a_{i}^{t}\) represents the semantic contribution of the \(t\) - th neighbor graph to the representation of spot \(i\) . A higher value of \(a_{i}^{t}\) means greater importance.  
+
+<|ref|>text<|/ref|><|det|>[[114, 563, 883, 601]]<|/det|>
+Subsequently, the final representation \(Y_{i}^{m}\) of spot \(i\) in the \(m\) - th modality can be generated by aggregating graph- specific representations according to their attention scores:  
+
+<|ref|>equation<|/ref|><|det|>[[114, 613, 260, 634]]<|/det|>
+\[Y_{i}^{t} = \sum_{t = 1}^{T}\alpha_{i}^{t}\cdot h_{i}^{t} (7)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 646, 883, 684]]<|/det|>
+such that \(Y_{i}^{m} \in R^{d_{3}}\) preserves the raw cell expressions, cell expression similarity, and spatial proximity within modality \(m\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 698, 506, 716]]<|/det|>
+## Between-Modality attention aggregation layer  
+
+<|ref|>text<|/ref|><|det|>[[114, 728, 883, 875]]<|/det|>
+Each individual omic modality provides a partial view of a complex tissue sample, thus requiring an integrated analysis to obtain a comprehensive picture. These views can contain both complementary and contradictory elements, and thus different importance should be assigned to each modality to achieve coherent data integration. Here we use a Between- Modality attention aggregation layer to adaptively integrate the different data modalities. This attention aggregation layer will focus on the more important omics modality by assigning greater weight values to the corresponding representation. Like the Within- Modality layer, we first learn the importance of modality \(m\) by calculating the following coefficient \(w_{i}^{m}\) :  
+
+<|ref|>equation<|/ref|><|det|>[[114, 888, 390, 909]]<|/det|>
+\[w_{i}^{m} = v^{T}\cdot \tanh \left(W_{Be}y_{i}^{m} + b_{Be}\right), (8)\]
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 87, 883, 145]]<|/det|>
+where \(w_{i}^{m}\) are attention coefficients that represent the importance of modality \(m\) to the representation of spot \(i\) . \(W_{Be}\) , \(b_{Be}\) , and \(\nu\) are learnable weight and bias variables, respectively. Similarly, we further normalize the attention coefficients using the softmax function:  
+
+<|ref|>equation<|/ref|><|det|>[[114, 156, 293, 190]]<|/det|>
+\[\beta_{i}^{m} = \frac{\exp{(w_{i}^{m})}}{\sum_{m = 1}^{M}\exp{(w_{i}^{m})}}, (9)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 201, 883, 240]]<|/det|>
+where \(\beta_{i}^{m}\) is the normalized attention score that represents the contribution of modality \(m\) to the representation of spot \(i\) .  
+
+<|ref|>text<|/ref|><|det|>[[114, 252, 883, 291]]<|/det|>
+Finally, we derive the final representation \(Z_{i}\) of spot \(i\) by aggregating each modality- specific representation according to their attention score \(\beta_{i}^{m}\) :  
+
+<|ref|>equation<|/ref|><|det|>[[114, 301, 293, 323]]<|/det|>
+\[Z_{i} = \sum_{m = 1}^{M}\beta_{i}^{m}\cdot y_{i}^{m}. (10)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 335, 883, 373]]<|/det|>
+After model training, the latent representation \(Z_{i} \in R^{d_{3}}\) can be used in various downstream analyses, including clustering, visualization, and DEG detection.  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 385, 241, 404]]<|/det|>
+## Model training  
+
+<|ref|>text<|/ref|><|det|>[[114, 415, 883, 453]]<|/det|>
+The resulting model is trained jointly with three different loss functions for reconstruction loss, correspondence loss, and adversarial loss. Each loss function is described as follows.  
+
+<|ref|>text<|/ref|><|det|>[[114, 465, 883, 578]]<|/det|>
+Reconstruction loss To enforce the learned latent representation to preserve the expression profiles from different modalities, we design an individual decoder for each modality to reverse the integrated representation \(Z_{i}\) back into the normalized expression space. Specifically, by taking output \(Z_{i}\) from the Between- Modality attention aggregation layer as input, the reconstructed representations \(\hat{H}_{1}^{l}\) and \(\hat{H}_{2}^{l}\) from the decoder at the \(l\) - th \((l \in \{1,2, \ldots , L - 1, L\})\) layer are formulated as follows:  
+
+<|ref|>equation<|/ref|><|det|>[[114, 588, 377, 610]]<|/det|>
+\[\hat{H}_{1}^{l} = \sigma (\tilde{A}_{s}Z_{1}^{l - 1}W_{d1}^{l - 1} + b_{d1}^{l - 1}), (11)\]  
+
+<|ref|>equation<|/ref|><|det|>[[114, 620, 377, 643]]<|/det|>
+\[\hat{H}_{2}^{l} = \sigma (\tilde{A}_{s}Z_{1}^{l - 1}W_{d2}^{l - 1} + b_{d2}^{l - 1}), (12)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 655, 883, 713]]<|/det|>
+where \(W_{d1}, W_{d2}, b_{d1},\) and \(b_{d2}\) are trainable weight matrices and bias vectors, respectively. \(\hat{H}_{1}^{l}\) and \(\hat{H}_{2}^{l}\) represent the reconstructed expression matrices for omics modalities 1 and 2, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[114, 726, 883, 763]]<|/det|>
+SpatialGlue's objective function to minimize the expression reconstruction loss is as follows:  
+
+<|ref|>equation<|/ref|><|det|>[[114, 775, 532, 802]]<|/det|>
+\[\mathcal{L}_{recon} = \gamma_{1}\sum_{i = 1}^{N}\left\| x_{i}^{1} - \hat{h}_{i}^{1}\right\|_{F}^{2} + \gamma_{2}\sum_{i = 1}^{N}\left\| x_{i}^{2} - \hat{h}_{i}^{2}\right\|_{F}^{2}, (13)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 814, 883, 890]]<|/det|>
+where \(x^{1}\) and \(x^{2}\) represent the original features of modalities 1 and 2, respectively. \(\gamma_{1}\) and \(\gamma_{2}\) are weight factors that are utilized to balance the contribution of different modalities. Due to the differences of sequencing technologies and molecular types, the feature distributions of different omics assays can vary significantly. As such, the weight factors also vary between different
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 882, 127]]<|/det|>
+spatial multi- omics technologies but are fixed for datasets obtained using the same omics technology.  
+
+<|ref|>text<|/ref|><|det|>[[114, 137, 883, 268]]<|/det|>
+Correspondence loss While reconstruction loss can enforce the learned latent representation to simultaneously capture the expression information of different modality data, it does not guarantee that the representation manifolds are fully aligned across modalities. To deal with the issue, we add a correspondence loss function. Correspondence loss aims to force consistency between a modality- specific representation \(Y_{m}\) and its corresponding representation \(\hat{Y}_{m}\) obtained through the decoder- encoder of another modality. Mathematically, the correspondence loss is defined as follows:  
+
+<|ref|>equation<|/ref|><|det|>[[113, 280, 528, 304]]<|/det|>
+\[\mathcal{L}_{corr} = \gamma_{3}\sum_{i = 1}^{N}\| y_{i}^{1} - \hat{y}_{i}^{1}\|_{F}^{2} + \gamma_{4}\sum_{i = 1}^{N}\| y_{i}^{2} - \hat{y}_{i}^{2}\|_{F}^{2},(14)\]  
+
+<|ref|>equation<|/ref|><|det|>[[113, 316, 546, 345]]<|/det|>
+\[\hat{Y}_{1}^{l} = \sigma \left(\tilde{A}_{s}\left(\sigma (\tilde{A}_{s}Y_{1}^{l - 1}W_{d2}^{l - 1} + b_{d2}^{l - 1})\right)W_{e2}^{l - 1} + b_{e2}^{l - 1}\right), (15)\]  
+
+<|ref|>equation<|/ref|><|det|>[[113, 357, 546, 386]]<|/det|>
+\[\hat{Y}_{2}^{l} = \sigma \left(\tilde{A}_{s}\left(\sigma (\tilde{A}_{s}Y_{2}^{l - 1}W_{d1}^{l - 1} + b_{d1}^{l - 1})\right)W_{e1}^{l - 1} + b_{e1}^{l - 1}\right). (16)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 399, 883, 439]]<|/det|>
+where \(\gamma_{3}\) and \(\gamma_{4}\) are hyper- parameters, controlling the influences of different modality data. We set \(\hat{Y}_{1}^{0} = Y_{1}\) and \(\hat{Y}_{2}^{0} = Y_{2}\) . \(\sigma (\cdot)\) is a nonlinear activation function, i.e., ReLU (Rectified Linear Unit).  
+
+<|ref|>text<|/ref|><|det|>[[114, 449, 883, 525]]<|/det|>
+Adversarial loss To constrain the representations across different modalities in a uniform latent space, we further add a discriminator, i.e., fully connected neural network, in our model. This discriminator tries to align the representations with \(\delta = 1\) denoting modality 1 and \(\delta = 0\) denoting modality 2. Formally, the adversarial loss of the discriminator is defined as:  
+
+<|ref|>equation<|/ref|><|det|>[[113, 536, 536, 562]]<|/det|>
+\[\mathcal{L}_{adv} = \frac{1}{N}\left[\sum_{i = 1}^{N}[\delta_{i}log(p_{i}) + (1 - \delta_{i})\log (1 - p_{i})]\right],(17)\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 575, 883, 612]]<|/det|>
+where \(p_{i}\) and \((1 - p_{i})\) represents the probability scores of the representations assigned to modalities 1 and 2, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[172, 624, 750, 643]]<|/det|>
+Therefore, the overall loss function used for model training is defined as:  
+
+<|ref|>equation<|/ref|><|det|>[[113, 655, 395, 676]]<|/det|>
+\[\mathcal{L}_{total} = \mathcal{L}_{recon} + \mathcal{L}_{corr} + \mathcal{L}_{adv}. (18)\]  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 688, 312, 705]]<|/det|>
+## Implementation details  
+
+<|ref|>text<|/ref|><|det|>[[114, 718, 883, 846]]<|/det|>
+For all datasets, a learning rate of 0.0001 was used. To account for differences in feature distribution across the datasets, a tailored group of weight factors \([y_{1}, y_{2}, y_{3}, y_{4}]\) was assigned to each one. The weight factors were [1, 50, 1, 5] for the SPOTS mouse spleen dataset, [1, 10, 1, 10] for the Stereo- CITE- seq mouse thymus dataset, [1, 2.5, 1, 1] for the spatial- ATAC- RNA- seq mouse brain dataset. The training epochs used for the SPOTS mouse spleen, Stereo- CITE- seq mouse thymus, and spatial- ATAC- RNA- seq mouse brain datasets were 900, 1500, and 1500, respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[114, 859, 300, 876]]<|/det|>
+## Seurat WNN Analysis
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 327]]<|/det|>
+For the SPOTS mouse spleen and Stereo- CITE- seq mouse thymus datasets, we employed the same data preprocessing steps as SpatialGlue and used the same PCA coefficients as input to analysis with Seurat WNN (FindMultiModalNeighbors function) \(^{13}\) . We then adjusted the resolution input to the Seurat FindCluster function to obtain the same number of clusters as SpatialGlue's output. For the Spatial- ATAC- RNA- seq mouse brain dataset, the output obtained from using the same preprocessing steps as SpatialGlue resulted in an excessive number of clusters (more than 20 at a resolution of 0.1). Therefore, we adjusted our preprocessing by first filtering out cells with less than 200 features. Thereafter, we employed SCTransform for normalization and PCA for dimension reduction of the RNA modality. Dimension reduction of the ATAC peaks was performed with Latent Semantic Indexing using the TFIDF and RunSVD functions from the Signac package \(^{20}\) . Seurat WNN analysis was then performed with the top 10 PCs of the RNA modality and 2- 10PCs of the ATAC modality. We discarded the top ATAC PC as it was correlated with sequencing depth.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 341, 290, 363]]<|/det|>
+## Data availability  
+
+<|ref|>text<|/ref|><|det|>[[114, 375, 883, 524]]<|/det|>
+All datasets used in this study are already published and were obtained from public data repositories. The SPOTS mouse spleen data was obtained from the GEO repository (accession no. GSE198353, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE198353) \(^{4}\) , the Stereo- CITE- seq mouse thymus data from BGI and the spatial- ATAC- RNA- seq mouse brain data from AtlasXplore (https://web.atlasxomics.com/visualization/Fan) \(^{3}\) . The details of all datasets used are available in the Methods section. The raw data used in this study have been uploaded to Zenodo and is freely available at https://zenodo.org/record/7879713#.ZE3aOnZBvUk.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 537, 298, 558]]<|/det|>
+## Code availability  
+
+<|ref|>text<|/ref|><|det|>[[116, 571, 883, 608]]<|/det|>
+An open- source Python implementation of the GraphST toolkit is accessible at https://github.com/JinmiaoChenLab/SpatialGlue.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 622, 345, 643]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[115, 656, 883, 750]]<|/det|>
+J.C. conceptualized and supervised the project. Y.L. designed the model. Y.L. developed the SpatialGlue software. Y.L., K.S.A., and J.C. wrote the manuscript. Y.L., J.C., R.S, C.Z, H.X., and K.S.A. performed the data analysis. C.Z. and R.S. ran the Seurat WNN algorithm. Y.L. prepared the figures. J.C., N.R.J.G, L.G.N., and N.H. annotated and interpreted the mouse thymus dataset. S.L., Y.H., M.J., A.C., and X.X generated the mouse thymus dataset.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 763, 332, 784]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[115, 797, 883, 907]]<|/det|>
+We thank Yingrou Tan for her assistance in mouse thymus data interpretation and Min Wu for his comments on the model. The research was supported by A\*STAR under its BMRC Central Research Fund (CRF, UIBR) Award; AI, Analytics and Informatics (AI3) Horizontal Technology Programme Office (HTPO) seed grant (Spatial transcriptomics ST in conjunction with graph neural networks for cell- cell interaction #C211118015) from A\*STAR, Singapore; Open Fund Individual Research Grant (Mapping hematopoietic lineages of healthy and high- risk acute
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 199]]<|/det|>
+myeloid leukemia patients with FLT3- ITD mutations using single- cell omics #OFIRG18nov- 0103) from Ministry of Health, Singapore; National Research Foundation (NRF), Award no. NRF- CRP26- 2021- 0001; the National Research Foundation, Singapore, and Singapore Ministry of Health's National Medical Research Council under its Open Fund- Large Collaborative Grant ("OF- LCG") (MOH- OFLCG18May- 0003). Singapore National Medical Research Council (#NMRC/OFLCG/003/2018).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 212, 328, 234]]<|/det|>
+## Ethics declarations  
+
+<|ref|>text<|/ref|><|det|>[[116, 248, 579, 266]]<|/det|>
+The authors declare that there are no competing interests.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 280, 240, 300]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[110, 309, 884, 890]]<|/det|>
+1. Liu, Y. et al. High-Spatial-Resolution Multi-Omics Sequencing via Deterministic Barcoding in Tissue. Cell 183, 1665-1681. e18 (2020).  
+2. Liu, Y. et al. High-plex protein and whole transcriptome co-mapping at cellular resolution with spatial CITE-seq. Nat. Biotechnol. (2023) doi:10.1038/s41587-023-01676-0.  
+3. Zhang, D. et al. Spatial epigenome-transcriptome co-profiling of mammalian tissues. Nature 616, 113-122 (2023).  
+4. Ben-Chetrit, N. et al. Integration of whole transcriptome spatial profiling with protein markers. Nat. Biotechnol. (2023) doi:10.1038/s41587-022-01536-3.  
+5. Vickovic, S. et al. SM-Omics is an automated platform for high-throughput spatial multi-omics. Nat. Commun. 13, 795 (2022).  
+6. Liao, S. et al. Integrated Spatial Transcriptomic and Proteomic Analysis of Fresh Frozen Tissue Based on Stereo-seq. bioRxiv 2023.04.28.538364 (2023) doi:10.1101/2023.04.28.538364.  
+7. Hudson, W. H. & Sudmeier, L. J. Localization of T cell clonotypes using the Visium spatial transcriptomics platform. STAR Protoc. 3, 101391 (2022).  
+8. Takei, Y. et al. Integrated spatial genomics reveals global architecture of single nuclei. Nature 590, 344-350 (2021).  
+9. Su, J.-H., Zheng, P., Kinrot, S. S., Bintu, B. & Zhuang, X. Genome-Scale Imaging of the 3D Organization and Transcriptional Activity of Chromatin. Cell 182, 1641-1659. e26 (2020).  
+10. Hu, J. et al. SpaGCN: Integrating gene expression, spatial location and histology to identify spatial domains and spatially variable genes by graph convolutional network. Nat. Methods 18, 1342-1351 (2021).  
+11. Long, Y. et al. Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST. Nat. Commun. 14, 1155 (2023).  
+12. Cao, Z.-J. & Gao, G. Multi-omics single-cell data integration and regulatory inference with graph-linked embedding. Nat. Biotechnol. 40, 1458-1466 (2022).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 876, 455]]<|/det|>
+13. Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573-3587. e29 (2021).14. Alexandre, Y. O. & Mueller, S. N. Splenic stromal niches in homeostasis and immunity. Nat. Rev. Immunol. (2023) doi:10.1038/s41577-023-00857-x.15. Borges da Silva, H. et al. Splenic Macrophage Subsets and Their Function during Blood-Borne Infections. Front. Immunol. 6, 480 (2015).16. Backer, R. et al. Effective collaboration between marginal metallophilic macrophages and CD8+ dendritic cells in the generation of cytotoxic T cells. Proc. Natl. Acad. Sci. U. S. A. 107, 216-221 (2010).17. Blackburn, C. C. & Manley, N. R. Developing a new paradigm for thymus organogenesis. Nat. Rev. Immunol. 4, 278-289 (2004).18. Wolf, F. A., Angerer, P. & Theis, F. J. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol. 19, 15 (2018).19. Kipf, T. N. & Welling, M. Semi-Supervised Classification with Graph Convolutional Networks. (2016) doi:10.48550/ARXIV.1609.02907.20. Stuart, T., Srivastava, A., Madad, S., Lareau, C. A. & Satija, R. Single-cell chromatin state analysis with Signac. Nat. Methods 18, 1333-1341 (2021).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 461, 245, 479]]<|/det|>
+## Figure legends  
+
+<|ref|>text<|/ref|><|det|>[[114, 491, 883, 896]]<|/det|>
+Figure 1: Interpretable deep dual-attention model enables the identification of fine-grained cell types in mouse spleen data generated using SPOTS. (a) Overview of the SpatialGlue framework. A spatial multi-omics technology simultaneously measures two distinct types of molecules, e.g., RNA and surface protein, while preserving spatial context of the tissue. SpatialGlue first uses the \(K\) -nearest neighbor (KNN) algorithm to construct a spatial neighbor graph using the spatial coordinates and a feature neighbor graph with the normalized expression data for each omics modality. Each modality has an implemented GNN- encoder that takes its normalized expressions and neighbor graph to learn two graph- specific representations by iteratively aggregating representations of neighbors. To capture the importance of different graphs, we designed a Within-Modality attention aggregation layer to adaptively integrate graph-specific representations and obtain a modality- specific representation. Finally, to preserve the importance of different modalities, SpatialGlue uses a Between-Modality attention aggregation layer to adaptively integrated modality- specific representations and output the final integrated representation of spots. (b) H&E image of the mouse spleen. (c) UMAP plots and spatial clustering of the RNA and protein expression data. (d) Comparison of UMAP plots of RNA and protein modalities and SpatialGlue integrated output, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of different modalities to each cluster in the mouse spleen dataset. (f) Heatmap showing the expression levels of differentially expressed ADTs for each cluster. (g) Normalized ADT levels of key surface markers for T cell (CD3, CD4, CD8), B cell (IgD, B220, CD19), RpMΦ (F4_80, CD68, CD163), MMMΦ (CD169), and Epithelia (EpCAM). (h) Violin plots indicating the expression distribution of two marker genes in the MMMΦ, MZMΦ, and RpMΦ clusters. (i) Neighborhood enrichment of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 882, 143]]<|/det|>
+cell type pairs. (j) Cluster co-occurrence score for each cluster at increasing distances. The full names of the abbreviations RpMΦ, MMMΦ, and MZMΦ are red pulp macro, CD169+ MMM, CD209a+ MZM, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[114, 157, 882, 486]]<|/det|>
+Figure 2: SpatialGlue accurately integrates different datasets of mouse thymus and mouse brain. The mouse thymus data of protein and RNA modalities was acquired with Stereo- CITE- seq, and the mouse brain data of RNA and ATAC modalities was acquired with spatial- ATAC- RNA- seq. (a) dsDNA image of the mouse thymus. (b) Total mRNA counts in the mouse thymus dataset. (c) UMAP plots and spatial clustering of the RNA and protein expression data in the mouse thymus dataset. (d) Comparison of UMAP plots of RNA and protein modalities and SpatialGlue integrated output in the mouse thymus dataset, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of each modality to each cluster in the mouse thymus dataset. (f) Annotated reference mouse brain coronal section from Allen Mouse Brain Atlas. (g) UMAP plots and spatial clustering of the RNA expression and ATAC data in the mouse brain dataset. (h) Comparison of UMAP plots of RNA and ATAC modalities and SpatialGlue integrated output in the mouse brain dataset, all colored by annotated clusters obtained from the integrated output. (i) Modality weights explaining the importance of each modality to each cluster in the mouse brain dataset. The full names of abbreviation used in the annotations of (f) and (h) are, ctx: cerebral cortex, cp: caudoputamen, vl: lateral ventricle, lpo: lateral preoptic area, aca: anterior cingulate area, ls: lateral septal nucleus, aco: anterior commissure, olfactory limb, acb: nucleus accumbens, cc: corpus callosum.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 500, 293, 518]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[115, 530, 882, 603]]<|/det|>
+Supplementary Figure S1: Results on mouse spleen. (a) Spatial plots of all clusters and each cluster identified by SpatialGlue. (b) DEGs of clusters found in tissue samples. (c) UMAP and spatial plots of clusters identified by Seurat WNN. (d) Modality weights from Seurat WNN explaining the importance of each modality to each cluster.  
+
+<|ref|>text<|/ref|><|det|>[[115, 616, 882, 707]]<|/det|>
+Supplementary Figure S2: Results on mouse thymus. (a) Spatial plots of all clusters and each cluster identified by SpatialGlue. (b) Expression of marker genes and proteins for each cell type. (c) DEGs of clusters found in tissue samples. (d) UMAP and spatial plots of clusters identified by Seurat WNN. (e) Modality weights from Seurat WNN explaining the importance of each modality to each cluster.  
+
+<|ref|>text<|/ref|><|det|>[[115, 720, 882, 792]]<|/det|>
+Supplementary Figure S3: Results on mouse brain. (a) Spatial plots of all clusters and each cluster identified by SpatialGlue. (b). DEGs of clusters found in tissue samples. (c) UMAP and spatial plots of clusters identified by Seurat WNN. (d) Modality weights from Seurat WNN explaining the importance of each modality to each cluster.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 45, 142, 70]]<|/det|>
+## Figures  
+
+<|ref|>image<|/ref|><|det|>[[45, 88, 740, 840]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 852, 115, 871]]<|/det|>
+<center>Figure 1 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 892, 928, 957]]<|/det|>
+Interpretable deep dual- attention model enables the identification of fine- grained cell types in mouse spleen data generated using SPOTS. (a) Overview of the SpatialGlue framework. A spatial multi- omics technology simultaneously measures two distinct types of molecules, e.g., RNA and surface protein,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[38, 44, 955, 504]]<|/det|>
+while preserving spatial context of the tissue. SpatialGlue first uses the K- nearest neighbor (KNN) algorithm to construct a spatial neighbor graph using the spatial coordinates and a feature neighbor graph with the normalized expression data for each omics modality. Each modality has an implemented GNN- encoder that takes its normalized expressions and neighbor graph to learn two graph- specific representations by iteratively aggregating representations of neighbors. To capture the importance of different graphs, we designed a Within- Modality attention aggregation layer to adaptively integrate graph- specific representations and obtain a modality- specific representation. Finally, to preserve the importance of different modalities, SpatialGlue uses a Between- Modality attention aggregation layer to adaptively integrated modality- specific representations and output the final integrated representation of spots. (b) H&E image of the mouse spleen. (c) UMAP plots and spatial clustering of the RNA and protein expression data. (d) Comparison of UMAP plots of RNA and protein modalities and SpatialGlue integrated output, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of different modalities to each cluster in the mouse spleen dataset. (f) Heatmap showing the expression levels of differentially expressed ADTs for each cluster. (g) Normalized ADT levels of key surface markers for T cell (CD3, CD4, CD8), B cell (IgD, B220, CD19), RpMΦ (F4_80, CD68, CD163), MMMΦ (CD169), and Epithelia (EpCAM). (h) Violin plots indicating the expression distribution of two marker genes in the MMMΦ, MZMΦ, and RpMΦ clusters. (i) Neighborhood enrichment of cell type pairs. (j) Cluster co-occurrence score for each cluster at increasing distances. The full names of the abbreviations RpMΦ, MMMΦ, and MZMΦ are red pulp macro, CD169<sup>+</sup> MMM, CD209a<sup>+</sup> MZM, respectively.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[45, 40, 800, 789]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 803, 117, 822]]<|/det|>
+<center>Figure 2 </center>  
+
+<|ref|>text<|/ref|><|det|>[[41, 844, 914, 956]]<|/det|>
+SpatialGlue accurately integrates different datasets of mouse thymus and mouse brain. The mouse thymus data of protein and RNA modalities was acquired with Stereo- CITE- seq, and the mouse brain data of RNA and ATAC modalities was acquired with spatial- ATAC- RNA- seq. (a) dsDNA image of the mouse thymus. (b) Total mRNA counts in the mouse thymus dataset. (c) UMAP plots and spatial clustering of the RNA and protein expression data in the mouse thymus dataset. (d) Comparison of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[40, 45, 945, 293]]<|/det|>
+UMAP plots of RNA and protein modalities and SpatialGlue integrated output in the mouse thymus dataset, all colored by annotated clusters obtained from the integrated output. (e) Modality weights explaining the importance of each modality to each cluster in the mouse thymus dataset. (f) Annotated reference mouse brain coronal section from Allen Mouse Brain Atlas. (g) UMAP plots and spatial clustering of the RNA expression and ATAC data in the mouse brain dataset. (h) Comparison of UMAP plots of RNA and ATAC modalities and SpatialGlue integrated output in the mouse brain dataset, all colored by annotated clusters obtained from the integrated output. (i) Modality weights explaining the importance of each modality to each cluster in the mouse brain dataset. The full names of abbreviation used in the annotations of (f) and (h) are, ctx: cerebral cortex, cp: caudoputamen, vl: lateral ventricle, lpo: lateral preoptic area, aca: anterior cingulate area, ls: lateral septal nucleus, aco: anterior commissure, olfactory limb, acb: nucleus accumbens, cc: corpus callosum.  
+
+<|ref|>sub_title<|/ref|><|det|>[[43, 315, 312, 343]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[43, 366, 768, 387]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[59, 404, 330, 556]]<|/det|>
+SupplementaryFigureS1. pdf SupplementaryFigureS2. pdf SupplementaryFigureS3. pdf SupplementaryTableS1. xlsx SupplementaryTableS2. xlsx SupplementaryTableS3. xlsx
+
+<--- Page Split --->

preprint/preprint__b43f149a50c18c934e9437c49ce087f0f79296e17429bb8dd31bfec88d2e934b/images_list.json

@@ -0,0 +1,92 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1: a. Global total MSW generation. b. Global MSW generation per capita. c. Global urban MSW generation per capita. d. Global rural MSW generation per capita",
+    "footnote": [],
+    "bbox": [
+      [
+        220,
+        95,
+        770,
+        430
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Fig. 2: Municipal solid waste (MSW) generation rates in urban and rural areas by scenario. For high-income regions as NAM and EU28, MSW per capita will remain pretty the same independent of the underlying socio-economic pathway. However, the different pathway trajectories have a strong influence on MSW per capita generation in low, and middle-income regions.",
+    "footnote": [],
+    "bbox": [
+      [
+        4,
+        175,
+        978,
+        575
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3: Municipal solid waste (MSW) management in 2015. Urban areas in low-middle income regions have increased MSW collection rates in last years. However, MSW treatment has not improved at the same pace, hence most of the waste is dumped, scattered or is subject to open burning. Rural areas face an even more challenging situation as in low-middle income regions collection rates are just about 35% - 45%. In general, high-income regions have established suitable MSW treatment systems in both urban and rural areas.",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        384,
+        936,
+        655
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4: Global \\(\\mathrm{CH_4}\\) emissions under CLE and MFR scenarios. Faster adoption of measures improving MSW systems will result in an early decrease of MSW ending up in dumpsites/uncontrolled landfills and therefore brings quicker reductions of future \\(\\mathrm{CH_4}\\) emissions from this source. Supplementary Results S2 presents a detailed analysis of the MFR scenarios.",
+    "footnote": [],
+    "bbox": [
+      [
+        90,
+        97,
+        916,
+        430
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5: Global amounts of MSW open burned and related emissions under CLE and MFR scenarios. Reduction fractions of MSW open burned result in the same reduction percentage of particulate matter and air pollutants. Supplementary Results S2 presents a detailed analysis of the MFR scenarios.",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        345,
+        940,
+        667
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6: Regional emissions of \\(\\mathrm{CH_4}\\) and BC from MSW. The target of all modelled scenarios is set to reach \\(\\sim 100\\%\\) of MSW collection and management by 2050. The environmental co-benefits will be obtained at different levels upon the level of socio-economic development and political and institutional arrangements. The different assumptions on policy interventions are then translated into a wide range of future emissions.",
+    "footnote": [],
+    "bbox": [
+      [
+        40,
+        111,
+        940,
+        699
+      ]
+    ],
+    "page_idx": 16
+  }
+]

preprint/preprint__b43f149a50c18c934e9437c49ce087f0f79296e17429bb8dd31bfec88d2e934b/preprint__b43f149a50c18c934e9437c49ce087f0f79296e17429bb8dd31bfec88d2e934b.mmd

@@ -0,0 +1,432 @@
+
+# Potentials for future reductions of global GHG and air pollutant emissions from circular municipal waste management systems  
+
+Adriana Gomez Sanabria ( \(\boxed{\text {gomezsa@iiasa.ac.at}}\) ) International Institute for Applied Systems Analysis https://orcid.org/0000- 0002- 2317- 3946  
+
+Gregor Kiesewetter International Institute for Applied Systems Analysis (IIASA) Zbigniew Klimont International Institute for Applied Systems Analysis https://orcid.org/0000- 0003- 2630- 198X Wolfgang Schöpp International Institute for Applied Systems Analysis (IIASA) Helmut Haberl University of Natural resources and Life Sciences  
+
+## Article  
+
+Keywords: Municipal waste, greenhouse gases, air pollution, methane, SDGs  
+
+Posted Date: June 22nd, 2021  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 512870/v1  
+
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Version of Record: A version of this preprint was published at Nature Communications on January 10th, 2022. See the published version at https://doi.org/10.1038/s41467- 021- 27624- 7.
+
+<--- Page Split --->
+
+# Potentials for future reductions of global GHG and air pollutant emissions from circular municipal waste management systems  
+
+4 Adriana Gomez- Sanabria\\*a,b, Gregor Kiesewetter, Zbigniew Klimont, Wolfgang Schoepp & Helmut 5 Haberlb 6 a Pollution Management Research Group, Energy, Climate and Environment Program, International Institute for Applied Systems Analysis, Laxenburg, Austria 7 b Institute of Social Ecology, University of Natural Resources and Life Sciences, Vienna, Austria  
+
+## 9 Contributions  
+
+10 AGS designed the study, performed the projections, emission simulations and analysis, and prepared the manuscript. GK performed the ambient air pollution concentration calculations. ZK provided expert guidance and contributed to the revision of the manuscript. WS prepared and imported the IAE- WEO activity drivers and provided methodological advice. HH participated in the development of the research and contributed to writing the manuscript. All authors were involved in the discussion during the process.  
+
+## 15 Corresponding author  
+
+16 Correspondence to: Adriana Gomez- Sanabria. Email: gomezsa@iiasa.ac.at  
+
+17 AGS: ORCID: 0000- 0002- 2317- 39  
+
+18 HH: ORCID: 0000- 0003- 2104- 5446
+
+<--- Page Split --->
+
+Recent trajectories of production and consumption patterns have resulted in massively rising quantities of municipal solid waste (MSW). In combination with the large global quantities of mismanaged MSW these increases cause detrimental effects on the environment and climate. Few analyses of the potential environmental co- benefits resulting from the implementation of circular MSW management systems exist. To our knowledge, no global study of possible future scenarios of MSW generation, composition, management, and associated burdens is available that explicitly considers the important differences between urban and rural settings. To help filling this gap, we here develop a systematic approach for evaluating the benefits of implementing circular MSW management systems in terms of their potentials to reduce greenhouse gas emissions (GHG) and air pollution. We also analyse their role in the pursuit of the Sustainable Development Goals (SDGs). Building on the Shared Socioeconomic Pathways (SSPs), we build two sets of global scenarios until 2050, namely baseline and mitigation scenarios. In these scenarios, we assess trajectories of future MSW generation and the impact of MSW management strategies on methane (CH4), carbon dioxide (CO2) and air pollutant emissions. We estimate that future MSW generation could increase to at least 3.7 Gt/yr and at most to 4.3 Gt/yr by 2050, depending on the respective SSP storyline. In 2050, we show that the adoption of mitigation strategies in the sustainability- oriented scenario yields earlier, and major, co- benefits compared to scenarios in which inequalities are reduced but that are focused solely on technical solutions. In 2050, the GHG emissions in the sustainability- oriented scenario amount to 182 Gg CO2eq/yr of CH4, to be released while CO2, particulate matter, and air pollutants from open burning of MSW can be virtually eliminated, indicating that this source of ambient air pollution can be entirely eradicated before 2050. We conclude that significant potentials exist to reduce GHG, and air pollution if circular MSW management systems are implemented. We also demonstrate that the 6.3 target of the SDG 6 can only be achieved through more ambitious sustainability- oriented scenarios that limit MSW generation and improve management.  
+
+Key words: Municipal waste, greenhouse gases, air pollution, methane, SDGs
+
+<--- Page Split --->
+
+Global quantities of municipal solid waste (MSW) generation have grown massively over the last decades, not only due to population growth but also as a result of economic growth and the consequent changes in production and consumption patterns<sup>1,2</sup>. Estimates suggest that the world population generated 1.9 Gt/yr of MSW in 2015 and is expected to generate about 3.5 Gt/yr of MSW in 2050<sup>3</sup>. High- income countries (in terms of the World Bank income classification) generate more waste per capita per year than low- income countries: they are responsible for 34% of the amount of MSW generated each year, even though they account for just 16% of the global population<sup>4</sup>. These large quantities of MSW generated each year necessitate the implementation of appropriate management systems if the additional associated environmental and health impacts should be avoided that would emerge in the absence of suitable treatment facilities<sup>5</sup>. High- income countries can deploy policies and instruments to cope with the rising MSW flows and hence have cleaner and better- organized waste management systems. Examples include the EU Waste Framework Directive 2008/98/EC<sup>6</sup>, the 3R’s strategy in Japan<sup>7</sup> and the Resource Conservation and Recovery Act 1976<sup>8</sup>, 1986 in the United States. However, high- income countries are still mostly not successful in reducing the amount of MSW generated each year<sup>9</sup>. By contrast, low- income countries often lack suitable management systems, which results from the shortage of funds, poor planning, poor implementation of law and lack of technology and expertise<sup>4,10,11</sup>. Additionally, the outsourcing of resource- intensive production and waste exports from high- income to low- income countries exacerbates the environmental problems resulting from inadequate waste management in many of these countries<sup>12</sup>. Often, open burning, littering and poorly managed landfills are the main ways of waste disposal in low- income countries<sup>4</sup>. Open waste burning results in the release of toxic pollutants, e.g., particulate matter (PM), black carbon (BC), organic carbon (OC), carbon oxide (CO), sulphur dioxide (SO<sub>2</sub>), among others, and greenhouse gases (GHG) including carbon dioxide (CO<sub>2</sub>) as well as smaller amounts of methane (CH<sub>4</sub>)<sup>13–15</sup>. Litter harms wildlife and ecosystems, especially marine life. Global marine litter is currently recognized as one of the biggest sources of ocean’s pollution<sup>16,17</sup>. Decomposition of organic matter in landfills can result in the release
+
+<--- Page Split --->
+
+of \(\mathrm{CH}_4^{18}\) , a greenhouse gas that is 28 times more potent per kg emitted than \(\mathrm{CO}_2\) in a 100 year timeframe<sup>19</sup>, and is also a precursor of tropospheric ozone which alters background ozone concentration and therefore impacts human health<sup>20- 22</sup>. In addition to the negative impacts on the environment and climate, these unsustainable practices have well documented adverse effects on human health<sup>23- 25</sup>. BC and OC, which are components of \(\mathrm{PM}_{2.5}\) , are associated with pulmonary disease, heart disease and acute lower respiratory infection<sup>26- 29</sup>. While reducing air pollution has positive health effects, the impact on the climate system is more difficult to assess. Given the complex interaction between air pollutants and GHGs in the atmosphere, polices that aim at reducing both air pollution and GHG emissions at the same time may succeed to reduce some GHG emission at the expense of reducing cooling effects from specific pollutants such as \(\mathrm{BC}^{30}\) . In the past years, research on waste has gone beyond disposal of wastes to assess the linkages between waste and resource use, climate change, air and water pollution. In that context, various studies have looked at emissions from landfills when assessing sectoral and regional contributions to GHG emissions and abatement potentials<sup>31- 34</sup>. Further assessments include the annual National Inventory Submissions of all Parties included in the Annex I of the Convention to the UNFCCC which comprise all reporting on GHG emissions and removals<sup>1</sup>. Current estimates are that landfills contribute about \(15\%\) to global anthropogenic \(\mathrm{CH}_4\) emissions<sup>31</sup>. Other studies show that open burning of MSW is an important contributor to particulate matter and air pollutant emissions<sup>14,35,36</sup>, specifically, it contributes \(11\%\) to total global \(\mathrm{PM}_{2.5}\) emissions and \(6 - 7\%\) to total global BC emissions<sup>35,36</sup>. BC from open burning of waste amounts to \(2 - 10\%\) of global \(\mathrm{CO}_{2\mathrm{eq}}\) emissions<sup>37</sup>.  
+
+However, very few studies comprehensively assess and model MSW at the global scale. A recent study estimates the global trends and environmental impacts of MSW up to \(2100^{3}\) in terms of MSW generation, composition, and treatment, as well as environmental impacts. Other studies look at MSW as a potential source of secondary materials and energy. It is estimated that the relative contribution of energy from waste and wastewater to the global primary energy demand could increase from \(2\%\) to \(9\%\) by 2040 and deliver 64
+
+<--- Page Split --->
+
+EJ of energy per year (1 EJ = \(10^{18}\) Joules) at the end of this period, if circular management systems are installed<sup>38</sup>. Current estimates are that only around \(13\%\) of the global MSW generated is recycled and \(5.5\%\) composted<sup>4</sup>. In a trend scenario perpetuating current conditions, this share is expected to increase to \(39\%\) in 2050 (includes composting and incineration)<sup>3</sup>. Recycling of waste, including composting and anaerobic digestion, can potentially be boosted in a sustainability-oriented scenario, but so far the extent to which that could be achieved has not been quantified.  
+
+Clearly, these assessments provide some insights on the contribution of MSW to GHG and air pollutants emissions as well as a source of energy and secondary materials. However, most of them focus on a single aspect of MSW (i.e., emissions from landfills and open burning) rather than on the MSW management system as such. Studies providing evidence of the potential environmental co- benefits resulting from the implementation of circular MSW management systems are still scarce. Furthermore, to our knowledge, no global analysis exists that considers differences between urban and rural settings and assesses how MSW generation, composition, management and associated environmental burdens might change under alternative, plausible future scenarios. We here fill that gap. Our main motivation is to contribute to improved understanding how different societal choices could transform MSW management practices in order to address global climate, pollution, and sustainability issues. To our knowledge, this is the first global study to show how the Shared Socioeconomic Pathways (SSPs) can be translated into emission baselines (CLE) and mitigation scenarios (MFR) for the MSW sector.  
+
+We present a new method to globally assess the current and future MSW generation in urban and rural areas and associated emissions as well as their implications for ambient \(\mathrm{PM}_{2.5}\) concentrations for a range of future population and macroeconomic developments to 2050 using the GAINS model as framework. These are represented by the five SSPs and a scenario consistent with the future macroeconomic and population pathways of the IEA's World Economic Outlook 2018<sup>39</sup>. Two variant scenarios are developed for each of the six future socioeconomic pathways; a 'Baseline - CLE' and a 'Maximum Technically Feasible Reduction - MFR', in which circular municipal waste management systems are implemented globally. This means that landfilling of MSW is restrained, material recycling rates are increased, technological improvements and
+
+<--- Page Split --->
+
+behavioral measures such as reduction of food and plastic waste generation are assumed to be implemented. Emissions of \(\mathrm{CH_4}\) , \(\mathrm{CO_2}\) (fossil fraction), PM2.5, BC, OC, CO, \(\mathrm{SO_2}\) , NOx, and NMVOCs are calculated for 184 countries/regions (differentiating urban and rural areas) for the period 2010 – 2050. Results are presented at the level of thirteen world regions and the global aggregate. Based on this comprehensive analysis, we quantify the potential reduction of GHG emissions as well as particulate matter and air pollution through circular MSW systems. We also assess which SDGs can be reached or will be failed under the different scenario assumptions. Our detailed representation of the MSW sector and associated emissions and mitigation potentials can be used as input to Integrated Assessments Models (IAMs) applied to develop emission scenarios for the IPCC, support regional and local scale air pollution studies, and inform local and national governments about the likely developments, environmental consequences, and mitigation opportunities in the MSW sector.  
+
+## Results  
+
+## Scenarios of MSW generation until 2050  
+
+Different socioeconomic assumptions underlying each of the SSPs lead to significant differences in future MSW flows (Fig. 1). The lowest quantities of MSW generation in 2050 are expected in SSP3 and SSP4 due to slow economic growth and inequalities between regions which is reflected in different consumption patterns. By contrast, in the SSP5 both income and urbanization rates increase strongly, resulting in a growth of the MSW generation quantities estimated at to 4296 Tg/yr. Interestingly, in a sustainability- oriented scenario (SSP1) MSW generation is expected to be just 10% lower than that in the SSP5 by 2050. However, when boosting the SSP1 with the adoption of measures targeted at reducing food and plastic waste (SSP1_MFR), it will be possible to reduce MSW generation by an additional 20% compared to SSP5 quantities by 2050.
+
+<--- Page Split --->
+
+The amount of MSW generated, its composition as well as prevalent management systems and policies strongly depend on the dynamics of population and economic activity. We parameterized the drivers of MSW as follows: the most important driver of future MSW generation is GDP. Separate elasticities that relate MSW/cap/yr to GDP/cap/yr are estimated for groups of countries representing four different average income levels under the assumption that MSW generation and its composition are highly dependent on average national income levels. The future composition of MSW is recalculated based on the estimated income elasticity of per-capita food waste generation to GDP/cap/yr. MSW composition fractions estimated separately include food, paper, plastic, glass, metal, wood, textile, and other mixed waste.  
+
+Quantities and composition of MSW generated differ between rural and urban populations. Data on rural waste generation are available for a limited number of countries. For countries where data on rural MSW generation are unavailable, rural waste generation is estimated by applying ratios of urban:rural MSW generation per capita for each region that were deriving from the available information for limited number of countries (see Methods). While the uncertainty of the estimate might be high, the split into urban and rural MSW quantities highlights where actions are needed to improve MSW management systems at local levels, allowing for better quantification of impacts and consequently serves better for policy design. Our estimates suggest that urban areas are currently responsible for \(70\%\) of the global MSW generated. In 2050 urban areas are expected to generated \(80\%\) of the total MSW while rural areas are expected generated the remaining \(20\%\) , i.e., MSW per capita in rural areas is expected to be \(50\%\) lower than in urban areas. In general, rural per capita MSW generation is much lower than those in urban areas due to their smaller purchasing power. However, in high-income countries these differences between urban and rural areas shrink over time.
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Fig. 1: a. Global total MSW generation. b. Global MSW generation per capita. c. Global urban MSW generation per capita. d. Global rural MSW generation per capita </center>  
+
+North America (NAM) is likely to continue having the highest average per capita MSW generation in both urban and rural areas by 2050, followed by Oceania and Europe. China is expected to have the highest growth in MSW generation per capita for urban and rural areas increasing by about \(45\%\) compared to 2015. The reason is the stronger economic growth expected in China over the next decade \(^{41}\) . India is expected to generate about \(13\%\) less MSW than China in 2050 across all scenarios. Even though South Asia (SASIA) and Latin America and Caribbean (LCAM) had similar average per capita MSW generation for both urban and rural areas in 2015, per capita MSW generation in Asia is expected to overtake LCAM in 2050 by about \(15\%\) . Even though Africa will experience the highest increase on MSW generation compared to 2015, it is
+
+<--- Page Split --->
+
+likely to continue having the lowest MSW generation per capita in the future (Fig. 2). Supplementary  
+
+Results. S1 displays total, urban, and rural waste generation by region and scenario.  
+
+![](images/Figure_2.jpg)
+
+<center>Fig. 2: Municipal solid waste (MSW) generation rates in urban and rural areas by scenario. For high-income regions as NAM and EU28, MSW per capita will remain pretty the same independent of the underlying socio-economic pathway. However, the different pathway trajectories have a strong influence on MSW per capita generation in low, and middle-income regions. </center>  
+
+Unfortunately, regions generating the highest amounts of MSW quantities per year have the lowest collection rates and the poorest MSW management systems. Average MSW collection rates in Africa, India, SASIA, and China are estimated to be in average of about \(50\% - 60\%\) , having urban areas collection rates of \(\sim 70\%\) and rural areas \(\sim 40\%\) . Moreover, the unsuitable management (i.e., disposed in dumpsites or burned without air pollution controls), of the collected fraction exacerbates the already precarious situation. Based on the detailed MSW activity and management strategies matrix of the GAINS model which comprises eight MSW streams and fourteen treatment technologies \(^{38}\) , our estimates suggest that in 2015, \(43\%\) of the global MSW collected ended up either in landfills ( \(13\%\) ) that are compacted and/or covered but not meeting
+
+<--- Page Split --->
+
+environmental standards to prevent leakage<sup>42</sup>, in unmanaged landfills without any type of management (hereafter referred as dumpsites) (21%), or was openly burned (9%) either directly at the dumpsites (including unintended fires) or in transfer stations. The remaining 29% of the collected waste was either disposed in sanitary landfills (10%), incinerated (high quality with air pollution controls and energy recovery) (7%), recycled (7%), or composted or anaerobically digested (4%), which is mostly happening in high-income countries. From the uncollected fraction, around 20% is estimated to be scattered MSW with a high probability of eventually reaching water courses and 10% openly burned (Fig. 3). The latter estimates are based on global assessments and detailed country-level studies presented in Table 1 in the methods section.  
+
+![](images/Figure_3.jpg)
+
+<center>Fig. 3: Municipal solid waste (MSW) management in 2015. Urban areas in low-middle income regions have increased MSW collection rates in last years. However, MSW treatment has not improved at the same pace, hence most of the waste is dumped, scattered or is subject to open burning. Rural areas face an even more challenging situation as in low-middle income regions collection rates are just about 35% - 45%. In general, high-income regions have established suitable MSW treatment systems in both urban and rural areas. </center>  
+
+Despite legislation banning open burning of MSW in most of the countries, our calculations indicate that around \(16\%\) of global MSW generated (whereof \(55\%\) collected and \(45\%\) uncollected), was openly burned, which is equivalent to \(380 \mathrm{Tg / yr}\) and \(394 \mathrm{Tg / yr}\) in 2010 and 2015, respectively. While in urban areas about \(60\%\) occurs either on transfer stations or dumpsites i.e., in the collected fraction, in rural areas is estimated that about \(80\%\) of the burning occurs in the uncollected fraction. Rural areas often lack appropriate MSW
+
+<--- Page Split --->
+
+management systems and therefore the uncollected waste is usually subject to be dumped, scattered or openly burned<sup>43</sup>.  
+
+If current MSW management strategies are maintained into the future, the expected quantities of MSW disposed of in dumpsites and openly burned would rise proportionally to the increase of MSW quantities. In contrast, in an ideal situation where a circular MSW management system (MFR), is implemented globally, it would be probable to avoid almost all dumping and open burning of MSW in 2050, thereby eliminating the environmental and health burdens associated with current management practices. Circular MSW management systems include restrained landfilling of MSW, increase material recycling rates, technological improvement, and implementation of behavioral measures such as reduction of food and plastic waste generation.  
+
+## Emissions to air  
+
+Our estimates indicate that current \(\mathrm{CH_4}\) emissions from MSW handling account for \(8\%\) (28 Tg/yr) of the global \(\mathrm{CH_4}\) anthropogenic emissions estimated at 344 Tg/yr in \(2015^{31}\) . Under the current management strategies, baseline \(\mathrm{CH_4}\) emissions in 2050 are projected to rise by a factor between 1.7 (SSP3_CLE) and 2 (SSP5_CLE) over the amount observed in 2015, increasing the contribution of MSW to \(13\%\) of the projected global \(\mathrm{CH_4}\) anthropogenic emissions estimated at 450 Tg/yr in \(2050^{31}\) . At the regional level, China, NAM, LCAM, and SASIA emitted the higher \(\mathrm{CH_4}\) from MSW in 2015. If current conditions are maintained until 2050, then India, Middle East, Africa and SASIA will face the highest growth in \(\mathrm{CH_4}\) emissions from MSW, with an increase of about \(60\%\) compared to 2015 levels. The expected rise of the \(\mathrm{CH_4}\) emissions on those regions is due to the increase of MSW generated, couple with the MSW (mis)management as scattered MSW, dumpsites and precarious landfills (cover or compacted without leakage controls or gas recovery) are the main options to deal with the MSW generated thereby increasing \(\mathrm{CH_4}\) emissions.  
+
+\(\mathrm{CH_4}\) emissions from waste deposited of in landfills today will be generated in future years as it depends on the degradability of the organic matter<sup>18</sup>. MSW generation quantities, composition and policy adoption at
+
+<--- Page Split --->
+
+early stages makes a significant difference in the trends of \(\mathrm{CH_4}\) emissions through the years. In a world implementing circular MSW management systems, the maximum diversion of MSW from dumpsites by 2030 is reached in SSP1_MFR with \(91\%\) less compared to the baseline. This is the result of the adoption of MSW reduction measures, speedy implementation of anaerobic digestion to treat organic waste and the establishment of source separated MSW collection systems to increase the recycling of materials. Total elimination of this practice is expected to happen around 2035 in this sustainability- oriented scenario. The adoption of measures is comparatively slower in scenarios depicting high inequalities between and within countries. Therefore, the diversion of MSW from dumpsites takes more time resulting in higher future \(\mathrm{CH_4}\) emissions. With the exception of SSP1_MFR in which \(\mathrm{CH_4}\) emissions are projected to decrease by \(4\%\) in 2030, an increase of about \(1\% - 2\%\) is expected to happen in all other MFR scenarios compared to the corresponding CLE. The maximum \(\mathrm{CH_4}\) emission reduction potential by 2050 will be reached in the SSP1_MFR in which \(\mathrm{CH_4}\) emissions are expected to decrease by \(87\%\) compared to the baseline, thus leaving still 182 CO2eq of \(\mathrm{CH_4}\) to be released in 2050. Other scenarios are expected to release more \(\mathrm{CH_4}\) , namely, SSP3_MFR will leave 646 CO2eq of \(\mathrm{CH_4}\) and SSP5_MFR 292 CO2eq of \(\mathrm{CH_4}\) to be emitted by 2050 which is \(50\%\) and \(80\%\) lower compared to the respective CLE counterparts (Fig. 4).
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Fig. 4: Global \(\mathrm{CH_4}\) emissions under CLE and MFR scenarios. Faster adoption of measures improving MSW systems will result in an early decrease of MSW ending up in dumpsites/uncontrolled landfills and therefore brings quicker reductions of future \(\mathrm{CH_4}\) emissions from this source. Supplementary Results S2 presents a detailed analysis of the MFR scenarios. </center>  
+
+Emissions of particulate matter and air pollutants depend on the quantities of MSW subject to open burning. Our results suggest that open burning of MSW is responsible for \(3.5\mathrm{Tg / yr}\) of \(\mathrm{PM}_{2.5}\) in 2015. BC emissions are estimated to be \(7\%\) and OC \(60\%\) of the \(\mathrm{PM}_{2.5}\) emissions. Overall, \(\mathrm{PM}_{2.5}\) emissions from MSW account for \(8\%\) of the total global anthropogenic \(\mathrm{PM}_{2.5}\) emissions. Global anthropogenic BC emissions are estimated at \(6.0\mathrm{Tg / yr}\) (GAINS) of which, following our results, \(6\%\) are from MSW burning (see supplement Table S3 for estimates for all pollutants). At the regional level, our calculations indicate that SASIA plus India, China, Africa, and LCAM emitted \(89\%\) of the particulate matter and air pollutants from MSW. India and China contributed about \(50\%\) and Africa \(21\%\) and LCAM the remaining \(18\%\) to those aggregate flows in 2015. Although open burning of MSW occurs in the collected and uncollected fraction in both urban and rural areas, most of emissions come from the collected MSW in urban areas. For example, in Indian cities waste handlers burn waste, despite being aware of the ban, mainly due to lack of infrastructure and to prevent accumulation<sup>44</sup>. Furthermore, with the projected growth of MSW generation and if the current conditions
+
+<--- Page Split --->
+
+prevail into the future then the anticipated global emissions of particulate matter and air pollutants from MSW are expected to nearly double in 2050 for all SSPs. SASIA, India, Africa, China and LCAM are expected to be responsible for \(93\%\) of the emissions. Future emissions in the CLE scenarios will increase proportionally to the quantities of MSW open burned. Consequently, the reduction of the fraction of MSW being openly burned translates directly into the same particulate matter and air pollutants emission reduction levels (Fig. 5). In that sense, in the SSP1_MFR, SSP5_MFR and ECLIPSE_V6b_MFR scenarios will be feasible to virtually eliminate open burning and therefore this source of air pollution already in 2030 while in the other scenarios this could potentially happen 10 to 15 years later.  
+
+![](images/Figure_5.jpg)
+
+<center>Fig. 5: Global amounts of MSW open burned and related emissions under CLE and MFR scenarios. Reduction fractions of MSW open burned result in the same reduction percentage of particulate matter and air pollutants. Supplementary Results S2 presents a detailed analysis of the MFR scenarios. </center>
+
+<--- Page Split --->
+
+At a regional level (Fig. 6), the pre- conditions of the MSW management systems in Europe, Oceania and to certain extent NAM show that the level of effort required to reduce emissions is similar across scenarios. This is the result of the historical evolution on MSW management systems together with the already high- income level and appropriate political arrangements in most of these regions. By contrast, all other regions show high variation across scenarios due to the different dynamics. When comparing the scenarios for regions such as China, India, SASIA, and LCAM, we see that in a sustainability- oriented scenario (SSP1_MFR) a speedier decrease in emissions is observed in urban and rural areas compared to the other scenarios. Moreover, the adoption of circular MSW management systems is slower in scenarios representing a world in which inequalities persist resulting in big differences between urban and rural areas. Consequently, higher emissions are expected across the years.
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Fig. 6: Regional emissions of \(\mathrm{CH_4}\) and BC from MSW. The target of all modelled scenarios is set to reach \(\sim 100\%\) of MSW collection and management by 2050. The environmental co-benefits will be obtained at different levels upon the level of socio-economic development and political and institutional arrangements. The different assumptions on policy interventions are then translated into a wide range of future emissions. </center>  
+
+As emissions from MSW burning contribute significantly to ambient \(\mathrm{PM}_{2.5}\) , particularly since the sources are often low- level and spatially located close to population, the improvement of MSW management will also have benefits in ambient \(\mathrm{PM}_{2.5}\) . To illustrate the possible contributions and mitigation potential from
+
+<--- Page Split --->
+
+this sector, we here quantify the contribution of MSW to \(\mathrm{PM}_{2.5}\) levels in different world regions. Calculations follow the approach applied in ref \(^{45}\) and are briefly described in the Methods section below. Differences between the scenarios are driven both by emission changes as well as urbanization trends. Concentrations are highest in India and other South Asia and are expected to grow further under CLE following the emission trends. Other developing regions show similar growth trends but lower absolute concentrations. In China, initial increases level off, peaking around 2035 (SSP1,2,3,4) or 2050 (SSP5). In Europe, North America and Oceania, contributions from MSW burning are much lower since the combustion happens in well-controlled installations and not as open burning. Gradual implementation of better practices and emission controls eventually decreases concentrations to \(\sim\) zero before 2050 in all MFR cases, although this is achievable at different points in time depending on the SSP storyline.  
+
+## Discussion  
+
+Here we present for the first time a systemic assessment of reduction potentials of GHGs and air pollutants emissions from implementing circular MSW management systems under six future socio- economic development pathways. The assessment includes the development of two scenarios, namely baseline (CLE) and maximum feasible mitigation potential (MFR) for each of the pathways. The explicit representation of urban and rural MSW generation, composition and management allows for a deeper analysis of future plausible management and emission trends. This study can assist national, regional, and local governments in developing strategies to limit the release of emissions into the environment as well as support assessments of feasibility and progress in achieving the UN Sustainable Development Goals (SDGs).  
+
+Our results show that future MSW generation quantities are expected to be between 1.7 to 2 times higher in 2050 compared to current levels in all scenarios. Our results also highlight that urban areas are responsible for about \(80\%\) and will continue being responsible for the higher share of MSW generated in the future. The generally high collection rates of MSW in urban areas does not necessarily imply appropriate management. In SASIA, India, China, LCAM and Africa about \(80\%\) of the collected MSW is either dumped or openly
+
+<--- Page Split --->
+
+burned. Furthermore, most of the MSW generated in rural areas is uncollected and thus ends up being illegally dumped, scattered, or openly burned resulting in several environmental impacts related to air pollution and greenhouse gas emissions and other health and environmental impacts out of the scope of this study. Our findings also indicate that in urban areas about \(60\%\) of the open burning occurs either on transfer stations or dumpsites i.e., in the collected fraction, while in rural areas is estimated that about \(80\%\) of the burning occurs in the uncollected fraction.  
+
+In the baseline (CLE), in which current MSW management practices persist without further policy implementation, emissions to air would increase proportionately to the growth in MSW generation. We then developed a set of mitigation scenarios (MFR) to assess the impacts of abatement measures compared to the corresponding baseline (CLE). The common target of our MFR scenarios is to achieve \(\sim 100\%\) of MSW collection and treatment by 2050 through the implementation of circular MSW management systems to simultaneously tackle emissions of \(\mathrm{CH_4}\) , \(\mathrm{CO_2}\) , particulate matter, and air pollutants. Co- benefits are obtained at different stages upon the level of socio- economic development and political and institutional arrangements. Evidently, all countries would benefit from reduced MSW generation and improved management in the sustainability- oriented scenario (SSP1_MFR), however, the additional benefit of respective measures are especially relevant for regions generating large MSW quantities and lacking suitable management systems. We show that the environmental co- benefits of avoided MSW generation combined with the speedy implementation of anaerobic digestion to treat organic waste and the establishment of source separated MSW collection to increase the recycling of materials (SSP1_MFR) yields major and earlier co- benefits in terms of reducing \(\mathrm{CH_4}\) , particulate matter, and air pollutants. However, more ambitious sustainability- oriented scenarios are crucial to meet the waste related SDGs, specially the 6.3 target which aims at "By 2030, improve water quality by reducing pollution, eliminating dumping and minimizing release of hazardous chemicals and materials, halving the proportion of untreated wastewater and substantially increasing recycling and safe reuse globally" \(^{46}\) . We demonstrate that under the current SSP1_MFR, it will
+
+<--- Page Split --->
+
+not be possible to totally eliminate scattered and open burning of MSW by 2030. Under this scenario the realization of the objective will be obtained five years later i.e., in the year 2035.  
+
+Our analysis also suggest that in 2030, 881 Gg \(\mathrm{CO}_{2}\mathrm{eq}\) of \(\mathrm{CH}_{4}(\mathrm{GWP}_{100}\) of \(28\mathrm{CO}_{2}\mathrm{eq}^{19}\) ) will still be released in the SSP1_CLE. Nonetheless, this is \(13\%\) lower compared to the \(\mathrm{CH}_{4}\) emissions expected in the SSP2_CLE, SSP3_CLE and SSP4_CLE and \(11\%\) lower in comparison to the SSP5_CLE and Eclipse_V6b_CLE. Considering that in 2030 high emissions of \(\mathrm{CO}_{2}\) from open burning of MSW would still be released in SSP2_MFR, SSP3_MFR, SSP4_MFR, the total average GHG emissions ( \(\mathrm{CH}_{4}\) , and \(\mathrm{CO}_{2}\) ) in these scenarios will sum up to an average of about \(1079\mathrm{CO}_{2}\mathrm{eq}\) , that is \(18\%\) higher than the emissions expected in the SSP1_MFR. In 2050, SSP1_MFR leaves \(182\mathrm{Gg}\mathrm{CO}_{2}\mathrm{eq}\) of \(\mathrm{CH}_{4}\) , to be released. That is \(37\%\) lower than the SSP5_MFR and Eclipse_V6b_MFR and 3.5 times lower than the expected emissions in the SSP3_MFR. These variation in emissions can make a substantial difference when considering that the world should stay below 1.5 degrees global warming i.e., the world can emit as maximum as \(10\mathrm{Pg}\mathrm{CO}_{2}\mathrm{eq}\) /yr of all GHGs in \(2050^{47}\) .  
+
+The reduction of MSW being openly burned translates into the same reduction level of emissions of particulate matter and air pollutants. Under the development of SSP1_MFR, SSP5_MFR and ECLIPSE_V6b_MFR, the maximum emission reduction potential will be realized in 2030 whereas in the SSP2_MFR will take 5 years more i.e., in 2040 and for the SSP3_MFR and SSP4_MFR 10 years more i.e., in 2045. At the same time, MSW combustion contributes to ambient \(\mathrm{PM}_{2.5}\) – in some world regions, this contribution is substantial. Most low-income countries, and particularly those with already high concentrations, show an increasing trend from this source under all SSPs, highlighting the importance of counteracting. The positive message is that mitigation is possible and the MSW contribution to ambient \(\mathrm{PM}_{2.5}\) can be virtually eliminated by 2050. However, this will not happen by itself.  
+
+Comparison to other studies: Our calculations suggest that the world generated 2289 Tg/yr of MSW in 2015. Estimates from other studies vary from \(1999^{3}\) to \(2010^{4}\) Tg/yr for the same year. Past assessments estimated
+
+<--- Page Split --->
+
+global MSW generation between \(2000^{48}\) to \(2400 \mathrm{Tg / yr}^{14}\) in 2010. Looking at MSW generation projections, our estimate for the SSP3 and SSP4 in 2050 are similar to the \(3539 \mathrm{Tg / yr}\) projected by Chen et al., 2020 (ref \(^{4}\) ). Our calculations suggest that although the SSP1 represents a sustainability- oriented pathway, MSW quantities in the baseline are foreseen to reach \(3901 \mathrm{Tg / yr}\) in 2050, which is only \(10\%\) lower than the expected MSW amounts in the SSP5. Our projection for MSW generation in the SSP2 is \(3801 \mathrm{Tg / yr}\) while ref \(^{3}\) estimated a MSW generation of about \(3500 \mathrm{Tg / yr}\) in 2050 for the same scenario. However, this estimate is more comparable with our SSP3 and SSP4 projection. The ECLIPSE_V6b_CLE (3948 \(\mathrm{Tg / yr}\) ) is comparable to the SSP1. At the regional level, we find that India is expected to generate about \(13\%\) less MSW than China in 2050 across all scenarios. This contrasts findings ref \(^{4}\) , in which projected MSW generation in India was about \(40\%\) higher than the projection for China in 2050. However, our finding for India is in line with the projection carried out by ref \(^{49}\) . Furthermore, the average per capita MSW generation in China is projected to be between \(30\%\) - \(40\%\) higher than those in India. The fact that estimates for 2010 are lower than those in 2015 and the variability of the results reflect on the one hand, the uncertainty of the data and on the other hand the differences of the methodologies used to derive these numbers. Furthermore, Our estimate of MSW openly burned is \(61\%\) lower than the estimate of ref \(^{14}\) , who estimated that \(40\%\) or an equivalent of \(970 \mathrm{Tg / yr}\) of total MSW generated in 2010 was openly burned (whereof \(64\%\) at residential sites and \(36\%\) at unmanaged dumpsites) and \(57\%\) higher than the estimate of ref \(^{36}\) , who estimated that about \(115 \mathrm{Tg / yr}\) - \(160 \mathrm{Tg / yr}\) of MSW was openly burned in 2010. Differences in estimated quantities can be attributed to variations in the per capita MSW generation rates adopted referring partly to different data sources, but also to differences in the methodology used to estimate the fraction of waste openly burned. While the assumption in ref \(^{14}\) refers to a fraction recommended in the IPCC (2006) guidelines, we develop our own method which we believe better represents the complexity of the MSW sector e.g., in terms of the urban- rural split and the country/region- specific MSW composition and MSW management pathways (see Methods). The differences of the estimates puts a magnifying glass on the urgency to develop national standardized MSW reporting systems, which in addition of being key to governments for the implementation
+
+<--- Page Split --->
+
+and evaluation of MSW treatment, can serve as part of the monitoring system of GHGs, air pollution and SDGs.  
+
+Our estimations indicate that current \(\mathrm{CH_4}\) emissions from MSW handling account for \(8\%\) (28 Tg) of the global \(\mathrm{CH_4}\) anthropogenic emissions estimated at \(344~\mathrm{Tg}\) in \(2015^{31}\) . Our estimate is \(17\%\) lower than the one estimated by ref \(^{35}\) and which has been adopted within the CMIP6 project \(^{50}\) . It is difficult to assess the level of agreement between both studies as estimates from ref \(^{35}\) include MSW and industrial waste while the focus of this study is on MSW and the importance to properly represent the sector for climate and air pollution assessments. However, comparing \(\mathrm{CH_4}\) emissions from MSW in the Eclipse_V5a \(^{36}\) to this study, we can see that the estimate in the latter is \(30~\mathrm{Tg / yr}\) or \(6\%\) higher.  
+
+Recent global \(\mathrm{CO_2}\) emissions area assessed at of 39153 Tg/yr in 2015, whereof 130 Tg/yr or \(0.33\%\) are generated from waste combustion (including industrial and municipal sources) \(^{35,51}\) . Ref \(^{14}\) calculates \(\mathrm{CO_2}\) emissions from open burning of MSW of \(1413~\mathrm{Tg / yr}\) in 2010, estimate that is around 10 to 15 times higher than that from ref \(^{35,51}\) and the one from this study.  
+
+In 2010, emissions of \(\mathrm{PM_{2.5}}\) , BC, and OC have been assessed at 6.1, 0.6 and 5.1 Tg, respectively \(^{14}\) . Our estimates are comparatively lower to those results. In contrast, our results for particulate matter are \(60\%\) higher than those from ref \(^{36}\) . In both cases the differences are related to the assumed quantities of MSW openly burned. Other studies \(^{35,51}\) have estimated BC and OC emissions from waste of \(0.7~\mathrm{Tg}\) and \(4.2~\mathrm{Tg}^{35}\) , respectively (Supplementary Results S3 show a comparison of different studies for different pollutants).  
+
+## Conclusions  
+
+Significant potentials exist to reduce GHG, and air pollution provided the implementation of circular MSW management systems. The 6.3 target of the SDG 6 can only be achieved through more ambitious sustainability- oriented scenarios that limit MSW generation and improve management. Similarly, these kinds of scenarios can directly contribute to the achievement of other SDGs, especially SDG 7, 9, 12, 14 and
+
+<--- Page Split --->
+
+15. Our results highlight the importance of acting at various fronts, namely, consumers behavior, technological development, technology transfer and institutional coordination. For instance, the benefits from reduction of MSW generation can be jeopardized by social and economic inequalities between and within regions which could restrain the adoption and implementation of measures to improve MSW management systems. Furthermore, for a world focused solely on end-of-pipe solutions will be also beneficial the implementation of policies targeted at reducing MSW generation. The finding is that the development of measures at the consumer side will not bring the expected benefits in terms of emissions reduction if quicker and responsible actions are not taken to bring MSW management systems as an important point in governmental agendas. Finally, we see that the majority of countries have developed some kind of legislation regarding the improvement of municipal solid waste management systems, however, the compliance is highly uncertain. A solid system for the reporting of MSW couple with a transparent systematic follow-up of policy enforcement will help to reduce the uncertainty of the estimates as well as will provide clearer insights into the efforts needed by countries to meet their climate, air pollution and SDGs commitments.  
+
+## Methods  
+
+The methodology for developing MSW generation scenarios and associated greenhouse gas and air pollutant emissions involves the following five elements: 1. Socioeconomic drivers are taken from the Shared Socioeconomic Pathway (SSP) Scenarios for the five SSPs \(^{52}\) and from the World Energy Outlook and UNDESA \(^{53}\) for the Eclipse_V6b_CLE (Supplementary Methods S4 presents a short description of the SSPs storylines). 2. The country-specific generation in per capita MSW is driven by expected growth in average per capita income as described in the Supplement of ref \(^{38}\) and further developed in this study (Supplementary Methods Fig. S2 and Fig.S3 show GDP per capita and urbanization rates). 3. Estimation of emissions draw on the methodologies presented in ref \(^{33,36,54}\) , but are extended to improve source-sector resolution and accommodate for new, MSW sector-specific, information. 4. Implementation of the current legislation for
+
+<--- Page Split --->
+
+waste management adopted before 2018. 5. Implementation of circular waste management systems are developed in accordance with the EU's waste management hierarchy - Directive 2008/98/EC<sup>6</sup>. The IIASA- GAINS model is used as a framework to carry out this assessment.  
+
+Municipal waste generation (MSW) activity and its characteristics.  
+
+Current MSW generation quantities, composition, collection rates, and waste management practices are retrieved from several sources, including national official statistics, peer- reviewed literature, and technical reports (see supplement of Gómez- Sanabria et al., 2018). The driver used to project future per capita MSW generation is GDP per capita. This is linked to MSW generation using elasticities estimated following the methodology first developed in ref<sup>33</sup> and further developed in ref<sup>55</sup>. This methodology is further developed in this study (Supplementary Methods S6). Separate elasticities are estimated for groups of countries representing four different average income levels under the assumption that MSW generation and its composition are highly dependent on average national income levels. Furthermore, MSW composition is recalculated based on the estimated income elasticity to per capita food waste generation. MSW composition fractions estimated separately include food, paper, plastic, glass, metal, wood, textile, and other waste. This last fraction includes ordinary mixed waste and may in some cases also include bulk waste.  
+
+Quantities and composition of MSW generated by rural and urban population are different. Data on rural waste generation is available for a limited number of countries, when underlying data on rural MSW generation is unavailable, rural waste generation is estimated by applying different shares related to the specific urban MSW generation rate per capita within specific region and using Eq. (1). This approach is likely to be an improved version of the one- half rural- urban waste generation ratio used by some studies<sup>4,56</sup> because it captures the differences between regions (Supplementary Methods S7 presents the adopted rural urban rates for different regions).  
+
+\[MSW_{u} = MSW_{t}*\left(\frac{P_{u}}{P_{u} + \left(R_{t^{\prime} / u}\right)^{*}P_{r}}\right) \quad (1)\]
+
+<--- Page Split --->
+
+where \(MSW_{t}\) is total MSW generated in a country/region, \(MSW_{u}\) and \(MSW_{r}\) are MSW generated in urban and rural areas, respectively, \(R_{(r / u)}\) represents rural per capita MSW generation as a fraction of the per capita urban MSW generation, and \(P_{u}\) and \(P_{r}\) is rural are urban and rural population, respectively.  
+
+## Open burning of MSW.  
+
+In countries without proper implementation of waste legislation, waste mismanagement is aggravated by poor waste separation at the source, low collection rates and low budget allocated to the waste sector \(^{40}\) . In the absence of reliable waste management systems, dumping and open burning of MSW, either at residential or dumpsites, become the only alternatives to reduce waste- volumes \(^{13,14}\) . Total MSW openly burned is estimated here as the sum of the fractions of uncollected MSW openly burned and collected MSW openly burned at dumpsites and transfer stations in urban and rural areas. The starting point to derive the quantities of MSW openly burned is the total MSW generated in urban and rural areas. Waste amounts are then split into collected and uncollected waste for urban and rural areas, respectively. Collected waste includes MSW collected by official authorities but also (recyclable) waste collected by the informal sector. Information on collection rates is gathered from sources presented in \(^{55}\) and complemented from information available in \(^{4,56}\) . The fraction of uncollected waste is then split into scattered waste or waste openly burned. The fraction of uncollected waste openly burned is assigned based on the information presented in Table 1, considering the current implementation of waste related legislation, income level, collection rates, and urbanization rate of each region. The fraction of collected MSW openly burned is estimated at 10% - 20% of the waste ending up in dumpsites, partly due to self- ignition resulting from poor management and partly due to deliberate burning to reduce waste volumes. In addition, a fraction of the collected waste is assumed to be burned at the transfer station or before reaching the disposal site, which is the case in several developing countries \(^{57}\) . Fractions of MSW openly burned, either on the streets or at dumpsites and transfer stations, are dependent
+
+<--- Page Split --->
+
+on the improvement of the MSW management systems and enforcement of the waste and air pollution legislation. Improvement of waste treatment systems results in reduction of the frequency of MSW openly burned \(^{58}\) . The quantification of these fractions is however highly uncertain. Literature provides a few different methodologies to estimate the amounts of waste openly burned (Table 1). The IPCC (2006) \(^{18}\) suggests 0.6 as a representative value for the fraction of total available waste to be burned that is actually openly burned. This assumption is used by Wiedinmyer et al., 2014 to estimate GHGs and air pollutants from open burning of waste. Bond et al., (2004) \(^{59}\) assumed lower rates of open burning of waste in rural areas in developing countries based on the statement that most of the waste in rural areas is biodegradable. Table 1 also shows that in many cases the default representative value of the IPCC maybe inadequate for several regions.  
+
+In general, the quantification of MSW openly burned in region \(i\) and year \(y - MSW_{(ob)iy}\) is calculated as the sum of MSW openly burned in urban areas \(MSW_{(obu)}\) and MSW openly burned in rural areas \(MSW_{(obr)}\) applying Eq (2).  
+
+\[MSW_{(ob)iy} = MSW_{(obu)iy} + MSW_{(obr)iy}\]  
+
+Where,  
+
+\[MSW_{(obu)iy} = [(MSW_{(u)iy}*C_{(u)iy}*(B_{0u} + B_{1u})) + (MSW_{(u)iy}*(1 - C_{(u)iy})*B_{2u})]\] \[MSW_{(obr)iy} = [(MSW_{(r)iy}*C_{(r)iy}*(B_{0r} + B_{1r})) + (MSW_{(r)iy}*(1 - C_{(r)iy})*B_{2r})]\]  
+
+Where, \(MSW_{(u)iy}\) and \(MSW_{(r)iy}\) are the total amounts of MSW generated in urban and rural areas, respectively. \(C_{(u)iy}\) and \(Coll_{(r)iy}\) are the MSW collection rates in urban and rural areas, respectively. \(\beta_{0u}\) and \(\beta_{0r}\) represent the fractions of collected MSW openly burned on transfer stations and \(\beta_{1u}\) and \(\beta_{1r}\) represent the fractions of collected MSW openly burned at dumpsites in urban and rural areas, respectively. \(\beta_{2u}\) and \(\beta_{2r}\) are the fractions of uncollected waste openly burned in urban and rural areas, respectively.  
+
+Emission estimations.
+
+<--- Page Split --->
+
+Emissions of non- \(\mathrm{CO}_{2}\) greenhouse gases and air pollutants \((E)\) by source \((s)\) and region \((i)\) are calculated in GAINS using Eq (3) \(^{54}\) :  
+
+\[E_{it} = \sum_{sit}A_{is}*ef_{sm}*Appl_{itsm}\]  
+
+where \(A_{is}\) is the activity data, i.e., the amount of MSW generated before management, \(ef_{sm}\) is the emission factor subject to technology \(m\) , and \(Appl_{itsm}\) is the application rate of the technology \(m\) to the activity \(A_{is}\) . The GAINS model matrix comprises fourteen different MSW waste management technologies including different types of source separation, recycling and treatment, different types of solid waste disposal sites and different types of incineration technologies and open burning of waste (Supplementary Methods 8). This extensive characterization of alternative treatment flows allows for a detailed representation of the solid waste management system and its emissions at the national/regional level. Emission factors for \(\mathrm{CH}_{4}\) and \(\mathrm{CO}_{2}\) are developed according to the 2006 IPCC Guidelines, Volume 5, Chapter 3 and Chapter \(5^{18}\) . PM emission factors are adopted from ref \(^{36}\) . These are 8.75 for \(\mathrm{PM}_{2.5}\) , 5.27 for OC and 0.65 g/kg for BC. Emission factors for \(\mathrm{SO}_{2}\) , NOx and NMVOC are adopted from ref \(^{60}\) and are consistent with ref \(^{14}\) . These are 0.5 for \(\mathrm{SO}_{2}\) , 3.74 for NOx, and 7.5 g/kg for NMVOC. The \(\mathrm{PM}_{2.5}\) concentrations are obtained using the annual \(\mathrm{PM}_{2.5}\) emissions applying a simplified version of the atmospheric calculation in the GAINS model \(^{45}\) . Those estimates build on a linearized representation of full atmospheric chemistry model simulations. Here, an atmospheric transfer coefficient is developed to related \(\mathrm{PM}_{2.5}\) emissions to ambient \(\mathrm{PM}_{2.5}\) concentrations from MSW burning.  
+
+## Description of the scenarios.  
+
+The baseline scenarios associated with the six socio- economic pathways describe the expected developments of municipal solid waste generation and management systems under current legislation 'CLE', hereafter baseline, i.e., assuming no further policies affecting the MSW sector are adopted until 2050. In addition, for each baseline an alternative scenario is constructed, which considers full implementation of circular MSW
+
+<--- Page Split --->
+
+management systems globally and is referred to as the maximum technically feasible reduction 'MFR' scenario, hereafter mitigation scenario. Note that the technical frontier is explored here without taking account of the cost to implement various waste management strategies.  
+
+The MFR scenario is developed according to the SSP narratives and assumes a maximum technically feasible phase- in of a waste management system that is fully consistent with the EU's waste management hierarchy (Directive 2008/98/EC). This means that a first priority is given to technologies that circulate materials, thereafter to technologies that recover energy, and only as a last resort to well managed landfills. The following maximum recycling potentials of waste streams are applied: \(90\%\) of municipal paper and textile waste and \(80\%\) of municipal plastic and wood waste can be recycled. It is further assumed that \(100\%\) of food waste can be source separated and treated in anaerobic digesters with biogas recovery. These MFR potentials are adopted in consonance with the socioeconomic development for each scenario. Supplementary Methods S9 presents a description of the MFR management narratives specified for each scenario along with the regional aggregation.  
+
+## Uncertainty  
+
+Regarding uncertainty, several data inputs (activity data, emission factors, type of management) go into the estimations and therefore is difficult to do a quantitative uncertainty estimation. Historical estimates of MSW generation, collection, management, and related emissions have associated uncertainties resulting from the different definitions of MSW coupled with contradictory reported values for generation and composition. The quality of the data suffers from inconsistencies in the definition of MSW generation across countries. In some cases, amounts reported for MSW generation correspond to the gross quantities of waste collected and in other cases to the MSW quantities left for landfill after quantities separated for treatment have been deducted. In developed countries, in particular in Europe, MSW covers household waste and waste that is similar in nature and composition. In developing countries, data on waste suffers from incomplete characterizations and clear definitions of the fractions and source sectors included in the
+
+<--- Page Split --->
+
+MSW are often lacking. These uncertainties are relatively high in developing countries compared to developed countries as in various cases data availability is quite limited in the former case \(^{3}\) . Additionally, some data reported for generation and collection refers to urban areas rather than national totals \(^{4,40}\) , which makes necessary to adopt assumptions based on dedicate studies for particular regions and expert knowledge to arrive at reasonable national MSW generation rates and attributions to urban and rural waste amounts. These uncertainties become bigger when estimating fractions of MSW openly burned as this information is in most of the cases not attainable. Moving to emission factors, \(\mathrm{CH_4}\) emission factors are based on the IPCC Guidelines \(2006^{18}\) , thereby carry out the uncertainties there described. Emissions factors for air pollutants and particulate matter depend on the composition of waste and burning conditions. Although we adopted the most recognized emission factors in the scientific arena, we acknowledge that large uncertainties are related to the values (uncertainties can be seen in ref \(^{14}\) ). Concerning uncertainty in projections, this is by some means assessed by adopting alternative activity scenarios which allows the comparison of the different estimates and reflect the sensitivities of the proposed measures to input assumptions \(^{63}\) . In general, there is a global need to improve information on MSW generation rates, treatment and level of policy implementation \(^{3}\) . Regardless of the uncertainties, we demonstrate the importance of improving global estimates of GHGs and air pollutant emissions from MSW and highlight the considerable role of this sector when assessing the respective mitigation potentials.  
+
+## Data Availability  
+
+The data used for this analysis is available in the Supplementary Information and excel spreadsheet.  
+
+## References  
+
+1. Krausmann, F. et al. Long-term trends in global material and energy use. in Social Ecology 199–216 (Springer, 2016).  
+2. Tisserant, A. et al. Solid waste and the circular economy: A global analysis of waste treatment and waste footprints. Journal of Industrial Ecology 21, 628–640 (2017).  
+3. Chen, D. M.-C., Bodirsky, B. L., Krueger, T., Mishra, A. & Popp, A. The world’s growing municipal solid waste: Trends and impacts. Environmental Research Letters (2020).  
+4. Kaza, S., Bhada-Tata, P. & Van Woerden, F. What a waste 2.0. A global snapshot of solid waste management to 2050. (2018).
+
+<--- Page Split --->
+
+5. Yadav, P. & Samadder, S. R. A global prospective of income distribution and its effect on life cycle assessment of municipal solid waste management: a review. Environmental Science and Pollution Research 24, 9123-9141 (2017).  
+
+6. Directive 2008/98/EC of the European parliament of the Council of European Union. (2008).  
+
+7. Ministry of the Environment. History and current state of waste management in Japan. (2012).  
+
+8. Resource Conservation and Recovery Act of 1976 in the United States. (1976).  
+
+9. Van Ewijk, S. & Stegemann, J. A. Limitations of the waste hierarchy for achieving absolute reductions in material throughput. Journal of Cleaner Production 132, 122-128 (2016).  
+
+10. Manaf, L. A., Samah, M. A. A. & Zukki, N. I. M. Municipal solid waste management in Malaysia: Practices and challenges. Waste Management 29, 2902-2906 (2009).  
+
+11. Ghanimeh, S. et al. Two-Level Comparison of Waste Management Systems in Low-, Middle-, and High-Income Cities. Environmental Engineering Science 36, 1281-1295 (2019).  
+
+12. Parajuly, K. & Fitzpatrick, C. Understanding the Impacts of Transboundary Waste Shipment Policies: The Case of Plastic and Electronic Waste. Sustainability 12, 2412 (2020).  
+
+13. Sharma, G. et al. Gridded Emissions of CO, NOx, SO2, CO2, NH3, HCl, CH4, PM2.5, PM10, BC, and NMVOC from Open Municipal Waste Burning in India. Environ. Sci. Technol. 53, 4765-4774 (2019).  
+
+14. Wiedinmyer, C., Yokelson, R. J. & Gullett, B. K. Global Emissions of Trace Gases, Particulate Matter, and Hazardous Air Pollutants from Open Burning of Domestic Waste. Environ. Sci. Technol. 48, 9523-9530 (2014).  
+
+15. Young Koo, Y. K. and K., Wooram and Jo, Young Min. Release of Harmful Air Pollutants from Open Burning of Domestic Municipal Solid Wastes in a Metropolitan Area of Korea. Aerosol and Air Quality Research 13, 1365-1372 (2013).  
+
+16. Walker, T. & Xanthos, D. A call for Canada to move toward zero plastic waste by reducing and recycling single-use plastics. Resources, Conservation and Recycling 133, 99-100 (2018).  
+
+17. Jambeck, J. R. et al. Marine pollution. Plastic waste inputs from land into the ocean. Science 347, 768-771 (2015).  
+
+18. IPCC. IPCC Guidelines for National Greenhouse Gas Inventories 2006. (2006).  
+
+19. IPCC. The Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Geneva, Switzerland (2014).  
+
+20. Melamed, M. L., Schmale, J. & von Schneidemesser, E. Sustainable policy—key considerations for air quality and climate change. Current Opinion in Environmental Sustainability 23, 85-91 (2016).  
+
+21. Zhang, Y. et al. Tropospheric ozone change from 1980 to 2010 dominated by equatorward redistribution of emissions. Nature Geoscience 9, 875-879 (2016).  
+
+22. Anenberg, S. et al. Global air quality and health benefits of mitigating short-lived climate forcers. (2011).
+
+<--- Page Split --->
+
+23. Andersson, K. et al. Sanitation, Wastewater Management and Sustainability: from Waste Disposal to Resource Recover. (2016).  
+
+24. Anenberg, S. C. et al. Global Air Quality and Health Co-benefits of Mitigating Near-Term Climate Change through Methane and Black Carbon Emission Controls. Environ Health Perspect 120, 831-839 (2012).  
+
+25. Das, B., Bhave, P. V., Sapkota, A. & Byanju, R. M. Estimating emissions from open burning of municipal solid waste in municipalities of Nepal. Waste Management 79, 481-490 (2018).  
+
+26. Arnold, C. Disease burdens associated with PM2.5 exposure: how a new model provided global estimates. Environmental health perspectives 122, A111-A111 (2014).  
+
+27. Beelen, R. et al. Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22 European cohorts within the multicentre ESCAPE project. The Lancet 383, 785-795 (2014).  
+
+28. Brauer, M., Amann, M. & Burnet, R. T. Global Burden of Disease Assessment (draft). (2012).  
+
+29. Janssen, N. A. H. et al. Black carbon as an additional indicator of the adverse health effects of airborne particles compared with PM10 and PM2.5. Environ Health Perspect 119, 1691-1699 (2011).  
+
+30. Harmsen, M. J. H. M. et al. Co-benefits of black carbon mitigation for climate and air quality. Climatic Change (2020) doi:10.1007/s10584-020-02800-8.  
+
+31. Höglund-Isaksson, L., Gómez-Sanabria, A., Klimont, Z., Rafaj, P. & Schöpp, W. Technical potentials and costs for reducing global anthropogenic methane emissions in the 2050 timeframe -results from the GAINS model. Environmental Research Communications 2, 025004 (2020).  
+
+32. Yusuf, R. O., Noor, Z. Z., Abba, A. H., Hassan, M. A. A. & Din, M. F. M. Methane emission by sectors: A comprehensive review of emission sources and mitigation methods. Renewable and Sustainable Energy Reviews 16, 5059-5070 (2012).  
+
+33. Höglund-Isaksson, L. Global anthropogenic methane emissions 2005-2030: technical mitigation potentials and costs. (2012) doi:10.5194/acpd-12-11275-2012.  
+
+34. Saunois, M., Jackson, R. B., Bousquet, P., Poulter, B. & Canadell, J. G. The growing role of methane in anthropogenic climate change. Environ. Res. Lett 11, 12 (2016).  
+
+35. Hoesly, R. M. et al. Historical (1750-2014) anthropogenic emissions of reactive gases and aerosols from the Community Emissions Data System (CEDS). Geoscientific Model Development 11, 369-408 (2018).  
+
+36. Klimont, Z. et al. Global anthropogenic emissions of particulate matter including black carbon. Atmospheric Chemistry and Physics 17, 8681-8723 (2017).  
+
+37. Reyna-Bensusan, N. et al. Experimental measurements of black carbon emission factors to estimate the global impact of uncontrolled burning of waste. Atmospheric Environment 213, 629-639 (2019).  
+
+38. Gómez Sanabria, A., Höglund Isaksson, L., Rafaj, P. & Schöpp, W. Carbon in global waste and wastewater flows-its potential as energy source under alternative future waste management regimes. Advances in Geosciences 45, 105-113 (2018).  
+
+39. IEA. World Energy Outlook 2018. (2018).
+
+<--- Page Split --->
+
+677 40. Hoornweg, D. & Bhada-Tata, P. What a waste. A global review of solid waste management. (2012).  
+
+679 41. OECD. OECD Economic Surveys. China. (2019).  
+
+680 42. Calvo, F., Moreno, B., Zamorano, M. & Szanto, M. Environmental diagnosis methodology for municipal waste landfills. Waste Management 25, 768- 779 (2005).  
+
+682 43. Mihai, F.- C. & Grozavu, A. Role of waste collection efficiency in providing a cleaner rural environment. Sustainability 11, 6855 (2019).  
+
+684 44. Ramaswami, A., Baidwan, N. K. & Nagpure, A. S. Exploring social and infrastructural factors affecting open burning of municipal solid waste (MSW) in Indian cities: A comparative case study of three neighborhoods of Delhi. Waste Management & Research 34, 1164- 1172 (2016).  
+
+687 45. Amann, M. et al. Reducing global air pollution: the scope for further policy interventions. Philosophical Transactions of the Royal Society A 378, 20190331 (2020).  
+
+689 46. United Nations. Resolution adopted by the General Assembly on 25 September 2015. (2015).  
+
+690 47. IPCC. Global Warming of \(1.5^{\circ}\mathrm{C}\) . An IPCC Special Report on the impacts of global warming of \(1.5^{\circ}\mathrm{C}\) above pre- industrial levels and related global greenhouse gas emission pathways, in the context of strengthening the global response to the threat of climate change, sustainable development, and efforts to eradicate poverty [Masson- Delmotte, V., P. Zhai, H.- O. Pörtner, D. Roberts, J. Skea, P.R. Shukla, A. Pirani, W. Moufouma- Okia, C. Pean, R. Pidcock, S. Connors, J.B.R. Matthews, Y. Chen, X. Zhou, M.I. Gomis, E. Lonny, T. Maycock, M. Tignor, and T. Waterfield (eds.)]. (2018).  
+
+696 48. UNEP & ISWA. Global Waste Management Outlook. (2015).  
+
+697 49. Joshi, R. & Ahmed, S. Status and challenges of municipal solid waste management in India: A review. Cogent Environmental Science 2, 1139434 (2016).  
+
+699 50. Gidden, M. et al. Global emissions pathways under different socioeconomic scenarios for use in CMIP6: A dataset of harmonized emissions trajectories through the end of the century. Geoscientific Model Development 12, 1443- 1475 (2019).  
+
+702 51. van Marle, M. J. E. et al. Historic global biomass burning emissions for CMIP6 (BB4CMIP) based on merging satellite observations with proxies and fire models (1750- 2015). Geosci. Model Dev. 10, 3329- 3357 (2017).  
+
+705 52. Riahi, K. et al. The Shared Socioeconomic Pathways and their energy, land use, and greenhouse gas emissions implications: An overview. Global Environmental Change 42, 153- 168 (2017).  
+
+707 53. UNDESA. World Urbanization Prospects: The 2018 Revision. https://population.un.org/wup/ (2018).  
+
+709 54. Amann, M. et al. Cost- effective control of air quality and greenhouse gases in Europe: Modeling and policy applications. Environmental Modelling & Software 26, 1489- 1501 (2011).  
+
+711 55. Gómez- Sanabria, A., Höglund- Isaksson, L., Rafaj, P. & Schöpp, W. Carbon in global waste and wastewater flows- its potential as energy source under alternative future waste management regimes. Advances in Geosciences 45, 105- 113 (2018).
+
+<--- Page Split --->
+
+56. Karak, T., Bhagat, R. M. & Bhattacharyya, P. Municipal Solid Waste Generation, Composition, and Management: The World Scenario. Critical Reviews in Environmental Science and Technology 43, 215–215 (2013).  
+
+57. Permadi, D. A. & Kim Oanh, N. T. Assessment of biomass open burning emissions in Indonesia and potential climate forcing impact. Atmospheric Environment 78, 250–258 (2013).  
+
+58. Hodzic, A., Wiedinmyer, C., Salcedo, D. & Jimenez, J. L. Impact of Trash Burning on Air Quality in Mexico City. Environ. Sci. Technol. 46, 4950–4957 (2012).  
+
+59. Bond, T. C. et al. A technology-based global inventory of black and organic carbon emissions from combustion. Journal of Geophysical Research: Atmospheres 109, (2004).  
+
+60. Akagi, S. K. et al. Emission factors for open and domestic biomass burning for use in atmospheric models. Atmospheric Chemistry and Physics 11, 4039–4072 (2011).  
+
+61. EEA. Managing municipal solid waste. A review of achievements in 32 European countries. (2013).  
+
+62. EUROSTAT. Guidance on municipal waste data collection. (2016).  
+
+63. Schöpp, W., Klimont, Z., Suutari, R. & Cofala, J. Uncertainty analysis of emission estimates in the RAINS integrated assessment model. Environmental Science & Policy 8, 601–613 (2005).  
+
+730  
+
+## Acknowledgements (optional)  
+
+The development of the ECLIPSE_V6b scenarios was supported by the European Union funded Action on Black Carbon in the Arctic.  
+
+## Ethics declarations  
+
+The authors declare that they have not conflict of interest.  
+
+## Supplementary Information  
+
+The supplement related to this article is available at
+
+<--- Page Split --->
+
+743
+
+# 744 Tables
+
+745 Table 1. Collection of studies quantifying municipal solid waste (MSW) openly burned.
+
+<table><tr><td>Source</td><td>Scale</td><td>Assumption</td><td>Results</td></tr><tr><td>Sharma et al., 2019</td><td>India</td><td>Calculation of waste burned at landfills was<br>based on a study in a landfill in Mumbai using<br>average FRP. Fraction open burning of waste 7%<br>-12%</td><td>68 Tg a-1 was open burned in<br>India in 2015</td></tr><tr><td>Wang et al., 2017</td><td>China</td><td>In reference to the limited literature, China's<br>averaged proportion of open MSW burning is set to 18.0% at residential and dumpsites and 38.0% at landfills.</td><td>The proportion of open burning<br>is estimated from 79.8% in<br>2000 to 57.0% in 2013</td></tr><tr><td>Klimont et al., 2017</td><td>Global</td><td>IPCC guidelines 2006; CEPMEIP, 2002;<br>EAWAG, 2008; Neurath, 2003. Fraction of open<br>burning of waste is 0.5% - 5% for developed<br>world and 10% -20% for developing world.</td><td>Global estimation of MSW<br>openly burned is estimated<br>115 Tg a-1 to 160 Tg a-1 in 2010</td></tr><tr><td>Weidimgyer at al., 2014</td><td>Global</td><td>Follows IPCC guidelines 2006 in which 60% of<br>the total waste available to be burned that is<br>actually burned</td><td>970 Tg a-1 of waste are globally<br>openly burned. 620 Tg a-1 at<br>residential level and 350 Tg a-1<br>at dumpsites.</td></tr><tr><td>Hodzic et al., 2012</td><td>Mexico<br>City</td><td>Assigned percentage of MSW burned according to socioeconomic status. Low and middle-low<br>60%, mid 30%, mid-high and high 20%. Based<br>on anecdotal evidence with Mexican researchers.</td><td>The burned fraction exceeds 4<br>Gg day-1</td></tr><tr><td>Bond et al., 2004</td><td>Global</td><td>Fraction of burned waste in urban areas base on<br>United Nations Human Settlement Programme,<br>2000</td><td>Worldwide 33 Tg a-1, including<br>14 Tg a-1 in Asia and 5 Tg a-1<br>in Africa</td></tr></table>
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+MunicipalwasteGHGreductionpotentialsSupplement.pdfGraphssupplementGOMEZetal.xlsx
+
+<--- Page Split --->

preprint/preprint__b43f149a50c18c934e9437c49ce087f0f79296e17429bb8dd31bfec88d2e934b/preprint__b43f149a50c18c934e9437c49ce087f0f79296e17429bb8dd31bfec88d2e934b_det.mmd

@@ -0,0 +1,570 @@
+<|ref|>title<|/ref|><|det|>[[44, 106, 920, 208]]<|/det|>
+# Potentials for future reductions of global GHG and air pollutant emissions from circular municipal waste management systems  
+
+<|ref|>text<|/ref|><|det|>[[44, 228, 860, 290]]<|/det|>
+Adriana Gomez Sanabria ( \(\boxed{\text {gomezsa@iiasa.ac.at}}\) ) International Institute for Applied Systems Analysis https://orcid.org/0000- 0002- 2317- 3946  
+
+<|ref|>text<|/ref|><|det|>[[44, 275, 861, 460]]<|/det|>
+Gregor Kiesewetter International Institute for Applied Systems Analysis (IIASA) Zbigniew Klimont International Institute for Applied Systems Analysis https://orcid.org/0000- 0003- 2630- 198X Wolfgang Schöpp International Institute for Applied Systems Analysis (IIASA) Helmut Haberl University of Natural resources and Life Sciences  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 499, 102, 516]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 535, 702, 556]]<|/det|>
+Keywords: Municipal waste, greenhouse gases, air pollution, methane, SDGs  
+
+<|ref|>text<|/ref|><|det|>[[44, 574, 305, 592]]<|/det|>
+Posted Date: June 22nd, 2021  
+
+<|ref|>text<|/ref|><|det|>[[44, 611, 463, 631]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 512870/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 649, 910, 691]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 727, 940, 770]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on January 10th, 2022. See the published version at https://doi.org/10.1038/s41467- 021- 27624- 7.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[58, 103, 883, 155]]<|/det|>
+# Potentials for future reductions of global GHG and air pollutant emissions from circular municipal waste management systems  
+
+<|ref|>text<|/ref|><|det|>[[58, 199, 884, 340]]<|/det|>
+4 Adriana Gomez- Sanabria\\*a,b, Gregor Kiesewetter, Zbigniew Klimont, Wolfgang Schoepp & Helmut 5 Haberlb 6 a Pollution Management Research Group, Energy, Climate and Environment Program, International Institute for Applied Systems Analysis, Laxenburg, Austria 7 b Institute of Social Ecology, University of Natural Resources and Life Sciences, Vienna, Austria  
+
+<|ref|>sub_title<|/ref|><|det|>[[60, 363, 238, 384]]<|/det|>
+## 9 Contributions  
+
+<|ref|>text<|/ref|><|det|>[[55, 401, 886, 550]]<|/det|>
+10 AGS designed the study, performed the projections, emission simulations and analysis, and prepared the manuscript. GK performed the ambient air pollution concentration calculations. ZK provided expert guidance and contributed to the revision of the manuscript. WS prepared and imported the IAE- WEO activity drivers and provided methodological advice. HH participated in the development of the research and contributed to writing the manuscript. All authors were involved in the discussion during the process.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 579, 311, 600]]<|/det|>
+## 15 Corresponding author  
+
+<|ref|>text<|/ref|><|det|>[[58, 618, 652, 636]]<|/det|>
+16 Correspondence to: Adriana Gomez- Sanabria. Email: gomezsa@iiasa.ac.at  
+
+<|ref|>text<|/ref|><|det|>[[58, 650, 366, 667]]<|/det|>
+17 AGS: ORCID: 0000- 0002- 2317- 39  
+
+<|ref|>text<|/ref|><|det|>[[58, 692, 374, 710]]<|/det|>
+18 HH: ORCID: 0000- 0003- 2104- 5446
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[100, 141, 886, 875]]<|/det|>
+Recent trajectories of production and consumption patterns have resulted in massively rising quantities of municipal solid waste (MSW). In combination with the large global quantities of mismanaged MSW these increases cause detrimental effects on the environment and climate. Few analyses of the potential environmental co- benefits resulting from the implementation of circular MSW management systems exist. To our knowledge, no global study of possible future scenarios of MSW generation, composition, management, and associated burdens is available that explicitly considers the important differences between urban and rural settings. To help filling this gap, we here develop a systematic approach for evaluating the benefits of implementing circular MSW management systems in terms of their potentials to reduce greenhouse gas emissions (GHG) and air pollution. We also analyse their role in the pursuit of the Sustainable Development Goals (SDGs). Building on the Shared Socioeconomic Pathways (SSPs), we build two sets of global scenarios until 2050, namely baseline and mitigation scenarios. In these scenarios, we assess trajectories of future MSW generation and the impact of MSW management strategies on methane (CH4), carbon dioxide (CO2) and air pollutant emissions. We estimate that future MSW generation could increase to at least 3.7 Gt/yr and at most to 4.3 Gt/yr by 2050, depending on the respective SSP storyline. In 2050, we show that the adoption of mitigation strategies in the sustainability- oriented scenario yields earlier, and major, co- benefits compared to scenarios in which inequalities are reduced but that are focused solely on technical solutions. In 2050, the GHG emissions in the sustainability- oriented scenario amount to 182 Gg CO2eq/yr of CH4, to be released while CO2, particulate matter, and air pollutants from open burning of MSW can be virtually eliminated, indicating that this source of ambient air pollution can be entirely eradicated before 2050. We conclude that significant potentials exist to reduce GHG, and air pollution if circular MSW management systems are implemented. We also demonstrate that the 6.3 target of the SDG 6 can only be achieved through more ambitious sustainability- oriented scenarios that limit MSW generation and improve management.  
+
+<|ref|>text<|/ref|><|det|>[[105, 881, 672, 899]]<|/det|>
+Key words: Municipal waste, greenhouse gases, air pollution, methane, SDGs
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[100, 140, 886, 900]]<|/det|>
+Global quantities of municipal solid waste (MSW) generation have grown massively over the last decades, not only due to population growth but also as a result of economic growth and the consequent changes in production and consumption patterns<sup>1,2</sup>. Estimates suggest that the world population generated 1.9 Gt/yr of MSW in 2015 and is expected to generate about 3.5 Gt/yr of MSW in 2050<sup>3</sup>. High- income countries (in terms of the World Bank income classification) generate more waste per capita per year than low- income countries: they are responsible for 34% of the amount of MSW generated each year, even though they account for just 16% of the global population<sup>4</sup>. These large quantities of MSW generated each year necessitate the implementation of appropriate management systems if the additional associated environmental and health impacts should be avoided that would emerge in the absence of suitable treatment facilities<sup>5</sup>. High- income countries can deploy policies and instruments to cope with the rising MSW flows and hence have cleaner and better- organized waste management systems. Examples include the EU Waste Framework Directive 2008/98/EC<sup>6</sup>, the 3R’s strategy in Japan<sup>7</sup> and the Resource Conservation and Recovery Act 1976<sup>8</sup>, 1986 in the United States. However, high- income countries are still mostly not successful in reducing the amount of MSW generated each year<sup>9</sup>. By contrast, low- income countries often lack suitable management systems, which results from the shortage of funds, poor planning, poor implementation of law and lack of technology and expertise<sup>4,10,11</sup>. Additionally, the outsourcing of resource- intensive production and waste exports from high- income to low- income countries exacerbates the environmental problems resulting from inadequate waste management in many of these countries<sup>12</sup>. Often, open burning, littering and poorly managed landfills are the main ways of waste disposal in low- income countries<sup>4</sup>. Open waste burning results in the release of toxic pollutants, e.g., particulate matter (PM), black carbon (BC), organic carbon (OC), carbon oxide (CO), sulphur dioxide (SO<sub>2</sub>), among others, and greenhouse gases (GHG) including carbon dioxide (CO<sub>2</sub>) as well as smaller amounts of methane (CH<sub>4</sub>)<sup>13–15</sup>. Litter harms wildlife and ecosystems, especially marine life. Global marine litter is currently recognized as one of the biggest sources of ocean’s pollution<sup>16,17</sup>. Decomposition of organic matter in landfills can result in the release
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 88, 886, 686]]<|/det|>
+of \(\mathrm{CH}_4^{18}\) , a greenhouse gas that is 28 times more potent per kg emitted than \(\mathrm{CO}_2\) in a 100 year timeframe<sup>19</sup>, and is also a precursor of tropospheric ozone which alters background ozone concentration and therefore impacts human health<sup>20- 22</sup>. In addition to the negative impacts on the environment and climate, these unsustainable practices have well documented adverse effects on human health<sup>23- 25</sup>. BC and OC, which are components of \(\mathrm{PM}_{2.5}\) , are associated with pulmonary disease, heart disease and acute lower respiratory infection<sup>26- 29</sup>. While reducing air pollution has positive health effects, the impact on the climate system is more difficult to assess. Given the complex interaction between air pollutants and GHGs in the atmosphere, polices that aim at reducing both air pollution and GHG emissions at the same time may succeed to reduce some GHG emission at the expense of reducing cooling effects from specific pollutants such as \(\mathrm{BC}^{30}\) . In the past years, research on waste has gone beyond disposal of wastes to assess the linkages between waste and resource use, climate change, air and water pollution. In that context, various studies have looked at emissions from landfills when assessing sectoral and regional contributions to GHG emissions and abatement potentials<sup>31- 34</sup>. Further assessments include the annual National Inventory Submissions of all Parties included in the Annex I of the Convention to the UNFCCC which comprise all reporting on GHG emissions and removals<sup>1</sup>. Current estimates are that landfills contribute about \(15\%\) to global anthropogenic \(\mathrm{CH}_4\) emissions<sup>31</sup>. Other studies show that open burning of MSW is an important contributor to particulate matter and air pollutant emissions<sup>14,35,36</sup>, specifically, it contributes \(11\%\) to total global \(\mathrm{PM}_{2.5}\) emissions and \(6 - 7\%\) to total global BC emissions<sup>35,36</sup>. BC from open burning of waste amounts to \(2 - 10\%\) of global \(\mathrm{CO}_{2\mathrm{eq}}\) emissions<sup>37</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[105, 697, 886, 845]]<|/det|>
+However, very few studies comprehensively assess and model MSW at the global scale. A recent study estimates the global trends and environmental impacts of MSW up to \(2100^{3}\) in terms of MSW generation, composition, and treatment, as well as environmental impacts. Other studies look at MSW as a potential source of secondary materials and energy. It is estimated that the relative contribution of energy from waste and wastewater to the global primary energy demand could increase from \(2\%\) to \(9\%\) by 2040 and deliver 64
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 88, 884, 268]]<|/det|>
+EJ of energy per year (1 EJ = \(10^{18}\) Joules) at the end of this period, if circular management systems are installed<sup>38</sup>. Current estimates are that only around \(13\%\) of the global MSW generated is recycled and \(5.5\%\) composted<sup>4</sup>. In a trend scenario perpetuating current conditions, this share is expected to increase to \(39\%\) in 2050 (includes composting and incineration)<sup>3</sup>. Recycling of waste, including composting and anaerobic digestion, can potentially be boosted in a sustainability-oriented scenario, but so far the extent to which that could be achieved has not been quantified.  
+
+<|ref|>text<|/ref|><|det|>[[102, 279, 885, 650]]<|/det|>
+Clearly, these assessments provide some insights on the contribution of MSW to GHG and air pollutants emissions as well as a source of energy and secondary materials. However, most of them focus on a single aspect of MSW (i.e., emissions from landfills and open burning) rather than on the MSW management system as such. Studies providing evidence of the potential environmental co- benefits resulting from the implementation of circular MSW management systems are still scarce. Furthermore, to our knowledge, no global analysis exists that considers differences between urban and rural settings and assesses how MSW generation, composition, management and associated environmental burdens might change under alternative, plausible future scenarios. We here fill that gap. Our main motivation is to contribute to improved understanding how different societal choices could transform MSW management practices in order to address global climate, pollution, and sustainability issues. To our knowledge, this is the first global study to show how the Shared Socioeconomic Pathways (SSPs) can be translated into emission baselines (CLE) and mitigation scenarios (MFR) for the MSW sector.  
+
+<|ref|>text<|/ref|><|det|>[[102, 662, 885, 907]]<|/det|>
+We present a new method to globally assess the current and future MSW generation in urban and rural areas and associated emissions as well as their implications for ambient \(\mathrm{PM}_{2.5}\) concentrations for a range of future population and macroeconomic developments to 2050 using the GAINS model as framework. These are represented by the five SSPs and a scenario consistent with the future macroeconomic and population pathways of the IEA's World Economic Outlook 2018<sup>39</sup>. Two variant scenarios are developed for each of the six future socioeconomic pathways; a 'Baseline - CLE' and a 'Maximum Technically Feasible Reduction - MFR', in which circular municipal waste management systems are implemented globally. This means that landfilling of MSW is restrained, material recycling rates are increased, technological improvements and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 886, 430]]<|/det|>
+behavioral measures such as reduction of food and plastic waste generation are assumed to be implemented. Emissions of \(\mathrm{CH_4}\) , \(\mathrm{CO_2}\) (fossil fraction), PM2.5, BC, OC, CO, \(\mathrm{SO_2}\) , NOx, and NMVOCs are calculated for 184 countries/regions (differentiating urban and rural areas) for the period 2010 – 2050. Results are presented at the level of thirteen world regions and the global aggregate. Based on this comprehensive analysis, we quantify the potential reduction of GHG emissions as well as particulate matter and air pollution through circular MSW systems. We also assess which SDGs can be reached or will be failed under the different scenario assumptions. Our detailed representation of the MSW sector and associated emissions and mitigation potentials can be used as input to Integrated Assessments Models (IAMs) applied to develop emission scenarios for the IPCC, support regional and local scale air pollution studies, and inform local and national governments about the likely developments, environmental consequences, and mitigation opportunities in the MSW sector.  
+
+<|ref|>sub_title<|/ref|><|det|>[[106, 457, 183, 476]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 501, 487, 523]]<|/det|>
+## Scenarios of MSW generation until 2050  
+
+<|ref|>text<|/ref|><|det|>[[105, 538, 886, 816]]<|/det|>
+Different socioeconomic assumptions underlying each of the SSPs lead to significant differences in future MSW flows (Fig. 1). The lowest quantities of MSW generation in 2050 are expected in SSP3 and SSP4 due to slow economic growth and inequalities between regions which is reflected in different consumption patterns. By contrast, in the SSP5 both income and urbanization rates increase strongly, resulting in a growth of the MSW generation quantities estimated at to 4296 Tg/yr. Interestingly, in a sustainability- oriented scenario (SSP1) MSW generation is expected to be just 10% lower than that in the SSP5 by 2050. However, when boosting the SSP1 with the adoption of measures targeted at reducing food and plastic waste (SSP1_MFR), it will be possible to reduce MSW generation by an additional 20% compared to SSP5 quantities by 2050.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 886, 330]]<|/det|>
+The amount of MSW generated, its composition as well as prevalent management systems and policies strongly depend on the dynamics of population and economic activity. We parameterized the drivers of MSW as follows: the most important driver of future MSW generation is GDP. Separate elasticities that relate MSW/cap/yr to GDP/cap/yr are estimated for groups of countries representing four different average income levels under the assumption that MSW generation and its composition are highly dependent on average national income levels. The future composition of MSW is recalculated based on the estimated income elasticity of per-capita food waste generation to GDP/cap/yr. MSW composition fractions estimated separately include food, paper, plastic, glass, metal, wood, textile, and other mixed waste.  
+
+<|ref|>text<|/ref|><|det|>[[102, 352, 886, 724]]<|/det|>
+Quantities and composition of MSW generated differ between rural and urban populations. Data on rural waste generation are available for a limited number of countries. For countries where data on rural MSW generation are unavailable, rural waste generation is estimated by applying ratios of urban:rural MSW generation per capita for each region that were deriving from the available information for limited number of countries (see Methods). While the uncertainty of the estimate might be high, the split into urban and rural MSW quantities highlights where actions are needed to improve MSW management systems at local levels, allowing for better quantification of impacts and consequently serves better for policy design. Our estimates suggest that urban areas are currently responsible for \(70\%\) of the global MSW generated. In 2050 urban areas are expected to generated \(80\%\) of the total MSW while rural areas are expected generated the remaining \(20\%\) , i.e., MSW per capita in rural areas is expected to be \(50\%\) lower than in urban areas. In general, rural per capita MSW generation is much lower than those in urban areas due to their smaller purchasing power. However, in high-income countries these differences between urban and rural areas shrink over time.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[220, 95, 770, 430]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[105, 440, 883, 473]]<|/det|>
+<center>Fig. 1: a. Global total MSW generation. b. Global MSW generation per capita. c. Global urban MSW generation per capita. d. Global rural MSW generation per capita </center>  
+
+<|ref|>text<|/ref|><|det|>[[105, 510, 884, 752]]<|/det|>
+North America (NAM) is likely to continue having the highest average per capita MSW generation in both urban and rural areas by 2050, followed by Oceania and Europe. China is expected to have the highest growth in MSW generation per capita for urban and rural areas increasing by about \(45\%\) compared to 2015. The reason is the stronger economic growth expected in China over the next decade \(^{41}\) . India is expected to generate about \(13\%\) less MSW than China in 2050 across all scenarios. Even though South Asia (SASIA) and Latin America and Caribbean (LCAM) had similar average per capita MSW generation for both urban and rural areas in 2015, per capita MSW generation in Asia is expected to overtake LCAM in 2050 by about \(15\%\) . Even though Africa will experience the highest increase on MSW generation compared to 2015, it is
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[48, 89, 884, 110]]<|/det|>
+likely to continue having the lowest MSW generation per capita in the future (Fig. 2). Supplementary  
+
+<|ref|>text<|/ref|><|det|>[[48, 122, 713, 140]]<|/det|>
+Results. S1 displays total, urban, and rural waste generation by region and scenario.  
+
+<|ref|>image<|/ref|><|det|>[[4, 175, 978, 575]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[102, 575, 884, 636]]<|/det|>
+<center>Fig. 2: Municipal solid waste (MSW) generation rates in urban and rural areas by scenario. For high-income regions as NAM and EU28, MSW per capita will remain pretty the same independent of the underlying socio-economic pathway. However, the different pathway trajectories have a strong influence on MSW per capita generation in low, and middle-income regions. </center>  
+
+<|ref|>text<|/ref|><|det|>[[102, 641, 884, 887]]<|/det|>
+Unfortunately, regions generating the highest amounts of MSW quantities per year have the lowest collection rates and the poorest MSW management systems. Average MSW collection rates in Africa, India, SASIA, and China are estimated to be in average of about \(50\% - 60\%\) , having urban areas collection rates of \(\sim 70\%\) and rural areas \(\sim 40\%\) . Moreover, the unsuitable management (i.e., disposed in dumpsites or burned without air pollution controls), of the collected fraction exacerbates the already precarious situation. Based on the detailed MSW activity and management strategies matrix of the GAINS model which comprises eight MSW streams and fourteen treatment technologies \(^{38}\) , our estimates suggest that in 2015, \(43\%\) of the global MSW collected ended up either in landfills ( \(13\%\) ) that are compacted and/or covered but not meeting
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 88, 885, 364]]<|/det|>
+environmental standards to prevent leakage<sup>42</sup>, in unmanaged landfills without any type of management (hereafter referred as dumpsites) (21%), or was openly burned (9%) either directly at the dumpsites (including unintended fires) or in transfer stations. The remaining 29% of the collected waste was either disposed in sanitary landfills (10%), incinerated (high quality with air pollution controls and energy recovery) (7%), recycled (7%), or composted or anaerobically digested (4%), which is mostly happening in high-income countries. From the uncollected fraction, around 20% is estimated to be scattered MSW with a high probability of eventually reaching water courses and 10% openly burned (Fig. 3). The latter estimates are based on global assessments and detailed country-level studies presented in Table 1 in the methods section.  
+
+<|ref|>image<|/ref|><|det|>[[55, 384, 936, 655]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[105, 666, 884, 740]]<|/det|>
+<center>Fig. 3: Municipal solid waste (MSW) management in 2015. Urban areas in low-middle income regions have increased MSW collection rates in last years. However, MSW treatment has not improved at the same pace, hence most of the waste is dumped, scattered or is subject to open burning. Rural areas face an even more challenging situation as in low-middle income regions collection rates are just about 35% - 45%. In general, high-income regions have established suitable MSW treatment systems in both urban and rural areas. </center>  
+
+<|ref|>text<|/ref|><|det|>[[105, 750, 884, 896]]<|/det|>
+Despite legislation banning open burning of MSW in most of the countries, our calculations indicate that around \(16\%\) of global MSW generated (whereof \(55\%\) collected and \(45\%\) uncollected), was openly burned, which is equivalent to \(380 \mathrm{Tg / yr}\) and \(394 \mathrm{Tg / yr}\) in 2010 and 2015, respectively. While in urban areas about \(60\%\) occurs either on transfer stations or dumpsites i.e., in the collected fraction, in rural areas is estimated that about \(80\%\) of the burning occurs in the uncollected fraction. Rural areas often lack appropriate MSW
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 883, 140]]<|/det|>
+management systems and therefore the uncollected waste is usually subject to be dumped, scattered or openly burned<sup>43</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[102, 161, 886, 406]]<|/det|>
+If current MSW management strategies are maintained into the future, the expected quantities of MSW disposed of in dumpsites and openly burned would rise proportionally to the increase of MSW quantities. In contrast, in an ideal situation where a circular MSW management system (MFR), is implemented globally, it would be probable to avoid almost all dumping and open burning of MSW in 2050, thereby eliminating the environmental and health burdens associated with current management practices. Circular MSW management systems include restrained landfilling of MSW, increase material recycling rates, technological improvement, and implementation of behavioral measures such as reduction of food and plastic waste generation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 430, 260, 450]]<|/det|>
+## Emissions to air  
+
+<|ref|>text<|/ref|><|det|>[[102, 476, 888, 841]]<|/det|>
+Our estimates indicate that current \(\mathrm{CH_4}\) emissions from MSW handling account for \(8\%\) (28 Tg/yr) of the global \(\mathrm{CH_4}\) anthropogenic emissions estimated at 344 Tg/yr in \(2015^{31}\) . Under the current management strategies, baseline \(\mathrm{CH_4}\) emissions in 2050 are projected to rise by a factor between 1.7 (SSP3_CLE) and 2 (SSP5_CLE) over the amount observed in 2015, increasing the contribution of MSW to \(13\%\) of the projected global \(\mathrm{CH_4}\) anthropogenic emissions estimated at 450 Tg/yr in \(2050^{31}\) . At the regional level, China, NAM, LCAM, and SASIA emitted the higher \(\mathrm{CH_4}\) from MSW in 2015. If current conditions are maintained until 2050, then India, Middle East, Africa and SASIA will face the highest growth in \(\mathrm{CH_4}\) emissions from MSW, with an increase of about \(60\%\) compared to 2015 levels. The expected rise of the \(\mathrm{CH_4}\) emissions on those regions is due to the increase of MSW generated, couple with the MSW (mis)management as scattered MSW, dumpsites and precarious landfills (cover or compacted without leakage controls or gas recovery) are the main options to deal with the MSW generated thereby increasing \(\mathrm{CH_4}\) emissions.  
+
+<|ref|>text<|/ref|><|det|>[[102, 830, 886, 881]]<|/det|>
+\(\mathrm{CH_4}\) emissions from waste deposited of in landfills today will be generated in future years as it depends on the degradability of the organic matter<sup>18</sup>. MSW generation quantities, composition and policy adoption at
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 886, 558]]<|/det|>
+early stages makes a significant difference in the trends of \(\mathrm{CH_4}\) emissions through the years. In a world implementing circular MSW management systems, the maximum diversion of MSW from dumpsites by 2030 is reached in SSP1_MFR with \(91\%\) less compared to the baseline. This is the result of the adoption of MSW reduction measures, speedy implementation of anaerobic digestion to treat organic waste and the establishment of source separated MSW collection systems to increase the recycling of materials. Total elimination of this practice is expected to happen around 2035 in this sustainability- oriented scenario. The adoption of measures is comparatively slower in scenarios depicting high inequalities between and within countries. Therefore, the diversion of MSW from dumpsites takes more time resulting in higher future \(\mathrm{CH_4}\) emissions. With the exception of SSP1_MFR in which \(\mathrm{CH_4}\) emissions are projected to decrease by \(4\%\) in 2030, an increase of about \(1\% - 2\%\) is expected to happen in all other MFR scenarios compared to the corresponding CLE. The maximum \(\mathrm{CH_4}\) emission reduction potential by 2050 will be reached in the SSP1_MFR in which \(\mathrm{CH_4}\) emissions are expected to decrease by \(87\%\) compared to the baseline, thus leaving still 182 CO2eq of \(\mathrm{CH_4}\) to be released in 2050. Other scenarios are expected to release more \(\mathrm{CH_4}\) , namely, SSP3_MFR will leave 646 CO2eq of \(\mathrm{CH_4}\) and SSP5_MFR 292 CO2eq of \(\mathrm{CH_4}\) to be emitted by 2050 which is \(50\%\) and \(80\%\) lower compared to the respective CLE counterparts (Fig. 4).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[90, 97, 916, 430]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[102, 445, 883, 504]]<|/det|>
+<center>Fig. 4: Global \(\mathrm{CH_4}\) emissions under CLE and MFR scenarios. Faster adoption of measures improving MSW systems will result in an early decrease of MSW ending up in dumpsites/uncontrolled landfills and therefore brings quicker reductions of future \(\mathrm{CH_4}\) emissions from this source. Supplementary Results S2 presents a detailed analysis of the MFR scenarios. </center>  
+
+<|ref|>text<|/ref|><|det|>[[102, 533, 884, 905]]<|/det|>
+Emissions of particulate matter and air pollutants depend on the quantities of MSW subject to open burning. Our results suggest that open burning of MSW is responsible for \(3.5\mathrm{Tg / yr}\) of \(\mathrm{PM}_{2.5}\) in 2015. BC emissions are estimated to be \(7\%\) and OC \(60\%\) of the \(\mathrm{PM}_{2.5}\) emissions. Overall, \(\mathrm{PM}_{2.5}\) emissions from MSW account for \(8\%\) of the total global anthropogenic \(\mathrm{PM}_{2.5}\) emissions. Global anthropogenic BC emissions are estimated at \(6.0\mathrm{Tg / yr}\) (GAINS) of which, following our results, \(6\%\) are from MSW burning (see supplement Table S3 for estimates for all pollutants). At the regional level, our calculations indicate that SASIA plus India, China, Africa, and LCAM emitted \(89\%\) of the particulate matter and air pollutants from MSW. India and China contributed about \(50\%\) and Africa \(21\%\) and LCAM the remaining \(18\%\) to those aggregate flows in 2015. Although open burning of MSW occurs in the collected and uncollected fraction in both urban and rural areas, most of emissions come from the collected MSW in urban areas. For example, in Indian cities waste handlers burn waste, despite being aware of the ban, mainly due to lack of infrastructure and to prevent accumulation<sup>44</sup>. Furthermore, with the projected growth of MSW generation and if the current conditions
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 884, 330]]<|/det|>
+prevail into the future then the anticipated global emissions of particulate matter and air pollutants from MSW are expected to nearly double in 2050 for all SSPs. SASIA, India, Africa, China and LCAM are expected to be responsible for \(93\%\) of the emissions. Future emissions in the CLE scenarios will increase proportionally to the quantities of MSW open burned. Consequently, the reduction of the fraction of MSW being openly burned translates directly into the same particulate matter and air pollutants emission reduction levels (Fig. 5). In that sense, in the SSP1_MFR, SSP5_MFR and ECLIPSE_V6b_MFR scenarios will be feasible to virtually eliminate open burning and therefore this source of air pollution already in 2030 while in the other scenarios this could potentially happen 10 to 15 years later.  
+
+<|ref|>image<|/ref|><|det|>[[120, 345, 940, 667]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[105, 686, 884, 732]]<|/det|>
+<center>Fig. 5: Global amounts of MSW open burned and related emissions under CLE and MFR scenarios. Reduction fractions of MSW open burned result in the same reduction percentage of particulate matter and air pollutants. Supplementary Results S2 presents a detailed analysis of the MFR scenarios. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 886, 397]]<|/det|>
+At a regional level (Fig. 6), the pre- conditions of the MSW management systems in Europe, Oceania and to certain extent NAM show that the level of effort required to reduce emissions is similar across scenarios. This is the result of the historical evolution on MSW management systems together with the already high- income level and appropriate political arrangements in most of these regions. By contrast, all other regions show high variation across scenarios due to the different dynamics. When comparing the scenarios for regions such as China, India, SASIA, and LCAM, we see that in a sustainability- oriented scenario (SSP1_MFR) a speedier decrease in emissions is observed in urban and rural areas compared to the other scenarios. Moreover, the adoption of circular MSW management systems is slower in scenarios representing a world in which inequalities persist resulting in big differences between urban and rural areas. Consequently, higher emissions are expected across the years.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[40, 111, 940, 699]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[105, 715, 884, 799]]<|/det|>
+<center>Fig. 6: Regional emissions of \(\mathrm{CH_4}\) and BC from MSW. The target of all modelled scenarios is set to reach \(\sim 100\%\) of MSW collection and management by 2050. The environmental co-benefits will be obtained at different levels upon the level of socio-economic development and political and institutional arrangements. The different assumptions on policy interventions are then translated into a wide range of future emissions. </center>  
+
+<|ref|>text<|/ref|><|det|>[[105, 813, 884, 898]]<|/det|>
+As emissions from MSW burning contribute significantly to ambient \(\mathrm{PM}_{2.5}\) , particularly since the sources are often low- level and spatially located close to population, the improvement of MSW management will also have benefits in ambient \(\mathrm{PM}_{2.5}\) . To illustrate the possible contributions and mitigation potential from
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 886, 397]]<|/det|>
+this sector, we here quantify the contribution of MSW to \(\mathrm{PM}_{2.5}\) levels in different world regions. Calculations follow the approach applied in ref \(^{45}\) and are briefly described in the Methods section below. Differences between the scenarios are driven both by emission changes as well as urbanization trends. Concentrations are highest in India and other South Asia and are expected to grow further under CLE following the emission trends. Other developing regions show similar growth trends but lower absolute concentrations. In China, initial increases level off, peaking around 2035 (SSP1,2,3,4) or 2050 (SSP5). In Europe, North America and Oceania, contributions from MSW burning are much lower since the combustion happens in well-controlled installations and not as open burning. Gradual implementation of better practices and emission controls eventually decreases concentrations to \(\sim\) zero before 2050 in all MFR cases, although this is achievable at different points in time depending on the SSP storyline.  
+
+<|ref|>sub_title<|/ref|><|det|>[[106, 425, 213, 444]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[102, 488, 886, 731]]<|/det|>
+Here we present for the first time a systemic assessment of reduction potentials of GHGs and air pollutants emissions from implementing circular MSW management systems under six future socio- economic development pathways. The assessment includes the development of two scenarios, namely baseline (CLE) and maximum feasible mitigation potential (MFR) for each of the pathways. The explicit representation of urban and rural MSW generation, composition and management allows for a deeper analysis of future plausible management and emission trends. This study can assist national, regional, and local governments in developing strategies to limit the release of emissions into the environment as well as support assessments of feasibility and progress in achieving the UN Sustainable Development Goals (SDGs).  
+
+<|ref|>text<|/ref|><|det|>[[102, 753, 886, 900]]<|/det|>
+Our results show that future MSW generation quantities are expected to be between 1.7 to 2 times higher in 2050 compared to current levels in all scenarios. Our results also highlight that urban areas are responsible for about \(80\%\) and will continue being responsible for the higher share of MSW generated in the future. The generally high collection rates of MSW in urban areas does not necessarily imply appropriate management. In SASIA, India, China, LCAM and Africa about \(80\%\) of the collected MSW is either dumped or openly
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 885, 268]]<|/det|>
+burned. Furthermore, most of the MSW generated in rural areas is uncollected and thus ends up being illegally dumped, scattered, or openly burned resulting in several environmental impacts related to air pollution and greenhouse gas emissions and other health and environmental impacts out of the scope of this study. Our findings also indicate that in urban areas about \(60\%\) of the open burning occurs either on transfer stations or dumpsites i.e., in the collected fraction, while in rural areas is estimated that about \(80\%\) of the burning occurs in the uncollected fraction.  
+
+<|ref|>text<|/ref|><|det|>[[102, 288, 886, 856]]<|/det|>
+In the baseline (CLE), in which current MSW management practices persist without further policy implementation, emissions to air would increase proportionately to the growth in MSW generation. We then developed a set of mitigation scenarios (MFR) to assess the impacts of abatement measures compared to the corresponding baseline (CLE). The common target of our MFR scenarios is to achieve \(\sim 100\%\) of MSW collection and treatment by 2050 through the implementation of circular MSW management systems to simultaneously tackle emissions of \(\mathrm{CH_4}\) , \(\mathrm{CO_2}\) , particulate matter, and air pollutants. Co- benefits are obtained at different stages upon the level of socio- economic development and political and institutional arrangements. Evidently, all countries would benefit from reduced MSW generation and improved management in the sustainability- oriented scenario (SSP1_MFR), however, the additional benefit of respective measures are especially relevant for regions generating large MSW quantities and lacking suitable management systems. We show that the environmental co- benefits of avoided MSW generation combined with the speedy implementation of anaerobic digestion to treat organic waste and the establishment of source separated MSW collection to increase the recycling of materials (SSP1_MFR) yields major and earlier co- benefits in terms of reducing \(\mathrm{CH_4}\) , particulate matter, and air pollutants. However, more ambitious sustainability- oriented scenarios are crucial to meet the waste related SDGs, specially the 6.3 target which aims at "By 2030, improve water quality by reducing pollution, eliminating dumping and minimizing release of hazardous chemicals and materials, halving the proportion of untreated wastewater and substantially increasing recycling and safe reuse globally" \(^{46}\) . We demonstrate that under the current SSP1_MFR, it will
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 884, 140]]<|/det|>
+not be possible to totally eliminate scattered and open burning of MSW by 2030. Under this scenario the realization of the objective will be obtained five years later i.e., in the year 2035.  
+
+<|ref|>text<|/ref|><|det|>[[102, 161, 886, 501]]<|/det|>
+Our analysis also suggest that in 2030, 881 Gg \(\mathrm{CO}_{2}\mathrm{eq}\) of \(\mathrm{CH}_{4}(\mathrm{GWP}_{100}\) of \(28\mathrm{CO}_{2}\mathrm{eq}^{19}\) ) will still be released in the SSP1_CLE. Nonetheless, this is \(13\%\) lower compared to the \(\mathrm{CH}_{4}\) emissions expected in the SSP2_CLE, SSP3_CLE and SSP4_CLE and \(11\%\) lower in comparison to the SSP5_CLE and Eclipse_V6b_CLE. Considering that in 2030 high emissions of \(\mathrm{CO}_{2}\) from open burning of MSW would still be released in SSP2_MFR, SSP3_MFR, SSP4_MFR, the total average GHG emissions ( \(\mathrm{CH}_{4}\) , and \(\mathrm{CO}_{2}\) ) in these scenarios will sum up to an average of about \(1079\mathrm{CO}_{2}\mathrm{eq}\) , that is \(18\%\) higher than the emissions expected in the SSP1_MFR. In 2050, SSP1_MFR leaves \(182\mathrm{Gg}\mathrm{CO}_{2}\mathrm{eq}\) of \(\mathrm{CH}_{4}\) , to be released. That is \(37\%\) lower than the SSP5_MFR and Eclipse_V6b_MFR and 3.5 times lower than the expected emissions in the SSP3_MFR. These variation in emissions can make a substantial difference when considering that the world should stay below 1.5 degrees global warming i.e., the world can emit as maximum as \(10\mathrm{Pg}\mathrm{CO}_{2}\mathrm{eq}\) /yr of all GHGs in \(2050^{47}\) .  
+
+<|ref|>text<|/ref|><|det|>[[102, 522, 886, 799]]<|/det|>
+The reduction of MSW being openly burned translates into the same reduction level of emissions of particulate matter and air pollutants. Under the development of SSP1_MFR, SSP5_MFR and ECLIPSE_V6b_MFR, the maximum emission reduction potential will be realized in 2030 whereas in the SSP2_MFR will take 5 years more i.e., in 2040 and for the SSP3_MFR and SSP4_MFR 10 years more i.e., in 2045. At the same time, MSW combustion contributes to ambient \(\mathrm{PM}_{2.5}\) – in some world regions, this contribution is substantial. Most low-income countries, and particularly those with already high concentrations, show an increasing trend from this source under all SSPs, highlighting the importance of counteracting. The positive message is that mitigation is possible and the MSW contribution to ambient \(\mathrm{PM}_{2.5}\) can be virtually eliminated by 2050. However, this will not happen by itself.  
+
+<|ref|>text<|/ref|><|det|>[[102, 845, 884, 894]]<|/det|>
+Comparison to other studies: Our calculations suggest that the world generated 2289 Tg/yr of MSW in 2015. Estimates from other studies vary from \(1999^{3}\) to \(2010^{4}\) Tg/yr for the same year. Past assessments estimated
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 87, 886, 876]]<|/det|>
+global MSW generation between \(2000^{48}\) to \(2400 \mathrm{Tg / yr}^{14}\) in 2010. Looking at MSW generation projections, our estimate for the SSP3 and SSP4 in 2050 are similar to the \(3539 \mathrm{Tg / yr}\) projected by Chen et al., 2020 (ref \(^{4}\) ). Our calculations suggest that although the SSP1 represents a sustainability- oriented pathway, MSW quantities in the baseline are foreseen to reach \(3901 \mathrm{Tg / yr}\) in 2050, which is only \(10\%\) lower than the expected MSW amounts in the SSP5. Our projection for MSW generation in the SSP2 is \(3801 \mathrm{Tg / yr}\) while ref \(^{3}\) estimated a MSW generation of about \(3500 \mathrm{Tg / yr}\) in 2050 for the same scenario. However, this estimate is more comparable with our SSP3 and SSP4 projection. The ECLIPSE_V6b_CLE (3948 \(\mathrm{Tg / yr}\) ) is comparable to the SSP1. At the regional level, we find that India is expected to generate about \(13\%\) less MSW than China in 2050 across all scenarios. This contrasts findings ref \(^{4}\) , in which projected MSW generation in India was about \(40\%\) higher than the projection for China in 2050. However, our finding for India is in line with the projection carried out by ref \(^{49}\) . Furthermore, the average per capita MSW generation in China is projected to be between \(30\%\) - \(40\%\) higher than those in India. The fact that estimates for 2010 are lower than those in 2015 and the variability of the results reflect on the one hand, the uncertainty of the data and on the other hand the differences of the methodologies used to derive these numbers. Furthermore, Our estimate of MSW openly burned is \(61\%\) lower than the estimate of ref \(^{14}\) , who estimated that \(40\%\) or an equivalent of \(970 \mathrm{Tg / yr}\) of total MSW generated in 2010 was openly burned (whereof \(64\%\) at residential sites and \(36\%\) at unmanaged dumpsites) and \(57\%\) higher than the estimate of ref \(^{36}\) , who estimated that about \(115 \mathrm{Tg / yr}\) - \(160 \mathrm{Tg / yr}\) of MSW was openly burned in 2010. Differences in estimated quantities can be attributed to variations in the per capita MSW generation rates adopted referring partly to different data sources, but also to differences in the methodology used to estimate the fraction of waste openly burned. While the assumption in ref \(^{14}\) refers to a fraction recommended in the IPCC (2006) guidelines, we develop our own method which we believe better represents the complexity of the MSW sector e.g., in terms of the urban- rural split and the country/region- specific MSW composition and MSW management pathways (see Methods). The differences of the estimates puts a magnifying glass on the urgency to develop national standardized MSW reporting systems, which in addition of being key to governments for the implementation
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 884, 139]]<|/det|>
+and evaluation of MSW treatment, can serve as part of the monitoring system of GHGs, air pollution and SDGs.  
+
+<|ref|>text<|/ref|><|det|>[[102, 160, 884, 373]]<|/det|>
+Our estimations indicate that current \(\mathrm{CH_4}\) emissions from MSW handling account for \(8\%\) (28 Tg) of the global \(\mathrm{CH_4}\) anthropogenic emissions estimated at \(344~\mathrm{Tg}\) in \(2015^{31}\) . Our estimate is \(17\%\) lower than the one estimated by ref \(^{35}\) and which has been adopted within the CMIP6 project \(^{50}\) . It is difficult to assess the level of agreement between both studies as estimates from ref \(^{35}\) include MSW and industrial waste while the focus of this study is on MSW and the importance to properly represent the sector for climate and air pollution assessments. However, comparing \(\mathrm{CH_4}\) emissions from MSW in the Eclipse_V5a \(^{36}\) to this study, we can see that the estimate in the latter is \(30~\mathrm{Tg / yr}\) or \(6\%\) higher.  
+
+<|ref|>text<|/ref|><|det|>[[104, 384, 884, 501]]<|/det|>
+Recent global \(\mathrm{CO_2}\) emissions area assessed at of 39153 Tg/yr in 2015, whereof 130 Tg/yr or \(0.33\%\) are generated from waste combustion (including industrial and municipal sources) \(^{35,51}\) . Ref \(^{14}\) calculates \(\mathrm{CO_2}\) emissions from open burning of MSW of \(1413~\mathrm{Tg / yr}\) in 2010, estimate that is around 10 to 15 times higher than that from ref \(^{35,51}\) and the one from this study.  
+
+<|ref|>text<|/ref|><|det|>[[104, 522, 884, 670]]<|/det|>
+In 2010, emissions of \(\mathrm{PM_{2.5}}\) , BC, and OC have been assessed at 6.1, 0.6 and 5.1 Tg, respectively \(^{14}\) . Our estimates are comparatively lower to those results. In contrast, our results for particulate matter are \(60\%\) higher than those from ref \(^{36}\) . In both cases the differences are related to the assumed quantities of MSW openly burned. Other studies \(^{35,51}\) have estimated BC and OC emissions from waste of \(0.7~\mathrm{Tg}\) and \(4.2~\mathrm{Tg}^{35}\) , respectively (Supplementary Results S3 show a comparison of different studies for different pollutants).  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 700, 229, 720]]<|/det|>
+## Conclusions  
+
+<|ref|>text<|/ref|><|det|>[[104, 761, 884, 877]]<|/det|>
+Significant potentials exist to reduce GHG, and air pollution provided the implementation of circular MSW management systems. The 6.3 target of the SDG 6 can only be achieved through more ambitious sustainability- oriented scenarios that limit MSW generation and improve management. Similarly, these kinds of scenarios can directly contribute to the achievement of other SDGs, especially SDG 7, 9, 12, 14 and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 886, 524]]<|/det|>
+15. Our results highlight the importance of acting at various fronts, namely, consumers behavior, technological development, technology transfer and institutional coordination. For instance, the benefits from reduction of MSW generation can be jeopardized by social and economic inequalities between and within regions which could restrain the adoption and implementation of measures to improve MSW management systems. Furthermore, for a world focused solely on end-of-pipe solutions will be also beneficial the implementation of policies targeted at reducing MSW generation. The finding is that the development of measures at the consumer side will not bring the expected benefits in terms of emissions reduction if quicker and responsible actions are not taken to bring MSW management systems as an important point in governmental agendas. Finally, we see that the majority of countries have developed some kind of legislation regarding the improvement of municipal solid waste management systems, however, the compliance is highly uncertain. A solid system for the reporting of MSW couple with a transparent systematic follow-up of policy enforcement will help to reduce the uncertainty of the estimates as well as will provide clearer insights into the efforts needed by countries to meet their climate, air pollution and SDGs commitments.  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 554, 196, 572]]<|/det|>
+## Methods  
+
+<|ref|>text<|/ref|><|det|>[[102, 614, 886, 890]]<|/det|>
+The methodology for developing MSW generation scenarios and associated greenhouse gas and air pollutant emissions involves the following five elements: 1. Socioeconomic drivers are taken from the Shared Socioeconomic Pathway (SSP) Scenarios for the five SSPs \(^{52}\) and from the World Energy Outlook and UNDESA \(^{53}\) for the Eclipse_V6b_CLE (Supplementary Methods S4 presents a short description of the SSPs storylines). 2. The country-specific generation in per capita MSW is driven by expected growth in average per capita income as described in the Supplement of ref \(^{38}\) and further developed in this study (Supplementary Methods Fig. S2 and Fig.S3 show GDP per capita and urbanization rates). 3. Estimation of emissions draw on the methodologies presented in ref \(^{33,36,54}\) , but are extended to improve source-sector resolution and accommodate for new, MSW sector-specific, information. 4. Implementation of the current legislation for
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 88, 884, 171]]<|/det|>
+waste management adopted before 2018. 5. Implementation of circular waste management systems are developed in accordance with the EU's waste management hierarchy - Directive 2008/98/EC<sup>6</sup>. The IIASA- GAINS model is used as a framework to carry out this assessment.  
+
+<|ref|>text<|/ref|><|det|>[[105, 195, 722, 218]]<|/det|>
+Municipal waste generation (MSW) activity and its characteristics.  
+
+<|ref|>text<|/ref|><|det|>[[105, 243, 886, 584]]<|/det|>
+Current MSW generation quantities, composition, collection rates, and waste management practices are retrieved from several sources, including national official statistics, peer- reviewed literature, and technical reports (see supplement of Gómez- Sanabria et al., 2018). The driver used to project future per capita MSW generation is GDP per capita. This is linked to MSW generation using elasticities estimated following the methodology first developed in ref<sup>33</sup> and further developed in ref<sup>55</sup>. This methodology is further developed in this study (Supplementary Methods S6). Separate elasticities are estimated for groups of countries representing four different average income levels under the assumption that MSW generation and its composition are highly dependent on average national income levels. Furthermore, MSW composition is recalculated based on the estimated income elasticity to per capita food waste generation. MSW composition fractions estimated separately include food, paper, plastic, glass, metal, wood, textile, and other waste. This last fraction includes ordinary mixed waste and may in some cases also include bulk waste.  
+
+<|ref|>text<|/ref|><|det|>[[105, 606, 886, 818]]<|/det|>
+Quantities and composition of MSW generated by rural and urban population are different. Data on rural waste generation is available for a limited number of countries, when underlying data on rural MSW generation is unavailable, rural waste generation is estimated by applying different shares related to the specific urban MSW generation rate per capita within specific region and using Eq. (1). This approach is likely to be an improved version of the one- half rural- urban waste generation ratio used by some studies<sup>4,56</sup> because it captures the differences between regions (Supplementary Methods S7 presents the adopted rural urban rates for different regions).  
+
+<|ref|>equation<|/ref|><|det|>[[371, 841, 870, 890]]<|/det|>
+\[MSW_{u} = MSW_{t}*\left(\frac{P_{u}}{P_{u} + \left(R_{t^{\prime} / u}\right)^{*}P_{r}}\right) \quad (1)\]
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[103, 158, 884, 252]]<|/det|>
+where \(MSW_{t}\) is total MSW generated in a country/region, \(MSW_{u}\) and \(MSW_{r}\) are MSW generated in urban and rural areas, respectively, \(R_{(r / u)}\) represents rural per capita MSW generation as a fraction of the per capita urban MSW generation, and \(P_{u}\) and \(P_{r}\) is rural are urban and rural population, respectively.  
+
+<|ref|>sub_title<|/ref|><|det|>[[106, 276, 328, 297]]<|/det|>
+## Open burning of MSW.  
+
+<|ref|>text<|/ref|><|det|>[[102, 325, 886, 897]]<|/det|>
+In countries without proper implementation of waste legislation, waste mismanagement is aggravated by poor waste separation at the source, low collection rates and low budget allocated to the waste sector \(^{40}\) . In the absence of reliable waste management systems, dumping and open burning of MSW, either at residential or dumpsites, become the only alternatives to reduce waste- volumes \(^{13,14}\) . Total MSW openly burned is estimated here as the sum of the fractions of uncollected MSW openly burned and collected MSW openly burned at dumpsites and transfer stations in urban and rural areas. The starting point to derive the quantities of MSW openly burned is the total MSW generated in urban and rural areas. Waste amounts are then split into collected and uncollected waste for urban and rural areas, respectively. Collected waste includes MSW collected by official authorities but also (recyclable) waste collected by the informal sector. Information on collection rates is gathered from sources presented in \(^{55}\) and complemented from information available in \(^{4,56}\) . The fraction of uncollected waste is then split into scattered waste or waste openly burned. The fraction of uncollected waste openly burned is assigned based on the information presented in Table 1, considering the current implementation of waste related legislation, income level, collection rates, and urbanization rate of each region. The fraction of collected MSW openly burned is estimated at 10% - 20% of the waste ending up in dumpsites, partly due to self- ignition resulting from poor management and partly due to deliberate burning to reduce waste volumes. In addition, a fraction of the collected waste is assumed to be burned at the transfer station or before reaching the disposal site, which is the case in several developing countries \(^{57}\) . Fractions of MSW openly burned, either on the streets or at dumpsites and transfer stations, are dependent
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 884, 400]]<|/det|>
+on the improvement of the MSW management systems and enforcement of the waste and air pollution legislation. Improvement of waste treatment systems results in reduction of the frequency of MSW openly burned \(^{58}\) . The quantification of these fractions is however highly uncertain. Literature provides a few different methodologies to estimate the amounts of waste openly burned (Table 1). The IPCC (2006) \(^{18}\) suggests 0.6 as a representative value for the fraction of total available waste to be burned that is actually openly burned. This assumption is used by Wiedinmyer et al., 2014 to estimate GHGs and air pollutants from open burning of waste. Bond et al., (2004) \(^{59}\) assumed lower rates of open burning of waste in rural areas in developing countries based on the statement that most of the waste in rural areas is biodegradable. Table 1 also shows that in many cases the default representative value of the IPCC maybe inadequate for several regions.  
+
+<|ref|>text<|/ref|><|det|>[[105, 408, 883, 498]]<|/det|>
+In general, the quantification of MSW openly burned in region \(i\) and year \(y - MSW_{(ob)iy}\) is calculated as the sum of MSW openly burned in urban areas \(MSW_{(obu)}\) and MSW openly burned in rural areas \(MSW_{(obr)}\) applying Eq (2).  
+
+<|ref|>equation<|/ref|><|det|>[[337, 520, 652, 541]]<|/det|>
+\[MSW_{(ob)iy} = MSW_{(obu)iy} + MSW_{(obr)iy}\]  
+
+<|ref|>text<|/ref|><|det|>[[106, 556, 159, 572]]<|/det|>
+Where,  
+
+<|ref|>equation<|/ref|><|det|>[[177, 586, 839, 641]]<|/det|>
+\[MSW_{(obu)iy} = [(MSW_{(u)iy}*C_{(u)iy}*(B_{0u} + B_{1u})) + (MSW_{(u)iy}*(1 - C_{(u)iy})*B_{2u})]\] \[MSW_{(obr)iy} = [(MSW_{(r)iy}*C_{(r)iy}*(B_{0r} + B_{1r})) + (MSW_{(r)iy}*(1 - C_{(r)iy})*B_{2r})]\]  
+
+<|ref|>text<|/ref|><|det|>[[105, 666, 884, 852]]<|/det|>
+Where, \(MSW_{(u)iy}\) and \(MSW_{(r)iy}\) are the total amounts of MSW generated in urban and rural areas, respectively. \(C_{(u)iy}\) and \(Coll_{(r)iy}\) are the MSW collection rates in urban and rural areas, respectively. \(\beta_{0u}\) and \(\beta_{0r}\) represent the fractions of collected MSW openly burned on transfer stations and \(\beta_{1u}\) and \(\beta_{1r}\) represent the fractions of collected MSW openly burned at dumpsites in urban and rural areas, respectively. \(\beta_{2u}\) and \(\beta_{2r}\) are the fractions of uncollected waste openly burned in urban and rural areas, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[106, 878, 311, 898]]<|/det|>
+Emission estimations.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 88, 884, 140]]<|/det|>
+Emissions of non- \(\mathrm{CO}_{2}\) greenhouse gases and air pollutants \((E)\) by source \((s)\) and region \((i)\) are calculated in GAINS using Eq (3) \(^{54}\) :  
+
+<|ref|>equation<|/ref|><|det|>[[375, 160, 612, 204]]<|/det|>
+\[E_{it} = \sum_{sit}A_{is}*ef_{sm}*Appl_{itsm}\]  
+
+<|ref|>text<|/ref|><|det|>[[105, 225, 886, 692]]<|/det|>
+where \(A_{is}\) is the activity data, i.e., the amount of MSW generated before management, \(ef_{sm}\) is the emission factor subject to technology \(m\) , and \(Appl_{itsm}\) is the application rate of the technology \(m\) to the activity \(A_{is}\) . The GAINS model matrix comprises fourteen different MSW waste management technologies including different types of source separation, recycling and treatment, different types of solid waste disposal sites and different types of incineration technologies and open burning of waste (Supplementary Methods 8). This extensive characterization of alternative treatment flows allows for a detailed representation of the solid waste management system and its emissions at the national/regional level. Emission factors for \(\mathrm{CH}_{4}\) and \(\mathrm{CO}_{2}\) are developed according to the 2006 IPCC Guidelines, Volume 5, Chapter 3 and Chapter \(5^{18}\) . PM emission factors are adopted from ref \(^{36}\) . These are 8.75 for \(\mathrm{PM}_{2.5}\) , 5.27 for OC and 0.65 g/kg for BC. Emission factors for \(\mathrm{SO}_{2}\) , NOx and NMVOC are adopted from ref \(^{60}\) and are consistent with ref \(^{14}\) . These are 0.5 for \(\mathrm{SO}_{2}\) , 3.74 for NOx, and 7.5 g/kg for NMVOC. The \(\mathrm{PM}_{2.5}\) concentrations are obtained using the annual \(\mathrm{PM}_{2.5}\) emissions applying a simplified version of the atmospheric calculation in the GAINS model \(^{45}\) . Those estimates build on a linearized representation of full atmospheric chemistry model simulations. Here, an atmospheric transfer coefficient is developed to related \(\mathrm{PM}_{2.5}\) emissions to ambient \(\mathrm{PM}_{2.5}\) concentrations from MSW burning.  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 716, 370, 737]]<|/det|>
+## Description of the scenarios.  
+
+<|ref|>text<|/ref|><|det|>[[105, 764, 886, 880]]<|/det|>
+The baseline scenarios associated with the six socio- economic pathways describe the expected developments of municipal solid waste generation and management systems under current legislation 'CLE', hereafter baseline, i.e., assuming no further policies affecting the MSW sector are adopted until 2050. In addition, for each baseline an alternative scenario is constructed, which considers full implementation of circular MSW
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[105, 88, 883, 172]]<|/det|>
+management systems globally and is referred to as the maximum technically feasible reduction 'MFR' scenario, hereafter mitigation scenario. Note that the technical frontier is explored here without taking account of the cost to implement various waste management strategies.  
+
+<|ref|>text<|/ref|><|det|>[[105, 194, 884, 502]]<|/det|>
+The MFR scenario is developed according to the SSP narratives and assumes a maximum technically feasible phase- in of a waste management system that is fully consistent with the EU's waste management hierarchy (Directive 2008/98/EC). This means that a first priority is given to technologies that circulate materials, thereafter to technologies that recover energy, and only as a last resort to well managed landfills. The following maximum recycling potentials of waste streams are applied: \(90\%\) of municipal paper and textile waste and \(80\%\) of municipal plastic and wood waste can be recycled. It is further assumed that \(100\%\) of food waste can be source separated and treated in anaerobic digesters with biogas recovery. These MFR potentials are adopted in consonance with the socioeconomic development for each scenario. Supplementary Methods S9 presents a description of the MFR management narratives specified for each scenario along with the regional aggregation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[107, 526, 218, 546]]<|/det|>
+## Uncertainty  
+
+<|ref|>text<|/ref|><|det|>[[105, 574, 884, 882]]<|/det|>
+Regarding uncertainty, several data inputs (activity data, emission factors, type of management) go into the estimations and therefore is difficult to do a quantitative uncertainty estimation. Historical estimates of MSW generation, collection, management, and related emissions have associated uncertainties resulting from the different definitions of MSW coupled with contradictory reported values for generation and composition. The quality of the data suffers from inconsistencies in the definition of MSW generation across countries. In some cases, amounts reported for MSW generation correspond to the gross quantities of waste collected and in other cases to the MSW quantities left for landfill after quantities separated for treatment have been deducted. In developed countries, in particular in Europe, MSW covers household waste and waste that is similar in nature and composition. In developing countries, data on waste suffers from incomplete characterizations and clear definitions of the fractions and source sectors included in the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 88, 886, 620]]<|/det|>
+MSW are often lacking. These uncertainties are relatively high in developing countries compared to developed countries as in various cases data availability is quite limited in the former case \(^{3}\) . Additionally, some data reported for generation and collection refers to urban areas rather than national totals \(^{4,40}\) , which makes necessary to adopt assumptions based on dedicate studies for particular regions and expert knowledge to arrive at reasonable national MSW generation rates and attributions to urban and rural waste amounts. These uncertainties become bigger when estimating fractions of MSW openly burned as this information is in most of the cases not attainable. Moving to emission factors, \(\mathrm{CH_4}\) emission factors are based on the IPCC Guidelines \(2006^{18}\) , thereby carry out the uncertainties there described. Emissions factors for air pollutants and particulate matter depend on the composition of waste and burning conditions. Although we adopted the most recognized emission factors in the scientific arena, we acknowledge that large uncertainties are related to the values (uncertainties can be seen in ref \(^{14}\) ). Concerning uncertainty in projections, this is by some means assessed by adopting alternative activity scenarios which allows the comparison of the different estimates and reflect the sensitivities of the proposed measures to input assumptions \(^{63}\) . In general, there is a global need to improve information on MSW generation rates, treatment and level of policy implementation \(^{3}\) . Regardless of the uncertainties, we demonstrate the importance of improving global estimates of GHGs and air pollutant emissions from MSW and highlight the considerable role of this sector when assessing the respective mitigation potentials.  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 648, 278, 668]]<|/det|>
+## Data Availability  
+
+<|ref|>text<|/ref|><|det|>[[108, 682, 828, 701]]<|/det|>
+The data used for this analysis is available in the Supplementary Information and excel spreadsheet.  
+
+<|ref|>sub_title<|/ref|><|det|>[[108, 714, 196, 729]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[101, 731, 880, 901]]<|/det|>
+1. Krausmann, F. et al. Long-term trends in global material and energy use. in Social Ecology 199–216 (Springer, 2016).  
+2. Tisserant, A. et al. Solid waste and the circular economy: A global analysis of waste treatment and waste footprints. Journal of Industrial Ecology 21, 628–640 (2017).  
+3. Chen, D. M.-C., Bodirsky, B. L., Krueger, T., Mishra, A. & Popp, A. The world’s growing municipal solid waste: Trends and impacts. Environmental Research Letters (2020).  
+4. Kaza, S., Bhada-Tata, P. & Van Woerden, F. What a waste 2.0. A global snapshot of solid waste management to 2050. (2018).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 90, 872, 145]]<|/det|>
+5. Yadav, P. & Samadder, S. R. A global prospective of income distribution and its effect on life cycle assessment of municipal solid waste management: a review. Environmental Science and Pollution Research 24, 9123-9141 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[50, 152, 840, 172]]<|/det|>
+6. Directive 2008/98/EC of the European parliament of the Council of European Union. (2008).  
+
+<|ref|>text<|/ref|><|det|>[[50, 179, 844, 198]]<|/det|>
+7. Ministry of the Environment. History and current state of waste management in Japan. (2012).  
+
+<|ref|>text<|/ref|><|det|>[[50, 205, 738, 224]]<|/det|>
+8. Resource Conservation and Recovery Act of 1976 in the United States. (1976).  
+
+<|ref|>text<|/ref|><|det|>[[50, 231, 835, 268]]<|/det|>
+9. Van Ewijk, S. & Stegemann, J. A. Limitations of the waste hierarchy for achieving absolute reductions in material throughput. Journal of Cleaner Production 132, 122-128 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 275, 880, 312]]<|/det|>
+10. Manaf, L. A., Samah, M. A. A. & Zukki, N. I. M. Municipal solid waste management in Malaysia: Practices and challenges. Waste Management 29, 2902-2906 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[50, 320, 880, 357]]<|/det|>
+11. Ghanimeh, S. et al. Two-Level Comparison of Waste Management Systems in Low-, Middle-, and High-Income Cities. Environmental Engineering Science 36, 1281-1295 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 364, 840, 401]]<|/det|>
+12. Parajuly, K. & Fitzpatrick, C. Understanding the Impacts of Transboundary Waste Shipment Policies: The Case of Plastic and Electronic Waste. Sustainability 12, 2412 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[50, 408, 859, 463]]<|/det|>
+13. Sharma, G. et al. Gridded Emissions of CO, NOx, SO2, CO2, NH3, HCl, CH4, PM2.5, PM10, BC, and NMVOC from Open Municipal Waste Burning in India. Environ. Sci. Technol. 53, 4765-4774 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 470, 865, 524]]<|/det|>
+14. Wiedinmyer, C., Yokelson, R. J. & Gullett, B. K. Global Emissions of Trace Gases, Particulate Matter, and Hazardous Air Pollutants from Open Burning of Domestic Waste. Environ. Sci. Technol. 48, 9523-9530 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[50, 532, 865, 587]]<|/det|>
+15. Young Koo, Y. K. and K., Wooram and Jo, Young Min. Release of Harmful Air Pollutants from Open Burning of Domestic Municipal Solid Wastes in a Metropolitan Area of Korea. Aerosol and Air Quality Research 13, 1365-1372 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[50, 594, 860, 631]]<|/det|>
+16. Walker, T. & Xanthos, D. A call for Canada to move toward zero plastic waste by reducing and recycling single-use plastics. Resources, Conservation and Recycling 133, 99-100 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 638, 870, 675]]<|/det|>
+17. Jambeck, J. R. et al. Marine pollution. Plastic waste inputs from land into the ocean. Science 347, 768-771 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[50, 682, 742, 702]]<|/det|>
+18. IPCC. IPCC Guidelines for National Greenhouse Gas Inventories 2006. (2006).  
+
+<|ref|>text<|/ref|><|det|>[[50, 709, 866, 746]]<|/det|>
+19. IPCC. The Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Geneva, Switzerland (2014).  
+
+<|ref|>text<|/ref|><|det|>[[50, 754, 866, 791]]<|/det|>
+20. Melamed, M. L., Schmale, J. & von Schneidemesser, E. Sustainable policy—key considerations for air quality and climate change. Current Opinion in Environmental Sustainability 23, 85-91 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 799, 825, 836]]<|/det|>
+21. Zhang, Y. et al. Tropospheric ozone change from 1980 to 2010 dominated by equatorward redistribution of emissions. Nature Geoscience 9, 875-879 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 843, 870, 881]]<|/det|>
+22. Anenberg, S. et al. Global air quality and health benefits of mitigating short-lived climate forcers. (2011).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 90, 884, 125]]<|/det|>
+23. Andersson, K. et al. Sanitation, Wastewater Management and Sustainability: from Waste Disposal to Resource Recover. (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 133, 884, 186]]<|/det|>
+24. Anenberg, S. C. et al. Global Air Quality and Health Co-benefits of Mitigating Near-Term Climate Change through Methane and Black Carbon Emission Controls. Environ Health Perspect 120, 831-839 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[50, 195, 864, 230]]<|/det|>
+25. Das, B., Bhave, P. V., Sapkota, A. & Byanju, R. M. Estimating emissions from open burning of municipal solid waste in municipalities of Nepal. Waste Management 79, 481-490 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 240, 864, 274]]<|/det|>
+26. Arnold, C. Disease burdens associated with PM2.5 exposure: how a new model provided global estimates. Environmental health perspectives 122, A111-A111 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[50, 283, 880, 319]]<|/det|>
+27. Beelen, R. et al. Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22 European cohorts within the multicentre ESCAPE project. The Lancet 383, 785-795 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[50, 327, 848, 345]]<|/det|>
+28. Brauer, M., Amann, M. & Burnet, R. T. Global Burden of Disease Assessment (draft). (2012).  
+
+<|ref|>text<|/ref|><|det|>[[50, 354, 850, 390]]<|/det|>
+29. Janssen, N. A. H. et al. Black carbon as an additional indicator of the adverse health effects of airborne particles compared with PM10 and PM2.5. Environ Health Perspect 119, 1691-1699 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[50, 399, 848, 434]]<|/det|>
+30. Harmsen, M. J. H. M. et al. Co-benefits of black carbon mitigation for climate and air quality. Climatic Change (2020) doi:10.1007/s10584-020-02800-8.  
+
+<|ref|>text<|/ref|><|det|>[[50, 443, 869, 496]]<|/det|>
+31. Höglund-Isaksson, L., Gómez-Sanabria, A., Klimont, Z., Rafaj, P. & Schöpp, W. Technical potentials and costs for reducing global anthropogenic methane emissions in the 2050 timeframe -results from the GAINS model. Environmental Research Communications 2, 025004 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[50, 505, 876, 558]]<|/det|>
+32. Yusuf, R. O., Noor, Z. Z., Abba, A. H., Hassan, M. A. A. & Din, M. F. M. Methane emission by sectors: A comprehensive review of emission sources and mitigation methods. Renewable and Sustainable Energy Reviews 16, 5059-5070 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[50, 567, 870, 602]]<|/det|>
+33. Höglund-Isaksson, L. Global anthropogenic methane emissions 2005-2030: technical mitigation potentials and costs. (2012) doi:10.5194/acpd-12-11275-2012.  
+
+<|ref|>text<|/ref|><|det|>[[50, 611, 835, 647]]<|/det|>
+34. Saunois, M., Jackson, R. B., Bousquet, P., Poulter, B. & Canadell, J. G. The growing role of methane in anthropogenic climate change. Environ. Res. Lett 11, 12 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 656, 844, 708]]<|/det|>
+35. Hoesly, R. M. et al. Historical (1750-2014) anthropogenic emissions of reactive gases and aerosols from the Community Emissions Data System (CEDS). Geoscientific Model Development 11, 369-408 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 718, 855, 753]]<|/det|>
+36. Klimont, Z. et al. Global anthropogenic emissions of particulate matter including black carbon. Atmospheric Chemistry and Physics 17, 8681-8723 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[50, 762, 844, 815]]<|/det|>
+37. Reyna-Bensusan, N. et al. Experimental measurements of black carbon emission factors to estimate the global impact of uncontrolled burning of waste. Atmospheric Environment 213, 629-639 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 824, 860, 877]]<|/det|>
+38. Gómez Sanabria, A., Höglund Isaksson, L., Rafaj, P. & Schöpp, W. Carbon in global waste and wastewater flows-its potential as energy source under alternative future waste management regimes. Advances in Geosciences 45, 105-113 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 886, 473, 904]]<|/det|>
+39. IEA. World Energy Outlook 2018. (2018).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 90, 830, 125]]<|/det|>
+677 40. Hoornweg, D. & Bhada-Tata, P. What a waste. A global review of solid waste management. (2012).  
+
+<|ref|>text<|/ref|><|det|>[[50, 134, 528, 151]]<|/det|>
+679 41. OECD. OECD Economic Surveys. China. (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 161, 850, 196]]<|/det|>
+680 42. Calvo, F., Moreno, B., Zamorano, M. & Szanto, M. Environmental diagnosis methodology for municipal waste landfills. Waste Management 25, 768- 779 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[50, 205, 828, 240]]<|/det|>
+682 43. Mihai, F.- C. & Grozavu, A. Role of waste collection efficiency in providing a cleaner rural environment. Sustainability 11, 6855 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 250, 880, 302]]<|/det|>
+684 44. Ramaswami, A., Baidwan, N. K. & Nagpure, A. S. Exploring social and infrastructural factors affecting open burning of municipal solid waste (MSW) in Indian cities: A comparative case study of three neighborhoods of Delhi. Waste Management & Research 34, 1164- 1172 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 312, 816, 346]]<|/det|>
+687 45. Amann, M. et al. Reducing global air pollution: the scope for further policy interventions. Philosophical Transactions of the Royal Society A 378, 20190331 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[50, 355, 840, 374]]<|/det|>
+689 46. United Nations. Resolution adopted by the General Assembly on 25 September 2015. (2015).  
+
+<|ref|>text<|/ref|><|det|>[[50, 383, 875, 488]]<|/det|>
+690 47. IPCC. Global Warming of \(1.5^{\circ}\mathrm{C}\) . An IPCC Special Report on the impacts of global warming of \(1.5^{\circ}\mathrm{C}\) above pre- industrial levels and related global greenhouse gas emission pathways, in the context of strengthening the global response to the threat of climate change, sustainable development, and efforts to eradicate poverty [Masson- Delmotte, V., P. Zhai, H.- O. Pörtner, D. Roberts, J. Skea, P.R. Shukla, A. Pirani, W. Moufouma- Okia, C. Pean, R. Pidcock, S. Connors, J.B.R. Matthews, Y. Chen, X. Zhou, M.I. Gomis, E. Lonny, T. Maycock, M. Tignor, and T. Waterfield (eds.)]. (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 497, 614, 515]]<|/det|>
+696 48. UNEP & ISWA. Global Waste Management Outlook. (2015).  
+
+<|ref|>text<|/ref|><|det|>[[50, 525, 852, 560]]<|/det|>
+697 49. Joshi, R. & Ahmed, S. Status and challenges of municipal solid waste management in India: A review. Cogent Environmental Science 2, 1139434 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 569, 864, 621]]<|/det|>
+699 50. Gidden, M. et al. Global emissions pathways under different socioeconomic scenarios for use in CMIP6: A dataset of harmonized emissions trajectories through the end of the century. Geoscientific Model Development 12, 1443- 1475 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[50, 630, 880, 682]]<|/det|>
+702 51. van Marle, M. J. E. et al. Historic global biomass burning emissions for CMIP6 (BB4CMIP) based on merging satellite observations with proxies and fire models (1750- 2015). Geosci. Model Dev. 10, 3329- 3357 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[50, 692, 864, 727]]<|/det|>
+705 52. Riahi, K. et al. The Shared Socioeconomic Pathways and their energy, land use, and greenhouse gas emissions implications: An overview. Global Environmental Change 42, 153- 168 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[50, 736, 844, 771]]<|/det|>
+707 53. UNDESA. World Urbanization Prospects: The 2018 Revision. https://population.un.org/wup/ (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 781, 866, 816]]<|/det|>
+709 54. Amann, M. et al. Cost- effective control of air quality and greenhouse gases in Europe: Modeling and policy applications. Environmental Modelling & Software 26, 1489- 1501 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[50, 826, 865, 878]]<|/det|>
+711 55. Gómez- Sanabria, A., Höglund- Isaksson, L., Rafaj, P. & Schöpp, W. Carbon in global waste and wastewater flows- its potential as energy source under alternative future waste management regimes. Advances in Geosciences 45, 105- 113 (2018).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 88, 870, 123]]<|/det|>
+56. Karak, T., Bhagat, R. M. & Bhattacharyya, P. Municipal Solid Waste Generation, Composition, and Management: The World Scenario. Critical Reviews in Environmental Science and Technology 43, 215–215 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[50, 120, 870, 156]]<|/det|>
+57. Permadi, D. A. & Kim Oanh, N. T. Assessment of biomass open burning emissions in Indonesia and potential climate forcing impact. Atmospheric Environment 78, 250–258 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[50, 163, 875, 199]]<|/det|>
+58. Hodzic, A., Wiedinmyer, C., Salcedo, D. & Jimenez, J. L. Impact of Trash Burning on Air Quality in Mexico City. Environ. Sci. Technol. 46, 4950–4957 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[50, 203, 875, 240]]<|/det|>
+59. Bond, T. C. et al. A technology-based global inventory of black and organic carbon emissions from combustion. Journal of Geophysical Research: Atmospheres 109, (2004).  
+
+<|ref|>text<|/ref|><|det|>[[50, 245, 875, 280]]<|/det|>
+60. Akagi, S. K. et al. Emission factors for open and domestic biomass burning for use in atmospheric models. Atmospheric Chemistry and Physics 11, 4039–4072 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[50, 286, 875, 321]]<|/det|>
+61. EEA. Managing municipal solid waste. A review of achievements in 32 European countries. (2013).  
+
+<|ref|>text<|/ref|><|det|>[[50, 327, 833, 363]]<|/det|>
+62. EUROSTAT. Guidance on municipal waste data collection. (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 369, 879, 404]]<|/det|>
+63. Schöpp, W., Klimont, Z., Suutari, R. & Cofala, J. Uncertainty analysis of emission estimates in the RAINS integrated assessment model. Environmental Science & Policy 8, 601–613 (2005).  
+
+<|ref|>text<|/ref|><|det|>[[50, 409, 83, 424]]<|/det|>
+730  
+
+<|ref|>sub_title<|/ref|><|det|>[[50, 490, 400, 510]]<|/det|>
+## Acknowledgements (optional)  
+
+<|ref|>text<|/ref|><|det|>[[50, 544, 879, 595]]<|/det|>
+The development of the ECLIPSE_V6b scenarios was supported by the European Union funded Action on Black Carbon in the Arctic.  
+
+<|ref|>sub_title<|/ref|><|det|>[[50, 625, 297, 645]]<|/det|>
+## Ethics declarations  
+
+<|ref|>text<|/ref|><|det|>[[50, 648, 525, 664]]<|/det|>
+The authors declare that they have not conflict of interest.  
+
+<|ref|>sub_title<|/ref|><|det|>[[50, 694, 385, 714]]<|/det|>
+## Supplementary Information  
+
+<|ref|>text<|/ref|><|det|>[[50, 750, 481, 766]]<|/det|>
+The supplement related to this article is available at
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[50, 94, 80, 106]]<|/det|>
+743
+
+<|ref|>title<|/ref|><|det|>[[50, 127, 173, 141]]<|/det|>
+# 744 Tables
+
+<|ref|>table_caption<|/ref|><|det|>[[50, 165, 745, 179]]<|/det|>
+745 Table 1. Collection of studies quantifying municipal solid waste (MSW) openly burned.
+
+<|ref|>table<|/ref|><|det|>[[105, 193, 880, 600]]<|/det|>
+
+<table><tr><td>Source</td><td>Scale</td><td>Assumption</td><td>Results</td></tr><tr><td>Sharma et al., 2019</td><td>India</td><td>Calculation of waste burned at landfills was<br>based on a study in a landfill in Mumbai using<br>average FRP. Fraction open burning of waste 7%<br>-12%</td><td>68 Tg a-1 was open burned in<br>India in 2015</td></tr><tr><td>Wang et al., 2017</td><td>China</td><td>In reference to the limited literature, China's<br>averaged proportion of open MSW burning is set to 18.0% at residential and dumpsites and 38.0% at landfills.</td><td>The proportion of open burning<br>is estimated from 79.8% in<br>2000 to 57.0% in 2013</td></tr><tr><td>Klimont et al., 2017</td><td>Global</td><td>IPCC guidelines 2006; CEPMEIP, 2002;<br>EAWAG, 2008; Neurath, 2003. Fraction of open<br>burning of waste is 0.5% - 5% for developed<br>world and 10% -20% for developing world.</td><td>Global estimation of MSW<br>openly burned is estimated<br>115 Tg a-1 to 160 Tg a-1 in 2010</td></tr><tr><td>Weidimgyer at al., 2014</td><td>Global</td><td>Follows IPCC guidelines 2006 in which 60% of<br>the total waste available to be burned that is<br>actually burned</td><td>970 Tg a-1 of waste are globally<br>openly burned. 620 Tg a-1 at<br>residential level and 350 Tg a-1<br>at dumpsites.</td></tr><tr><td>Hodzic et al., 2012</td><td>Mexico<br>City</td><td>Assigned percentage of MSW burned according to socioeconomic status. Low and middle-low<br>60%, mid 30%, mid-high and high 20%. Based<br>on anecdotal evidence with Mexican researchers.</td><td>The burned fraction exceeds 4<br>Gg day-1</td></tr><tr><td>Bond et al., 2004</td><td>Global</td><td>Fraction of burned waste in urban areas base on<br>United Nations Human Settlement Programme,<br>2000</td><td>Worldwide 33 Tg a-1, including<br>14 Tg a-1 in Asia and 5 Tg a-1<br>in Africa</td></tr></table>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 43, 311, 71]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[44, 93, 765, 113]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 570, 177]]<|/det|>
+MunicipalwasteGHGreductionpotentialsSupplement.pdfGraphssupplementGOMEZetal.xlsx
+
+<--- Page Split --->

preprint/preprint__b44761917413727d7f3c9d8df505481475d97925823bb47b71c6242dca1bf619/images_list.json

@@ -0,0 +1,92 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1: The data formatting used to represent each visit (left) as a multi-hot vector containing indices for medical codes, static label codes to cover demographics and disease phenotypes, and special codes describing the shape and temporal ordering of the patient's visit. Additionally, the matrix representation of each EHR (right) as a series of temporally ordered visit vectors.",
+    "footnote": [],
+    "bbox": [
+      [
+        50,
+        102,
+        940,
+        345
+      ]
+    ],
+    "page_idx": 4
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2: The architecture of HALO utilizing an autoregressive multi-granularity approach which analyzes at both the visit and code level to generate next code probabilities based on the history of all previous visits as generated through a stack of transformer decoder layers and the previous codes in the current visit through a series of masked linear layers.",
+    "footnote": [],
+    "bbox": [
+      [
+        80,
+        103,
+        940,
+        456
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3: These plots show the Unigram, Sequential Visit Bigram, and Same Record Bigram probabilities for each synthetic dataset. With the exception of SynTEG, all models exhibit some correlation in the unigram and temporal bigram evaluations, but many have weak correlation or consistently yield higher synthetic probabilities due to a lack of temporal consistency and repetition across visits in the records. HALO and to a lesser extent, HALO – Coarse perform the best in all settings, while HALO is the only one that can realistically produce pairs of codes within and across visits and achieve state-of-the-art results.",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        100,
+        833,
+        810
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Figure 4: We plotted the probabilities of each chronic disease label in the original outpatient EHR training dataset against their corresponding probabilities in each synthetic dataset. The \\(R^2\\) value is shown in parentheses in the legend. The SynTEG and LSTM baselines both struggle with temporal consistency as manifested through their weak ability to create these chronic disease labels in the \"label\" visit, so they are omitted from the plot. In contrast, the EVA, HALO - Coarse, and HALO architectures all closely mirror the training data with HALO and EVA performing the best overall on average.",
+    "footnote": [],
+    "bbox": [
+      [
+        39,
+        102,
+        480,
+        330
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5: Two demonstrations of HALO being able to capture the distribution of the gaps between visits in the outpatient EHR dataset variables once the model is augmented to support it. First, examining the mean visit gap by visit number across both the real and synthetic datasets shows that HALO is able to effectively capture the pattern of patients with many records having shorter gaps in their later visits. Second, the probability density of the visits gaps as a whole shows HALO approximating the true shape overall as well.",
+    "footnote": [],
+    "bbox": [
+      [
+        55,
+        115,
+        920,
+        338
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Figure 6: Two demonstrations of HALO being able to capture the distribution of labs in the inpatient EHR dataset. Both the binary presence of the lab probabilities and the average value of the labs when they are present closely approximate that of the real dataset.",
+    "footnote": [],
+    "bbox": [
+      [
+        87,
+        430,
+        892,
+        664
+      ]
+    ],
+    "page_idx": 12
+  }
+]

preprint/preprint__b44761917413727d7f3c9d8df505481475d97925823bb47b71c6242dca1bf619/preprint__b44761917413727d7f3c9d8df505481475d97925823bb47b71c6242dca1bf619.mmd

@@ -0,0 +1,547 @@
+
+# Synthesize Extremely High-dimensional Longitudinal Electronic Health Records via Hierarchical Autoregressive Language Model  
+
+Brandon Theodorou  University of Illinois Urbana- Champaign  
+
+Cao Xiao  Relativity  
+
+Jimeng Sun ( jimeng@illinois.edu )  University of Illinois Urbana- Champaign  https://orcid.org/0000- 0003- 1512- 6426  
+
+## Article  
+
+# Keywords:  
+
+Posted Date: March 10th, 2023  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 2644725/v1  
+
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License.  Read Full License  
+
+Additional Declarations: There is NO Competing Interest.  
+
+Version of Record: A version of this preprint was published at Nature Communications on August 31st, 2023. See the published version at https://doi.org/10.1038/s41467- 023- 41093- 0.
+
+<--- Page Split --->
+
+# Synthesize Extremely High-dimensional Longitudinal Electronic Health Records via Hierarchical Autoregressive Language Model  
+
+Brandon Theodorou \(^{1}\) , Cao Xiao \(^{2}\) , Jimeng Sun \(^{1*}\) University of Illinois Urbana- Champaign. \(^{1}\) Relativity Inc. \(^{2}\) \* To whom correspondence should be addressed: jimeng@illinois.edu  
+
+## ABSTRACT  
+
+Synthetic electronic health records (EHRs) that are both realistic and preserve privacy can serve as an alternative to real EHRs for machine learning (ML) modeling and statistical analysis. However, generating high- fidelity and granular electronic health record (EHR) data in its original, highly- dimensional form poses challenges for existing methods due to the complexities inherent in high- dimensional data. In this paper, we propose Hierarchical Autoregressive Language mOdel (HALO) for generating longitudinal high- dimensional EHR, which preserve the statistical properties of real EHR and can be used to train accurate ML models without privacy concerns. Our HALO method, designed as a hierarchical autoregressive model, generates a probability density function of medical codes, clinical visits, and patient records, allowing for the generation of realistic EHR data in its original, unaggregated form without the need for variable selection or aggregation. Additionally, our model also produces high- quality continuous variables in a longitudinal and probabilistic manner.  
+
+We conducted extensive experiments and demonstrate that HALO can generate high- fidelity EHR data with high- dimensional disease code probabilities ( \(d \approx 10,000\) ), disease code co- occurrence probabilities within a visit ( \(d \approx 1,000,000\) ), and conditional probabilities across consecutive visits ( \(d \approx 5,000,000\) ) and achieve above 0.9 \(R^2\) correlation in comparison to real EHR data. In comparison to the leading baseline, HALO improves predictive modeling by over \(17\%\) in its predictive accuracy and perplexity on a hold- off test set of real EHR data. This performance then enables downstream ML models trained on its synthetic data to achieve comparable accuracy to models trained on real data (0.938 area under the ROC curve with HALO data vs. 0.943 with real data). Finally, using a combination of real and synthetic data enhances the accuracy of ML models beyond that achieved by using only real EHR data.  
+
+## 1 INTRODUCTION  
+
+The widespread adoption of electronic health record (EHR) systems has established the foundation for machine learning (ML) and artificial intelligence (AI) applications in healthcare. The EHR data is highly complex, comprising over 10,000 unique medical codes for diagnoses, procedures, and medications, as well as thousands of lab measurements. Each patient record can include multiple visits with combinations of diagnoses, procedures, medications, and labs. These combinations create intricate relationships and complex patterns across tens of thousands of medical codes. AI and ML techniques are used to learn and model complex patterns in EHR data, enabling applications such as clinical predictive modeling [1, 2], health monitoring [3, 4], computational phenotyping [5, 6], treatment recommendations [7- 9], and more. However, the progress  
+
+of AI and ML in healthcare is often impeded by the difficulty of accessing and sharing large real EHR datasets. Sharing EHR data is challenging due to privacy, security, and legal constraints. While patient de- identification can alleviate some of these concerns by removing obvious patient identifiers such as name, address, and birth date [10, 11], studies have shown that the risk of re- identification remains high even after thorough de- identification [12- 14].  
+
+Using synthetic patient data can offer a safer alternative to sharing real EHR data. Generative models can produce synthetic datasets as substitutes for real patient data [15- 21]. Various methods have been proposed in the literature, including structured patient record generation [19, 20, 22- 24] and longitudinal record generation [15, 16, 21].  
+
+To date, existing methods have not been able to generate realistic EHR data in its original, high- dimensional form. The high dimensionality of EHR data, along with rare and sparse variables and complex relationships among variables, makes the generation task extremely difficult. Consequently, existing approaches all concede to creating lower- dimensional data by either aggregating variables or using a subset of more common variables of a manageable size. For example, the MedGAN method [19] modeled 615 disease categories without longitudinal information; the SynTEG model [15] aggregates codes to higher level phenotypes and then removes rare phenotypes, resulting in only 1,276 variables; the ehrMGAN approach [21] reduced the variable dimension to be less than 100, and EVA [16] models frequent co- occurrence patterns in the original EHR data as one- hot vectors, limiting its ability to generate diverse and novel co- occurrence patterns. Table 1 shows that previous approaches have been limited in their ability to model the full dimensionality of real patient data. While these low- dimensional approaches may capture the proper statistics on a small number of variables and support narrow ML use cases relying solely on those variables, the resulting synthetic data is inadequate for broader applications that require high- dimensional data including comprehensive statistical analysis, patient phenotyping, billing prediction and analysis, disease staging, and comprehensive data sharing.  
+
+We propose a new approach for generating high- dimensional EHR data in its native form: the Hierarchical Autoregressive Language Model (HALO). This model takes an autoregressive and probabilistic approach, and can capture the hierarchical distribution of EHR records and their temporal relationships. By using a hierarchical approach to model binary sequences of over a million variables, HALO is able to efficiently learn and represent complex patterns in EHR data. We evaluate the performance of HALO by training it on a comprehensive outpatient claims dataset, as well as the MIMIC- III inpatient EHR data [25], and compare the results with a diverse set of existing synthetic EHR data generation techniques
+
+<--- Page Split --->
+
+
+Table 1: Previous ML approaches for generating synthetic EHR data and their respective dimensionality. \\* signifies a non-longitudinal output (producing either a patient embedding or a single aggregated vector instead of a series of visits) while \\* signifies the special case of one-hot vector output that can only generate a limited number of common code combinations per visit predefined based on patterns from the training EHR data. No past approaches have ever produced synthetic health record data matching the highdimensionality (on the order of 10,o00+ medical codes).   
+
+<table><tr><td>Method</td><td>Dimensionality</td></tr><tr><td>CONAN [18]</td><td>128*</td></tr><tr><td>CorGAN [17]</td><td>1,071*</td></tr><tr><td>EHR-M-GAN [21]</td><td>98</td></tr><tr><td>EMR-WGAN [22]</td><td>944*</td></tr><tr><td>EVA [16]</td><td>-^</td></tr><tr><td>HGAN [23]</td><td>926*</td></tr><tr><td>MedGan [19]</td><td>615*</td></tr><tr><td>MedWGAN [20]</td><td>1,651*</td></tr><tr><td>SynTEG [15]</td><td>1,276</td></tr><tr><td>HALO</td><td>9,882</td></tr></table>  
+
+such as [15, 16, 26]. We evaluate the data quality based on its utility in modeling the statistical data distribution and for supporting ML models. HALO can accurately synthesize high- dimensional EHR data via modeling disease code probabilities ( \(d \approx 10,000\) ), disease code co- occurrence probabilities within a visit ( \(d \approx 1,000,000\) ), and conditional probabilities across consecutive visits ( \(d \approx 5,000,000\) ). In our experiments, we found that HALO achieves a correlation coefficient of above \(0.9 R^2\) when compared to real EHR data, demonstrating its ability to generate realistic data.  
+
+In addition to generating high- fidelity and granular EHR data, we show that HALO improves predictive modeling on our EHR dataset by more than \(17\%\) compared to the leading baseline. We evaluate the predictive accuracy and perplexity of HALO on a hold- off test set, demonstrating its superiority. Furthermore, the synthetic data generated by HALO enable downstream ML models to achieve comparable accuracy to models trained on real data, with an AUC of 0.938 for HALO data versus 0.943 for real data. We then demonstrate that combining real and synthetic data generated by HALO can improve the accuracy of ML models even more compared to using just real EHR data. Furthermore, we show that HALO generates realistic data while simultaneously protecting the privacy of patients in the training data, as evaluated by a series of privacy metrics.  
+
+## 2 RELATED WORK  
+
+Structured EHRs are multi- level longitudinal records, where each patient is represented by a sequence of visits. Each visit is characterized by a set of medical codes, reflecting the diagnoses, procedures, and medications administered during that visit. Additional patient information, such as demographics, disease phenotype labels, lab test results, and inter- visit time, can also be included.  
+
+Of all the EHR generation methods, rule- based approaches, such as Synthea [27] or SynPUF [28], have proven to be the most effective in delivering practical value. These simple approaches either offer de- identification of real records by combining data across multiple patients in a sufficiently privacy- preserving way [28], simulation of patients within a complex yet constrained rule- based system [27], Bayesian probabilistic modeling of aggregated, non- temporal patient records [29], or proprietary method without detailed explanation [30- 32]. Many of these systems can only produce synthetic patient data with limited capacity in realism and utility. We focus instead on ML methods that have the potential to generate realistic high- dimensional synthetic patient data.   
+
+### 2.1 GAN-based Methods  
+
+Many synthetic data generation methods use Generative Adversarial Networks (GANs), which involve a generator that creates realistic data, and a discriminator that decides if the data is real or fake [33]. The GANs has been applied to patient record generation first in [19] followed by many other GAN- based approaches [15, 17, 18, 20- 24, 34]. However, GANs have limitations when generating sequential data like EHRs. They usually only produce one output (no time connections) and so most EHR generation methods aggregate EHR data into one time step [22- 24], create a representation of EHR data [18], or do both [19, 20].  
+
+GANs also struggle with high dimensional and sparse data like real- world EHR, limiting all existing synthetic EHR GAN approaches to produce relatively low dimensional data through the aggregation of visits and medical codes or removal of rare codes. For example, there are a few methods in this category which do generate longitudinal data. LongGAN [34] and EHR- M- GAN [21] both focus only on dense lab time series of under a hundred dimensions. CorGAN [17] generates records with 1,071 distinct codes, and the current state of the art GAN approach that we baseline against, SynTEG [15], both combines and removes rare codes before arriving at a final dimensionality of 1,276.  
+
+While GANs have the potential to be conditioned on external factors and labels, such as demographics or disease phenotype labels, the ability to do so has not been extensively explored in existing works on EHR generation. Moreover, there are only a limited number of approaches that can generate synthetic EHR data tailored to specific diseases. For example, SmoothGAN [24] focuses on aggregated lab and medication information and does not model individual visits; EHR- M- GAN [21] offers conditional and sequential capabilities, but for low dimensional (under 100 dimensions) lab time- series information; CONAN and MaskEHR [18, 35] model only a single rare- disease population for data augmentation; and EMR- WGAN and HGAN [22, 23] can only model low- dimensional (both under 1000 dimensions) aggregated EHRs.  
+
+### 2.2 Deep Sequential Methods  
+
+Accurately modeling the longitudinal nature of EHRs is crucial for realistic EHR generation. In recent years, two methods have shown progress in generating sequential EHRs by using either a GAN or a VAE to condition on representations of past patient visits to generate current visits [15, 16]. Specifically, SynTEG [15] models the time between visits, and EVA [16] offers a conditional variant. In our experiments, we compare HALO to these two models. However, both SynTEG and EVA often need to perform preprocessing steps to
+
+<--- Page Split --->
+
+reduce the dimensionality of the vocabulary by aggregating medical codes and removing rare codes.  
+
+### 2.3 Language Models  
+
+Our objective is to develop an improved method for generating realistic and high- dimensional EHR data by drawing inspiration from natural language generation. Language generation models predict the next word based on the preceding words, thereby learning a probability distribution of languages. Similarly, EHR models predict the next visit based on past visits. Also our proposed method provides an explicit probability output that allows for direct modeling and evaluation of the underlying data distribution. This approach is particularly beneficial in accurately capturing the complex and high- dimensional nature of EHR data.  
+
+The Transformer architecture, introduced in [36], has revolutionized natural language processing and enabled the development of large, attention- based models like BERT [37] and GPT [26, 38, 39]. Among these models, we draw inspiration from GPT, which relies on a stack of Transformer decoder blocks that use masking to predict the next set of probabilities in parallel, allowing for fast training and scalability. However, applying language models directly to EHR data poses unique challenges. Unlike natural language sequences, EHR data exhibits a hierarchical structure that must be captured, with medical codes associated with specific patient visits, and visits associated with individual patients. Additionally, EHR data contains heterogeneous elements, including demographic variables, structured medical codes, and numeric lab measures, not all of which are discrete tokens. Addressing these challenges requires novel approaches that leverage the strengths of language models while adapting them to the peculiarities of EHR data.  
+
+## 3 METHOD  
+
+### 3.1 Problem Formulation  
+
+We first formalize the problem and introduce key notations.  
+
+EHR Data We represent a patient record \(\mathcal{R}\) as a sequence of visits over time such that  
+
+\[\mathcal{R} = \mathcal{V}^{(1)},\mathcal{V}^{(2)},\ldots ,\mathcal{V}^{(T)} \quad (1)\]  
+
+where each visit \(\mathcal{V}^{(t)}\) contains a varying number of medical codes \(m_{1}^{(t)},m_{2}^{(t)},\dots ,m_{|\mathcal{V}_{C}^{(t)}|}^{(t)}\in C\) , lab values \(l_{1}^{(t)},\dots ,l_{|\mathcal{V}_{L}^{(t)}|}^{(t)}\in \mathcal{L}\) , and the inter- visit time gap \(g^{(t)}\) . \(C\) is then the set of all medical codes in our vocabulary, including diagnoses, procedures, and medications and \(\mathcal{L}\) is the set of all labs. Beyond the longitudinal records, a patient record also possesses some static information \(S\) containing demographics such as gender, race, and birth year and disease phenotype label \(\mathcal{D}\) indicating major and persistent disease conditions.  
+
+Matrix Representation To allow input to HALO and other machine learning models, we then convert \(\mathcal{R}\) , \(S\) , and \(\mathcal{D}\) into a matrix representation \(\mathbf{R}\) . Specifically, we build \(\mathbf{R} = [\mathbf{v}_s,\mathbf{v}_l,\mathbf{v}_1,\dots ,\mathbf{v}_T,\mathbf{v}_e]\) , a matrix containing a sequence of the vector representations for each of the patient's \(T\) visits, a preceding "start visit", "label visit" and a succeeding "end visit".  
+
+Table 2: Table of Notations   
+
+<table><tr><td>Notation</td><td>Description</td></tr><tr><td>R</td><td>A patient&#x27;s EHR medical record</td></tr><tr><td>V(t)</td><td>The t-th visit in R</td></tr><tr><td>m(t)</td><td>The i-th medical code in V(t)</td></tr><tr><td>l(t)</td><td>The j-th lab value in V(t)</td></tr><tr><td>g(t)</td><td>The gap between the t - 1 and t-th visits</td></tr><tr><td>S</td><td>A patient&#x27;s static demographic information</td></tr><tr><td>D</td><td>A patient&#x27;s chronic disease information</td></tr><tr><td>L</td><td>The set of all labs</td></tr><tr><td>T ∈ N</td><td>The number of visits in R</td></tr><tr><td>C</td><td>The set of all medical codes</td></tr><tr><td>R ∈ R(T+3)×C</td><td>The matrix representation of R, S, and D</td></tr><tr><td>vL ∈ R|C|</td><td>The vector representation of the t-th visit in R</td></tr><tr><td>cL ∈ {0,1}</td><td>The binary presence of the i-th code in C in vL</td></tr></table>  
+
+The start visit \(\mathbf{v}_s\) is a one- hot vector containing a special start code added to \(C\) to signify the start of the record often required for certain model architectures.  
+
+The label visit \(\mathbf{v}_l\) similarly contains special codes added to \(C\) representing demographic and chronic disease phenotypes from \(S\) and \(D\) , respectively. For example, this label visit will have codes representing the patient's gender, racial and ethnic groups, birth year, and any chronic labels.  
+
+Each subsequent visit \(\mathbf{v}_t\in \mathbb{R}^{|C|}\) is then represented as a multi- . hot binary vector representing medical codes, lab values, and intervisit gaps present in that visit. To represent continuous lab values and visit gaps in a discrete form, we employ a granular discretization. This is achieved by adding multiple range codes to \(C\) for each lab test and for the intervals between visits. By converting all medical information into binary variables, \(c_{t}^{i}\) represents the presence of the \(i\) - th code in \(C\) in the \(t\) - th visit of the patient record \(\mathcal{R}\)  
+
+Finally, to signal the end of the patient record in \(\mathbf{v}_e\) , a special last visit code is added to \(C\) , serving a similar purpose to a stop token in natural language generation. This not only enables generative models to learn when to terminate records but also allows for \(\mathbf{R}\) to be padded through additional columns into a constant length for batch input without altering its content.  
+
+Figure 1 depicts the format of the visit vector and the EHR representation. Table 2 lists relevant notations used in the paper.  
+
+Generation task is to create \(\mathbf{R}^{\prime}\) , a synthetic patient record that is statistically similar to and offers the utility of \(\mathbf{R}\) without any one- to- one mapping to a real patient. Our HALO method does this by learning distribution \(P(\mathbf{R})\) .  
+
+### 3.2 Hierarchical Autoregressive Language Model (HALO)  
+
+We model the probability of patient record \(\mathbf{R}\) , \(P(\mathbf{R})\) , via a hierarchical autoregressive model, which utilizes both visit- and code- level structures of a patient record. First, it factorizes the probability along the visit level using the autoregressive identity by  
+
+\[\begin{array}{rl} & P(\mathbf{R}) = P(\mathbf{v}_s,\mathbf{v}_l,\dots ,\mathbf{v}_T,\mathbf{v}_e)\\ & \qquad = P(\mathbf{v}_s)P(\mathbf{v}_l|\mathbf{v}_s)P(\mathbf{v}_1|\mathbf{v}_s,\mathbf{v}_l)\dots P(\mathbf{v}_e|\mathbf{v}_s,\mathbf{v}_l,\dots ,\mathbf{v}_T) \end{array} \quad (2)\]
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1: The data formatting used to represent each visit (left) as a multi-hot vector containing indices for medical codes, static label codes to cover demographics and disease phenotypes, and special codes describing the shape and temporal ordering of the patient's visit. Additionally, the matrix representation of each EHR (right) as a series of temporally ordered visit vectors. </center>  
+
+to produce what we call our coarse autoregressive sequence. We then continue to factorize the probability of visits further along the code level by converting  
+
+\[\begin{array}{r l} & {P(\mathbf{v}_{t}|\mathbf{v}_{s},\dots ,\mathbf{v}_{t - 1}) = P(c_{t}^{1}|\mathbf{v}_{s},\dots ,\mathbf{v}_{t - 1})P(c_{t}^{2}|\mathbf{v}_{s},\dots ,\mathbf{v}_{t - 1},c_{t}^{1})}\\ & {\qquad \dots P(c_{t}^{C}|\mathbf{v}_{s},\dots ,\mathbf{v}_{t - 1},c_{t}^{1},\dots ,c_{t}^{C - 1})} \end{array} \quad (3)\]  
+
+into what we call our fine autoregressive sequence. This final probability is then rewritten as the product  
+
+\[P(\mathbf{R}) = \prod_{t}\prod_{i}^{C}P(c_{t}^{i}|\mathbf{v}_{s},\dots ,\mathbf{v}_{t - 1},c_{t}^{1},\dots ,c_{t}^{i - 1}) \quad (4)\]  
+
+where the probability of each code is based on each of the previous visits and each of the previous codes in the current visit. Our multigranularity approach enables the modeling of sequences of many binary variables per record. This is achieved by grouping prior information into significantly fewer multivariate time steps for previous visits, while retaining the full autoregressive modeling capability for each current visit. Our HALO architecture is designed to reflect this powerful yet compact model, with a structure divided into two distinct granularity levels: visit level and code level. This allows for each code to be conditioned on all previous visits and the past codes of the current visit.  
+
+3.2.1 Visit- Level Module. We begin with the coarse, visit- level granularity. We use a stack of \(M\) transformer decoder blocks to generate a sequence of visit- level histories, where the \(t\) - th element in the sequence, \(\mathbf{h}_t^{(M)} \in \mathbb{R}^{n_{\mathrm{emb}}}\) , is an embedding that represents all of a patient's medical history through their \(t\) - th visit. Those histories then combine to form \(\mathbf{H}^{(M)} \in \mathbb{R}^{(T + 3) \times n_{\mathrm{emb}}}\) (where the 3 in \(T + 3\) includes the start, label, and end visits), the output of the first module which serves of the purpose of the \(\mathbf{v}_s, \mathbf{v}_l, \mathbf{v}_1, \dots , \mathbf{v}_{t - 1}\) priors in Equation 4.  
+
+To encode each of the multi- hot visit representations \([\mathbf{v}_1 \dots \mathbf{v}_n]\) into a fixed- length vector in \(\mathbb{R}^{n_{\mathrm{emb}}}\) , we employ an embedding layer  
+
+that includes two trainable parameter matrices: a code embedding matrix \(\mathbf{W}_c\) and a positional embedding matrix \(\mathbf{W}_p\) . The code embedding matrix maps each visit code to a dense vector representation, while the positional embedding matrix captures the relative position of each visit in the sequence. Next, we use a decoder model consisting of \(M = 12\) transformer decoder blocks to generate a series of visit history representations, which summarize the information contained in all previous visits in the coarse, visit- level sequence. The transformer decoder blocks employ masked multihead self- attention, which allows the model to attend to all previous visits while preventing information leakage from future visits. This process is written more formally as  
+
+\[\begin{array}{rl} & {\mathbf{H}^{(0)} = \mathbf{R}\mathbf{W}_c + \mathbf{W}_p}\\ & {\mathbf{H}^{(m)} = \mathrm{transformer\_block}(\mathbf{H}^{(m - 1)})\quad \forall m\in [1,M]} \end{array} \quad (5)\]  
+
+where \(\mathbf{R} \in \mathbb{R}^{(T + 3) \times C}\) is the patient record matrix representation, \(\mathbf{W}_c \in \mathbb{R}^{C \times n_{\mathrm{emb}}}\) is the code embedding matrix, \(\mathbf{W}_p \in \mathbb{R}^{(T + 2) \times n_{\mathrm{emb}}}\) is the positional embedding matrix (to recapture the position and order of the sequence of visits), and each transformer block is based on a decoder block from the original transformer architecture [36] which we describe in more detail in our supplemental material.  
+
+Summary: Having processed the multi- hot patient visits through the initial, coarse visit- level module of our architecture, we obtain a sequence of visit history representations \(\mathbf{H}^{(M)}\) , which capture the collective information of all previous visits up to each time step. These representations provide a compressed summary of the patient's visit history, enabling downstream modules to make predictions based on the patient's medical trajectory.  
+
+3.2.2 Code- Level Module. However, we still need to add in the code- level priors and generate output probabilities. To construct the input for the fine, code- level module, we offset and concatenate the previous module's visit history embedding outputs with the
+
+<--- Page Split --->
+
+original record input, \(\mathbf{R}\) . Specifically, we append the first \(T + 2\) visit histories with the last \(T + 2\) visit representations \([\mathbf{v}_I,\mathbf{v}_1,\dots ,\mathbf{v}_T,\mathbf{v}_e]\) to create \(\mathbf{H}^{\prime (0)}\) . Each of the \(T + 2\) inputs in \(\mathbf{H}^{\prime (0)}\) has a representation of the history of all the previous visits and the codes of the current visit, mirroring both the visit and code priors in Equation 4. The final input representation \(\mathbf{H}^{\prime (0)}\) has size \(\mathbb{R}^{(T + 2)\times (n_{\mathrm{emb}} + C)}\)  
+
+To model the distribution of each \(P(c_{i}^{l})\) , this \(\mathbf{H}^{\prime (0)}\) is then fed through \(N = 2\) masked linear layers which maintain the same dimensionality and use upper triangular masking of the weight matrix to ensure that they preserve the autoregressive property of the probabilities (and have a ReLU activation function between layers). The probabilities are generated formally by  
+
+\[\begin{array}{rl} & {\mathbf{H}^{\prime (0)} = \mathrm{offset\_and\_concat}(\mathbf{H}^{(M)},\mathbf{R})}\\ & {\mathbf{H}^{\prime (n)} = \mathrm{masked\_linear}(\mathbf{H}^{\prime (n - 1)})\quad \forall n\in [1,N]}\\ & {\mathbf{O} = \mathrm{sigmoid}(\mathbf{H}^{(N)}[\cdot ,n_{\mathrm{emb}}:\cdot ])} \end{array} \quad (6)\]  
+
+where the submatrix indexing at the end removes the visit- level history embedding portions of each vector to extract just the code probabilities, and the masked linear layers are achieved by  
+
+\[\mathbf{H}^{\prime (n)} = \max (0,\mathbf{H}^{\prime (n - 1)}(\mathbf{W}^{(n)}\odot \mathbf{M}) + \mathbf{b}^{(n)}) \quad (7)\]  
+
+where the max function is omitted for the final fine layer (sigmoid is used instead), \(\odot\) is element- wise matrix multiplication, \(\mathbf{M}\in \mathbf{R}^{(n_{\mathrm{emb}} + C)\times (n_{\mathrm{emb}} + C)}\) is the upper triangular masking matrix (with ones in the upper triangular portion and zeros in the lower portion) to preserve the autoregressive property, and \(\mathbf{W}^{(n)}\in\) \(\mathbb{R}^{(n_{\mathrm{emb}} + C)\times (n_{\mathrm{emb}} + C)}\) and \(\mathbf{b}^{(n)}\in \mathbb{R}^{n_{\mathrm{emb}} + C}\) are the trainable parameters of the module.  
+
+The output \(\mathbf{O}\in \mathbb{R}^{(T + 2)\times C}\) is then a matrix of probabilities of each code for each visit after the start visit built from the visit histories and each previous code in the same visit. Each code corresponds to a conditional probability in the product from Equation 4.  
+
+We train our model using the binary cross- entropy loss function over each medical code (treating the problem as a multi- label classification problem) with masking applied such that the start visit as well as any padded visits (of all zeros) do not contribute to the loss. The architecture of our model is shown in Figure 2.  
+
+### 3.3 Additional Features and Considerations  
+
+Finally, We discuss different variants and add- on features of HALO.  
+
+3.3.1 Conditional Generation. Our method generates electronic health record (EHR) data by using demographics \(S\) and chronic disease phenotypes \(\mathcal{D}\) as labels, which are represented in our label vocabulary and applied to individual visits, as shown in Figure 1. We selected these labels based on their relevance to downstream use cases. Each label is represented as a binary variable in \(\mathbf{v}_I\) , indicating the presence of the corresponding disease or demographics group indicator. These indicators are defined by concepts such as specific categories of genders, races, ethnicity, age groups, and more. We can easily extend this strategy to include other labels of interest, such as various biomarkers, patient outcomes, or even abstract patient embeddings.  
+
+3.3.2 Unconditional Generation. Our setup generates electronic health record (EHR) data with conditional labels by incorporating a "label visit" in the data format, as illustrated in Figure 1. This format enables easy generation of labeled and conditional data, which are highly valuable for using synthetic data in machine learning tasks and as an augmentation tool, particularly for rare cohorts. However, it's important to note that this formatting is optional. If desired, the "label visit" component can be removed from the EHR representation, and the architecture can be trained to generate unconditioned EHRs without any modification.   
+
+3.3.3 Generation of Continuous Variables. Our model can generate not only medical codes but also continuous variables, such as lab values and temporal gaps between visits. However, the availability of these additional variables in the generated data depends on their presence in the original dataset used for training. For example, the outpatient EHR dataset used in our study includes the time between visits, while the inpatient EHR dataset includes lab values.  
+
+In previous models, continuous values were typically generated using either GANs, which lack the autoregressive probabilistic modeling that we employ, or value predictors (such as time series analysis models), which we often found to produce average values with insufficient variance. To overcome these limitations, we model continuous variables within the healthcare domain by discretizing lab values and temporal gaps into clinically equivalent buckets. The resulting binary variables are included in the model's context, denoted as \(C\) , before being converted back to continuous values through random uniform sampling within the corresponding bucket range. By using this approach, our model generates more realistic and diverse continuous variables than previous methods.  
+
+More specifically, to generate discrete versions of continuous variables, such as lab values and temporal gaps, we divide the range of each variable into several "buckets", as represented by the values \(b_{1},b_{2},\dots ,b_{|f_{j}^{(r)}|}\) , where \(|f_{j}^{(r)}|\) refers to the number of buckets required. We determine the bucket ranges by either seeking advice from clinicians on practical ranges, creating granular but equivalent groupings, or using a histogram construction algorithm [40]. The same approach is applied to temporal gaps as well.  
+
+For example, the heart rate lab test with possible values ranging from 0 to 400 beats per minute down could be broken down into twenty different buckets splitting the overall span into smaller ranges which offer the same medical meaning for all their contained values. This breakdown could have \(b_{1} = (0,40)\) and \(b_{7} = (90,100)\) . These buckets then convert the single continuous variable into many binary variables. Whenever the continuous variable is present in the original EHR, a single one of those variable representing the corresponding bucket is set to 1 with the rest remaining 0. For instance, if a patient has a heart rate lab measurement of 93 bpm in their seventh visit, the seventh of the new heart rate variables would be 1 and the rest would remain 0. If there was no such lab measurement in the visit, they would all be 0.  
+
+These new binary variables are added into the wider code vocabulary \(C\) and treated in the same way as all of the other medical codes in the vocabulary by our HALO model during learning and generation. After generation, the specific lab values and inter- visit gaps are converted back into a continuous value by uniformly sampling from the corresponding bucket range at the very end.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2: The architecture of HALO utilizing an autoregressive multi-granularity approach which analyzes at both the visit and code level to generate next code probabilities based on the history of all previous visits as generated through a stack of transformer decoder layers and the previous codes in the current visit through a series of masked linear layers. </center>  
+
+This discretization allows us to maintain the same powerful and probabilistic modeling process, matching the probabilistic variance of real continuous values in the same way we match the variance of medical code presences. However, by building appropriately granular buckets, we can avoid losing meaningful information and maintain a full representation of a patient. We explore the performance of this approach further in our experiments.  
+
+## 4 EXPERIMENTAL RESULTS  
+
+We evaluate our method and compare it to several baselines comprising both recently proposed models and other logical autoregressive model architectures on a series of experiments on both outpatient and inpatient EHR datasets. To maintain the fidelity of the original EHR data, our experiments focus on synthesizing original granular medical codes without aggregating or combining codes. Specifically, we seek to answer the following questions.  
+
+- Is HALO effective at modeling the underlying data distribution of electronic health records? [Section 4.3]  
+
+- Can HALO produce a synthetic dataset that is statistically similar to real EHR data? [Section 4.4]  
+
+- Can HALO augment real data for more accurate disease phenotyping prediction? [Section 4.5]  
+
+- Can HALO generate realistic continuous variables such as lab results and visit time gap? [Section 4.6]  
+
+- Can HALO preserve patient privacy in the training? [Section 4.7]  
+
+### 4.1 Datasets and Experimental Setup  
+
+Datasets We use two datasets for our experiments:  
+
+(1) The outpatient EHR is from a large real-world US claims data. It contains 929,268 patients and binary labels for 11 chronic diseases (specific diseases and patient counts are included in the supplementary material). This yields a final real-world outpatient EHR dataset with an average of 34.16 visits per record and 3.52 codes per visit with 9,882 unique ICD-10 codes.  
+
+(2) The inpatient EHR is from the MIMIC-III ICU stay dataset [25]. It contains 46,520 patients with 25 disease phenotype labels as defined by the MIMIC benchmark [41]. This dataset has an average of 1.26 visits per record and 15.11 codes per visit with 6,841 unique ICD-9 codes. Note that this includes patients with just a single visit (and as we will show, HALO's Code-Level Module allow it to be very effective on those patients).  
+
+Both datasets share the same patient representation as a series of visits along with chronic disease phenotype labels. We keep the ICD codes in the data without code aggregation or removing any infrequent codes.  
+
+Experiment setup: We use a 0.8- 0.2 training- test split with an additional 0.9- 0.1 training- validation split during training for both outpatient and inpatient datasets. We use the Adam optimizer with learning rate 1e- 4 (which was arrived upon through experimentation). We use a batch size of 48 and train for 50 epochs. Finally, we implement the model and train using the PyTorch framework [42].
+
+<--- Page Split --->
+
+### 4.2 Baseline Methods  
+
+Below we outline the baseline methods and the necessary alterations to those baselines to adapt to our problem setting.  
+
+HALO - Coarse This baseline is an ablation baseline consisting of just the coarse, visit- level granularity module of the full HALO architecture. It generates each code probability based on all previous visits (grouped into a multi- hot representation) but without the fine, inter- visit modeling such that \(P(c_{i}^{t})\) is modeled by \(P(c_{i}^{t}|\mathbf{v}_{1},\dots ,\mathbf{v}_{t - 1})\) instead of \(P(c_{i}^{t}|\mathbf{v}_{1},\dots ,\mathbf{v}_{i - 1},c_{i}^{1},\dots ,c_{i}^{t - 1})\) It consists predominantly of 12 transformer decoder blocks in the model of [38] augmented to support multi- hot as opposed to one- hot inputs and outputs within the embedding layer and final activation layer.  
+
+GPT Model [38]. We applied the GPT model without any augmentation to support multi- hot inputs and outputs but instead with the conversion of EHRs to a fully one- hot sequential representation. However, this model had to be shrunk down to 3 blocks from 12 to fit into memory because this greatly expanded the length of the sequences.  
+
+LSTM EHR Model is a deep, autoregressive LSTM model, which is directly analogous to the HALO - Coarse model but uses LSTM blocks instead of transformer decoder blocks.  
+
+SynTEG [15] is a GAN- based model that uses a transformer and LSTM- based encoder model to generate embeddings of EHRs up to a given visit before feeding those embeddings into a conditional GAN which generates the next visit.  
+
+EVA [16] is a VAE- based model which uses a bidirectional- LSTM encoder and CNN- based decoder (using deconvolutions to expand the latent encoding to the proper temporal dimension and then masked, diluted 1D convolutions to build the records in an autoregressive manner). The only change we made was to convert the output from one- hot code combinations to multi- hot code probabilities to allow for greater representative power.  
+
+### 4.3 Evaluating EHR Language Modeling  
+
+The first evaluation is conducted by predicting the probabilities and outputs of the test set. In this phase, we assess the performance of HALO against two multi- hot language model baselines, namely HALO - Coarse and LSTM. These baselines explicitly generate a probability distribution without accessing the entire input. It's worth noting that other baseline models, such as the GAN- based SynTEG model, the VAE- based EVA model, and the GPT model, cannot be directly compared in this task. This is because these methods sequentially add elements within visits and/or do not make a single probability prediction for each code within the visit.  
+
+Our first evaluation aims to assess the capability of the models to predict the presence of potential medical codes, given a patient's past medical history and the previous codes from the current visit. Note that we explore different orderings of codes (such as most to least prevalent, alphanumeric, random, etc.) but find no noticeable effect, settling on a random ordering throughout our experiments. This evaluation is crucial in showcasing a model's ability to learn patterns from the patient population, as well as its potential to perform well in various patient simulation and extension applications. We show the results in Table 3 where we see that HALO outperforms  
+
+<table><tr><td rowspan="2"></td><td colspan="2">Outpatient EHR</td><td colspan="2">Inpatient EHR</td></tr><tr><td>BCE Loss</td><td>F1 Score</td><td>BCE Loss</td><td>F1 Score</td></tr><tr><td>LSTM</td><td>7.744 × 10-4</td><td>0</td><td>2.600 × 10-4</td><td>0.193</td></tr><tr><td>HALO - Coarse</td><td>1.631 × 10-4</td><td>0.829</td><td>2.019 × 10-4</td><td>0.343</td></tr><tr><td>HALO</td><td>1.624 × 10-4</td><td>0.828</td><td>1.932 × 10-4</td><td>0.414</td></tr></table>
+
+Table 3: The results of binary classification metrics on the test set for each of our autoregressive, predictive models. HALO outperforms both of the baselines, achieving up to an \(4\%\) decrease in test BCE loss and a \(17\%\) increase in F1 score.  
+
+the two compared language model architectures. Upon closer examination, we observed that the LSTM baseline model struggled with the complexity and size of the outpatient EHR dataset, while our proposed model HALO performed comparably to the HALO - Coarse ablation baseline. In contrast, in the inpatient EHR setting, where the visits are shorter but contain more codes, HALO's multigranularity approach proved to be highly effective. Specifically, the model achieved a notable \(4\%\) reduction in test BCE loss and a \(17\%\) increase in F1 Score when compared to the single granularity HALO - Coarse model. Notably, both HALO models significantly outperformed the LSTM baseline in this setting. These results highlight the significant value of our multi- granularity approach in handling the complex and diverse nature of medical codes in different EHR settings.  
+
+Additionally, we present perplexity, which evaluates the probability or likelihood of the test set as quantified by a model trained on the training set, normalized by the unit of consideration that we are interested in. In our case, this normalizing unit is the number of medical codes in a patient's medical record (or equivalently number of ones in R). Perplexity is defined mathematically by  
+
+\[\begin{array}{rl} & PP(D) = \sqrt[N]{\frac{1}{P(D)}}\\ & \qquad = \sqrt[N]{\frac{1}{P(\mathbf{R}^{(1)},\dots,\mathbf{R}^{(|D|)})}}\\ & \qquad = \sqrt[N]{\frac{1}{P(\mathbf{R}^{(1)})\cdot\dots\cdot P(\mathbf{R}^{(|D|)})}} \end{array} \quad (8)\]  
+
+where \(D\) is the test dataset and \(\mathbf{R}^{(t)}\) is the \(t\) - th record in \(D\) . In practice we calculate the values by summing their log probabilities, using the equivalent form  
+
+\[PP(D) = \exp \left(-\frac{1}{N}\sum_{\mathbf{R}\in D}\log P(\mathbf{R})\right) \quad (9)\]  
+
+The normalized value then also corresponds to how many of the different normalizing units (medical codes) one would have to randomly pick between on average to achieve the same probability. The results of the perplexity evaluation are shown in Table 4. We see similar results as with the classification evaluation with both HALO and HALO - Coarse performing very well on the outpatient EHR dataset (with HALO performing slightly better) as the LSTM baseline struggles, and HALO easily outpacing both baseline methods in this likelihood evaluation for the inpatient EHR dataset, producing a \(13\%\) lower perplexity per present code as compared to the HALO - Coarse architecture without the inter- visit modeling. Thus, in
+
+<--- Page Split --->
+
+
+Table 4: The perplexity results on the test set for each of our three likelihood-based models. Baseline methods SynTEG, EVA, and GPT are all omitted here because they either do not produce a probability distribution, peek at the outputs, or utilize a different, non-comparable data representation. HALO outperforms both of the compared methods, yielding up to a \(13\%\) lower perplexity per present code as compared to the leading HALO - Coarse baseline.   
+
+<table><tr><td></td><td>Outpatient EHR PP Per Code</td><td>Inpatient EHR PP Per Code</td></tr><tr><td>LSTM</td><td>660.204</td><td>74.565</td></tr><tr><td>HALO - Coarse</td><td>3.927</td><td>28.448</td></tr><tr><td>HALO</td><td>3.903</td><td>24.664</td></tr></table>  
+
+both of these test set evaluations, we see that HALO is much more effective in terms of modeling the underlying distribution of EHRs.  
+
+### 4.4 Statistical Similarity to real EHRs  
+
+The second analysis evaluates the statistical similarity of the generated and real data. For each methods, we generate a synthetic dataset of the same size as the training dataset. We then compare the unigram and bigram (both within the same visit and across consecutive visits) probabilities for each unique code and pair of codes within the true and synthetic datasets. We perform this evaluation normalized at both the visit and record level, analyzing roughly 10,000 individual codes and over a million pairs of codes. Finally, we also compare the means, standard deviations, and probabilities of certain aggregate statistics such as the number of visits per record, number of medical codes per visit, and the prevalence of each chronic disease label. We show plots of the code probabilities normalized at the record level and a figure containing the chronic disease label probabilities for the outpatient EHR dataset in Figure 3 and Figure 4 respectively. we offer an interactive visualization (allows zoom, pan, and hover over points for specific disease names) of the "HALO vs. Real" disease prevalence plot at https://vega.github.io/. We also provide a table containing the aggregate statistics for both datasets in Table 5. Furthermore, we offer the \(R^2\) values for each of the three types of code probabilities normalized at the visit level in both our core high- dimensional outpatient EHR dataset as well as a lower- dimensional setting (with code aggregation and rare code removal down to around 1,300 different prevalent code phenotypes) in Table 6. Finally, we provide the full visit level code probability plots, probability densities underlying the aggregated statistics, and a discussion of the various failure modes of our baseline methods for that evaluation in our supplementary material. HALO again outperforms the baseline methods in each evaluation.  
+
+Specifically, we see that besides the GPT baseline struggling with the complexity of the outpatient EHR dataset in terms of stopping the record generation (as is common to many language models in the text generation domain as their overall quality decays for long sequences, and the lack of visit level grouping in its data representation causes its sequences to be considerably longer), the language model architectures (GPT, LSTM, HALO - Coarse, and HALO) are able to model both the shape of the synthetic records as well as the temporal dependencies much better on average than  
+
+the VAE and especially GAN- based baselines. While each of the compared methods model the unigram code probabilities relatively well, this better temporal modeling is shown in the overall synthetic record and visit lengths, the generation of chronic disease labels in the second visit, and the sequential bigram evaluation. However, the LSTM and HALO - Coarse language model baselines falter with respect to same- visit bigram probabilities due to their lack of intervisit dependency modeling while the GPT baseline which models each code individually and so offers that inter- visit modeling is able to maintain relatively stronger performance there. HALO is able to combine and build on each baseline's strengths without any of the weaknesses, using the compact multi- hot representation to offer an extremely powerful model that does not struggle with any length or feature of data while simultaneously maintaining the inter- visit modeling in an even more powerful and structured way. As such, it is able to best maintain performance in this high- dimensional setting and produces state of the art results which closely model the true training data in all settings from record and visit lengths, label probabilities, and finally all combinations of code probabilities. This signifies that HALO is capable of generating data which looks incredibly realistic, at least at the surface level.  
+
+### 4.5 Accurate Disease Phenotyping Using Synthetic EHRs  
+
+The final evaluation explores the utility of the synthetic datasets for training disease classifiers. To this end, we utilize two different synthetically- supplemented data setups and machine learning classifiers to predict chronic disease labels based on patients' visits in each. In each of the two data setups we use a simple bidirectional LSTM with a single- layer fully connected head classifier to predict chronic disease label(s) based on a patients' visits.  
+
+Accurate Disease Phenotyping: The first of the two data setups explores how models perform in real world settings when the training data is either completely synthetic or augmented with synthetic data. We repeat the experiments for each of the 11 chronic disease labels in the outpatient EHR dataset which originate from the list identified by the Centers for Medicare and Medicaid Services and used in the SynPUF dataset [28] and also for each of the 25 chronic disease in the inpatient EHR dataset which originates from the popular benchmark proposed in [41]. For each chronic disease, we randomly extract 2,500 records for training that both do and do not possess that chronic disease phenotype label from each of our 6 synthetic datasets and the real training data, forming 7 balanced training datasets. The number 2,500 was chosen to be large enough for training machine learning models but small enough that each dataset had enough positive labels for each disease. We then train classifiers on each of these datasets for each label. We select the best model for each dataset using a validation set of 250 records of either class from the original validation dataset, and we evaluate on test sets of 500 records of either class from the original test set. We display the average accuracy, F1 score, and rank for each synthetic dataset from each of the compared models across each chronic disease labels in the inpatient EHR dataset in Table 8. For the outpatient EHR dataset we then additionally explore models trained on a training set of real data additionally augmented with
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3: These plots show the Unigram, Sequential Visit Bigram, and Same Record Bigram probabilities for each synthetic dataset. With the exception of SynTEG, all models exhibit some correlation in the unigram and temporal bigram evaluations, but many have weak correlation or consistently yield higher synthetic probabilities due to a lack of temporal consistency and repetition across visits in the records. HALO and to a lesser extent, HALO – Coarse perform the best in all settings, while HALO is the only one that can realistically produce pairs of codes within and across visits and achieve state-of-the-art results. </center>
+
+<--- Page Split --->
+
+
+Table 5: We computed aggregate statistics on the number of visits per record and the number of codes per visit in the training datasets (outpatient and inpatient), and each of the synthetic datasets generated by the different methods. Our proposed model HALO outperformed all the baselines while closely approximating the distribution of the true training data. In the inpatient EHR dataset, HALO continued to exhibit strong performance, surpassing the other models and accurately replicating the distribution of the training data. In the outpatient EHR dataset, GPT struggled with the length of the sequences, leading to difficulty in generating synthetic records. In comparison, HALO outperformed the EVA and SynTEG baselines, with the GAN-based SynTEG model struggling the most.   
+
+<table><tr><td rowspan="2"></td><td rowspan="2"></td><td colspan="2">Outpatient EHR</td><td colspan="2">Inpatient EHR</td></tr><tr><td>Record Length Mean (Std. Dev.)</td><td>Visit Length Mean (Std. Dev.)</td><td>Record Length Mean (Std. Dev.)</td><td>Visit Length Mean (Std. Dev.)</td></tr><tr><td>EVA</td><td>29.49 (28.88)</td><td>3.35 (1.71)</td><td>1.20 (0.723)</td><td>11.92 (3.665)</td><td></td></tr><tr><td>SynTEG</td><td>93.00 (2.30)</td><td>3.70 (4.10)</td><td>27.55 (3.34)</td><td>5.93 (10.96)</td><td></td></tr><tr><td>LSTM</td><td>32.04 (27.14)</td><td>3.22 (1.64)</td><td>1.30 (0.56)</td><td>9.53 (2.91)</td><td></td></tr><tr><td>GPT</td><td>95.72 (3.37)</td><td>2.70 (1.73)</td><td>1.26 (0.73)</td><td>9.67 (5.45)</td><td></td></tr><tr><td>HALO - Coarse</td><td>35.26 (31.87)</td><td>3.77 (2.23)</td><td>1.13 (0.39)</td><td>11.21 (3.91)</td><td></td></tr><tr><td>HALO</td><td>36.19 (33.41)</td><td>3.93 (2.72)</td><td>1.31 (0.84)</td><td>11.93 (6.45)</td><td></td></tr><tr><td>Train Data</td><td>34.18 (32.35)</td><td>3.52 (2.18)</td><td>1.27 (0.92)</td><td>15.11 (8.64)</td><td></td></tr></table>  
+
+Table 6: We calculated \(R^2\) values to measure the correlations of the three types of code probabilities for different synthetic datasets against the training data in both high-dimensional and low-dimensional settings. Although the results showed a drop in performance for each method in the high-dimensional setting, HALO was able to maintain strong performance with minimal decline. Overall, our proposed method achieved state-of-the-art performance, outperforming the baselines in both bigram evaluations in low and high dimensional settings.   
+
+<table><tr><td rowspan="2"></td><td colspan="3">High-Dimensional Outpatient EHR</td><td colspan="3">Low-Dimensional Outpatient EHR</td></tr><tr><td>Unigram Code Probabilities</td><td>Sequential Visits Bigram Probabilities</td><td>Same Visit Bigram Probabilities</td><td>Unigram Code Probabilities</td><td>Sequential Visits Bigram Probabilities</td><td>Same Visits Bigram Probabilities</td></tr><tr><td>EVA</td><td>0.910</td><td>0.082</td><td>0.128</td><td>0.957</td><td>0.134</td><td>0.225</td></tr><tr><td>SynTEG</td><td>0.915</td><td>0.355</td><td>0.082</td><td>0.784</td><td>0.315</td><td>0.211</td></tr><tr><td>LSTM</td><td>0.900</td><td>0.077</td><td>0.127</td><td>0.962</td><td>0.135</td><td>0.225</td></tr><tr><td>GPT</td><td>0.743</td><td>0.382</td><td>0.262</td><td>0.924</td><td>0.626</td><td>0.515</td></tr><tr><td>HALO - Coarse</td><td>0.794</td><td>0.357</td><td>0.176</td><td>0.882</td><td>0.503</td><td>0.247</td></tr><tr><td>HALO</td><td>0.914</td><td>0.508</td><td>0.362</td><td>0.949</td><td>0.686</td><td>0.562</td></tr></table>  
+
+each of those synthetic datasets, and we show those aggregated results of mean test set classification performance across the 11 label- based tasks are shown in Table 7. We provide a full set of results by chronic disease label in our supplementary material. In both datasets, we can see that each of GPT, HALO - Coarse, and HALO's data largely maintain the performance of real training data and offer large improvements over the SynTEG, EVA, and LSTM baselines. HALO then offers the best results in having the least drop off among the three on average when used to train in the absence of real data and also the most improvement in performance over just the real training data when used as an augmentation technique.  
+
+Phenotyping of Rare Conditions: We conducted a simulation to demonstrate the usefulness of synthetic EHR data in identifying uncommon conditions. We extracted a highly imbalanced dataset of patients labeled with the cancer chronic disease from the outpatient EHR dataset. The dataset consisted of 50,000 EHR records from the original outpatient EHR training data without the cancer chronic disease label and just 1,000 with the label.  
+
+We trained a classifier on this imbalanced data and compared their performance to classifiers trained on data balanced by adding 49,000 positively labeled records from each of our synthetic datasets.  
+
+We then also baselined with a classifier trained on an upper bound ideal dataset balanced using real data.  
+
+The results of the evaluation are shown in Table 9. In particular, HALO outperforms each of the baselines, offering large gains on the original unbalanced dataset as well as the other synthetically augmented datasets and approaching the upper bound performance of the ideal balanced dataset.  
+
+This simulation shows the potential of synthetic EHR data to support the identification of uncommon conditions and highlights the value of using balanced data for training classifiers.  
+
+### 4.6 Realistic Continuous Variables in Synthetic EHRs  
+
+We conclude with a brief exploration to demonstrate the viability of our discretized representation of continuous values, and HALO's effectiveness in using it to model those variables. We build new training datasets including visit gaps in the outpatient EHR dataset and lab values in the inpatient EHR dataset. We use these dataset to train a new version of our model and generate another synthetic dataset of 250,000 and 45,000 records respectively.  
+
+We then show that the distributions of those variables match the real values. In Figure 5 and Table 10, we show that HALO accurately
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Figure 4: We plotted the probabilities of each chronic disease label in the original outpatient EHR training dataset against their corresponding probabilities in each synthetic dataset. The \(R^2\) value is shown in parentheses in the legend. The SynTEG and LSTM baselines both struggle with temporal consistency as manifested through their weak ability to create these chronic disease labels in the "label" visit, so they are omitted from the plot. In contrast, the EVA, HALO - Coarse, and HALO architectures all closely mirror the training data with HALO and EVA performing the best overall on average. </center>  
+
+Table 7: We compare the performance of chronic disease classification models trained on different types of training data in the outpatient setting - real data, synthetic data generated by different methods, and real data augmented by synthetic data. GPT, HALO - Coarse, and HALO's synthetic data perform better than the other methods, and are comparable to using real data as training data. Augmenting real data with HALO's synthetic data leads to better performance than just using real data. HALO has the best results, with little drop-off in performance compared to real data and the largest gain when used to augment the training set.   
+
+<table><tr><td></td><td>Avg. Accuracy</td><td>Avg. F1 Score</td><td>Avg. AUROC</td></tr><tr><td>EVA</td><td>0.508</td><td>0.283</td><td>0.471</td></tr><tr><td>SynTEG</td><td>0.507</td><td>0.514</td><td>0.506</td></tr><tr><td>LSTM</td><td>0.506</td><td>0.467</td><td>0.495</td></tr><tr><td>GPT</td><td>0.851</td><td>0.854</td><td>0.914</td></tr><tr><td>HALO - Coarse</td><td>0.867</td><td>0.863</td><td>0.920</td></tr><tr><td>HALO</td><td>0.879</td><td>0.878</td><td>0.938</td></tr><tr><td>Real Data</td><td>0.891</td><td>0.895</td><td>0.943</td></tr><tr><td>EVA + Real</td><td>0.844</td><td>0.852</td><td>0.921</td></tr><tr><td>SynTEG + Real</td><td>0.846</td><td>0.850</td><td>0.915</td></tr><tr><td>LSTM + Real</td><td>0.853</td><td>0.857</td><td>0.923</td></tr><tr><td>GPT + Real</td><td>0.904</td><td>0.906</td><td>0.953</td></tr><tr><td>HALO - Coarse + Real</td><td>0.910</td><td>0.910</td><td>0.958</td></tr><tr><td>HALO + Real</td><td>0.912</td><td>0.912</td><td>0.959</td></tr></table>  
+
+replicates the gaps between patient visits and the pattern of shorter gaps for longer records. In Figure 6, we demonstrate that HALO replicates not only the presence but also the average values of performed lab tests. Specific labs included (corresponding to points in  
+
+Table 8: The comparison of average performance by accuracy, F1 Score, and rank of chronic disease classification models across each of the 25 chronic disease labels in our inpatient dataset trained on each of our synthetic datasets and tested on real data. GPT, HALO - Coarse, and HALO's data offer large improvements over the other baselines and maintain similar performance to real training data. HALO's synthetic data performs the best with the highest average performance of all of the synthetic methods.   
+
+<table><tr><td>Method</td><td>Avg. Acc.</td><td>Avg. F1 Score</td><td>Avg. Rank</td></tr><tr><td>EVA</td><td>0.536</td><td>0.580</td><td>4.94</td></tr><tr><td>SynTEG</td><td>0.539</td><td>0.438</td><td>4.86</td></tr><tr><td>LSTM</td><td>0.522</td><td>0.565</td><td>5.20</td></tr><tr><td>GPT</td><td>0.877</td><td>0.880</td><td>2.00</td></tr><tr><td>HALO - Coarse</td><td>0.863</td><td>0.865</td><td>2.24</td></tr><tr><td>HALO</td><td>0.882</td><td>0.884</td><td>1.76</td></tr></table>  
+
+Table 9: The results of a variety of binary classification metrics on the test set for the simulated rare-disease detection task comparing models trained on datasets balanced using each of the synthetic datasets and baselined against models trained on the original imbalanced data (representing the rare disease dataset) an upper bound ideal dataset balanced using real data. EVA and SynTEG fail to offer much utility while the language model architectures LSTM, GPT, and HALO - Coarse offer a lot of value. However, HALO achieves state of the art results and closely approximates the performance of a true, balanced dataset.   
+
+<table><tr><td></td><td>BCE Loss</td><td>Accuracy</td><td>F1 Score</td><td>AUROC</td></tr><tr><td>Original Imbalanced Ideal Balanced</td><td>0.693</td><td>0.497</td><td>0.013</td><td>0.417</td></tr><tr><td></td><td>0.127</td><td>0.951</td><td>0.951</td><td>0.989</td></tr><tr><td>EVA</td><td>0.615</td><td>0.695</td><td>0.705</td><td>0.730</td></tr><tr><td>SynTEG</td><td>0.598</td><td>0.735</td><td>0.758</td><td>0.786</td></tr><tr><td>LSTM</td><td>0.593</td><td>0.702</td><td>0.714</td><td>0.743</td></tr><tr><td>GPT</td><td>0.472</td><td>0.880</td><td>0.869</td><td>0.956</td></tr><tr><td>HALO - Coarse</td><td>0.265</td><td>0.918</td><td>0.916</td><td>0.959</td></tr><tr><td>HALO</td><td>0.192</td><td>0.931</td><td>0.931</td><td>0.976</td></tr></table>  
+
+<table><tr><td></td><td>Gap Mean (Days)</td></tr><tr><td>Real Outpatient EHR Data</td><td>33.53</td></tr><tr><td>HALO</td><td>35.77</td></tr></table>  
+
+Table 10: The average gap between visits in number of days in the outpatient EHR training dataset and the synthetic HALO dataset created using the augmented method to handle additional continuous variables. The full probability distributions underlying these numbers can be seen in Figure ??  
+
+those two plots) are included in our supplemental material. Overall, HALO's approach to continuous variables is effective, and it has the potential to generate comprehensive synthetic patient records with multiple variables of different types.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Figure 5: Two demonstrations of HALO being able to capture the distribution of the gaps between visits in the outpatient EHR dataset variables once the model is augmented to support it. First, examining the mean visit gap by visit number across both the real and synthetic datasets shows that HALO is able to effectively capture the pattern of patients with many records having shorter gaps in their later visits. Second, the probability density of the visits gaps as a whole shows HALO approximating the true shape overall as well. </center>  
+
+![](images/Figure_6.jpg)
+
+<center>Figure 6: Two demonstrations of HALO being able to capture the distribution of labs in the inpatient EHR dataset. Both the binary presence of the lab probabilities and the average value of the labs when they are present closely approximate that of the real dataset. </center>  
+
+### 4.7 Privacy Protection of Synthetic EHRs  
+
+In addition to demonstrating the high fidelity of synthetic EHRs generated by HALO, we want to ensure that the privacy of the patients within the original training dataset has not been compromised. To that end, we conducted three commonly used privacy evaluations to test its robustness. Our results show that the outstanding performance of HALO is not due to memorization or any other violation of patient privacy.  
+
+4.7.1 Membership Inference Attack. The first evaluation is the ability to thwart a membership inference attack. These attacks aim to determine whether a real patient record was used in the training dataset to generate the synthetic records. Membership inference attacks are a well- known privacy test in the field of synthetic EHR generation, and addressing them is crucial to ensure the privacy and confidentiality of patient identities.   
+
+To demonstrate that HALO is not susceptible to such an attack, we show that we can prevent two different attempts at a membership inference attack based on the synthetic data generator and the synthetic dataset itself. We generate an attack dataset by first selecting 100,000 records from each real dataset that were used for training and assigning them a positive label. Then we select 100,000 records
+
+<--- Page Split --->
+
+from the remaining records not used for training as the negative label set.  
+
+Next, we conduct two attacks:  
+
+In the Model Attack, we label the 100,000 records with the highest log probability from the model as positive, predicting that they were part of the training dataset. In the Dataset Attack, we label the 100,000 records with the lowest hamming distance to the closest record in synthetic dataset as positive. We pick hamming distance (equivalent to Manhattan Distance in our binary setting) as our distance metric between patient records throughout our privacy evaluations in accordance with [43], but any distance metric could be substituted interchangeably.  
+
+These two attacks allow us to test the ability of the synthetic dataset to prevent an attacker from inferring whether a real record was used in the training dataset.  
+
+We show the results of the classifications from the attacks in Table 11. The accuracy of both attacks on both datasets is approximately \(50\%\) , which is similar to a random guess. This shows that neither the model nor synthetic dataset reveal any meaningful or compromising information about patient identity of the training dataset. We also perform the dataset attack with each of our baseline datasets and see that each similarly thwarts it, achieving very similar accuracies of around \(50\%\) as well. Note that we don't perform the model attack with the baseline models because most of them don't offer a probability output of input patient records, and the dataset- based attack is the standard one used throughout literature in this domain.  
+
+4.7.2 Attribute Inference Attack. The second evaluation is the ability to thwart a typical attribute inference attack. This attack determines whether the synthetic dataset leaks specific and sensitive patient attributes based on correlations from demographic and other more common, less sensitive attributes of the patient. Consequently, it tests whether the synthetic dataset can be used to learn individual attributes of real patient data.  
+
+To demonstrate that HALO is not susceptible to such an attack, we show that it thwarts the nearest neighbor- based attribute inference attack. In this attack, we use subsets of the synthetic dataset and the original training dataset, randomly sampled to match the size of the original test dataset. We define demographic information, chronic disease labels, and the binary presence of the 500 most common medical codes (determined by the training dataset) as the conditional attributes. The sensitive attributes to be identified are the binary presence of the remaining uncommon medical codes.  
+
+To conduct the attack, we find the closest patient in the synthetic dataset for each patient in the training set based on having the most shared conditional attributes. We then predict each of the uncommon attributes to be the same as that closest synthetic patient. Those predicted attributes are compared with the ground truth sensitive patient attributes and graded using F1 Score. We then repeat this attack with real patients from the test dataset in place of the synthetic dataset and use the results as a baseline for acceptable attribute inference.  
+
+We show the results of the classifications from the nearest neighbor attacks in Table 12. There we see that not only are the prediction F1 Scores incredibly low on both datasets (4.7% for the outpatient  
+
+dataset and \(3.3\%\) for the inpatient dataset), they are crucially lower than the baseline attack from the test set. This attack, labeled "Real Data Attack" in the table, sets the threshold for the amount of information revealed by the patterns of real data. So, staying below that level means incurring only an acceptable amount of attack success. So, we see that the synthetic dataset does not reveal any meaningful insight into the attributes of real patient data. We then see that each of the baseline synthetic datasets pass the test as well by having lower F1 Scores than the real data attack. GPT and HALO- Coarse allow similar F1 Scores to HALO while all of the rest have extremely low scores, likely because they do not capture the real patterns as effectively.  
+
+4.7.3 Nearest Neighbor Adversarial Accuracy Risk. The final evaluation, first proposed in [44], measures the degree to which a model overfits to its training dataset by looking at the relative likelihood of a patient's nearest neighbor being in the same or different datasets. As such, passing this test ensures that a generative model is generating wholly new synthetic patients rather than copying or performing simple augmentation on real training patients.  
+
+The evaluation is performed by calculating the metric Nearest Neighbor Adversarial Accuracy (NNAA). Let \(S_{T}\) , \(S_{S}\) , and \(S_{E}\) be random subsets of \(n\) records (we use \(n = 5,000\) records in our experiment) from the training, synthetic, and evaluation datasets respectively. NNAA risk is then the difference  
+
+\[AAES - AATs \quad (10)\]  
+
+where  
+
+\[\begin{array}{l}{{A A E S=\frac{1}{2}\left(\frac{1}{n}\sum_{i=1}^{n}1\left(d_{E S}(i)>d_{E E}(i)\right)+\frac{1}{n}\sum_{i=1}^{n}1\left(d_{S E}(i)>d_{S S}(i)\right)\right)}}\\ {{A A T S=\frac{1}{2}\left(\frac{1}{n}\sum_{i=1}^{n}1\left(d_{T S}(i)>d_{T T}(i)\right)+\frac{1}{n}\sum_{i=1}^{n}1\left(d_{S T}(i)>d_{S S}(i)\right)\right)}}\end{array} \quad (11)\]  
+
+where the \(E\) subscript throughout refers to the evaluation (test) dataset, \(S\) refers to the synthetic dataset, and \(T\) refers to the training dataset. \(1(\cdot)\) is then the indicator function and \(d_{ES}(i)\) is the distance from the \(i\) - th record in the evaluation dataset to its closest record (as determined by hamming distance in accordance with [43]) in the synthetic dataset. Each of \(E\) and \(S\) in \(d_{ES}(i)\) can also be replaced interchangeably with any of \(E\) , \(S\) , and \(T\) , where the calculation just omits the record in question if the two datasets are the same. So, each \(\frac{1}{n}\sum_{i = 1}^{n}1(d_{AB}(i) > d_{AA}(i))\) component is the probability of a record in dataset \(A\) being closer to another record in its own dataset than any record in dataset \(B\) . If they are randomly drawn from the same or similar distributions, we would expect that probability to be \(\frac{1}{2}\) , but it could be much lower if one of the datasets were copying from the other. We baseline this likelihood of the synthetic dataset copying from both its training and testing datasets, comparing the two to produce our overall risk.  
+
+[44] set 0.03 as the threshold for an acceptable NNAA risk. We show in Table 13 that the NNAA values for both our inpatient and outpatient datasets are easily below that mark. Furthermore, we show that as more data is added as with the outpatient EHR dataset, the risk decreases to an extremely small value. So, we show that our HALO method is not overfitting to or copying from its training
+
+<--- Page Split --->
+
+
+Table 11: The results of the two different membership inference attacks using the HALO model. For each record in the attack dataset, we find both the log probability of the record from the trained model (Model attack) and the hamming distance to the closest record in the synthetic dataset (Dataset attack). The attacks then label the half of the records with the highest probability or lowest distance records respectively as in the training set. We see that the accuracy for either attack is right around \(50\%\) which is similar to a random guess. This indicates that the synthetic dataset and the model do not reveal any patient identifying information about the original training datasets. We also find that each of the baseline synthetic datasets similarly thwart the dataset attack.   
+
+<table><tr><td rowspan="2"></td><td colspan="3">Outpatient EHR</td><td colspan="3">Inpatient EHR</td></tr><tr><td>Acc.</td><td>Precision</td><td>Recall</td><td>Acc.</td><td>Precision</td><td>Recall</td></tr><tr><td>HALO Dataset Attack</td><td>0.501</td><td>0.501</td><td>0.501</td><td>0.492</td><td>0.491</td><td>0.477</td></tr><tr><td>HALO Model Attack</td><td>0.509</td><td>0.509</td><td>0.509</td><td>0.515</td><td>0.515</td><td>0.515</td></tr><tr><td>EVA Dataset Attack</td><td>0.498</td><td>0.498</td><td>0.496</td><td>0.493</td><td>0.493</td><td>0.477</td></tr><tr><td>SynTEG Dataset Attack</td><td>0.500</td><td>0.500</td><td>0.500</td><td>0.491</td><td>0.491</td><td>0.467</td></tr><tr><td>LSTM Dataset Attack</td><td>0.499</td><td>0.499</td><td>0.496</td><td>0.494</td><td>0.494</td><td>0.481</td></tr><tr><td>GPT Dataset Attack</td><td>0.500</td><td>0.500</td><td>0.500</td><td>0.492</td><td>0.491</td><td>0.455</td></tr><tr><td>HALO - Coarse Dataset Attack</td><td>0.500</td><td>0.500</td><td>0.499</td><td>0.491</td><td>0.491</td><td>0.462</td></tr></table>  
+
+Table 12: The results of a nearest neighbor attribute inference attack. The results showed that the F1 Score on both the inpatient and output datasets was below 0.05, and crucially lower than the baseline attacks using real data from the test set. This baseline attack sets the threshold for the amount of information revealed by the patterns of real data and so staying below it means incurring only an acceptable amount of attack success. This suggests that the synthetic dataset does not reveal any significant insights into the attributes of real patient data, and that HALO is effective in preventing an attacker from inferring sensitive information. We then see that each of the baseline synthetic datasets pass the test as well by having lower F1 Scores than the real data attack. GPT and HALO- Coarse allow similar F1 Scores to HALO while all of the rest have extremely low scores, likely because they do not capture the real patterns as effectively.   
+
+<table><tr><td></td><td>Outpatient EHR F1 Score</td><td>Inpatient EHR F1 Score</td></tr><tr><td>Synthetic Data Attack</td><td>0.0397</td><td>0.0335</td></tr><tr><td>Real Data Attack</td><td>0.0503</td><td>0.0473</td></tr><tr><td>EVA</td><td>0.0108</td><td>0.0078</td></tr><tr><td>SynTEG</td><td>0.0162</td><td>0.0094</td></tr><tr><td>LSTM</td><td>0.0119</td><td>0.0068</td></tr><tr><td>GPT</td><td>0.0447</td><td>0.0324</td></tr><tr><td>HALO - Coarse</td><td>0.0330</td><td>0.0202</td></tr></table>  
+
+dataset and instead is producing wholly new synthetic records. We repeat the evaluation with each of the baseline synthetic datasets and show that they pass as well. HALO thus passes all three privacy evaluations and shows that its impressive performance does not come at the expense of patient privacy.  
+
+## 5 LIMITATIONS  
+
+While we have shown the impressive performance of HALO in both producing high- quality, high- fidelity, and privacy- preserving, we now briefly discuss some remaining limitations. First, the architecture is designed in the model of a large language model. While the multi- modal setup allows the model to condition on more patterns  
+
+Table 13: The Nearest Neighbor Adversarial Accuracy (NNAA) risk values for our two datasets. These values are calculated through the likelihood of data in the synthetic dataset being overly similar to records in the training set, normalized by their baseline likelihood of being close to unseen test set data. The metric was proposed in [44] where they set 0.03 as the acceptable risk threshold, a value that both the inpatient and outpatient synthetic datasets are well below. HALO and other baselines all achieve much lower NNAA risk.   
+
+<table><tr><td>Method</td><td>Outpatient NNAA</td><td>Inpatient NNAA</td></tr><tr><td>HALO</td><td>0.0104</td><td>0.0211</td></tr><tr><td>EVA</td><td>0.0040</td><td>0.0018</td></tr><tr><td>SynTEG</td><td>-0.0002</td><td>-0.0080</td></tr><tr><td>LSTM</td><td>0.0178</td><td>0.0082</td></tr><tr><td>GPT</td><td>0.0045</td><td>0.0221</td></tr><tr><td>HALO - Coarse</td><td>0.0047</td><td>0.0301</td></tr></table>  
+
+per data point and learn more efficiently, our high- performing generator still requires relatively large training datasets which might not be available in some settings.  
+
+Another important aspect of our model is that it generates synthetic records through a probabilistic process. While it learns realworld patterns during training, there is still a chance that some generated records may not be clinically meaningful. However, this risk can be mitigated through postprocessing with clinical rules that validate the synthetic records. If our model is deployed in the real world, it is important to consider implementing such postprocessing steps to ensure that only clinically relevant synthetic records are produced.  
+
+Finally, our HALO model focuses on generating longitudinal EHR data, such as medical codes and lab results. However, other crucial data modalities, such as clinical notes and medical images, are not yet covered by the model. To generate fully comprehensive patient records that include all modalities, it will be necessary to use diverse training data and develop multiple models to handle each modality. This exciting avenue of research is a promising future direction.
+
+<--- Page Split --->
+
+## 6 CONCLUSION  
+
+In this paper, we proposed a new method HALO for generating high- dimensional synthetic longitudinal EHR data. Our method is specifically designed to handle the sequential, multi- granular, and extremely high- dimensional nature of electronic health records by generating an explicit probability distribution over the codes, visits, and records, and HALO can generate realistic data so without needing to aggregate or remove any codes as past approaches have unanimously done. We then showed that HALO can produce incredibly realistic synthetic EHR data. Specifically, we showed that HALO can capture the probability distribution underlying the records better than other language model baselines and then produce a synthetic dataset that both looks similar to and offers the utility of real patient records as measured by medical code occurrence probabilities and machine learning classification tasks augmented with synthetic data. Finally, we also show that our method offers this performance without compromising privacy through several privacy evaluations.  
+
+In conclusion, one of the key advantages of HALO is its ability to generate binary sequences that are over a million variables in length. Its impressive performance makes it a promising avenue for developing and sharing realistic but synthetic EHR datasets that can support diverse applications. This represents an exciting opportunity to expand the use of synthetic data in the healthcare field and could help to address some of the challenges associated with data privacy and security.  
+
+## 7 DATA AVAILABILITY  
+
+While the outpatient EHR dataset is proprietary, the MIMIC- III inpatient EHR dataset [25] that we use is publicly available and may be downloaded and used freely after performing training and applying on PhysioNet.  
+
+## REFERENCES  
+
+[1] Choi, E., Bahadori, M. T., Song, L., Stewart, W. F. & Sun, J. Gram: graph- based attention model for healthcare representation learning. In Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, 787- 795 (2017). [2] Choi, E. et al. Retain: An interpretable predictive model for healthcare using reverse time attention mechanism. Advances in neural information processing systems 29 (2016). [3] Farrar, C. R. & Worden, K. Structural health monitoring: a machine learning perspective (John Wiley & Sons, 2012). [4] Duggal, R., Freitas, S., Xiao, C., Chau, D. H. & Sun, J. Rest: Robust and efficient neural networks for sleep monitoring in the wild. In Proceedings of The Web Conference 2020, 1704- 1714 (2020). [5] Fu, T., Hoang, T. N., Xiao, C. & Sun, J. Ddl: Deep dictionary learning for predictive phenotyping. In IJCAI: proceedings of the conference, vol. 2019, 5857 (NIH Public Access, 2019). [6] Che, Z. & Liu, Y. Deep learning solutions to computational phenotyping in health care. In 2017 IEEE International Conference on Data Mining Workshops (ICDMW), 1100- 1109 (IEEE, 2017). [7] Shang, J., Xiao, C., Ma, T., Li, H. & Sun, J. Gamenet: Graph augmented memory networks for recommending medication combination. In proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, 1126- 1133 (2019). [8] Wang, L., Zhang, W., He, X. & Zha, H. Supervised reinforcement learning with recurrent neural network for dynamic treatment recommendation. In Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, 2447- 2456 (2018). [9] Shang, J., Ma, T., Xiao, C. & Sun, J. Pre- training of graph augmented transformers for medication recommendation. arXiv preprint arXiv:1906.00346 (2019). [10] Neamatullah, I. et al. Automated de- identification of free- text medical records. BMC medical informatics and decision making 8, 1- 17 (2008).  
+
+[11] Kushida, C. A. et al. Strategies for de- identification and anonymization of electronic health record data for use in multicenter research studies. Medical care 50, 582 (2012). [12] El Emam, K. et al. The re- identification risk of canadians from longitudinal demographics. BMC medical informatics and decision making 11, 1- 12 (2011). [13] Benitez, K. & Malin, B. Evaluating re- identification risks with respect to the hipaa privacy rule. Journal of the American Medical Informatics Association 17, 169- 177 (2010). [14] El Emam, K., Jonker, E., Arbuckle, L. & Malin, B. A systematic review of re- identification attacks on health data. PloS one 6, e28071 (2011). [15] Zhang, Z., Yan, C., Lasko, T. A., Sun, J. & Malin, B. A. Synteg: a framework for temporal structured electronic health data simulation. Journal of the American Medical Informatics Association 28, 596- 604 (2021). [16] Biswal, S. et al. Eva: Generating longitudinal electronic health records using conditional variational autoencoders. In Machine Learning for Healthcare Conference, 260- 282 (PMLR, 2021). [17] Torfi, A. & Fox, E. A. Corgan: Correlation- capturing convolutional generative adversarial networks for generating synthetic healthcare records. In The Thirty- Third International Flairs Conference (2020). [18] Cui, L. et al. Conan: Complementary pattern augmentation for rare disease detection. In Proceedings of the AAAI Conference on Artificial Intelligence, vol. 34, 614- 621 (2020). [19] Choi, E. et al. Generating multi- label discrete patient records using generative adversarial networks. In Machine learning for healthcare conference, 286- 305 (PMLR, 2017). [20] Baowaly, M. K., Lin, C.- C., Liu, C.- L. & Chen, K.- T. Synthesizing electronic health records using improved generative adversarial networks. Journal of the American Medical Informatics Association 26, 228- 241 (2019). [21] Li, J., Cairns, B. J., Li, J. & Zhu, T. Generating synthetic mixed- type longitudinal electronic health records for artificial intelligent applications. arXiv preprint arXiv:2112.12047 (2021). [22] Zhang, Z., Yan, C., Mesa, D. A., Sun, J. & Malin, B. A. Ensuring electronic medical record simulation through better training, modeling, and evaluation. Journal of the American Medical Informatics Association 27, 99- 108 (2020). [23] Yan, C., Zhang, Z., Nyemba, S. & Malin, B. A. Generating electronic health records with multiple data types and constraints. In AMIA Annual Symposium Proceedings, vol. 2020, 1335 (American Medical Informatics Association, 2020). [24] Rashidian, S. et al. Smooth- gan: towards sharp and smooth synthetic ethr data generation. In International Conference on Artificial Intelligence in Medicine, 37- 48 (Springer, 2020). [25] Johnson, A. E. et al. Mimic- iii, a freely accessible critical care database. Scientific data 3, 1- 9 (2016). [26] Radford, A., Narasimhan, K., Salimans, T. & Sutskever, I. Improving language understanding by generative pre- training (2018). [27] Walonoski, J. et al. Synthea: An approach, method, and software mechanism for generating synthetic patients and the synthetic electronic health care record. Journal of the American Medical Informatics Association 25, 230- 238 (2018). [28] Borton, J. et al. Data entrepreneurs' synthetic puf: A working puf as an alternative to traditional synthetic and non- synthetic pufs. JSM Proceedings, Survey Research Methods Section (2010). [29] Tucker, A., Wang, Z., Rotalinti, Y. & Myles, P. Generating high- fidelity synthetic patient data for assessing machine learning healthcare software. NPJ digital medicine 3, 1- 13 (2020). [30] Kartoun, U. Advancing informatics with electronic medical records bots (embots). Software Impacts 2, 100006 (2019). URL https://www.sciencedirect.com/science/article/pii/S2665963819300065. [31] Foraker, R. et al. Analyses of original and computationally- derived electronic health record data: The national covid cohort collaborative. Journal of Medical Internet Research (2021). [32] Philippidis, A. Synthetic data for a real pandemic: Syntegra applying machine learning- based engine to create replica of nih's national covid cohort collaboration (n3c) dataset. GEN Edge 3, 42- 47 (2021). [33] Goodfellow, I. et al. Generative adversarial nets. Advances in neural information processing systems 27 (2014). [34] Sun, S. et al. Generating longitudinal synthetic ethr data with recurrent autoencoders and generative adversarial networks. In Heterogeneous Data Management, Polystores, and Analytics for Healthcare, 153- 165 (Springer, 2021). [35] Ma, F., Wang, Y., Gao, J., Xiao, H. & Zhou, J. Rare disease prediction by generating quality- assured electronic health records. In Proceedings of the 2020 SIAM International Conference on Data Mining, 514- 522 (SIAM, 2020). [36] Vaswani, A. et al. Attention is all you need. In Advances in neural information processing systems, 5998- 6008 (2017). [37] Devlin, J., Chang, M.- W., Lee, K. & Toutanova, K. Bert: Pre- training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805 (2018). [38] Radford, A. et al. Language models are unsupervised multitask learners. OpenAI blog 1, 9 (2019).
+
+<--- Page Split --->
+
+[39] Brown, T. B. et al. Language models are few-shot learners. CoRR abs/2005.14165 (2020). URL https://arxiv.org/abs/2005.14165. 2005.14165. [40] Guha, S., Koudas, N. & Shim, K. Approximation and streaming algorithms for histogram construction problems. ACM Transactions on Database Systems (TODS) 31, 396- 438 (2006).[41] Harutyunyan, H., Khachatrian, H., Kale, D. C., Ver Steeg, G. & Galstyan, A. Multitask learning and benchmarking with clinical time series data. Scientific Data 6, 96 (2019). URL https://doi.org/10.1038/s41597- 019- 0103- 9. [42] Paszke, A. et al. Pytorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems 32, 8024- 8035 (Curran Associates, Inc., 2019). URL http://papers.neurips.cc/paper/9015-pytorch- an- imperative- style- high- performance- deep- learning- library.pdf.[43] Yan, C. et al. A multifaceted benchmarking of synthetic electronic health record generation models. Nature Communications 13, 7609 (2022).[44] Yale, A. et al. Generation and evaluation of privacy preserving synthetic health data. Neurocomputing 416, 244- 255 (2020).  
+
+## 8 ADDENDUM  
+
+Acknowledgments This work is in part supported by National Science Foundation award SCH- 2014438, IIS- 1418511, CCF- 1533768, IIS- 2034479, the National Institute of Health award NIH R01 1R01NS107291- 01 and R56HL138415. Author Contributions BT and JS proposed the method, BT and conducted all the experiments, BT, CX and JS wrote the manuscript. Competing Interests The authors declare that they have no competing financial interests. Correspondence Correspondence and requests for materials should be addressed to jimeng@illinois.edu.  
+
+1753  
+
+1754  
+
+1755  
+
+1756  
+
+1757  
+
+1758  
+
+1759  
+
+1760  
+
+1761  
+
+1762  
+
+1763  
+
+1764  
+
+1765  
+
+1766  
+
+1767  
+
+1768  
+
+1769  
+
+1770  
+
+1771  
+
+1772  
+
+1773  
+
+1774  
+
+1775  
+
+1776  
+
+1777  
+
+1778  
+
+1779  
+
+1780  
+
+1781  
+
+1782  
+
+1783  
+
+1784  
+
+1785  
+
+1786  
+
+1787  
+
+1788  
+
+1789  
+
+1790  
+
+1791  
+
+1792  
+
+1793  
+
+1794  
+
+1795  
+
+1796  
+
+1797  
+
+1798
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+MedisynSupplementalMaterial.pdf
+
+<--- Page Split --->

preprint/preprint__b45a5e7f99daa68aef16db2d23be2a87229cf48cb4f0b92666b594da915a95c1/images_list.json

@@ -0,0 +1,55 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1",
+    "footnote": [],
+    "bbox": [
+      [
+        40,
+        90,
+        770,
+        831
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2",
+    "footnote": [],
+    "bbox": [
+      [
+        42,
+        216,
+        952,
+        438
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3",
+    "footnote": [],
+    "bbox": [
+      [
+        40,
+        40,
+        592,
+        789
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Figure 5",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 14
+  }
+]

preprint/preprint__b45a5e7f99daa68aef16db2d23be2a87229cf48cb4f0b92666b594da915a95c1/preprint__b45a5e7f99daa68aef16db2d23be2a87229cf48cb4f0b92666b594da915a95c1.mmd

@@ -0,0 +1,271 @@
+
+# Ctenophore extracellular DNA traps demonstrate conserved antimicrobial behaviors present at the emergence of animals  
+
+Lauren Vandepas LVANDEPASEBENAROYARESEARCH.ORG  
+
+University of Miami  Caroline Stefani  Benaroya Research Institute https://orcid.org/0000- 0003- 3420- 432X  
+
+Frederick Goetz  University of Wisconsin- Milwaukee  
+
+Nikki Traylor- Knowles  University of Miami  
+
+William Browne  University of Miami https://orcid.org/0000- 0001- 8200- 6489  
+
+Adam Lacy- Hulbert  University of Washington https://orcid.org/0000- 0003- 2162- 0156  
+
+## Article  
+
+Keywords: Ctenophore, immune cell evolution, Crassostrea gigas, extracellular DNA traps  
+
+Posted Date: April 22nd, 2022  
+
+DOI: https://doi.org/10.21203/rs.3. rs- 1535199/v1  
+
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Additional Declarations: There is NO Competing Interest.  
+
+Version of Record: A version of this preprint was published at Nature Communications on April 6th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 46807- 6.
+
+<--- Page Split --->
+
+## Abstract  
+
+The formation of extracellular DNA traps (ETosis) is a first response mechanism by specific immune cells following exposure to microbes \(^{1,2}\) . First characterized in vertebrate neutrophils, cells capable of ETosis have been recently discovered in several invertebrate taxa and the formation of invertebrate DNA traps has been most thoroughly examined in bivalves \(^{3 - 5}\) . Here we report that ctenophores – thought to have diverged very early from the metazoan stem lineage \(^{6,7}\) – possess immune cell types capable of ETosis, suggesting that this cellular immune response behavior was likely present early in metazoan evolution. To assess conservation of ET activation between evolutionarily distant phyla, we deployed a comparative approach integrating data from the model ctenophore Mnemiopsis leidyi and the oyster Crassostrea gigas to develop a novel imaging analysis pipeline to quantify ETosis in large numbers of cells. We demonstrate that both Mnemiopsis and Crassostrea immune cells can undergo ETosis after exposure to diverse microbes and pharmacological stimuli. Our results suggest that the range of cellular immune behaviors and signaling cascades that produce extracellular DNA traps were likely present prior to the divergence of extant metazoan lineages and thus ETosis represents an evolutionarily ancient defense against pathogens existing at the dawn of animal multicellularity.  
+
+## Main  
+
+The release of extracellular DNA traps (ETs) is a morphologically and molecularly distinct immune- based process of cell death during which ETosis- competent immune cells cast nuclear chromatin material into the surrounding extracellular space in filamentous nets, trapping and killing invading microbes \(^{8,9}\) . ETosis can be initiated by activation of specific signaling cascades following exposure to cytokines, microbes, pathogen- associated molecular patterns (PAMPs), or pharmacological agents \(^{10,11}\) . Initially believed to be a behavior exclusive to vertebrate immune cells, a number of recent studies have highlighted ETosis as an anti- microbial behavior also present in non- vertebrate taxa \(^{3,4,12 - 15}\) . Importantly, it remains unknown whether non- vertebrate immune cells competent for ETosis function via conserved molecular pathways. Thus the homology of non- vertebrate anti- microbial behavior to vertebrate leukocyte ETosis is unclear \(^{15}\) .  
+
+Prior to the production of ETs in vertebrate and bivalve mollusc immune cells, a reactive oxygen species (ROS) burst is required to activate signaling cascades (reviewed in Auguste et al 2020). Critically, ROS production can be generated through several distinct pathways depending on the stimulus \(^{9}\) . For example, NADPH- dependent ETosis involves NOX (NADPH- oxidase) generated by the plasma membrane, protein kinase C (PKC), and calcium, resulting in increased cytosolic ROS production following incubation with phorbol 12- myristate 13- acetate (PMA), fungi, or gram- positive bacteria, \(^{10}\) . Alternatively, ETosis can be induced via a ROS burst from mitochondria independent of NOX \(^{9,10,16,17}\) . Induction of ETosis can also be triggered through initiation of a calcium flux across the membrane in ETosis- competent cells via exposure to the calcium ionophore A23187, the potassium ionophore nigericin, or UV light \(^{10,16}\) . Notably, induction of ETosis using calcium ionophores has been tested in relatively few taxa, and the stimulation of ETosis via the potassium ionophore nigericin has not been tested outside of vertebrate model systems.
+
+<--- Page Split --->
+
+Ctenophores (also known as "comb jellies") are a phylum of gelatinous planktonic marine invertebrates that diverge very early from the animal stem lineage and may represent the most ancient extant metazoan phylum \(^{6,7}\) . Ctenophores have two distinct germ layers – ectoderm and endomesoderm ��� separated by a jelly- like layer of collagenous mesoglea, and lack a circulatory system where immune cells are typically concentrated (Fig. 1A). Understanding functional attributes of ctenophore physiological systems has provided insights into fundamental conservation of animal cell types and signaling pathways, as well as revealing mechanisms for emergence of evolutionary novelties \(^{18 - 21}\) . Currently, the ctenophore immune system remains almost entirely undescribed \(^{20,22}\) .  
+
+Specific immune cell types have not been explicitly identified in ctenophores, though they do possess mobile amoebocyte- like or stellate cells that reside in the mesoglea and are capable of phagocytosis \(^{20,23}\) . Whether ctenophores have other specialized mechanisms of cellular immune defense when exposed to classic microbial PAMPs has not been reported. Here we demonstrate that the model ctenophore Mnemiopsis leidyi possesses cell types capable of ETosis. Using a comparative approach with the oyster Crassostrea gigas, we further show in both distantly related species that ETosis is initiated in response to diverse microbial and pharmacological stimuli. To quantify the ETosis response, we developed a novel unbiased automated imaging pipeline to functionally identify and compare ETotic cells between species. Our data suggest that cellular immune behaviors and signaling cascades that trigger ETs under a range of stimuli represent an evolutionarily ancient defense against pathogens that was likely present prior to the divergence of extant metazoan lineages.  
+
+## Results  
+
+## Ctenophore immune cells undergo ETosis in response to pathogen exposure  
+
+To assess cellular immune responses to microbial challenge in vitro, we isolated live cells from whole ctenophore preparations (Fig. 1B). After incubation with fluorescent Escherichia coli, we observed motile, stellate cells competent for phagocytosing large amounts of bacteria (Supp. Video 1; Supp. Figure 1). We further observed that some stellate cells changed their morphology dramatically by retracting their processes, undergoing nuclear rotation, and subsequently rapidly extruding nuclear material (Fig. 1C; Supp. Video 2). This behavior is remarkably similar to that of vertebrate monocytes during the cytoskeletal rearrangements preceding extracellular DNA trap (ET) formation \(^{10,24}\) . This led us to speculate that some ctenophore immune cells were producing ETs in response to the presence of microbes.  
+
+We examined microbially- challenged Mnemiopsis cells using confocal microscopy and observed cells with decondensed DNA cast in large areas surrounding individual cell bodies. These networks of extracellular DNA were closely associated with individual E. coli (Fig. 1D). Three- dimensional rendering of confocal z- stacks revealed that E. coli bacteria were entangled in the extruded Mnemiopsis DNA (Fig. 1E;
+
+<--- Page Split --->
+
+Supp. Video 3). To further characterize the extruded DNA using immunofluorescence, we stained cells with an antibody that recognizes an array of histone proteins (H1, H2A, H2B, H3, H4). We observed that Mnemiopsis immune cell ETs are composed of chromatin, typical of ETs described in other taxa (Fig. 1F). ETotic cells show diffuse staining over a large area, whereas non- ETotic cells display intact nuclei with stereotypical complements of concentrated DNA and histone labeling (Fig. 1D- F; 9). These data identify specific centophore cell types that produce bona fide extracellular chromatin traps when exposed to a microbial signature.  
+
+## Diverse microbial signatures induce ET formation in Mnemiopsis leidyi independently of Non-ETotic cell death  
+
+To simultaneously and accurately quantify ETosis and non- ETotic cell death we developed a semiautomated imaging pipeline using a combination of two DNA dyes: Hoechst, a membrane- permeable stain which labels all cell nuclei, and SytoxGreen, a non- permeable stain, which selectively labels nuclei of dying or dead cells with compromised cell membranes. Both dyes label extracellular DNA nets (Fig. 2). Importantly, the nuclear envelopes of necrotic and apoptotic cells remain relatively intact, displaying concentrated fluorescent labeling 25. Using our novel image analysis pipeline, we performed automated comparisons of thousands of images to accurately calculate percentages of live, dead, or ETotic cells. Critically, the development of this pipeline allowed us to accurately identify and discriminate between ETosis and non- ETotic cell death following treatments.  
+
+After exposing isolated Mnemiopsis cells to E. coli (Fig. 2A, B), we analyzed the stained nuclei in each image. Using image segmentation analysis, we defined individual cell masks based on Hoechst signal and measured relative fluorescence intensities of Hoechst and SytoxGreen inside each cell mask (Fig. 2C). Using those measurements, we defined 3 distinct cell populations: live cells, ETotic cells and non- ETotic cell death (Fig. 2D). Live cells are negative for SytoxGreen fluorescence with no dispersion of Hoechst signals (Hoechst<sup>high</sup>/SytoxGreen<sup>low</sup>) (Fig. 2B, 2D). Cells that have ETotic nuclei exhibit high dispersion of Hoechst fluorescence associated with extracellular DNA net formation (Hoechst<sup>low</sup>). In contrast, cells that have condensed Hoechst fluorescence and high SytoxGreen fluorescent signals (Hoechst<sup>high</sup>/SytoxGreen<sup>high</sup>) are dying from non- ETotic cell death processes. This approach allowed us to accurately identify and quantify large numbers of cells, including those undergoing ETosis after incubation with E. coli.  
+
+We then expanded our analyses of ET production to include responses to additional microbial stimuli ( https://github.com/carolinestefani/ETosis- and- death- automated- pipeline). We examined whether Mnemiopsis immune cells undergo ETosis when exposed to a diverse array of microbes, including the following: heat- killed gram- negative bacteria E. coli, heat- killed gram- positive bacteria Staphylococcus aureus, and cell wall extract from yeast Saccharomyces cerevisiae (zymosan). ETotic cells display diffuse Hoechst signal characteristic of filamentous DNA nets following exposure to E. coli, S. aureus, or zymosan (Fig. 3A). In contrast non- ETotic cells maintain intact nuclear material with concentrated
+
+<--- Page Split --->
+
+Hoechst labeling. We analyzed both ETosis and total cell death in Mnemiopsis cells after four hours of microbial exposure (N = 24 individual animals). Contour plots representing Hoechst and SytoxGreen signal intensities show three distinct clusters: ETotic cells, dying cells, and living cells (Fig. 3B).  
+
+ETosis events significantly increased in Mnemiopsis cells when exposed to zymosan or E. coli in vitro, with mean increases in detected ETotic events of \(17\%\) and \(7\%\) over untreated cells (Fig. 3C). Incubation with S. aureus also elicited a strong ETotic response in Mnemiopsis cells, with a mean increase in detected ETotic events of \(69\%\) over untreated cells. Notably, non- ETotic cell death increased significantly after incubation with E. coli but did not change after exposure to zymosan or S. aureus (Fig. 3C). Our data demonstrate that a panel of microbes stimulates ETosis in ctenophore cells and that we can accurately and efficiently measure ETosis and non- ETotic cell death events simultaneously.  
+
+## Extracellular trap formation in Mnemiopsis is induced by classic pharmacological stimuli  
+
+ETosis can be induced in vertebrate granulocytes and leukocytes with pharmacological agents that activate production of ROS via discrete intracellular signaling pathways \(^{10,11,16}\) . Though ctenophore genomes appear to lack predicted gene homologs to many classic innate immunity pathway cell surface receptors and signaling intermediaries, many components of metazoan stress response pathways, as well as secondary messenger machinery, are present \(^{20,26,27}\) . We hypothesized that ETosis in Mnemiopsis immune cells could be stimulated by classic chemical agents broadly used in studies of vertebrate ETs.  
+
+We observed a significant induction of ETosis in Mnemiopsis after a four- hour exposure to PMA, with a mean increase of \(4.3\%\) over untreated control cells (Fig. 3D- F). Nigericin, a potassium ionophore, induces ET formation in vertebrates by initiating ROS release from mitochondria \(^{10,28}\) . Initiation of a calcium influx following stimulation by calcium ionophore A23817 (calcium) can also induce ET formation independent of NOX \(^{16}\) . We observed a significant induction of ETosis following incubation of Mnemiopsis cells with both the potassium ionophore nigericin and the calcium ionophore A23187, with a mean increase of \(33\%\) and \(20\%\) over untreated control cells, respectively (Fig. 3D- F). Nigericin exposure elicited the highest amount of ET formation (Fig. 3F; Supp. Video 4; Supp. Figure 2).  
+
+While we did not observe a significant change in levels of non- ETotic cell death following PMA stimulus, our results did indicate a significant increase in non- ETotic cell death following nigericin and A23187 treatments (Fig. 3F). Exposure of Mnemiopsis cells to all three classical pharmacological stimuli – PMA, the potassium ionophore nigericin, and the calcium ionophore A23187 – induced significant ET formation. These results implicate diverse signaling pathways involved in production of ETs that also retain deep evolutionary conservation across extant metazoan phyla.
+
+<--- Page Split --->
+
+## ETosis in the model bivalve, Crassostrea gigas, is induced by diverse microbial and chemical stimuli  
+
+We sought another non- vertebrate model to assess conservation of ET formation across phyla because ET stimulation in Mnemiopsis immune cells is intriguingly similar to ETosis in mammalian systems. The formation of invertebrate extracellular DNA traps has been characterized most extensively in molluscs \(^{3 - }\) \(^{5,13,29}\) . Bivalve molluscs, like the Pacific oyster Crassostrea gigas, have blood- like circulatory cells, collectively called hemocytes, that have immune functions \(^{30}\) . However, prior attempts to stimulate production of ETs in oyster via exposure to pathogen signatures and PMA has varied between studies \(^{5}\) . For example previous studies observed a robust ETotic response following challenge with Vibrio, a virulent bivalve pathogen, in oyster hemocytes \(^{3}\) , however the induction of ETosis with other microbes has been only modest or not observed in previous studies \(^{5}\) . Thus, conservation of ET induction pathways in bivalves remains unclear \(^{3 - 5}\) .  
+
+We measured ETs and cell death in Crassostrea hemocytes after exposure to zymosan, E. coli, and S. aureus (Fig. 4A- D). We found that ETosis is significantly stimulated in hemocytes exposed to all three pathogens (Fig. 4D; Supp. Figure 3), with mean increases of \(13\%\) , \(19\%\) , and \(22\%\) over untreated hemocytes, respectively. Hemocyte non- ETotic cell death (apoptosis, necrosis) increased significantly with E. coli exposure but not for zymosan challenge (Fig. 4D).  
+
+We also assayed Crassostrea hemocytes for ET production after exposure to pharmacological reagents that engage distinct signaling pathways in vertebrate monocytes. We analyzed the behavior of Crassostrea hemocytes after four- hour incubation with PMA, the potassium ionophore nigericin, and the calcium ionophore A23187 (Fig. 4E). Exposure to PMA and the calcium ionophore A23187 induced significant ET production in isolated Crassostrea hemocytes (Fig. 4F, 4G). However, the potassium ionophore nigericin showed no significant induction (Fig. 4F, 4G). In contrast, a large proportion of Crassostrea hemocytes produce ETs after 4 hours of exposure to A23187, similar to prior studies \(^{5}\) .  
+
+## Discussion  
+
+Identifying the core molecular mechanisms of immune cells and their evolutionary origins is critical for understanding both the evolution of metazoan immunity function and the evolution of multicellularity. Initially described in 2004 as a unique antimicrobial behavior of vertebrate neutrophils, cells capable of extracellular trap production have recently subsequently been discovered in multiple non- vertebrate bilaterian taxa \(^{4,12,14}\) . Here we report that ctenophore immune cells are competent for ETosis in response to microbial challenge and assess potential mechanisms associated with ETosis pathway activation with well- characterized pharmacological reagents. ETosis in both the model ctenophore Mnemiopsis leidyi and the model oyster Crassostrea gigas is initiated after exposure to diverse microbial signatures as well as pharmaceutical compounds that are known to activate distinct signaling cascades in vertebrates.
+
+<--- Page Split --->
+
+Many non- vertebrates lack gene homologs for major proteins known to be essential for ETosis in vertebrate neutrophils, such as PAD4, neutrophil elastase, and pannexin- 1. It is currently unknown what signaling molecules are necessary for ET formation in non- vertebrate taxa. While the involvement of NADPH per se has not been directly assessed in invertebrate ET production, there is strong evidence for the presence of multiple ETosis pathways that can be activated via stimulation with microbial signatures, parasites, PMA, A23187, and UV light \(^{3 - 5,12 - 14}\) . Prior studies in bivalves conflict as to whether ETosis can be stimulated with the NADPH- dependent pathway stimulus, PMA, or with microbes commonly used in immunological studies \(^{4,5}\) . Our results demonstrate that Crassostrea hemocytes produce ETs in response to stimulation by diverse PAMPs and distinct pharmacological reagents that, in vertebrate monocytes, engage both NADPH- dependent signaling (PMA stimulus) and NADPH- independent signaling (A23187 stimulus). Notably, Crassostrea hemocytes did not produce ETs following stimulation by the potassium ionophore nigericin, suggesting a potential loss of nigericin stimulating signal transduction in this mollusc species.  
+
+Prior to ET production in vitro, Mnemiopsis immune cells and Crassostrea hemocytes that underwent ETosis after exposure to E. coli did not seem to have phagocytosed bacteria in significant amounts. Intriguingly, we observed other motile cells that phagocytosed large amounts of bacteria without undergoing ETosis (Supp. Video 1, Supp. Video 5; Supp. Figure 3). Our functional characterization of ETosis competent cells in both taxa suggest that future studies should address whether the ETosis- competent and the highly phagocytic non- ETotic cells represent discrete immune cell types in these non- vertebrate taxa. In contrast to vertebrate immune cell nuclear morphologies that correspond with discrete immune cell functions \(^{31}\) , there do not seem to be notable differences in nuclear morphologies in either Crassostrea hemocytes or Mnemiopsis cells.  
+
+Ctenophores, which diverged very early from the metazoan stem lineage, have not had specific immune cell antimicrobial behaviors described beyond phagocytosis \(^{20,23}\) . We demonstrate that Mnemiopsis leidyi possesses cells functionally competent for ETosis in response to a range of pathogen challenges. The production of extracellular traps in Mnemiopsis suggests this immune cell type specific antimicrobial defense behavior was likely present very early in metazoan evolution (Fig. 5). Our data further demonstrates that both Mnemiopsis leidyi and Crassostrea gigas immune cells are capable of ETosis via stimulation with pharmaceutical agents that, in vertebrate immune cells, differentially induce either NADPH- dependent or NADPH- independent pathways to activate ET production. These data suggest that ETosis is an ancient and fundamental immune cell defense behavior that is not only present among distantly related metazoans but is also triggered by similar mechanisms.  
+
+## References  
+
+1. Brinkmann, V. et al. Neutrophil extracellular traps kill bacteria. Science 303, 1532–1535, doi:10.1126/science.1092385 (2004).
+
+<--- Page Split --->
+
+2. Wartha, F., Beiter, K., Normark, S. & Henriques-Normark, B. Neutrophil extracellular traps: casting the NET over pathogenesis. Curr Opin Microbiol 10, 52–56, doi:10.1016/j.mib.2006.12.005 (2007).  
+
+3. Poirier, A. C. et al. Antimicrobial histones and DNA traps in invertebrate immunity: evidences in Crassostrea gigas. J Biol Chem 289, 24821–24831, doi:10.1074/jbc.M114.576546 (2014).  
+
+4. Robb, C. T., Dyrrynda, E. A., Gray, R. D., Rossi, A. G. & Smith, V. J. Invertebrate extracellular phagocyte traps show that chromatin is an ancient defence weapon. Nat Commun 5, 4627, doi:10.1038/ncomms5627 (2014).  
+
+5. Romero, A., Novoa, B. & Figueras, A. Extracellular traps (ETosis) can be activated through NADPH-dependent and -independent mechanisms in bivalve mollusks. Dev Comp Immunol 106, 103585, doi:10.1016/j.dci.2019.103585 (2020).  
+
+6. Dunn, C. W. et al. Broad phylogenomic sampling improves resolution of the animal tree of life. Nature 452, 745–749, doi:10.1038/nature06614 (2008).  
+
+7. Li, Y., Shen, X. X., Evans, B., Dunn, C. W. & Rokas, A. Rooting the Animal Tree of Life. Mol Biol Evol 38, 4322–4333, doi:10.1093/molbev/msab170 (2021).  
+
+8. Fuchs, T. A. et al. Novel cell death program leads to neutrophil extracellular traps. J Cell Biol 176, 231–241, doi:10.1083/jcb.200606027 (2007).  
+
+9. Sollberger, G., Tilley, D. O. & Zychlinsky, A. Neutrophil Extracellular Traps: The Biology of Chromatin Externalization. Dev Cell 44, 542–553, doi:10.1016/j.devcel.2018.01.019 (2018).  
+
+10. Kenny, E. F. et al. Diverse stimuli engage different neutrophil extracellular trap pathways. Elife 6, doi:10.7554/eLife.24437 (2017).  
+
+11. Auguste, M., Balbi, T., Ciacci, C. & Canesi, L. Conservation of Cell Communication Systems in Invertebrate Host-Defence Mechanisms: Possible Role in Immunity and Disease. Biology (Basel) 9, doi:10.3390/biology9080234 (2020).  
+
+12. Homa, J., Ortmann, W. & Kolaczkowska, E. Conservative Mechanisms of Extracellular Trap Formation by Annelida Eisenia andrei: Serine Protease Activity Requirement. PLoS One 11, e0159031, doi:10.1371/journal.pone.0159031 (2016).  
+
+13. Lange, M. K. et al. Gastropod-derived haemocyte extracellular traps entrap metastrongyloid larval stages of Angiostrongylus vasorum, Aelurostrongylus abstrusus and Troglostrongylus brevior. Parasit Vectors 10, 50, doi:10.1186/s13071-016-1961-z (2017).  
+
+14. Nascimento, M. T. C. et al. DNA extracellular traps are part of the immune repertoire of Periplaneta americana. Dev Comp Immunol 84, 62–70, doi:10.1016/j.dci.2018.01.012 (2018).  
+
+15. Ramos-Martinez, E. et al. Multiple Origins of Extracellular DNA Traps. Front Immunol 12, 621311, doi:10.3389/fimmu.2021.621311 (2021).  
+
+16. Douda, D. N., Khan, M. A., Grasemann, H. & Palaniyar, N. SK3 channel and mitochondrial ROS mediate NADPH oxidase-independent NETosis induced by calcium influx. Proc Natl Acad Sci U S A 112, 2817–2822, doi:10.1073/pnas.1414055112 (2015).
+
+<--- Page Split --->
+
+17. Papayannopoulos, V. Neutrophil extracellular traps in immunity and disease. Nat Rev Immunol 18, 134-147, doi:10.1038/nri.2017.105 (2018).  
+
+18. Babonis, L. S., Ryan, J. F., Enjolras, C. & Martindale, M. Q. Genomic analysis of the tryptome reveals molecular mechanisms of gland cell evolution. Evodevo 10, 23, doi:10.1186/s13227-019-0138-1 (2019).  
+
+19. Jokura, K. et al. CTENO64 Is Required for Coordinated Paddling of Ciliary Comb Plate in Ctenophores. Curr Biol 29, 3510-3516 e3514, doi:10.1016/j.cub.2019.08.059 (2019).  
+
+20. Traylor-Knowles, N., Vandepas, L. E. & Browne, W. E. Still Enigmatic: Innate Immunity in the Ctenophore Mnemiopsis leidyi. Integr Comp Biol 59, 811-818, doi:10.1093/icb/icz116 (2019).  
+
+21. Presnell, J. S. & Browne, W. E. Kruppel-like factor gene function in the ctenophore Mnemiopsis leidyi assessed by CRISPR/Cas9-mediated genome editing. Development 148, doi:10.1242/dev.199771 (2021).  
+
+22. Bolte, S. et al. Specific immune priming in the invasive ctenophore Mnemiopsis leidyi. Biol Lett 9, 20130864, doi:10.1098/rsbl.2013.0864 (2013).  
+
+23. Franc, J. M. La matrice mésogléenne des Ctenaires. Bulletin de la Société zoologique de France 113, 347-447 (1988).  
+
+24. Boe, D. M., Curtis, B. J., Chen, M. M., Ippolito, J. A. & Kovacs, E. J. Extracellular traps and macrophages: new roles for the versatile phagocyte. J Leukoc Biol 97, 1023-1035, doi:10.1189/jlb.4R1014-521R (2015).  
+
+25. Guimaraes-Costa, A. B., Nascimento, M. T., Wardini, A. B., Pinto-da-Silva, L. H. & Saraiva, E. M. ETosis: A Microbicidal Mechanism beyond Cell Death. J Parasitol Res 2012, 929743, doi:10.1155/2012/929743 (2012).  
+
+26. Ryan, J. F. et al. The genome of the ctenophore Mnemiopsis leidyi and its implications for cell type evolution. Science 342, 1242592, doi:10.1126/science.1242592 (2013).  
+
+27. Moroz, L. L. et al. The ctenophore genome and the evolutionary origins of neural systems. Nature 510, 109-114, doi:10.1038/nature13400 (2014).  
+
+28. Meier, A. et al. Inhibition of Human Neutrophil Extracellular Trap (NET) Production by Propofol and Lipid Emulsion. Front Pharmacol 10, 323, doi:10.3389/fphar.2019.00323 (2019).  
+
+29. Bachere, E. et al. The new insights into the oyster antimicrobial defense: Cellular, molecular and genetic view. Fish Shellfish Immunol 46, 50-64, doi:10.1016/j.fsi.2015.02.040 (2015).  
+
+30. Jiang, S. et al. Functional characterisation of phagocytes in the Pacific oyster Crassostrea gigas. PeerJ 4, e2590, doi:10.7717/peerj.2590 (2016).  
+
+31. Skinner, B. M. & Johnson, E. E. Nuclear morphologies: their diversity and functional relevance. Chromosoma 126, 195-212, doi:10.1007/s00412-016-0614-5 (2017).  
+
+32. Presnell, J. S. et al. The Presence of a Functionally Tripartite Through-Gut in Ctenophora Has Implications for Metazoan Character Trait Evolution. Curr Biol 26, 2814-2820, doi:10.1016/j.cub.2016.08.019 (2016).
+
+<--- Page Split --->
+
+33. Vandepas, L. E., Warren, K. J., Amemiya, C. T. & Browne, W. E. Establishing and maintaining primary cell cultures derived from the ctenophore Mnemiopsis leidyi. J Exp Biol 220, 1197–1201, doi:10.1242/jeb.152371 (2017).  
+34. Acharya, M. et al. alphav Integrins combine with LC3 and atg5 to regulate Toll-like receptor signalling in B cells. Nat Commun 7, 10917, doi:10.1038/ncomms10917 (2016).  
+35. Dieter, A. C., Vandepas, L. E. & Browne, W. E. Isolation and Maintenance of In Vitro Cell Cultures from the Ctenophore Mnemiopsis leidyi. Methods Mol Biol 2450, 347–358, doi:10.1007/978-1-0716-2172-1_18 (2022).  
+
+## Methods  
+
+## Animal maintenance  
+
+Laboratory cultures of Mnemiopsis leidyi were maintained as previously described \(^{32}\) . Adult Mnemiopsis cells were isolated following established protocols \(^{20,32,33}\) . Crassostrea gigas were maintained under flowing seawater at approximately \(13^{\circ}\mathrm{C}\) and hemolymph was extracted from the adductor muscle with a syringe. Cells from both taxa were maintained in vitro under sterile conditions in filtered seawater (FSW) + 1% penicillin/streptomycin.  
+
+## Stimulation of ETosis and imaging  
+
+For stimulation and quantification of ETosis, cells were isolated from 24 individual animals, plated in 96 well plates, and exposed to pHrodo- E. coli, Staphylococcus aureus, zymosan particles (Sigma Aldrich), 25 uM nigericin (Thermo Fisher Scientific), 1 mg/mL PMA (Sigma Aldrich), or 4 uM A23187 (Sigma). Each experimental condition for each animal was performed in triplicate. Live cell staining was performed following (2) and live imaging was performed using a JuLI Stage (NanoEntek).  
+
+For immunofluorescence, ctenophore cells were prepared following \(^{33- 35}\) and labeled with mouse anti- histone H11- 4 (EMD Millipore) and anti- mouse Alexa Fluor 488 (Thermo Fisher Scientific). The H11- 4 histone antibody was selected because it recognizes histones H1, H2A/B, H3, and H4 proteins across diverse species. Cells were imaged using \(\times 60\) objective and \(\times 100\) oil objective, on a Nikon Ti (Eclipse) inverted microscope with Ultraviolet Spinning Disc (CSU- X1) confocal scanner (Perkin Elmer). Images were captured with an Orca- ER Camera using Velocity (Quorum technologies). Post- acquisition analysis such as contrast adjustment, deconvolution through iterative restoration and colocalization were performed using Velocity software.  
+
+For quantification of ETosis and total cell death, cells were treated with Hoechst 33342 (Sigma Aldrich) and SytoxGreen (Invitrogen) for 20 min, before imaging at 20X on an automated imaging plate reader, Cytation 3 (Biotek, software Gen5 v4.2).  
+
+Automated image- based profiling
+
+<--- Page Split --->
+
+We analyzed over 28,000 total images from our experiments using CellProfiler (v4.1.3). Image quality was assessed by calculating a focus score using two classes Otsu thresholding method, weighted variance on 20x20 pixel measurements. We calculated and applied an illumination correction for each fluorescent channel (SytoxGreen and Hoechst) using a background illumination function of 50 pixels block size, without smoothing. Each corrected image was then segmented using a global robust background method (0.05- 50), with a smoothing scale of 1.3488 and a correction factor of 0.89. Clumped objects were identified and split by shape. For each segmented object we measured the number and intensity of pixels in each fluorescent channel.  
+
+Each image and each segmented object, along with Metadata, were exported as csv files by experiment. R (v4.0.5) software with tidyverse (v1.3.1), dplyr (v1.0.7) and readr (v1.4.0) packages were then used to transform the datasets. Data from images and objects were merged, and measurements from individual images with a Focus Score \(< 0.2\) were removed from further analysis. This allowed us to identify and select only images that were in focus. Surface area, Hoechst intensity and SytoxGreen intensity per object (nucleus) and per individual animal were then imported into FlowJo (v10.8.0), and percentages of cells per delineated population (dead/dying cell, live cell, and ETotic cell) were calculated. Dying cells were gated using Hoechst Intensity \(>0.35\) and SytoxGreen intensity \(>0.1\) ; live cells were gated using Hoechst Intensity \(>0.35\) and Sytox Green intensity \(< 0.1\) ; and ETotic cells were gated using Hoechst Intensity \(< 0.35\) . Finally, percentages per individual animal surveyed were combined and tested for statistical significance using GraphPad Prism (v9.2.0). All statistical tests were performed using two- tailed unpaired student t- test \(^{*}p< 0.05\) , \(^{**}p< 0.01\) , \(^{***}p< 0.001\) , \(^{****}p< 0.0001\) .  
+
+## Declarations  
+
+## Acknowledgements  
+
+The authors are grateful to Anna Bruchez and Rachel Prins for imaging support. This research was supported by the National Oceanographic and Atmospheric Administration, a National Research Council Postdoctoral Fellowship to LEV, and the National Science Foundation under Grant No. 2013692. CS and ALH were supported by National Institutes of Health Grants R33AI119341 and R01GM102482.  
+
+## Author contributions  
+
+Conceptualization, C.S. and L.E.V.; Methodology, L.E.V. and C.S.; Formal Analysis, C.S.; Investigation, L.E.V. and C.S.; Writing – Original Draft, L.E.V. and C.S.; Writing – Review & Editing, C.S., A.L.H., W.E.B., N.T.K., F.W.G., and L.E.V.; Funding Acquisition, A.L.H., W.E.B., N.T.K., F.W.G., and L.EV.; Resources, A.L.H., W.E.B., and F.W.G.; Supervision, A.L.H., W.E.B., and F.W.G.  
+
+## Competing interests  
+
+The authors declare no competing interests.
+
+<--- Page Split --->
+
+## Figures  
+
+![](images/Figure_1.jpg)
+
+<center>Figure 1 </center>  
+
+Mnemiopsis immune cells produce extracellular DNA traps. (A) Adult lobate ctenophore Mnemiopsis leidyi. (B) Schematic of isolation of cells from whole Mnemiopsis leidyi. (C) Still images from Movie S1 showing a motile stellate cell retracting its processes, spinning, and extruding its nuclear contents after
+
+<--- Page Split --->
+
+exposure to E. coli. (D) Merged confocal image. Left – Nuclei of unstimulated Mnemiopsis cells. Right – A cell exposed to TxRed- E. coli has undergone ETosis; chromatin has been extruded from the cell in a large web- like pattern. E. coli are entrapped by the chromatin filaments (white arrowheads). (E) 3- dimensional image of Mnemiopsis extracellular DNA nets with E. coli entrapped. (F) Merged confocal image. Ctenophore extracellular traps are composed of DNA and histones. Histone 11- 4 antibody (green) and Hoechst (white) staining are visible in intact and ETosed Mnemiopsis cells treated with the K+ ionophore nigericin. White arrowhead marks DNA+histone nets. (C- E scale bar: \(10 \mu \mathrm{m}\) ).  
+
+![](images/Figure_2.jpg)
+
+<center>Figure 2 </center>  
+
+Novel high- throughput imaging pipeline. Schematic representation of experimental and imaging workflow. (A) Treated and untreated cells in culture were labeled with vital dye Hoechst and membrane impermeable Sytox Green and assessed for ETosis and/or cell death. (B) Example of the segmentation output and masking for identifying individual intact or ETosed nuclei. (C) Individual cells were scored for viability, ETosis, or death using FlowJo software. Scoring is based on fluorescent intensity and nuclear material area. Cells that have undergone ETosis exhibit dispersion and decreased intensity of Hoechst fluorescence associated with extracellular DNA net formation (Hoechstlow). In contrast, dead and dying cells have condensed Hoechst fluorescence and high SytoxGreen fluorescent signals (Hoechsthigh/SytoxGreenhigh). High intensity blue fluorescence signals (Hoechsthigh) denote intact nuclei of live cells.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+
+<center>Figure 3 </center>  
+
+Diverse pathogen signatures and classic pharmaceuticals stimulate ETosis in Mnemiopsis immune cells. (A) Representative images of ETotic Mnemiopsis cells exposed to fungal or bacterial pathogen signatures. (B) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following pathogen exposures. (C) ETosis is significantly stimulated in Mnemiopsis cells exposed to zymosan, gram- negative E. coli, and gram- positive S. aureus after 4 hours. Apoptosis
+
+<--- Page Split --->
+
+increases slightly with E. coli treatment but is not significant for other pathogen signatures. (D) Representative images of ETotic Mnemiopsis cells exposed to PMA, calcium ionophore A23187, or potassium ionophore nigericin. (E) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following treatment with established ETosis-inducing pharmaceuticals. (F) ETosis is significantly stimulated in hemocytes exposed to PMA, nigericin, and A23187. Non-ETosis cell death increases significantly after incubation with ionophores nigericin and A23187, but not PMA.  
+
+![](images/Figure_5.jpg)
+
+
+<--- Page Split --->
+
+## Figure 4  
+
+Some pathogen signatures and classic pharmaceuticals induce ETosis in Crassostrea hemocytes. (A) Schematic of hemocyte isolation from Pacific oyster Crassostrea gigas. (B) Representative images of ETotic C. gigas hemocytes exposed to fungal or bacterial pathogen signatures. PMA, calcium ionophore A23187, and potassium ionophore nigericin significantly induce ETosis in Crassostrea hemocytes. (C) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following pathogen exposures. (D) ETosis is significantly stimulated in Crassostrea cells exposed to zymosan, gram- negative E. coli, and gram- positive S. aureus after 4 hours. Apoptosis increases slightly with E. coli treatment and decreases following S. aureus incubation, but is not significant for zymosan. (E) Representative images of ETotic C. gigas hemocytes stimulated with PMA, A23187, or nigericin. (F) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following treatment with established ETosis- inducing pharmaceuticals. (G) PMA and A23187, but not nigericin, significantly stimulates ETosis in hemocytes. Apoptosis of hemocytes decreases when treated with PMA or A23187.
+
+<--- Page Split --->
+![PLACEHOLDER_16_0]
+
+<center>Figure 5 </center>  
+
+ETosis is an ancient metazoan immune response. Summary of ETosis phenomena across Metazoa. The presence of cells competent for ETosis in protostome, deuterostome, and non- bilaterian taxa indicate that production of extracellular DNA traps is an ancient animal immune defense.  
+
+## Supplementary Files
+
+<--- Page Split --->
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- SupplementalFigures.docx- SuppVideo1MnemiopsisPhagocytosis.mov- SuppVideo2MnemiopsisETosisEcoli.avi- SuppVideo33DMnemiopsisETConfocalStack.avi- SuppVideo4MnemiopsisETosisNigericin.avi- SuppVideo5CrassostreaPhagocytosis.avi
+
+<--- Page Split --->

preprint/preprint__b45a5e7f99daa68aef16db2d23be2a87229cf48cb4f0b92666b594da915a95c1/preprint__b45a5e7f99daa68aef16db2d23be2a87229cf48cb4f0b92666b594da915a95c1_det.mmd

@@ -0,0 +1,362 @@
+<|ref|>title<|/ref|><|det|>[[44, 108, 900, 208]]<|/det|>
+# Ctenophore extracellular DNA traps demonstrate conserved antimicrobial behaviors present at the emergence of animals  
+
+<|ref|>text<|/ref|><|det|>[[44, 230, 390, 276]]<|/det|>
+Lauren Vandepas LVANDEPASEBENAROYARESEARCH.ORG  
+
+<|ref|>text<|/ref|><|det|>[[50, 304, 660, 363]]<|/det|>
+University of Miami  Caroline Stefani  Benaroya Research Institute https://orcid.org/0000- 0003- 3420- 432X  
+
+<|ref|>text<|/ref|><|det|>[[44, 370, 360, 410]]<|/det|>
+Frederick Goetz  University of Wisconsin- Milwaukee  
+
+<|ref|>text<|/ref|><|det|>[[44, 417, 234, 457]]<|/det|>
+Nikki Traylor- Knowles  University of Miami  
+
+<|ref|>text<|/ref|><|det|>[[44, 464, 633, 504]]<|/det|>
+William Browne  University of Miami https://orcid.org/0000- 0001- 8200- 6489  
+
+<|ref|>text<|/ref|><|det|>[[44, 511, 633, 553]]<|/det|>
+Adam Lacy- Hulbert  University of Washington https://orcid.org/0000- 0003- 2162- 0156  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 595, 102, 612]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 632, 810, 652]]<|/det|>
+Keywords: Ctenophore, immune cell evolution, Crassostrea gigas, extracellular DNA traps  
+
+<|ref|>text<|/ref|><|det|>[[44, 671, 303, 690]]<|/det|>
+Posted Date: April 22nd, 2022  
+
+<|ref|>text<|/ref|><|det|>[[44, 709, 474, 728]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3. rs- 1535199/v1  
+
+<|ref|>text<|/ref|><|det|>[[44, 746, 910, 789]]<|/det|>
+License: © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[44, 808, 530, 827]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 862, 949, 905]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on April 6th, 2024. See the published version at https://doi.org/10.1038/s41467- 024- 46807- 6.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 159, 68]]<|/det|>
+## Abstract  
+
+<|ref|>text<|/ref|><|det|>[[40, 81, 955, 407]]<|/det|>
+The formation of extracellular DNA traps (ETosis) is a first response mechanism by specific immune cells following exposure to microbes \(^{1,2}\) . First characterized in vertebrate neutrophils, cells capable of ETosis have been recently discovered in several invertebrate taxa and the formation of invertebrate DNA traps has been most thoroughly examined in bivalves \(^{3 - 5}\) . Here we report that ctenophores – thought to have diverged very early from the metazoan stem lineage \(^{6,7}\) – possess immune cell types capable of ETosis, suggesting that this cellular immune response behavior was likely present early in metazoan evolution. To assess conservation of ET activation between evolutionarily distant phyla, we deployed a comparative approach integrating data from the model ctenophore Mnemiopsis leidyi and the oyster Crassostrea gigas to develop a novel imaging analysis pipeline to quantify ETosis in large numbers of cells. We demonstrate that both Mnemiopsis and Crassostrea immune cells can undergo ETosis after exposure to diverse microbes and pharmacological stimuli. Our results suggest that the range of cellular immune behaviors and signaling cascades that produce extracellular DNA traps were likely present prior to the divergence of extant metazoan lineages and thus ETosis represents an evolutionarily ancient defense against pathogens existing at the dawn of animal multicellularity.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 428, 112, 453]]<|/det|>
+## Main  
+
+<|ref|>text<|/ref|><|det|>[[41, 467, 955, 679]]<|/det|>
+The release of extracellular DNA traps (ETs) is a morphologically and molecularly distinct immune- based process of cell death during which ETosis- competent immune cells cast nuclear chromatin material into the surrounding extracellular space in filamentous nets, trapping and killing invading microbes \(^{8,9}\) . ETosis can be initiated by activation of specific signaling cascades following exposure to cytokines, microbes, pathogen- associated molecular patterns (PAMPs), or pharmacological agents \(^{10,11}\) . Initially believed to be a behavior exclusive to vertebrate immune cells, a number of recent studies have highlighted ETosis as an anti- microbial behavior also present in non- vertebrate taxa \(^{3,4,12 - 15}\) . Importantly, it remains unknown whether non- vertebrate immune cells competent for ETosis function via conserved molecular pathways. Thus the homology of non- vertebrate anti- microbial behavior to vertebrate leukocyte ETosis is unclear \(^{15}\) .  
+
+<|ref|>text<|/ref|><|det|>[[40, 694, 955, 952]]<|/det|>
+Prior to the production of ETs in vertebrate and bivalve mollusc immune cells, a reactive oxygen species (ROS) burst is required to activate signaling cascades (reviewed in Auguste et al 2020). Critically, ROS production can be generated through several distinct pathways depending on the stimulus \(^{9}\) . For example, NADPH- dependent ETosis involves NOX (NADPH- oxidase) generated by the plasma membrane, protein kinase C (PKC), and calcium, resulting in increased cytosolic ROS production following incubation with phorbol 12- myristate 13- acetate (PMA), fungi, or gram- positive bacteria, \(^{10}\) . Alternatively, ETosis can be induced via a ROS burst from mitochondria independent of NOX \(^{9,10,16,17}\) . Induction of ETosis can also be triggered through initiation of a calcium flux across the membrane in ETosis- competent cells via exposure to the calcium ionophore A23187, the potassium ionophore nigericin, or UV light \(^{10,16}\) . Notably, induction of ETosis using calcium ionophores has been tested in relatively few taxa, and the stimulation of ETosis via the potassium ionophore nigericin has not been tested outside of vertebrate model systems.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 44, 955, 230]]<|/det|>
+Ctenophores (also known as "comb jellies") are a phylum of gelatinous planktonic marine invertebrates that diverge very early from the animal stem lineage and may represent the most ancient extant metazoan phylum \(^{6,7}\) . Ctenophores have two distinct germ layers – ectoderm and endomesoderm – separated by a jelly- like layer of collagenous mesoglea, and lack a circulatory system where immune cells are typically concentrated (Fig. 1A). Understanding functional attributes of ctenophore physiological systems has provided insights into fundamental conservation of animal cell types and signaling pathways, as well as revealing mechanisms for emergence of evolutionary novelties \(^{18 - 21}\) . Currently, the ctenophore immune system remains almost entirely undescribed \(^{20,22}\) .  
+
+<|ref|>text<|/ref|><|det|>[[41, 247, 960, 499]]<|/det|>
+Specific immune cell types have not been explicitly identified in ctenophores, though they do possess mobile amoebocyte- like or stellate cells that reside in the mesoglea and are capable of phagocytosis \(^{20,23}\) . Whether ctenophores have other specialized mechanisms of cellular immune defense when exposed to classic microbial PAMPs has not been reported. Here we demonstrate that the model ctenophore Mnemiopsis leidyi possesses cell types capable of ETosis. Using a comparative approach with the oyster Crassostrea gigas, we further show in both distantly related species that ETosis is initiated in response to diverse microbial and pharmacological stimuli. To quantify the ETosis response, we developed a novel unbiased automated imaging pipeline to functionally identify and compare ETotic cells between species. Our data suggest that cellular immune behaviors and signaling cascades that trigger ETs under a range of stimuli represent an evolutionarily ancient defense against pathogens that was likely present prior to the divergence of extant metazoan lineages.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 520, 144, 545]]<|/det|>
+## Results  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 559, 905, 621]]<|/det|>
+## Ctenophore immune cells undergo ETosis in response to pathogen exposure  
+
+<|ref|>text<|/ref|><|det|>[[41, 635, 950, 840]]<|/det|>
+To assess cellular immune responses to microbial challenge in vitro, we isolated live cells from whole ctenophore preparations (Fig. 1B). After incubation with fluorescent Escherichia coli, we observed motile, stellate cells competent for phagocytosing large amounts of bacteria (Supp. Video 1; Supp. Figure 1). We further observed that some stellate cells changed their morphology dramatically by retracting their processes, undergoing nuclear rotation, and subsequently rapidly extruding nuclear material (Fig. 1C; Supp. Video 2). This behavior is remarkably similar to that of vertebrate monocytes during the cytoskeletal rearrangements preceding extracellular DNA trap (ET) formation \(^{10,24}\) . This led us to speculate that some ctenophore immune cells were producing ETs in response to the presence of microbes.  
+
+<|ref|>text<|/ref|><|det|>[[42, 857, 953, 947]]<|/det|>
+We examined microbially- challenged Mnemiopsis cells using confocal microscopy and observed cells with decondensed DNA cast in large areas surrounding individual cell bodies. These networks of extracellular DNA were closely associated with individual E. coli (Fig. 1D). Three- dimensional rendering of confocal z- stacks revealed that E. coli bacteria were entangled in the extruded Mnemiopsis DNA (Fig. 1E;
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 45, 951, 202]]<|/det|>
+Supp. Video 3). To further characterize the extruded DNA using immunofluorescence, we stained cells with an antibody that recognizes an array of histone proteins (H1, H2A, H2B, H3, H4). We observed that Mnemiopsis immune cell ETs are composed of chromatin, typical of ETs described in other taxa (Fig. 1F). ETotic cells show diffuse staining over a large area, whereas non- ETotic cells display intact nuclei with stereotypical complements of concentrated DNA and histone labeling (Fig. 1D- F; 9). These data identify specific centophore cell types that produce bona fide extracellular chromatin traps when exposed to a microbial signature.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 234, 921, 296]]<|/det|>
+## Diverse microbial signatures induce ET formation in Mnemiopsis leidyi independently of Non-ETotic cell death  
+
+<|ref|>text<|/ref|><|det|>[[41, 310, 955, 515]]<|/det|>
+To simultaneously and accurately quantify ETosis and non- ETotic cell death we developed a semiautomated imaging pipeline using a combination of two DNA dyes: Hoechst, a membrane- permeable stain which labels all cell nuclei, and SytoxGreen, a non- permeable stain, which selectively labels nuclei of dying or dead cells with compromised cell membranes. Both dyes label extracellular DNA nets (Fig. 2). Importantly, the nuclear envelopes of necrotic and apoptotic cells remain relatively intact, displaying concentrated fluorescent labeling 25. Using our novel image analysis pipeline, we performed automated comparisons of thousands of images to accurately calculate percentages of live, dead, or ETotic cells. Critically, the development of this pipeline allowed us to accurately identify and discriminate between ETosis and non- ETotic cell death following treatments.  
+
+<|ref|>text<|/ref|><|det|>[[41, 532, 950, 785]]<|/det|>
+After exposing isolated Mnemiopsis cells to E. coli (Fig. 2A, B), we analyzed the stained nuclei in each image. Using image segmentation analysis, we defined individual cell masks based on Hoechst signal and measured relative fluorescence intensities of Hoechst and SytoxGreen inside each cell mask (Fig. 2C). Using those measurements, we defined 3 distinct cell populations: live cells, ETotic cells and non- ETotic cell death (Fig. 2D). Live cells are negative for SytoxGreen fluorescence with no dispersion of Hoechst signals (Hoechst<sup>high</sup>/SytoxGreen<sup>low</sup>) (Fig. 2B, 2D). Cells that have ETotic nuclei exhibit high dispersion of Hoechst fluorescence associated with extracellular DNA net formation (Hoechst<sup>low</sup>). In contrast, cells that have condensed Hoechst fluorescence and high SytoxGreen fluorescent signals (Hoechst<sup>high</sup>/SytoxGreen<sup>high</sup>) are dying from non- ETotic cell death processes. This approach allowed us to accurately identify and quantify large numbers of cells, including those undergoing ETosis after incubation with E. coli.  
+
+<|ref|>text<|/ref|><|det|>[[41, 803, 955, 960]]<|/det|>
+We then expanded our analyses of ET production to include responses to additional microbial stimuli ( https://github.com/carolinestefani/ETosis- and- death- automated- pipeline). We examined whether Mnemiopsis immune cells undergo ETosis when exposed to a diverse array of microbes, including the following: heat- killed gram- negative bacteria E. coli, heat- killed gram- positive bacteria Staphylococcus aureus, and cell wall extract from yeast Saccharomyces cerevisiae (zymosan). ETotic cells display diffuse Hoechst signal characteristic of filamentous DNA nets following exposure to E. coli, S. aureus, or zymosan (Fig. 3A). In contrast non- ETotic cells maintain intact nuclear material with concentrated
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 44, 930, 112]]<|/det|>
+Hoechst labeling. We analyzed both ETosis and total cell death in Mnemiopsis cells after four hours of microbial exposure (N = 24 individual animals). Contour plots representing Hoechst and SytoxGreen signal intensities show three distinct clusters: ETotic cells, dying cells, and living cells (Fig. 3B).  
+
+<|ref|>text<|/ref|><|det|>[[42, 128, 951, 285]]<|/det|>
+ETosis events significantly increased in Mnemiopsis cells when exposed to zymosan or E. coli in vitro, with mean increases in detected ETotic events of \(17\%\) and \(7\%\) over untreated cells (Fig. 3C). Incubation with S. aureus also elicited a strong ETotic response in Mnemiopsis cells, with a mean increase in detected ETotic events of \(69\%\) over untreated cells. Notably, non- ETotic cell death increased significantly after incubation with E. coli but did not change after exposure to zymosan or S. aureus (Fig. 3C). Our data demonstrate that a panel of microbes stimulates ETosis in ctenophore cells and that we can accurately and efficiently measure ETosis and non- ETotic cell death events simultaneously.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 315, 902, 377]]<|/det|>
+## Extracellular trap formation in Mnemiopsis is induced by classic pharmacological stimuli  
+
+<|ref|>text<|/ref|><|det|>[[42, 392, 953, 530]]<|/det|>
+ETosis can be induced in vertebrate granulocytes and leukocytes with pharmacological agents that activate production of ROS via discrete intracellular signaling pathways \(^{10,11,16}\) . Though ctenophore genomes appear to lack predicted gene homologs to many classic innate immunity pathway cell surface receptors and signaling intermediaries, many components of metazoan stress response pathways, as well as secondary messenger machinery, are present \(^{20,26,27}\) . We hypothesized that ETosis in Mnemiopsis immune cells could be stimulated by classic chemical agents broadly used in studies of vertebrate ETs.  
+
+<|ref|>text<|/ref|><|det|>[[42, 546, 951, 732]]<|/det|>
+We observed a significant induction of ETosis in Mnemiopsis after a four- hour exposure to PMA, with a mean increase of \(4.3\%\) over untreated control cells (Fig. 3D- F). Nigericin, a potassium ionophore, induces ET formation in vertebrates by initiating ROS release from mitochondria \(^{10,28}\) . Initiation of a calcium influx following stimulation by calcium ionophore A23817 (calcium) can also induce ET formation independent of NOX \(^{16}\) . We observed a significant induction of ETosis following incubation of Mnemiopsis cells with both the potassium ionophore nigericin and the calcium ionophore A23187, with a mean increase of \(33\%\) and \(20\%\) over untreated control cells, respectively (Fig. 3D- F). Nigericin exposure elicited the highest amount of ET formation (Fig. 3F; Supp. Video 4; Supp. Figure 2).  
+
+<|ref|>text<|/ref|><|det|>[[42, 748, 947, 882]]<|/det|>
+While we did not observe a significant change in levels of non- ETotic cell death following PMA stimulus, our results did indicate a significant increase in non- ETotic cell death following nigericin and A23187 treatments (Fig. 3F). Exposure of Mnemiopsis cells to all three classical pharmacological stimuli – PMA, the potassium ionophore nigericin, and the calcium ionophore A23187 – induced significant ET formation. These results implicate diverse signaling pathways involved in production of ETs that also retain deep evolutionary conservation across extant metazoan phyla.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 42, 925, 105]]<|/det|>
+## ETosis in the model bivalve, Crassostrea gigas, is induced by diverse microbial and chemical stimuli  
+
+<|ref|>text<|/ref|><|det|>[[41, 119, 955, 357]]<|/det|>
+We sought another non- vertebrate model to assess conservation of ET formation across phyla because ET stimulation in Mnemiopsis immune cells is intriguingly similar to ETosis in mammalian systems. The formation of invertebrate extracellular DNA traps has been characterized most extensively in molluscs \(^{3 - }\) \(^{5,13,29}\) . Bivalve molluscs, like the Pacific oyster Crassostrea gigas, have blood- like circulatory cells, collectively called hemocytes, that have immune functions \(^{30}\) . However, prior attempts to stimulate production of ETs in oyster via exposure to pathogen signatures and PMA has varied between studies \(^{5}\) . For example previous studies observed a robust ETotic response following challenge with Vibrio, a virulent bivalve pathogen, in oyster hemocytes \(^{3}\) , however the induction of ETosis with other microbes has been only modest or not observed in previous studies \(^{5}\) . Thus, conservation of ET induction pathways in bivalves remains unclear \(^{3 - 5}\) .  
+
+<|ref|>text<|/ref|><|det|>[[42, 375, 920, 488]]<|/det|>
+We measured ETs and cell death in Crassostrea hemocytes after exposure to zymosan, E. coli, and S. aureus (Fig. 4A- D). We found that ETosis is significantly stimulated in hemocytes exposed to all three pathogens (Fig. 4D; Supp. Figure 3), with mean increases of \(13\%\) , \(19\%\) , and \(22\%\) over untreated hemocytes, respectively. Hemocyte non- ETotic cell death (apoptosis, necrosis) increased significantly with E. coli exposure but not for zymosan challenge (Fig. 4D).  
+
+<|ref|>text<|/ref|><|det|>[[42, 504, 941, 662]]<|/det|>
+We also assayed Crassostrea hemocytes for ET production after exposure to pharmacological reagents that engage distinct signaling pathways in vertebrate monocytes. We analyzed the behavior of Crassostrea hemocytes after four- hour incubation with PMA, the potassium ionophore nigericin, and the calcium ionophore A23187 (Fig. 4E). Exposure to PMA and the calcium ionophore A23187 induced significant ET production in isolated Crassostrea hemocytes (Fig. 4F, 4G). However, the potassium ionophore nigericin showed no significant induction (Fig. 4F, 4G). In contrast, a large proportion of Crassostrea hemocytes produce ETs after 4 hours of exposure to A23187, similar to prior studies \(^{5}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 686, 191, 711]]<|/det|>
+## Discussion  
+
+<|ref|>text<|/ref|><|det|>[[41, 726, 952, 931]]<|/det|>
+Identifying the core molecular mechanisms of immune cells and their evolutionary origins is critical for understanding both the evolution of metazoan immunity function and the evolution of multicellularity. Initially described in 2004 as a unique antimicrobial behavior of vertebrate neutrophils, cells capable of extracellular trap production have recently subsequently been discovered in multiple non- vertebrate bilaterian taxa \(^{4,12,14}\) . Here we report that ctenophore immune cells are competent for ETosis in response to microbial challenge and assess potential mechanisms associated with ETosis pathway activation with well- characterized pharmacological reagents. ETosis in both the model ctenophore Mnemiopsis leidyi and the model oyster Crassostrea gigas is initiated after exposure to diverse microbial signatures as well as pharmaceutical compounds that are known to activate distinct signaling cascades in vertebrates.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[39, 44, 954, 343]]<|/det|>
+Many non- vertebrates lack gene homologs for major proteins known to be essential for ETosis in vertebrate neutrophils, such as PAD4, neutrophil elastase, and pannexin- 1. It is currently unknown what signaling molecules are necessary for ET formation in non- vertebrate taxa. While the involvement of NADPH per se has not been directly assessed in invertebrate ET production, there is strong evidence for the presence of multiple ETosis pathways that can be activated via stimulation with microbial signatures, parasites, PMA, A23187, and UV light \(^{3 - 5,12 - 14}\) . Prior studies in bivalves conflict as to whether ETosis can be stimulated with the NADPH- dependent pathway stimulus, PMA, or with microbes commonly used in immunological studies \(^{4,5}\) . Our results demonstrate that Crassostrea hemocytes produce ETs in response to stimulation by diverse PAMPs and distinct pharmacological reagents that, in vertebrate monocytes, engage both NADPH- dependent signaling (PMA stimulus) and NADPH- independent signaling (A23187 stimulus). Notably, Crassostrea hemocytes did not produce ETs following stimulation by the potassium ionophore nigericin, suggesting a potential loss of nigericin stimulating signal transduction in this mollusc species.  
+
+<|ref|>text<|/ref|><|det|>[[41, 359, 950, 565]]<|/det|>
+Prior to ET production in vitro, Mnemiopsis immune cells and Crassostrea hemocytes that underwent ETosis after exposure to E. coli did not seem to have phagocytosed bacteria in significant amounts. Intriguingly, we observed other motile cells that phagocytosed large amounts of bacteria without undergoing ETosis (Supp. Video 1, Supp. Video 5; Supp. Figure 3). Our functional characterization of ETosis competent cells in both taxa suggest that future studies should address whether the ETosis- competent and the highly phagocytic non- ETotic cells represent discrete immune cell types in these non- vertebrate taxa. In contrast to vertebrate immune cell nuclear morphologies that correspond with discrete immune cell functions \(^{31}\) , there do not seem to be notable differences in nuclear morphologies in either Crassostrea hemocytes or Mnemiopsis cells.  
+
+<|ref|>text<|/ref|><|det|>[[41, 580, 956, 809]]<|/det|>
+Ctenophores, which diverged very early from the metazoan stem lineage, have not had specific immune cell antimicrobial behaviors described beyond phagocytosis \(^{20,23}\) . We demonstrate that Mnemiopsis leidyi possesses cells functionally competent for ETosis in response to a range of pathogen challenges. The production of extracellular traps in Mnemiopsis suggests this immune cell type specific antimicrobial defense behavior was likely present very early in metazoan evolution (Fig. 5). Our data further demonstrates that both Mnemiopsis leidyi and Crassostrea gigas immune cells are capable of ETosis via stimulation with pharmaceutical agents that, in vertebrate immune cells, differentially induce either NADPH- dependent or NADPH- independent pathways to activate ET production. These data suggest that ETosis is an ancient and fundamental immune cell defense behavior that is not only present among distantly related metazoans but is also triggered by similar mechanisms.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 831, 196, 857]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[61, 872, 839, 917]]<|/det|>
+1. Brinkmann, V. et al. Neutrophil extracellular traps kill bacteria. Science 303, 1532–1535, doi:10.1126/science.1092385 (2004).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 44, 941, 92]]<|/det|>
+2. Wartha, F., Beiter, K., Normark, S. & Henriques-Normark, B. Neutrophil extracellular traps: casting the NET over pathogenesis. Curr Opin Microbiol 10, 52–56, doi:10.1016/j.mib.2006.12.005 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[57, 95, 901, 140]]<|/det|>
+3. Poirier, A. C. et al. Antimicrobial histones and DNA traps in invertebrate immunity: evidences in Crassostrea gigas. J Biol Chem 289, 24821–24831, doi:10.1074/jbc.M114.576546 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[57, 144, 937, 210]]<|/det|>
+4. Robb, C. T., Dyrrynda, E. A., Gray, R. D., Rossi, A. G. & Smith, V. J. Invertebrate extracellular phagocyte traps show that chromatin is an ancient defence weapon. Nat Commun 5, 4627, doi:10.1038/ncomms5627 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[57, 215, 928, 282]]<|/det|>
+5. Romero, A., Novoa, B. & Figueras, A. Extracellular traps (ETosis) can be activated through NADPH-dependent and -independent mechanisms in bivalve mollusks. Dev Comp Immunol 106, 103585, doi:10.1016/j.dci.2019.103585 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[57, 287, 954, 333]]<|/det|>
+6. Dunn, C. W. et al. Broad phylogenomic sampling improves resolution of the animal tree of life. Nature 452, 745–749, doi:10.1038/nature06614 (2008).  
+
+<|ref|>text<|/ref|><|det|>[[57, 337, 945, 382]]<|/det|>
+7. Li, Y., Shen, X. X., Evans, B., Dunn, C. W. & Rokas, A. Rooting the Animal Tree of Life. Mol Biol Evol 38, 4322–4333, doi:10.1093/molbev/msab170 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[57, 386, 916, 431]]<|/det|>
+8. Fuchs, T. A. et al. Novel cell death program leads to neutrophil extracellular traps. J Cell Biol 176, 231–241, doi:10.1083/jcb.200606027 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[57, 435, 933, 480]]<|/det|>
+9. Sollberger, G., Tilley, D. O. & Zychlinsky, A. Neutrophil Extracellular Traps: The Biology of Chromatin Externalization. Dev Cell 44, 542–553, doi:10.1016/j.devcel.2018.01.019 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 484, 910, 529]]<|/det|>
+10. Kenny, E. F. et al. Diverse stimuli engage different neutrophil extracellular trap pathways. Elife 6, doi:10.7554/eLife.24437 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[50, 533, 934, 600]]<|/det|>
+11. Auguste, M., Balbi, T., Ciacci, C. & Canesi, L. Conservation of Cell Communication Systems in Invertebrate Host-Defence Mechanisms: Possible Role in Immunity and Disease. Biology (Basel) 9, doi:10.3390/biology9080234 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[50, 604, 954, 671]]<|/det|>
+12. Homa, J., Ortmann, W. & Kolaczkowska, E. Conservative Mechanisms of Extracellular Trap Formation by Annelida Eisenia andrei: Serine Protease Activity Requirement. PLoS One 11, e0159031, doi:10.1371/journal.pone.0159031 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[50, 676, 925, 744]]<|/det|>
+13. Lange, M. K. et al. Gastropod-derived haemocyte extracellular traps entrap metastrongyloid larval stages of Angiostrongylus vasorum, Aelurostrongylus abstrusus and Troglostrongylus brevior. Parasit Vectors 10, 50, doi:10.1186/s13071-016-1961-z (2017).  
+
+<|ref|>text<|/ref|><|det|>[[50, 749, 928, 794]]<|/det|>
+14. Nascimento, M. T. C. et al. DNA extracellular traps are part of the immune repertoire of Periplaneta americana. Dev Comp Immunol 84, 62–70, doi:10.1016/j.dci.2018.01.012 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[50, 798, 920, 844]]<|/det|>
+15. Ramos-Martinez, E. et al. Multiple Origins of Extracellular DNA Traps. Front Immunol 12, 621311, doi:10.3389/fimmu.2021.621311 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[50, 848, 934, 916]]<|/det|>
+16. Douda, D. N., Khan, M. A., Grasemann, H. & Palaniyar, N. SK3 channel and mitochondrial ROS mediate NADPH oxidase-independent NETosis induced by calcium influx. Proc Natl Acad Sci U S A 112, 2817–2822, doi:10.1073/pnas.1414055112 (2015).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[48, 44, 940, 94]]<|/det|>
+17. Papayannopoulos, V. Neutrophil extracellular traps in immunity and disease. Nat Rev Immunol 18, 134-147, doi:10.1038/nri.2017.105 (2018).  
+
+<|ref|>text<|/ref|><|det|>[[48, 95, 936, 163]]<|/det|>
+18. Babonis, L. S., Ryan, J. F., Enjolras, C. & Martindale, M. Q. Genomic analysis of the tryptome reveals molecular mechanisms of gland cell evolution. Evodevo 10, 23, doi:10.1186/s13227-019-0138-1 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[48, 166, 844, 211]]<|/det|>
+19. Jokura, K. et al. CTENO64 Is Required for Coordinated Paddling of Ciliary Comb Plate in Ctenophores. Curr Biol 29, 3510-3516 e3514, doi:10.1016/j.cub.2019.08.059 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[48, 216, 900, 260]]<|/det|>
+20. Traylor-Knowles, N., Vandepas, L. E. & Browne, W. E. Still Enigmatic: Innate Immunity in the Ctenophore Mnemiopsis leidyi. Integr Comp Biol 59, 811-818, doi:10.1093/icb/icz116 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[48, 264, 940, 332]]<|/det|>
+21. Presnell, J. S. & Browne, W. E. Kruppel-like factor gene function in the ctenophore Mnemiopsis leidyi assessed by CRISPR/Cas9-mediated genome editing. Development 148, doi:10.1242/dev.199771 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[48, 336, 920, 381]]<|/det|>
+22. Bolte, S. et al. Specific immune priming in the invasive ctenophore Mnemiopsis leidyi. Biol Lett 9, 20130864, doi:10.1098/rsbl.2013.0864 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[48, 385, 945, 430]]<|/det|>
+23. Franc, J. M. La matrice mésogléenne des Ctenaires. Bulletin de la Société zoologique de France 113, 347-447 (1988).  
+
+<|ref|>text<|/ref|><|det|>[[48, 434, 840, 501]]<|/det|>
+24. Boe, D. M., Curtis, B. J., Chen, M. M., Ippolito, J. A. & Kovacs, E. J. Extracellular traps and macrophages: new roles for the versatile phagocyte. J Leukoc Biol 97, 1023-1035, doi:10.1189/jlb.4R1014-521R (2015).  
+
+<|ref|>text<|/ref|><|det|>[[48, 505, 950, 574]]<|/det|>
+25. Guimaraes-Costa, A. B., Nascimento, M. T., Wardini, A. B., Pinto-da-Silva, L. H. & Saraiva, E. M. ETosis: A Microbicidal Mechanism beyond Cell Death. J Parasitol Res 2012, 929743, doi:10.1155/2012/929743 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[48, 578, 935, 623]]<|/det|>
+26. Ryan, J. F. et al. The genome of the ctenophore Mnemiopsis leidyi and its implications for cell type evolution. Science 342, 1242592, doi:10.1126/science.1242592 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[48, 627, 925, 672]]<|/det|>
+27. Moroz, L. L. et al. The ctenophore genome and the evolutionary origins of neural systems. Nature 510, 109-114, doi:10.1038/nature13400 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[48, 676, 932, 721]]<|/det|>
+28. Meier, A. et al. Inhibition of Human Neutrophil Extracellular Trap (NET) Production by Propofol and Lipid Emulsion. Front Pharmacol 10, 323, doi:10.3389/fphar.2019.00323 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[48, 725, 910, 771]]<|/det|>
+29. Bachere, E. et al. The new insights into the oyster antimicrobial defense: Cellular, molecular and genetic view. Fish Shellfish Immunol 46, 50-64, doi:10.1016/j.fsi.2015.02.040 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[48, 775, 919, 820]]<|/det|>
+30. Jiang, S. et al. Functional characterisation of phagocytes in the Pacific oyster Crassostrea gigas. PeerJ 4, e2590, doi:10.7717/peerj.2590 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[48, 825, 888, 870]]<|/det|>
+31. Skinner, B. M. & Johnson, E. E. Nuclear morphologies: their diversity and functional relevance. Chromosoma 126, 195-212, doi:10.1007/s00412-016-0614-5 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[48, 874, 884, 941]]<|/det|>
+32. Presnell, J. S. et al. The Presence of a Functionally Tripartite Through-Gut in Ctenophora Has Implications for Metazoan Character Trait Evolution. Curr Biol 26, 2814-2820, doi:10.1016/j.cub.2016.08.019 (2016).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[47, 44, 944, 232]]<|/det|>
+33. Vandepas, L. E., Warren, K. J., Amemiya, C. T. & Browne, W. E. Establishing and maintaining primary cell cultures derived from the ctenophore Mnemiopsis leidyi. J Exp Biol 220, 1197–1201, doi:10.1242/jeb.152371 (2017).  
+34. Acharya, M. et al. alphav Integrins combine with LC3 and atg5 to regulate Toll-like receptor signalling in B cells. Nat Commun 7, 10917, doi:10.1038/ncomms10917 (2016).  
+35. Dieter, A. C., Vandepas, L. E. & Browne, W. E. Isolation and Maintenance of In Vitro Cell Cultures from the Ctenophore Mnemiopsis leidyi. Methods Mol Biol 2450, 347–358, doi:10.1007/978-1-0716-2172-1_18 (2022).  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 255, 163, 280]]<|/det|>
+## Methods  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 297, 227, 315]]<|/det|>
+## Animal maintenance  
+
+<|ref|>text<|/ref|><|det|>[[42, 335, 953, 450]]<|/det|>
+Laboratory cultures of Mnemiopsis leidyi were maintained as previously described \(^{32}\) . Adult Mnemiopsis cells were isolated following established protocols \(^{20,32,33}\) . Crassostrea gigas were maintained under flowing seawater at approximately \(13^{\circ}\mathrm{C}\) and hemolymph was extracted from the adductor muscle with a syringe. Cells from both taxa were maintained in vitro under sterile conditions in filtered seawater (FSW) + 1% penicillin/streptomycin.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 467, 348, 486]]<|/det|>
+## Stimulation of ETosis and imaging  
+
+<|ref|>text<|/ref|><|det|>[[42, 504, 952, 617]]<|/det|>
+For stimulation and quantification of ETosis, cells were isolated from 24 individual animals, plated in 96 well plates, and exposed to pHrodo- E. coli, Staphylococcus aureus, zymosan particles (Sigma Aldrich), 25 uM nigericin (Thermo Fisher Scientific), 1 mg/mL PMA (Sigma Aldrich), or 4 uM A23187 (Sigma). Each experimental condition for each animal was performed in triplicate. Live cell staining was performed following (2) and live imaging was performed using a JuLI Stage (NanoEntek).  
+
+<|ref|>text<|/ref|><|det|>[[41, 634, 956, 815]]<|/det|>
+For immunofluorescence, ctenophore cells were prepared following \(^{33- 35}\) and labeled with mouse anti- histone H11- 4 (EMD Millipore) and anti- mouse Alexa Fluor 488 (Thermo Fisher Scientific). The H11- 4 histone antibody was selected because it recognizes histones H1, H2A/B, H3, and H4 proteins across diverse species. Cells were imaged using \(\times 60\) objective and \(\times 100\) oil objective, on a Nikon Ti (Eclipse) inverted microscope with Ultraviolet Spinning Disc (CSU- X1) confocal scanner (Perkin Elmer). Images were captured with an Orca- ER Camera using Velocity (Quorum technologies). Post- acquisition analysis such as contrast adjustment, deconvolution through iterative restoration and colocalization were performed using Velocity software.  
+
+<|ref|>text<|/ref|><|det|>[[42, 832, 931, 899]]<|/det|>
+For quantification of ETosis and total cell death, cells were treated with Hoechst 33342 (Sigma Aldrich) and SytoxGreen (Invitrogen) for 20 min, before imaging at 20X on an automated imaging plate reader, Cytation 3 (Biotek, software Gen5 v4.2).  
+
+<|ref|>text<|/ref|><|det|>[[44, 916, 336, 935]]<|/det|>
+Automated image- based profiling
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[41, 45, 958, 224]]<|/det|>
+We analyzed over 28,000 total images from our experiments using CellProfiler (v4.1.3). Image quality was assessed by calculating a focus score using two classes Otsu thresholding method, weighted variance on 20x20 pixel measurements. We calculated and applied an illumination correction for each fluorescent channel (SytoxGreen and Hoechst) using a background illumination function of 50 pixels block size, without smoothing. Each corrected image was then segmented using a global robust background method (0.05- 50), with a smoothing scale of 1.3488 and a correction factor of 0.89. Clumped objects were identified and split by shape. For each segmented object we measured the number and intensity of pixels in each fluorescent channel.  
+
+<|ref|>text<|/ref|><|det|>[[41, 241, 958, 513]]<|/det|>
+Each image and each segmented object, along with Metadata, were exported as csv files by experiment. R (v4.0.5) software with tidyverse (v1.3.1), dplyr (v1.0.7) and readr (v1.4.0) packages were then used to transform the datasets. Data from images and objects were merged, and measurements from individual images with a Focus Score \(< 0.2\) were removed from further analysis. This allowed us to identify and select only images that were in focus. Surface area, Hoechst intensity and SytoxGreen intensity per object (nucleus) and per individual animal were then imported into FlowJo (v10.8.0), and percentages of cells per delineated population (dead/dying cell, live cell, and ETotic cell) were calculated. Dying cells were gated using Hoechst Intensity \(>0.35\) and SytoxGreen intensity \(>0.1\) ; live cells were gated using Hoechst Intensity \(>0.35\) and Sytox Green intensity \(< 0.1\) ; and ETotic cells were gated using Hoechst Intensity \(< 0.35\) . Finally, percentages per individual animal surveyed were combined and tested for statistical significance using GraphPad Prism (v9.2.0). All statistical tests were performed using two- tailed unpaired student t- test \(^{*}p< 0.05\) , \(^{**}p< 0.01\) , \(^{***}p< 0.001\) , \(^{****}p< 0.0001\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 536, 213, 561]]<|/det|>
+## Declarations  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 577, 216, 595]]<|/det|>
+## Acknowledgements  
+
+<|ref|>text<|/ref|><|det|>[[44, 614, 944, 702]]<|/det|>
+The authors are grateful to Anna Bruchez and Rachel Prins for imaging support. This research was supported by the National Oceanographic and Atmospheric Administration, a National Research Council Postdoctoral Fellowship to LEV, and the National Science Foundation under Grant No. 2013692. CS and ALH were supported by National Institutes of Health Grants R33AI119341 and R01GM102482.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 720, 223, 739]]<|/det|>
+## Author contributions  
+
+<|ref|>text<|/ref|><|det|>[[44, 757, 953, 844]]<|/det|>
+Conceptualization, C.S. and L.E.V.; Methodology, L.E.V. and C.S.; Formal Analysis, C.S.; Investigation, L.E.V. and C.S.; Writing – Original Draft, L.E.V. and C.S.; Writing – Review & Editing, C.S., A.L.H., W.E.B., N.T.K., F.W.G., and L.E.V.; Funding Acquisition, A.L.H., W.E.B., N.T.K., F.W.G., and L.EV.; Resources, A.L.H., W.E.B., and F.W.G.; Supervision, A.L.H., W.E.B., and F.W.G.  
+
+<|ref|>sub_title<|/ref|><|det|>[[45, 864, 220, 882]]<|/det|>
+## Competing interests  
+
+<|ref|>text<|/ref|><|det|>[[45, 902, 425, 920]]<|/det|>
+The authors declare no competing interests.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 44, 144, 69]]<|/det|>
+## Figures  
+
+<|ref|>image<|/ref|><|det|>[[40, 90, 770, 831]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 850, 115, 869]]<|/det|>
+<center>Figure 1 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 890, 933, 956]]<|/det|>
+Mnemiopsis immune cells produce extracellular DNA traps. (A) Adult lobate ctenophore Mnemiopsis leidyi. (B) Schematic of isolation of cells from whole Mnemiopsis leidyi. (C) Still images from Movie S1 showing a motile stellate cell retracting its processes, spinning, and extruding its nuclear contents after
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[40, 44, 944, 202]]<|/det|>
+exposure to E. coli. (D) Merged confocal image. Left – Nuclei of unstimulated Mnemiopsis cells. Right – A cell exposed to TxRed- E. coli has undergone ETosis; chromatin has been extruded from the cell in a large web- like pattern. E. coli are entrapped by the chromatin filaments (white arrowheads). (E) 3- dimensional image of Mnemiopsis extracellular DNA nets with E. coli entrapped. (F) Merged confocal image. Ctenophore extracellular traps are composed of DNA and histones. Histone 11- 4 antibody (green) and Hoechst (white) staining are visible in intact and ETosed Mnemiopsis cells treated with the K+ ionophore nigericin. White arrowhead marks DNA+histone nets. (C- E scale bar: \(10 \mu \mathrm{m}\) ).  
+
+<|ref|>image<|/ref|><|det|>[[42, 216, 952, 438]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[42, 464, 117, 483]]<|/det|>
+<center>Figure 2 </center>  
+
+<|ref|>text<|/ref|><|det|>[[40, 505, 944, 715]]<|/det|>
+Novel high- throughput imaging pipeline. Schematic representation of experimental and imaging workflow. (A) Treated and untreated cells in culture were labeled with vital dye Hoechst and membrane impermeable Sytox Green and assessed for ETosis and/or cell death. (B) Example of the segmentation output and masking for identifying individual intact or ETosed nuclei. (C) Individual cells were scored for viability, ETosis, or death using FlowJo software. Scoring is based on fluorescent intensity and nuclear material area. Cells that have undergone ETosis exhibit dispersion and decreased intensity of Hoechst fluorescence associated with extracellular DNA net formation (Hoechstlow). In contrast, dead and dying cells have condensed Hoechst fluorescence and high SytoxGreen fluorescent signals (Hoechsthigh/SytoxGreenhigh). High intensity blue fluorescence signals (Hoechsthigh) denote intact nuclei of live cells.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[40, 40, 592, 789]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 800, 117, 820]]<|/det|>
+<center>Figure 3 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 841, 951, 953]]<|/det|>
+Diverse pathogen signatures and classic pharmaceuticals stimulate ETosis in Mnemiopsis immune cells. (A) Representative images of ETotic Mnemiopsis cells exposed to fungal or bacterial pathogen signatures. (B) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following pathogen exposures. (C) ETosis is significantly stimulated in Mnemiopsis cells exposed to zymosan, gram- negative E. coli, and gram- positive S. aureus after 4 hours. Apoptosis
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[42, 45, 916, 201]]<|/det|>
+increases slightly with E. coli treatment but is not significant for other pathogen signatures. (D) Representative images of ETotic Mnemiopsis cells exposed to PMA, calcium ionophore A23187, or potassium ionophore nigericin. (E) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following treatment with established ETosis-inducing pharmaceuticals. (F) ETosis is significantly stimulated in hemocytes exposed to PMA, nigericin, and A23187. Non-ETosis cell death increases significantly after incubation with ionophores nigericin and A23187, but not PMA.  
+
+<|ref|>image<|/ref|><|det|>[[42, 210, 528, 959]]<|/det|>
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[44, 43, 117, 62]]<|/det|>
+## Figure 4  
+
+<|ref|>text<|/ref|><|det|>[[40, 82, 951, 378]]<|/det|>
+Some pathogen signatures and classic pharmaceuticals induce ETosis in Crassostrea hemocytes. (A) Schematic of hemocyte isolation from Pacific oyster Crassostrea gigas. (B) Representative images of ETotic C. gigas hemocytes exposed to fungal or bacterial pathogen signatures. PMA, calcium ionophore A23187, and potassium ionophore nigericin significantly induce ETosis in Crassostrea hemocytes. (C) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following pathogen exposures. (D) ETosis is significantly stimulated in Crassostrea cells exposed to zymosan, gram- negative E. coli, and gram- positive S. aureus after 4 hours. Apoptosis increases slightly with E. coli treatment and decreases following S. aureus incubation, but is not significant for zymosan. (E) Representative images of ETotic C. gigas hemocytes stimulated with PMA, A23187, or nigericin. (F) Representative FlowJo graph showing distributions of fluorescent signals from Hoechst and Sytox Green following treatment with established ETosis- inducing pharmaceuticals. (G) PMA and A23187, but not nigericin, significantly stimulates ETosis in hemocytes. Apoptosis of hemocytes decreases when treated with PMA or A23187.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[62, 45, 600, 780]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[44, 800, 118, 820]]<|/det|>
+<center>Figure 5 </center>  
+
+<|ref|>text<|/ref|><|det|>[[42, 842, 955, 909]]<|/det|>
+ETosis is an ancient metazoan immune response. Summary of ETosis phenomena across Metazoa. The presence of cells competent for ETosis in protostome, deuterostome, and non- bilaterian taxa indicate that production of extracellular DNA traps is an ancient animal immune defense.  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 932, 311, 959]]<|/det|>
+## Supplementary Files
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[43, 45, 765, 64]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 83, 504, 235]]<|/det|>
+- SupplementalFigures.docx- SuppVideo1MnemiopsisPhagocytosis.mov- SuppVideo2MnemiopsisETosisEcoli.avi- SuppVideo33DMnemiopsisETConfocalStack.avi- SuppVideo4MnemiopsisETosisNigericin.avi- SuppVideo5CrassostreaPhagocytosis.avi
+
+<--- Page Split --->

preprint/preprint__b46b8d2eb3030c20fb151d48a71b46752fa49c1c102bc842028043ad26768782/images_list.json

@@ -0,0 +1,47 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1. Surface state of Co(0001) sample at indicated temperatures in an atmosphere of CO:H2 1:2 at 1 bar studied by hard X-ray photoelectron spectroscopy (HAXPES) utilizing 4600eV photons at \\(0.3^{\\circ}\\) incidence. a shows Co \\(2\\mathrm{p}_{3 / 2}\\) core-level spectra b displays the C 1s region and c depicts the O 1s region. Subplots d, e, f show the same conditions as a, b, c but for a Co(1014) surface. The columns of XPS data have constant scaling of the vertical axis. Subplot g shows a representative figure of the high energy surface X-ray diffraction HESXRD data at 67.4keV of the Co(0001) crystal (full set is shown in SI). The detector is protected by beam stops at the bulk Bragg peak positions. h X-ray structure factor extracted from the 2D diffraction data shown in g at reaction conditions at 496K and 1 bar reaction mixture (1:2 CO:H2), data from the hcp part of the surface used for the fit (orange circles with vertical lines indicating an estimation 10% relative error), data from the fcc part (grey circles), fit result (solid line).",
+    "footnote": [],
+    "bbox": [
+      [
+        125,
+        88,
+        866,
+        617
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2. Isobar C 1s spectra for Co(0001) at a 500mbar and b 200mbar. In c we show 150mbar on Co(10\\bar{1}4). All mixtures are 1:2 CO:H2. Subplot d shows a comparison of 406 K data as function of pressure on the Co(0001) crystal, and subplot e displays a direct comparison of stepped Co(10\\bar{1}4) and flat Co(0001) crystals at 406 K. All y-axes have the same scaling. Data has been normalized according to SI Section S6c. The color code is the same as in Fig. 1 b/e.",
+    "footnote": [],
+    "bbox": [
+      [
+        210,
+        90,
+        785,
+        547
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Figure 3. Dynamic study of a CO addition and removal experiment. a 2D time-resolved spectrum of the C 1s region at 406 K (b for 506 K) total pressure under CO flow is 200mbar with a 1:2 CO:H₂ mixture. In the beginning and end the sample is subjected to a H₂ flow alone. The lines on the right are extracted from a and b at indicated times and resemble these reaction conditions from bottom to top: pure H₂, first 6 minutes under CO:H₂ mixture, last 6min under CO:H₂ mixture, only H₂ after removal of CO from the mixture (red) and at the end of the experiment after at least 15min in H₂(light blue).",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        140,
+        820,
+        655
+      ]
+    ],
+    "page_idx": 13
+  }
+]

preprint/preprint__b46b8d2eb3030c20fb151d48a71b46752fa49c1c102bc842028043ad26768782/preprint__b46b8d2eb3030c20fb151d48a71b46752fa49c1c102bc842028043ad26768782.mmd

@@ -0,0 +1,320 @@
+
+# Operando Probing of the Fischer-Tropsch Reaction on Co Single Crystal Surfaces up to 1 bar  
+
+Anders Nilsson andersn@fysik.su.se  
+
+Stockholm University https://orcid.org/0000- 0003- 1968- 8696  
+
+Patrick Lömker Stockholm University https://orcid.org/0000- 0002- 5297- 710X  
+
+David Degerman Stockholm University  
+
+Christopher Goodwin ALBA Synchrotron https://orcid.org/0000- 0002- 0062- 0643  
+
+Mikhail Shipilin Stockholm University  
+
+Peter Amann Stockholm University  
+
+Gabriel Rodrigues Stockholm University  
+
+Fernando Garcia Martinez Universidad del Pais Vasco  
+
+Raffael Rameshan Montan University Leoben  
+
+Jörgen Gladh Swedish Defence Research Agency https://orcid.org/0000- 0002- 5389- 5675  
+
+Hsin-Yi Wang Stockholm University  
+
+Alexander Holm Stockholm University https://orcid.org/0000- 0002- 3660- 4389  
+
+Steffen Tober Forschungszentrum Jülich https://orcid.org/0000- 0002- 6563- 6072  
+
+Jan-Christian Schober Deutsches Elektronen- Synchrotron  
+
+Leon Jacobse Deutsches Elektronen- Synchrotron  
+
+Markus Soldemo
+
+<--- Page Split --->
+
+Stockholm University  
+
+Vedran Vonk Centre for X- ray and Nano Science CXNS, Deutsches Elektronen- Synchrotron DESY https://orcid.org/0000- 0001- 9854- 1101  
+
+Robert Gleissner Deutsches Elektronen- Synchrotron  
+
+Heshmatt Noei Deutsches Elektronen- Synchrotron (DESY)  
+
+Zoltan Hegedüs Deutsches Elektronen- Synchrotron (DESY) https://orcid.org/0000- 0003- 2691- 8111  
+
+Andreas Stierle Deutsches Elektronen- Synchrotron (DESY) https://orcid.org/0000- 0002- 0303- 6282  
+
+Christoph Schlueter Photon Science, Deutsches Elektronen- Synchrotron DESY  
+
+## Article  
+
+Keywords: Fischer- Tropsch, Cobalt, Hydrogenation, Heterogeneous Catalysis  
+
+Posted Date: April 3rd, 2024  
+
+DOI: https://doi.org/10.21203/rs.3.rs- 3970719/v1  
+
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+Additional Declarations: There is NO Competing Interest.  
+
+Version of Record: A version of this preprint was published at Nature Communications on January 24th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 56082- 8.
+
+<--- Page Split --->
+
+# Operando Probing of the Fischer-Tropsch Reaction on Co Single Crystal Surfaces up to 1 bar  
+
+Patrick Lönker \(^{1,2}*\) , David Degerman \(^{1}\) , Christopher M. Goodwin \(^{1,7}\) , Mikhail Shipilin \(^{1}\) , Peter Amann \(^{1,\#}\) , Gabriel L.S. Rodrigues \(^{1}\) , Fernando Garcia- Martinez \(^{2}\) , Raffael Rameshan \(^{4}\) , Jörgen Gladh \(^{1,5}\) , Hsin- Yi Wang \(^{1}\) , Markus Soldemo \(^{1}\) , Alexander Holm \(^{1,6,7}\) , Steffen Tober \(^{8}\) , Jan- Christian Schober \(^{8}\) , Leon Jacobse \(^{8}\) , Vedran Vonk \(^{8}\) , Robert Gleissner \(^{8}\) , Heshmat Noei \(^{8}\) , Zoltan Hegedues \(^{2}\) , Andreas Stierle \(^{8,9}\) , Christoph Schlueter \(^{2}\) , Anders Nilsson \(^{1*}\)  
+
+\(^{1}\) Department of Physics, Stockholm University, 10691 Stockholm, Sweden \(^{2}\) Photon Science, Deutsches Elektronen- Synchrotron DESY, 22607 Hamburg, Germany \(^{3}\) ALBA Synchrotron Light Facility, Carrer de la Llum 2, 26, 08290 Cerdanyola del Vallés, Spain \(^{4}\) Lehrstuhl für Physikalische Chemie, Montanuniversität Leoben, 8700 Leoben, Austria \(^{5}\) PULSE Institute, SLAC National Accelerator Laboratory, Menlo Park, 94305 California, USA \(^{6}\) Department of Materials and Environmental Chemistry, Stockholm University, 106 91 Stockholm, Sweden. \(^{7}\) Laboratory of Organic Electronics, Department of Science and Technology (ITN), Linköping University, SE- 60174, Norrköping, Sweden \(^{8}\) Centre for X- Ray and Nanoscience CXNS, Deutsches Elektronen- Synchrotron DESY, 22607, Hamburg, Germany \(^{9}\) Physics Department, University of Hamburg, 20148 Hamburg, Germany \(^{\#}\) present address: Scienta Omicron, Limburgerstrasse 75, 65232 Taunusstein, Germany  
+
+Keywords: Fischer- Tropsch, Cobalt, Hydrogenation, Heterogeneous Catalysis.  
+
+## Abstract:  
+
+The surface chemistry of the Fischer- Tropsch catalytic reaction over Co has still several unknowns. Here, we report an operando X- ray photoelectron spectroscopy study of Co(0001) and Co(1014), and operando high energy surface X- ray diffraction of Co(0001), during the Fischer- Tropsch reaction at 0.15 bar - 1 bar and 406 K - 548 K in a \(\mathrm{H}_2 / \mathrm{CO}\) gas mixture. We find that the Co surfaces remain metallic under all conditions and that the coverage of chemisorbed species ranges from 0.4 - 1.7 monolayers depending on pressure and temperature. The adsorbates include CO on- top, C/CxHy and various longer hydrocarbon molecules, indicating a rate- limiting direct CO dissociation pathway and that only hydrocarbon species participate in the chain growth. The accumulation of hydrocarbon species points to the termination step being rate- limiting as well. Furthermore, we demonstrate that the intermediate surface species are highly dynamic, appearing and disappearing with time delays after rapid changes in the reactants' composition.
+
+<--- Page Split --->
+
+## I. Introduction  
+
+The Fischer- Tropsch (FT) reaction is an important industrial process, as it produces higher hydrocarbons from synthesis gas (syngas, \(\approx 1:2\) CO:H2 gas mixture) over Co, is an important industrial process'. The FT reaction was used during earlier time as a way to avoid oil embargos for some countries during world war II and the Apartheid regime in South Africa. In the current era it can become an important avenue for a sustainable chemical industry if CO is generated from \(\mathrm{CO_2}\) via the reverse water gas shift reaction where the \(\mathrm{CO_2}\) has been captured either directly from the atmosphere or at an intense carbon source. The hydrogen can be produced, not from the current steam reforming process of methane, but instead through electrolysis of water where the electricity is coming from a renewable source such as wind and solar. Presently, the Fischer- Tropsch reaction utilizes Fe, Ru or Co based catalysts' that yield different hydrocarbon distributions. The Co based FT reaction typically generates long chain hydrocarbons and waxes and operates at a temperature of 470- 510 K and pressures of a few tens of bars'.  
+
+The chemical state of the Co catalyst has previously been investigated through postreaction analysis of single crystal surfaces to be in a metallic state \(^{2 - 4}\) . However, bulk sensitive measurements under high temperatures and pressures during operando of the FT reaction of Co nanoparticles have shown the existence of small amounts of oxides \(^{5 - 10}\) . Furthermore, it has also been proposed that a partially oxidized Co catalyst can be responsible for a high activity \(^{11}\) . Although no operando measurements during the FT reaction has detected any major presence of a Co carbide bulk phase it has been demonstrated that \(\mathrm{CoC_2}\) nano prisms shows a high selectivity to olefin formation \(^{12}\) . Recent in- situ surface sensitive measurements of the FT reaction on Fe show a growing carbide phase starting immediately after the reaction is initiated \(^{13}\) and on Ni at low temperatures dissolution of carbon into the bulk as a dilute carbide phase has been observed \(^{14}\) . An open key question is if the state of the Co catalyst in the surface region remains fully in a metallic state or if surface oxide and near surface carbide can be present during the reaction conditions. Addressing this question necessitates detection using surface sensitive techniques performed while the reaction is turning over.  
+
+The reaction mechanism of the FT reaction consists of a sequence of elementary reaction steps \(^{15}\) . The first step after CO adsorption is the dissociation of CO generating carbon monomeric species. Afterwards such C can either attach to other carbon atoms and thus grow the hydrocarbon chain or attach to hydrogen atoms resulting in a weakening of the bond between the carbons and the surface, ultimately leading to desorption. The CO activation has
+
+<--- Page Split --->
+
+resulted in two major hypotheses based on theoretical calculations: there is either a direct dissociation, often denoted carbide mechanism<sup>16,17</sup>, or hydrogen assisted dissociation via the generation of a \(\mathrm{CH_xO}\) species<sup>18,19</sup>. It has been proposed that the hydrogenation of adsorbed \(\mathrm{C^{20}}\) and the termination step are partly rate limiting<sup>21</sup> as well as hydrogenation of atomic O and \(\mathrm{OH^{22,23}}\) . Here it would be essential to probe the adsorbates on the surface, to determine intermediates that accumulate as the reaction proceeds, as a pointer towards specific rate limiting steps.  
+
+All chemical sensitive studies over the FT reaction of Co under operando conditions have been conducted with methods mostly probing the bulk, such as X- ray absorption spectroscopy (XAS) and X- ray powder diffraction (XRD)<sup>5- 10</sup>. There have been efforts with Infrared Spectroscopy but here the observation is limited to exclusively probing adsorbed CO since no other species in the Co surface region could be detected<sup>24</sup>. Scanning tunneling microscopy (STM) have probed Co single crystal surfaces under FT at atmospheric conditions where the morphology of steps and terraces could be followed but without chemical sensitivity<sup>3,4</sup>. In one STM study conducted at 4 bar and 492 K on the Co(0001) surface stripes were observed during the FT reaction interpreted as the appearance of long chain hydrocarbon molecules<sup>25</sup>. A number of surface science studies of model molecules under vacuum have been conducted on Co single crystal surfaces<sup>22,26- 28</sup> but it is unclear if the model molecules are relevant for reactions occurring at many orders of magnitude higher pressures and temperatures.  
+
+X- ray photoelectron spectroscopy (XPS) is a unique surface sensitive method to investigate the chemical state of catalytic surfaces and adsorbed intermediates through core- level shifts. The high inelastic scattering cross- section of photoelectrons in the gas phase makes vacuum conditions necessary. Post analysis with XPS has been conducted of Co single crystal surfaces that have been in a reactor with atmospheric pressure<sup>3,4</sup> or 4 bar<sup>2</sup>, at temperatures where the reaction is turning- over, followed by evacuating the reactor to vacuum and then transferring the sample to the spectrometer chamber, where the measurement was conducted. Although adsorbed CO, adsorbed carbidic carbon and hydrocarbon species were observed it is unclear if species may decompose or desorb when the system is evacuated and the temperature reduced. Near- ambient XPS (NAPXPS) studies of Co foil have been restricted to 0.1 mbar<sup>23</sup> — far from the conditions of atmospheric pressure where the FT reaction occurs — have been detected to produce methane and some small hydrocarbons on Co single crystal surfaces<sup>3,4,29</sup>.  
+
+Here, we used an ambient- pressure XPS (APXPS) instrument called POLARIS operating at pressures up to 1 bar for \(\mathrm{CO / H_2}\) mixtures and as high temperatures as 506 K. The POLARIS instrument is based on the virtual pressure cell, where we create a \(\sim 30\) micron thick local high
+
+<--- Page Split --->
+
+pressure cushion and utilize grazing incidence of the incoming hard X- rays to provide surface sensitivity, despite high kinetic energy of the photoelectrons<sup>30</sup>. We have used flat Co(0001) and stepped Co(10¯14) single crystal substrates that have been shown previously to turn- over the FT reaction towards mainly methane but also minor fractions of C<sub>2+</sub> hydrocarbon species at close to 1 bar and 500 K<sup>4</sup>. Since the FT reaction is known to be structure sensitive<sup>31- 34</sup> (i.e. a Co stepped crystal, Co(10¯15) gives much higher turn- over than the terrace surface<sup>3</sup>) we have thus directly compared the Co(0001) with the Co(10¯14) surface to elucidate the influence of steps on the reaction. As FT reactions have been demonstrated at the same conditions as in the current study we will denote the experiments as operando. Furthermore, we show the facile appearance and disappearance of C<sub>x</sub>H<sub>y</sub> adsorbates as seen in section II.d as indicators of an operando state. Our complementary operando surface X- ray diffraction experiments yield atomic surface structure information under reaction conditions.  
+
+## II. Results and Discussion  
+
+a. Chemical and Structural State of Co Single Crystals and Adsorbates at 1bar  
+
+First, we address the chemical state of Co: whether it is metallic, oxidic or carbidic in the near surface region. This information would not necessary be observable in bulk sensitive measurements, as this active phase exists only close to the surface (i.e. the first few monolayers). Fig. 1a-c shows the operando Co 2p<sub>3/2</sub>, C 1s, and O 1s signal of the Co(0001) catalyst at a pressure of 1 bar with a reaction mixture of 1:2 CO:H<sub>2</sub> and a temperature of 406 K and 506 K (the higher corresponding to the typical FT high yield conditions) measured in the POLARIS instrument at a photon energy of 4600 eV (for samples and gases description turn to S1, for experimental setup description turn to S2). The Co 2p<sub>3/2</sub> spectrum is composed of a single peak at 778.0eV that shows a completely metallic state with no shoulder at 780.0 eV<sup>35</sup> that would indicate an oxide. A carbide modification of Co would be expected at somewhat higher binding energy compared to the Co<sup>0</sup> metal peak if the shift follows as seen in Ni 2p<sub>3/2</sub> upon carbon dissolution into bulk Ni<sup>14</sup> and the formation of Fe carbides<sup>13</sup>. The peaks in the C 1s region at 285.7 eV and in the O 1s region at 531.7 eV (blue) correspond to adsorbed CO in top position<sup>28</sup>. The C1s feature at 284.1 eV at 506 K is related to hydrocarbon species and exemplifies the reaction intermediates. Other states are only observable at 406 K which will be further
+
+<--- Page Split --->
+
+discussed below. A carbide species would be seen at around \(283.0 \mathrm{eV}\) and oxides at around \(529.3 \mathrm{eV}^{36}\) and none are detected in the spectra.  
+
+The stepped \(\mathrm{Co}(10\bar{1} 4)\) surface is probed at the same conditions as the \(\mathrm{Co}(0001)\) and the results there of are presented in Fig. 1d- f for the \(\mathrm{Co} 2\mathrm{p}_{3 / 2}\) , C 1s, and O 1s core- levels. Co is metallic during the reaction also on this facet. Adsorbates in the C 1s region, however show a striking difference. Peaks at \(284.1 \mathrm{eV}\) and \(284.7 \mathrm{eV}\) are now observed at \(406 \mathrm{K}\) and as well at \(506 \mathrm{K}\) . Also, there is intensity in the region of \(283.5 \mathrm{eV}\) for both temperatures. Overall the total C coverage is higher than for the flat surface. The O 1s intensity again shows only significant contributions of \(\mathrm{CO}_{\mathrm{top}}\) adsorbates at \(506 \mathrm{K}\) and additionally adsorbed \(\mathrm{H}_2\mathrm{O}\) at \(406 \mathrm{K}\) .
+
+<--- Page Split --->
+![](images/Figure_1.jpg)
+
+<center>Figure 1. Surface state of Co(0001) sample at indicated temperatures in an atmosphere of CO:H2 1:2 at 1 bar studied by hard X-ray photoelectron spectroscopy (HAXPES) utilizing 4600eV photons at \(0.3^{\circ}\) incidence. a shows Co \(2\mathrm{p}_{3 / 2}\) core-level spectra b displays the C 1s region and c depicts the O 1s region. Subplots d, e, f show the same conditions as a, b, c but for a Co(1014) surface. The columns of XPS data have constant scaling of the vertical axis. Subplot g shows a representative figure of the high energy surface X-ray diffraction HESXRD data at 67.4keV of the Co(0001) crystal (full set is shown in SI). The detector is protected by beam stops at the bulk Bragg peak positions. h X-ray structure factor extracted from the 2D diffraction data shown in g at reaction conditions at 496K and 1 bar reaction mixture (1:2 CO:H2), data from the hcp part of the surface used for the fit (orange circles with vertical lines indicating an estimation 10% relative error), data from the fcc part (grey circles), fit result (solid line). </center>  
+
+As XPS measurements are sensitive to the chemical state we have completed this data set with structure sensitive high energy surface X- ray diffractometry (HESXRD) on a Co(0001) single crystal at 200 mbar and 456 K with 1:2 CO:H2 under flow conditions (further conditions
+
+<--- Page Split --->
+
+are shown in S5). In Fig. 1g we display a maximum intensity per pixel from an angular scan rotating the sample around the surface normal in the range of the Co(0001) (1,0) crystal truncation rod. Full experimental details are given in the supporting information (S4). Our key observation is the appearance of a single surface rod at (1,0) reciprocal lattice units that indicates an unreconstructed hcp surface. A more detailed analysis shows that the behavior at partial pressures of 200mbar and 1 bar of \(\mathrm{CO:H_2}\) 1:2 mixtures the surfaces neither reconstruct (S5), which is in line with previous findings of operando STM observations<sup>4</sup>. No indication for the formation of ordered carbide formation is found. In Fig. 1h we present the X- ray structure factor extracted from the data in g. From the fit we can deduce, that the surface is atomically smooth under all gas mixtures studied with a slight inward relaxation of the topmost layer of \(\sim 0.04 \mathrm{\AA}\) (S2). The fit improves by including CO molecules on the surface, but due to the small data set available, the occupancies and position could not be further refined. The full data set gives also evidence, that a few percent of the surface is fcc (111) terminated. Due to the low number of fcc- terminated sites the contribution from fcc can therefore be neglected for the XPS data analysis.  
+
+We can thereby conclude that on a \(\mathrm{Co(0001)}\) single crystal at 1 bar and around \(500\mathrm{K}\) during the operation of the FT reaction the Co surface remains fully metallic and retains an ordered, flat surface which exhibits considerable crystal truncation rod signal. Furthermore, there are no signs of a surface carbide or surface oxide indicating that the C 1s and O 1s spectral intensities are related to chemisorbed species.  
+
+### b. Detailed C 1s Spectral Interpretation  
+
+When inspecting the XPS spectra from the adsorbate at different conditions we have curve fitted the data into specific components. Since many different conditions in terms of pressure and temperature are measured an assigned spectroscopic component should at least be clearly visible as a peak or strong shoulder in one spectrum. The chemical assignment is based either on experimental spectra obtained from model compounds in ultrahigh vacuum (UHV) on \(\mathrm{Co(0001)^{26,28,37,38}}\) or on density functional theory (DFT) binding energy calculations (S8). The binding energy scale potentially could differ by two tenths of an eV due to recoil effects at high kinetic energies that depends on the bonding strength (for discussion the reader is referred to S6.3), however, we estimate these effects to be negligible.
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+
+<center>Figure 2. Isobar C 1s spectra for Co(0001) at a 500mbar and b 200mbar. In c we show 150mbar on Co(10\bar{1}4). All mixtures are 1:2 CO:H2. Subplot d shows a comparison of 406 K data as function of pressure on the Co(0001) crystal, and subplot e displays a direct comparison of stepped Co(10\bar{1}4) and flat Co(0001) crystals at 406 K. All y-axes have the same scaling. Data has been normalized according to SI Section S6c. The color code is the same as in Fig. 1 b/e. </center>  
+
+Fig. 2a and 2b shows the C 1s spectra from the reaction of CO and \(\mathrm{H}_{2}\) with a mix ratio of 1:2 at a pressure of 500 mbar and 200 mbar, respectively, at temperatures in the range of 406 K to 523K over Co(0001). The corresponding O 1s spectra are shown in the supplementary material (S6.1). We assign the 283.2 eV (green) feature to chemisorbed C or CH on the surface based on XPS spectra obtained from either decomposition of ethylene on Co(0001) as observed at 283.2 eV \(^{37,38}\) or at 282.8 eV \(^{28}\) and seen in exposure of CO and \(\mathrm{H}_{2}\) at 4 bar followed by evacuation at 283.3 eV \(^{2}\) . The DFT calculation (S8) gives a binding energy of 283.1 eV for adsorbed C, (energy scale corrected against experimental value of adsorbed CO in on- top
+
+<--- Page Split --->
+
+position) whereas adsorbed CH has a somewhat lower value of \(283.0 \mathrm{eV}\) . With only such a small difference in C 1s binding energy between adsorbed C and CH and since there is a variation of the experimental value between \(282.8 - 283.3 \mathrm{eV}\) it is not possible to distinguish the two adsorbates we thus denote this peak C/CH at \(283.2 \mathrm{eV}\) (green).  
+
+The component observed at \(283.5 \mathrm{eV}\) (yellow) we assign to chemisorbed \(\mathrm{CH}_2\) species based on DFT calculations (S8). This binding energy has previously been reported as related to the \(\mathrm{CH}_3\) group in ethylidyne \(^{28}\) but spectra of the chemisorbed ethylidyne molecule also include a peak corresponding C to the group bonded to the surface at \(282.9 \mathrm{eV}\) . Since we observe the \(283.5 \mathrm{eV}\) feature at several conditions without any low binding energy component the \(283.5 \mathrm{eV}\) peak cannot be associated with ethylidyne. With a similar argument, the \(283.5 \mathrm{eV}\) feature cannot be one of the carbons in adsorbed ethylene, where the adsorption site generates two inequivalent C atoms, since then it should be accompanied by a second carbon peak at \(283.9 \mathrm{eV}^{28}\) . Next component, located at \(284.1 \mathrm{eV}\) (light red) we assign to hydrogen- saturated monomers of longer chain hydrocarbons, such as \(\mathrm{- CH}_2\) - and \(\mathrm{- CH}_3\) groups on the surface based on DFT calculations and previous post- analysis experiments \(^{2}\) . While some part of these hydrocarbon chains are in contact or in the proximity of the surface through undersaturated monomers, fully saturated parts are most likely pointing away from the surface and would correspond to the \(284.7 \mathrm{eV}\) (light blue) feature \(^{39}\) . The energy difference between the initial and final states in a photoionization event is much smaller when the hydrocarbon group is directly bonded to the surface allowing for metallic screening of the core- hole state resulting in a lower binding energy for the parts of the hydrocarbon in direct contact with the surface (light red peak) than for the parts pointing away from the surface (light blue) \(^{39}\) . Lastly, a clear peak originating from CO adsorbed in on- top configuration, denoted \(\mathrm{CO}_{\mathrm{top}}\) , is observed at \(285.7 \mathrm{eV}\) (dark blue). We find no indication of CO at other adsorption sites, such as the bridge or hollow sites, as commonly reported in UHV studies at liquid nitrogen temperatures \(^{22}\) or in AP- XPS studies at \(\sim 1000 \mathrm{x}\) lower pressures \(^{40}\) , yet we observe only \(\mathrm{CO}_{\mathrm{top}}\) in our experiments. The O 1s spectra contains mostly a feature associated with adsorbed CO in on- top position (S6.1). No clear indication of any significant amount adsorbed O, OH, CHO, COH or \(\mathrm{CH}_3\mathrm{O}\) species are detected.
+
+<--- Page Split --->
+
+Fig. 2a shows the C 1s spectra at a pressure of \(500\mathrm{mbar}\) from \(406\mathrm{K}\) to \(523\mathrm{K}\) on \(\mathrm{Co(0001)}\) . The coverage of the adsorbates has been determined through a specific normalization procedure (details: S6.3). We observe the largest total coverage of carbon containing species at the lowest temperature of \(406\mathrm{K}\) corresponding to \(1.5\mathrm{ML}\) with the \(\mathrm{- CH_2}\) - peak (yellow) clearly dominating the spectrum, but also intensity is observed in the region of non- screened hydrocarbon chains (light blue). A "monolayer" is here defined relative to the surface atoms of the Co substrate. What is clearly noted is that the total coverage decreases with increasing temperature to below monolayer coverage for \(\mathrm{T} > 480\mathrm{K}\) . We observe at \(406\mathrm{K}\) a large amount of the hydrocarbon species with C atoms both bonded to the surface and with \(\mathrm{CH_2}\) and \(\mathrm{CH_3}\) groups away from the surface as well as surface bound \(\mathrm{CH_2}\) groups. As we reach the highest temperature of \(523\mathrm{K}\) there is almost only CO on the surface and some small amount of adsorbed \(\mathrm{C / CH}\) . Chain growth requires a considerable coverage of carbon species which are not fully saturated by hydrogen, which on the (0001) surface occurs below \(485\mathrm{K}\) . This process can consequently occur at the lower temperature where there is a higher coverage of \(\mathrm{CH_2}\) groups and various adsorbed hydrocarbon species.  
+
+Fig. 2b shows the same trend of temperatures but with a total pressure of \(200\mathrm{mbar}\) on \(\mathrm{Co(0001)}\) . At the lowest temperature we have an almost similar total coverage of \(1.3\mathrm{ML}\) . We notice the same trend where the amount of species decreased with increasing temperature. What is mainly different is that the amount of \(\mathrm{CH_2}\) adsorbed species is now much higher in comparison to hydrocarbon molecules. The chain growth becomes less efficient with lower coverage. Again, the CO coverage \((\sim 1 / 3\mathrm{ML})\) is almost independent of temperature.  
+
+Fig. 2c shows the trend with the stepped \(\mathrm{Co(1014)}\) surface at total pressure of \(150\mathrm{mbar}\) . In general, we again observe an almost constant CO coverage but an increase in the hydrocarbon content and decrease of \(\mathrm{CH_2}\) adsorbed species indicating more efficient chain growth at steps compared to terraces. The total coverage at the lowest temperature of \(425\mathrm{K}\) is \(1.7\mathrm{ML}\) and compares well with our previous observations on the \(\mathrm{Co(0001)}\) surface.  
+
+Fig. 2d shows a comparison of spectra at different pressure on the \(\mathrm{Co(0001)}\) surface at the FT temperature of \(406\mathrm{K}\) . We observe an increase in the total coverage of adsorbed hydrocarbon species on the surface going to 1 bar, however, the relative distribution of different molecular fragments is somewhat similar at this low temperature. We can relate that the production of hydrocarbon at this temperature is limited by the desorption of products and only
+
+<--- Page Split --->
+
+becomes more efficient to a smaller degree with the increase in pressure since the surface is blocking its active sites due to kinetic hindrance.  
+
+Fig. 2e shows a comparison of the C 1s spectra at the temperature of \(506\mathrm{K}\) between the flat \(\mathrm{Co(0001)}\) and the stepped \(\mathrm{Co(10\bar{1}4)}\) surfaces. At this temperature the catalyst is expected to be active for the FT reaction. There is a striking increase in the hydrocarbon content with the presence of steps. It is interesting to note that also the production of methane and minor hydrocarbon species increased by almost an order of magnitude between flat and stepped Co surfaces in a recent STM reactor study. This increased reactivity was attributed to the lowering of the energy barriers for a rate limiting CO dissociation, and the increased hydrocarbon presence at all examined temperatures on the stepped crystal is fully consistent with this hypothesis.  
+
+Our data supports the view that CO dissociates most efficiently on the steps through a direct dissociation route and not the hydrogen assisted mechanism. The direct route hypothesis is strengthened by not observing any CHO species on the surface that would be visible at 285.0 eV and \(530.1\mathrm{eV}\) (S6). Although a weak component at the C 1s position could possibly be overlapping with adsorbed CO and hydrocarbon species, but there is no appreciable intensity at the low binding energy in the O 1s spectra. Furthermore, ultrafast measurements using X- ray lasers have demonstrated that the CHO species could only exist in an extremely short lived transient regime with a life time of only a few picoseconds and could never build up any appreciable coverage during steady- state reaction conditions. We do not observe any significant \(\mathrm{CH_2O}\) species with a calculated binding energy of \(286.5\mathrm{eV}\) and \(530.2\mathrm{eV}\) or \(\mathrm{CH_3O}\) species at \(286.5\mathrm{eV}\) and \(531.2\mathrm{eV}\) (S6) pointing to that non- dissociated CO does not significantly contribute to the chain growth. Finally, as the coverage of adsorbed hydrocarbon species with more than two attached hydrogens per carbon is quite high the hydrogenation termination step leading to desorption would also be rate limiting. We therefore predict that both CO dissociation and the final hydrogenation leading to desorption is rate limiting under the current conditions on the stepped surface. On the flat surface the different hydrocarbon hydrogenation steps seem to be limiting as well, indicating an overall less active surface.
+
+<--- Page Split --->
+
+d. Dynamics upon Changes in Reactant Composition  
+
+![](images/Figure_3.jpg)
+
+<center>Figure 3. Dynamic study of a CO addition and removal experiment. a 2D time-resolved spectrum of the C 1s region at 406 K (b for 506 K) total pressure under CO flow is 200mbar with a 1:2 CO:H₂ mixture. In the beginning and end the sample is subjected to a H₂ flow alone. The lines on the right are extracted from a and b at indicated times and resemble these reaction conditions from bottom to top: pure H₂, first 6 minutes under CO:H₂ mixture, last 6min under CO:H₂ mixture, only H₂ after removal of CO from the mixture (red) and at the end of the experiment after at least 15min in H₂(light blue). </center>  
+
+Fig. 3 depicts the time dependence of the C 1s spectra related to the FT reaction of the Co(0001) surface by applying and removing CO while keeping a constant H₂ flow on the sample at 200 mbar total pressure. We performed experiments at 406 K (panel a with line extracts shown in b) and at 506 K (panel c with line extracts shown in d). At 406 K the sample surface is covered with a tiny amount of species at 284.5eV initially. Upon exposure with CO
+
+<--- Page Split --->
+
+there immediately appears intensity in the \(\mathrm{CH}_2\) and \(\mathrm{CO_{top}}\) regions (283.5 and \(285.7\mathrm{eV}\) respectively) indicating that some CO is dissociated and hydrogenated. Over an interval of approximately \(30\mathrm{min}\) there is a continuous growth of the \(284.5\mathrm{eV}\) state indicating the appearance of longer chain hydrocarbons due to chain growth. Upon removal of the CO in the reaction mixture this component remains for a certain time while \(\mathrm{CO_{top}}\) and \(\mathrm{CH}_2\) are reacted away within the time resolution of this experiment. Continuing in this configuration, the \(284.5\mathrm{eV}\) hydrocarbon peak intensity reduces, indicating facile reaction and departure from the surface aided by the presence of \(\mathrm{H}_2\) , which supports our operando claim. We are here observing the rate limitation of the final hydrogenation step that removes the hydrocarbon species on the surface. A competing reaction to the hydrocarbon chain growth is the fast \(\mathrm{CH}_2\) hydrogenation into methane, which also explains the swift vanishing of the \(\mathrm{CH}_2\) contribution upon CO gas removal from reaction mixture<sup>26</sup>.  
+
+At \(506\mathrm{K}\) , we observe mainly low \(\mathrm{CO_{top}}\) surface coverage and only to a negligible degree \(\mathrm{C_xH_y}\) species as compared to the \(406\mathrm{K}\) experiment during reaction conditions, which is in line with the trends in the static measurements shown in Fig. 2. After CO is removed from the reaction mixture the intermediate hydrocarbons desorb or react away and the corresponding peak diminishes to baseline intensity. The dynamic response is much faster at the higher temperature. From these temperatures we derive that the surface is highly dynamic (i.e. turning over and in operando) and that changes in the conditions needs time to establish a steady state. Moreover, we observe that the rate limiting step changes from the formation of carbon chains at lower temperatures to the dissociation of CO at higher temperatures.  
+
+## III. Conclusion  
+
+We have studied the two Co single crystal surfaces of (0001) and \((10\bar{1} 4)\) using operando XPS at almost 100 times higher pressures than traditional NAPXPS and can directly probe adsorbates on the surface during the reaction. The C 1s and O 1s spectra shows only adsorbed species even at pressures close to 1 bar and the Co \(2\mathrm{p}_{3 / 2}\) spectra have no sign of an oxide or a carbide component. The surface X- ray diffraction results on Co(0001) demonstrate that the surface stays atomically smooth under reaction conditions. Thereby, there is no indication of any chemical or structural changes of the Co substrate surface region as the reaction proceeds. Our observations tip the scales in the discussion regarding the nature of the CO dissociation towards the direct or often denoted carbide mechanism since no sign of hydrogen assisted
+
+<--- Page Split --->
+
+dissociation in terms of CHO detected species in the C or O 1s spectra. Furthermore, the chain growth involves only hydrocarbon species, since undissociated CO participation should show up as detected \(\mathrm{CH}_2\mathrm{O}\) spectral components, which we did not observe. There are also no ethylidene or adsorbed ethylene intermediates detected pointing to simple chain growth of - \(\mathrm{CH}_x\) - species resulting in an increasing amount of hydrocarbon species with groups both bonded directly to the surface but also pointing away towards the gas phase. The increasing appearance of hydrocarbon species at \(406\mathrm{K}\) as well as on the stepped surface where CO dissociation is more facile shows that the final termination step in terms of hydrogenation is also rate- limiting. Finally, our observation of the Co based Fischer- Tropsch reaction is highly dynamic meaning that the involved species (despite a potentially long residence time) show changing adsorbate compositions as a direct consequence of changes in the reactant mixtures. The time for the delay is strongly temperature dependent and are on the tens of minutes scale.
+
+<--- Page Split --->
+
+## ASSOCIATED CONTENT Supporting Information.  
+
+PDF file contains details of (i) sample preparation and assessment, (ii) the beam damage assessment, (iii) sample heating and surface temperature measurements, (iv) XPS data corrections. It also contains more data sets for the readers' reference.  
+
+## AUTHOR INFORMATION  
+
+## Corresponding Authors  
+
+\*E- mail for P.L.: patrick.loemker@fysik.su.se  \*E- Mail for A.N.: andersn@fysik.su.se  
+
+## Author Contributions  
+
+P.L. with input from A.N. planned the experiments at Petra III. P.L., D.D., M.S., C.M.G., J.G., H.- Y.W., A.H., R.R. A.S., C.S., A.N., P.A. participated in the XPS experimental work, while Z.H., A.S., H.N., V.V., J.- C.S., R.G., S.T. and L.J. participated in the SXRD measurements. P.L. extracted and plotted the SXRD data and A.S. fitted it. G.L.S.R. performed theoretical calculations. P.L. and M.S. developed the data analysis software and P.L. did the data analysis. P.L. and AN wrote the manuscript. All authors contributed to the literature research, result discussion and manuscript improvement.  
+
+## Notes  
+
+The authors declare no competing financial interests.  
+
+## ACKNOWLEDGEMENTS  
+
+The work was supported by the Swedish Research Council (Vetenskapsrådet, VR, project 2017- 00559 and project 2013- 8823), the Knut & Alice Wallenberg (KAW, grant nr. 2016.0042) foundation as well as the Swedish Foundation for strategic research (Stiftelsen för Strategisk Forskning, SSF, Proj. ITM 17- 0034). The research leading to this result has also
+
+<--- Page Split --->
+
+been supported by the project CALIPSO plus under Grant Agreement 730872 from the EU Framework Program for Research and Innovation HORIZON 2020. The experimental part of this research was carried out at P22 and P21 beamlines at DESY, a member of the Helmholtz Association (HGF). Beamtime was given for in house research proposals. The DFT calculations were performed using resources provided by the Swedish National Infrastructure for Computing (SNIC) at the HPC2N center. The authors would like to acknowledge the help of the P22 beamline engineer Katrin Ederer, and the Technical Division at Stockholm University.
+
+<--- Page Split --->
+
+## References  
+
+1. Dry, M. E. The Fischer–Tropsch process: 1950–2000. Catal. Today 71, 227–241 (2002).  
+
+2. Jin, Y. et al. Elementary Surface Reactions on Co(0001) under Fischer–Tropsch Synthesis Conditions. J. Phys. Chem. C 121, 21535–21540 (2017).  
+
+3. Golder, K. M. & Wintterlin, J. In Situ/Operando STM of the Fischer–Tropsch Synthesis on a Co(10f15) Surface—A Study to Bridge the Materials Gap between Single-Crystal Models and Supported Catalysts. ACS Catal. 12, 7199–7209 (2022).  
+
+4. Boller, B., Durner, K. M. & Wintterlin, J. The active sites of a working Fischer–Tropsch catalyst revealed by operando scanning tunnelling microscopy. Nat. Catal. 2, 1027 (2019).  
+
+5. Ernst, B., Bensaddik, A., Hilaire, L., Chaumette, P. & Kiennemann, A. Study on a cobalt silica catalyst during reduction and Fischer–Tropsch reaction: In situ EXAFS compared to XPS and XRD. Catal. Today 39, 329–341 (1998).  
+
+6. Cats, K. H. et al. X-ray nanoscopy of cobalt Fischer–Tropsch catalysts at work. Chem. Commun. 49, 4622–4624 (2013).  
+
+7. Price, S. W. T. et al. Chemical imaging of Fischer-Tropsch catalysts under operating conditions. Sci. Adv. 3, e1602838 (2017).  
+
+8. Ronning, M. et al. Combined XRD and XANES studies of a Re-promoted Co/γ-Al2O3 catalyst at Fischer–Tropsch synthesis conditions. Catal. Today 155, 289–295 (2010).  
+
+9. Rochet, A. et al. In situ and operando structural characterisation of a Fischer–Tropsch supported cobalt catalyst. Catal. Today 171, 186–191 (2011).  
+
+10. Huffman, G. P. et al. In-Situ XAFS Investigation of K-Promoted Co Catalysts. J. Catal. 151, 17–25 (1995).
+
+<--- Page Split --->
+
+11. Melaet, G. et al. Evidence of Highly Active Cobalt Oxide Catalyst for the Fischer–Tropsch Synthesis and CO₂ Hydrogenation. J. Am. Chem. Soc. 136, 2260–2263 (2014).  
+
+12. Zhong, L. et al. Cobalt carbide nanoprisms for direct production of lower olefins from syngas. Nature 538, 84–87 (2016).  
+
+13. Shipilin, M. et al. In Situ Surface-Sensitive Investigation of Multiple Carbon Phases on Fe(110) in the Fischer–Tropsch Synthesis. ACS Catal. 12, 7609–7621 (2022).  
+
+14. Degerman, D. et al. State of the Surface During CO Hydrogenation over Ni(111) and Ni(211) Probed by Operando X-ray Photoelectron Spectroscopy. J. Phys. Chem. C acs.jpcc.2c07650 (2023) doi:10.1021/acs.jpcc.2c07650.  
+
+15. Dry, M. E. The Fischer-Tropsch (FT) Synthesis Processes. in Handbook of Heterogeneous Catalysis (eds. Ertl, G., Knözinger, H., Schüth, F. & Weitkamp, J.) hetcat0150 (Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, Germany, 2008). doi:10.1002/9783527610044. hetcat0150.  
+
+16. Shetty, S. & Santen, R. A. van. Hydrogen induced CO activation on open Ru and Co surfaces. Phys. Chem. Chem. Phys. 12, 6330–6332 (2010).  
+
+17. Petersen, M. A., van den Berg, J.-A., Ciobica, I. M. & van Helden, P. Revisiting CO Activation on Co Catalysts: Impact of Step and Kink Sites from DFT. ACS Catal. 7, 1984–1992 (2017).  
+
+18. Storsæter, S., Chen, D. & Holmen, A. Microkinetic modelling of the formation of C1 and C2 products in the Fischer–Tropsch synthesis over cobalt catalysts. Surf. Sci. 600, 2051–2063 (2006).  
+
+19. Inderwildi, O. R., Jenkins, S. J. & King, D. A. Fischer–Tropsch Mechanism Revisited: Alternative Pathways for the Production of Higher Hydrocarbons from Synthesis Gas. J. Phys. Chem. C 112, 1305–1307 (2008).
+
+<--- Page Split --->
+
+20. Banerjee, A., van Bavel, A. P., Kuipers, H. P. C. E. & Saeys, M. CO Activation on Realistic Cobalt Surfaces: Kinetic Role of Hydrogen. ACS Catal. 7, 5289–5293 (2017).  
+
+21. Yang, J., Tveten, E. Z., Chen, D. & Holmen, A. Understanding the Effect of Cobalt Particle Size on Fischer–Tropsch Synthesis: Surface Species and Mechanistic Studies by SSITKA and Kinetic Isotope Effect. Langmuir 26, 16558–16567 (2010).  
+
+22. Weststrate, C. J., van de Loosdrecht, J. & Niemantsverdriet, J. W. Spectroscopic insights into cobalt-catalyzed Fischer-Tropsch synthesis: A review of the carbon monoxide interaction with single crystalline surfaces of cobalt. J. Catal. 342, 1–16 (2016).  
+
+23. Wu, L., Liu, H., Reese, J. M. & Zhang, Y. Non-equilibrium dynamics of dense gas under tight confinement. J. Fluid Mech. 794, 252–266 (2016).  
+
+24. Oosterbeek, H. Bridging the pressure and material gap in heterogeneous catalysis: cobalt Fischer–Tropsch catalysts from surface science to industrial application. Phys. Chem. Chem. Phys. 9, 3570–3576 (2007).  
+
+25. Navarro, V., van Spronsen, M. A. & Frenken, J. W. M. In situ observation of self-assembled hydrocarbon Fischer–Tropsch products on a cobalt catalyst. Nat. Chem. 8, 929–934 (2016).  
+
+26. Weststrate, C. J., van Helden, P. & Niemantsverdriet, J. W. Reflections on the Fischer-Tropsch synthesis: Mechanistic issues from a surface science perspective. Catal. Today 275, 100–110 (2016).  
+
+27. Weststrate, C. J. & Niemantsverdriet, J. W. Understanding FTS selectivity: the crucial role of surface hydrogen. Faraday Discuss. 197, 101–116 (2017).
+
+<--- Page Split --->
+
+28. Weststrate, C. J. (Kees-J. et al. Mechanistic insight into carbon-carbon bond formation on cobalt under simulated Fischer-Tropsch synthesis conditions. Nat. Commun. 11, 750 (2020).  
+
+29. Geerlings, I. J. C., Zonneyville, M. C. & De Groot, C. P. M. The Fischer-Tropsch reaction on a cobalt (0001) single crystal. Catal. Lett. 5, 309-314 (1990).  
+
+30. Amann, P. et al. A high-pressure x-ray photoelectron spectroscopy instrument for studies of industrially relevant catalytic reactions at pressures of several bars. Rev. Sci. Instrum. 90, 103102 (2019).  
+
+31. Bezemer, G. L. et al. Cobalt Particle Size Effects in the Fischer-Tropsch Reaction Studied with Carbon Nanofiber Supported Catalysts. J. Am. Chem. Soc. 128, 3956-3964 (2006).  
+
+32. Prieto, G., Martínez, A., Concepción, P. & Moreno-Tost, R. Cobalt particle size effects in Fischer-Tropsch synthesis: structural and in situ spectroscopic characterisation on reverse micelle-synthesised Co/ITQ-2 model catalysts. J. Catal. 266, 129-144 (2009).  
+
+33. den Breejen, J. P. et al. On the Origin of the Cobalt Particle Size Effects in Fischer-Tropsch Catalysis. J. Am. Chem. Soc. 131, 7197-7203 (2009).  
+
+34. Xiong, H., Motchelaho, M. A. M., Moyo, M., Jewell, L. L. & Coville, N. J. Correlating the preparation and performance of cobalt catalysts supported on carbon nanotubes and carbon spheres in the Fischer-Tropsch synthesis. J. Catal. 278, 26-40 (2011).  
+
+35. Biesinger, M. C. et al. Resolving surface chemical states in XPS analysis of first row transition metals, oxides and hydroxides: Cr, Mn, Fe, Co and Ni. Appl. Surf. Sci. 257, 2717-2730 (2011).
+
+<--- Page Split --->
+
+36. Weststrate, C. J., van Helden, P., van de Loosdrecht, J. & Niemantsverdriet, J. W. Elementary steps in Fischer–Tropsch synthesis: CO bond scission, CO oxidation and surface carbiding on Co(0001). Surf. Sci. 648, 60–66 (2016).  
+
+37. Xu, L. et al. A Photoemission Study of Ethylene Decomposition on a Co(0001) Surface: Formation of Different Types of Carbon Species. J. Phys. Chem. C 116, 4167–4174 (2012).  
+
+38. Xu, L. et al. Transformation of Carbon Monomers and Dimers to Graphene Islands on Co(0001): Thermodynamics and Kinetics. J. Phys. Chem. C 117, 2952–2958 (2013).  
+
+39. Weiss, K. et al. XPS and XAS investigation of condensed and adsorbed n-octane on a Cu(110) surface. J. Electron Spectrosc. Relat. Phenom. 128, 179–191 (2003).  
+
+40. Chai, P. et al. Near Ambient-Pressure X-ray Photoelectron Spectroscopy Study of CO Activation and Hydrogenation on Co(0001). J. Phys. Chem. C 125, 22223–22230 (2021).  
+
+41. LaRue, J. et al. Real-Time Elucidation of Catalytic Pathways in CO Hydrogenation on Ru. J. Phys. Chem. Lett. 8, 3820–3825 (2017).
+
+<--- Page Split --->
+
+## Supplementary Files  
+
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+- CoFTsynthesisSupplementaryMaterials.docx
+
+<--- Page Split --->

preprint/preprint__b46b8d2eb3030c20fb151d48a71b46752fa49c1c102bc842028043ad26768782/preprint__b46b8d2eb3030c20fb151d48a71b46752fa49c1c102bc842028043ad26768782_det.mmd

@@ -0,0 +1,429 @@
+<|ref|>title<|/ref|><|det|>[[43, 106, 925, 175]]<|/det|>
+# Operando Probing of the Fischer-Tropsch Reaction on Co Single Crystal Surfaces up to 1 bar  
+
+<|ref|>text<|/ref|><|det|>[[44, 195, 220, 240]]<|/det|>
+Anders Nilsson andersn@fysik.su.se  
+
+<|ref|>text<|/ref|><|det|>[[45, 268, 601, 288]]<|/det|>
+Stockholm University https://orcid.org/0000- 0003- 1968- 8696  
+
+<|ref|>text<|/ref|><|det|>[[44, 294, 601, 335]]<|/det|>
+Patrick Lömker Stockholm University https://orcid.org/0000- 0002- 5297- 710X  
+
+<|ref|>text<|/ref|><|det|>[[44, 340, 241, 381]]<|/det|>
+David Degerman Stockholm University  
+
+<|ref|>text<|/ref|><|det|>[[44, 387, 576, 428]]<|/det|>
+Christopher Goodwin ALBA Synchrotron https://orcid.org/0000- 0002- 0062- 0643  
+
+<|ref|>text<|/ref|><|det|>[[44, 433, 241, 474]]<|/det|>
+Mikhail Shipilin Stockholm University  
+
+<|ref|>text<|/ref|><|det|>[[44, 480, 241, 520]]<|/det|>
+Peter Amann Stockholm University  
+
+<|ref|>text<|/ref|><|det|>[[44, 526, 241, 566]]<|/det|>
+Gabriel Rodrigues Stockholm University  
+
+<|ref|>text<|/ref|><|det|>[[44, 572, 293, 612]]<|/det|>
+Fernando Garcia Martinez Universidad del Pais Vasco  
+
+<|ref|>text<|/ref|><|det|>[[44, 618, 279, 658]]<|/det|>
+Raffael Rameshan Montan University Leoben  
+
+<|ref|>text<|/ref|><|det|>[[44, 664, 725, 705]]<|/det|>
+Jörgen Gladh Swedish Defence Research Agency https://orcid.org/0000- 0002- 5389- 5675  
+
+<|ref|>text<|/ref|><|det|>[[44, 710, 241, 750]]<|/det|>
+Hsin-Yi Wang Stockholm University  
+
+<|ref|>text<|/ref|><|det|>[[44, 756, 601, 797]]<|/det|>
+Alexander Holm Stockholm University https://orcid.org/0000- 0002- 3660- 4389  
+
+<|ref|>text<|/ref|><|det|>[[44, 803, 650, 844]]<|/det|>
+Steffen Tober Forschungszentrum Jülich https://orcid.org/0000- 0002- 6563- 6072  
+
+<|ref|>text<|/ref|><|det|>[[44, 849, 359, 890]]<|/det|>
+Jan-Christian Schober Deutsches Elektronen- Synchrotron  
+
+<|ref|>text<|/ref|><|det|>[[44, 896, 359, 936]]<|/det|>
+Leon Jacobse Deutsches Elektronen- Synchrotron  
+
+<|ref|>text<|/ref|><|det|>[[44, 942, 194, 959]]<|/det|>
+Markus Soldemo
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 46, 241, 64]]<|/det|>
+Stockholm University  
+
+<|ref|>text<|/ref|><|det|>[[44, 70, 773, 135]]<|/det|>
+Vedran Vonk Centre for X- ray and Nano Science CXNS, Deutsches Elektronen- Synchrotron DESY https://orcid.org/0000- 0001- 9854- 1101  
+
+<|ref|>text<|/ref|><|det|>[[44, 140, 360, 182]]<|/det|>
+Robert Gleissner Deutsches Elektronen- Synchrotron  
+
+<|ref|>text<|/ref|><|det|>[[44, 187, 425, 228]]<|/det|>
+Heshmatt Noei Deutsches Elektronen- Synchrotron (DESY)  
+
+<|ref|>text<|/ref|><|det|>[[44, 233, 781, 276]]<|/det|>
+Zoltan Hegedüs Deutsches Elektronen- Synchrotron (DESY) https://orcid.org/0000- 0003- 2691- 8111  
+
+<|ref|>text<|/ref|><|det|>[[44, 280, 781, 323]]<|/det|>
+Andreas Stierle Deutsches Elektronen- Synchrotron (DESY) https://orcid.org/0000- 0002- 0303- 6282  
+
+<|ref|>text<|/ref|><|det|>[[44, 327, 545, 368]]<|/det|>
+Christoph Schlueter Photon Science, Deutsches Elektronen- Synchrotron DESY  
+
+<|ref|>sub_title<|/ref|><|det|>[[44, 406, 103, 425]]<|/det|>
+## Article  
+
+<|ref|>text<|/ref|><|det|>[[44, 444, 709, 465]]<|/det|>
+Keywords: Fischer- Tropsch, Cobalt, Hydrogenation, Heterogeneous Catalysis  
+
+<|ref|>text<|/ref|><|det|>[[44, 483, 289, 502]]<|/det|>
+Posted Date: April 3rd, 2024  
+
+<|ref|>text<|/ref|><|det|>[[44, 521, 474, 541]]<|/det|>
+DOI: https://doi.org/10.21203/rs.3.rs- 3970719/v1  
+
+<|ref|>text<|/ref|><|det|>[[42, 558, 914, 601]]<|/det|>
+License: © © This work is licensed under a Creative Commons Attribution 4.0 International License. Read Full License  
+
+<|ref|>text<|/ref|><|det|>[[42, 618, 533, 639]]<|/det|>
+Additional Declarations: There is NO Competing Interest.  
+
+<|ref|>text<|/ref|><|det|>[[42, 674, 944, 718]]<|/det|>
+Version of Record: A version of this preprint was published at Nature Communications on January 24th, 2025. See the published version at https://doi.org/10.1038/s41467- 025- 56082- 8.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[184, 84, 815, 141]]<|/det|>
+# Operando Probing of the Fischer-Tropsch Reaction on Co Single Crystal Surfaces up to 1 bar  
+
+<|ref|>text<|/ref|><|det|>[[117, 149, 875, 262]]<|/det|>
+Patrick Lönker \(^{1,2}*\) , David Degerman \(^{1}\) , Christopher M. Goodwin \(^{1,7}\) , Mikhail Shipilin \(^{1}\) , Peter Amann \(^{1,\#}\) , Gabriel L.S. Rodrigues \(^{1}\) , Fernando Garcia- Martinez \(^{2}\) , Raffael Rameshan \(^{4}\) , Jörgen Gladh \(^{1,5}\) , Hsin- Yi Wang \(^{1}\) , Markus Soldemo \(^{1}\) , Alexander Holm \(^{1,6,7}\) , Steffen Tober \(^{8}\) , Jan- Christian Schober \(^{8}\) , Leon Jacobse \(^{8}\) , Vedran Vonk \(^{8}\) , Robert Gleissner \(^{8}\) , Heshmat Noei \(^{8}\) , Zoltan Hegedues \(^{2}\) , Andreas Stierle \(^{8,9}\) , Christoph Schlueter \(^{2}\) , Anders Nilsson \(^{1*}\)  
+
+<|ref|>text<|/ref|><|det|>[[115, 270, 869, 515]]<|/det|>
+\(^{1}\) Department of Physics, Stockholm University, 10691 Stockholm, Sweden \(^{2}\) Photon Science, Deutsches Elektronen- Synchrotron DESY, 22607 Hamburg, Germany \(^{3}\) ALBA Synchrotron Light Facility, Carrer de la Llum 2, 26, 08290 Cerdanyola del Vallés, Spain \(^{4}\) Lehrstuhl für Physikalische Chemie, Montanuniversität Leoben, 8700 Leoben, Austria \(^{5}\) PULSE Institute, SLAC National Accelerator Laboratory, Menlo Park, 94305 California, USA \(^{6}\) Department of Materials and Environmental Chemistry, Stockholm University, 106 91 Stockholm, Sweden. \(^{7}\) Laboratory of Organic Electronics, Department of Science and Technology (ITN), Linköping University, SE- 60174, Norrköping, Sweden \(^{8}\) Centre for X- Ray and Nanoscience CXNS, Deutsches Elektronen- Synchrotron DESY, 22607, Hamburg, Germany \(^{9}\) Physics Department, University of Hamburg, 20148 Hamburg, Germany \(^{\#}\) present address: Scienta Omicron, Limburgerstrasse 75, 65232 Taunusstein, Germany  
+
+<|ref|>text<|/ref|><|det|>[[118, 544, 654, 560]]<|/det|>
+Keywords: Fischer- Tropsch, Cobalt, Hydrogenation, Heterogeneous Catalysis.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 573, 202, 588]]<|/det|>
+## Abstract:  
+
+<|ref|>text<|/ref|><|det|>[[116, 596, 883, 888]]<|/det|>
+The surface chemistry of the Fischer- Tropsch catalytic reaction over Co has still several unknowns. Here, we report an operando X- ray photoelectron spectroscopy study of Co(0001) and Co(1014), and operando high energy surface X- ray diffraction of Co(0001), during the Fischer- Tropsch reaction at 0.15 bar - 1 bar and 406 K - 548 K in a \(\mathrm{H}_2 / \mathrm{CO}\) gas mixture. We find that the Co surfaces remain metallic under all conditions and that the coverage of chemisorbed species ranges from 0.4 - 1.7 monolayers depending on pressure and temperature. The adsorbates include CO on- top, C/CxHy and various longer hydrocarbon molecules, indicating a rate- limiting direct CO dissociation pathway and that only hydrocarbon species participate in the chain growth. The accumulation of hydrocarbon species points to the termination step being rate- limiting as well. Furthermore, we demonstrate that the intermediate surface species are highly dynamic, appearing and disappearing with time delays after rapid changes in the reactants' composition.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[147, 84, 317, 101]]<|/det|>
+## I. Introduction  
+
+<|ref|>text<|/ref|><|det|>[[116, 121, 883, 435]]<|/det|>
+The Fischer- Tropsch (FT) reaction is an important industrial process, as it produces higher hydrocarbons from synthesis gas (syngas, \(\approx 1:2\) CO:H2 gas mixture) over Co, is an important industrial process'. The FT reaction was used during earlier time as a way to avoid oil embargos for some countries during world war II and the Apartheid regime in South Africa. In the current era it can become an important avenue for a sustainable chemical industry if CO is generated from \(\mathrm{CO_2}\) via the reverse water gas shift reaction where the \(\mathrm{CO_2}\) has been captured either directly from the atmosphere or at an intense carbon source. The hydrogen can be produced, not from the current steam reforming process of methane, but instead through electrolysis of water where the electricity is coming from a renewable source such as wind and solar. Presently, the Fischer- Tropsch reaction utilizes Fe, Ru or Co based catalysts' that yield different hydrocarbon distributions. The Co based FT reaction typically generates long chain hydrocarbons and waxes and operates at a temperature of 470- 510 K and pressures of a few tens of bars'.  
+
+<|ref|>text<|/ref|><|det|>[[116, 440, 883, 780]]<|/det|>
+The chemical state of the Co catalyst has previously been investigated through postreaction analysis of single crystal surfaces to be in a metallic state \(^{2 - 4}\) . However, bulk sensitive measurements under high temperatures and pressures during operando of the FT reaction of Co nanoparticles have shown the existence of small amounts of oxides \(^{5 - 10}\) . Furthermore, it has also been proposed that a partially oxidized Co catalyst can be responsible for a high activity \(^{11}\) . Although no operando measurements during the FT reaction has detected any major presence of a Co carbide bulk phase it has been demonstrated that \(\mathrm{CoC_2}\) nano prisms shows a high selectivity to olefin formation \(^{12}\) . Recent in- situ surface sensitive measurements of the FT reaction on Fe show a growing carbide phase starting immediately after the reaction is initiated \(^{13}\) and on Ni at low temperatures dissolution of carbon into the bulk as a dilute carbide phase has been observed \(^{14}\) . An open key question is if the state of the Co catalyst in the surface region remains fully in a metallic state or if surface oxide and near surface carbide can be present during the reaction conditions. Addressing this question necessitates detection using surface sensitive techniques performed while the reaction is turning over.  
+
+<|ref|>text<|/ref|><|det|>[[117, 785, 881, 903]]<|/det|>
+The reaction mechanism of the FT reaction consists of a sequence of elementary reaction steps \(^{15}\) . The first step after CO adsorption is the dissociation of CO generating carbon monomeric species. Afterwards such C can either attach to other carbon atoms and thus grow the hydrocarbon chain or attach to hydrogen atoms resulting in a weakening of the bond between the carbons and the surface, ultimately leading to desorption. The CO activation has
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 882, 250]]<|/det|>
+resulted in two major hypotheses based on theoretical calculations: there is either a direct dissociation, often denoted carbide mechanism<sup>16,17</sup>, or hydrogen assisted dissociation via the generation of a \(\mathrm{CH_xO}\) species<sup>18,19</sup>. It has been proposed that the hydrogenation of adsorbed \(\mathrm{C^{20}}\) and the termination step are partly rate limiting<sup>21</sup> as well as hydrogenation of atomic O and \(\mathrm{OH^{22,23}}\) . Here it would be essential to probe the adsorbates on the surface, to determine intermediates that accumulate as the reaction proceeds, as a pointer towards specific rate limiting steps.  
+
+<|ref|>text<|/ref|><|det|>[[116, 255, 882, 546]]<|/det|>
+All chemical sensitive studies over the FT reaction of Co under operando conditions have been conducted with methods mostly probing the bulk, such as X- ray absorption spectroscopy (XAS) and X- ray powder diffraction (XRD)<sup>5- 10</sup>. There have been efforts with Infrared Spectroscopy but here the observation is limited to exclusively probing adsorbed CO since no other species in the Co surface region could be detected<sup>24</sup>. Scanning tunneling microscopy (STM) have probed Co single crystal surfaces under FT at atmospheric conditions where the morphology of steps and terraces could be followed but without chemical sensitivity<sup>3,4</sup>. In one STM study conducted at 4 bar and 492 K on the Co(0001) surface stripes were observed during the FT reaction interpreted as the appearance of long chain hydrocarbon molecules<sup>25</sup>. A number of surface science studies of model molecules under vacuum have been conducted on Co single crystal surfaces<sup>22,26- 28</sup> but it is unclear if the model molecules are relevant for reactions occurring at many orders of magnitude higher pressures and temperatures.  
+
+<|ref|>text<|/ref|><|det|>[[117, 552, 882, 840]]<|/det|>
+X- ray photoelectron spectroscopy (XPS) is a unique surface sensitive method to investigate the chemical state of catalytic surfaces and adsorbed intermediates through core- level shifts. The high inelastic scattering cross- section of photoelectrons in the gas phase makes vacuum conditions necessary. Post analysis with XPS has been conducted of Co single crystal surfaces that have been in a reactor with atmospheric pressure<sup>3,4</sup> or 4 bar<sup>2</sup>, at temperatures where the reaction is turning- over, followed by evacuating the reactor to vacuum and then transferring the sample to the spectrometer chamber, where the measurement was conducted. Although adsorbed CO, adsorbed carbidic carbon and hydrocarbon species were observed it is unclear if species may decompose or desorb when the system is evacuated and the temperature reduced. Near- ambient XPS (NAPXPS) studies of Co foil have been restricted to 0.1 mbar<sup>23</sup> — far from the conditions of atmospheric pressure where the FT reaction occurs — have been detected to produce methane and some small hydrocarbons on Co single crystal surfaces<sup>3,4,29</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[118, 845, 881, 914]]<|/det|>
+Here, we used an ambient- pressure XPS (APXPS) instrument called POLARIS operating at pressures up to 1 bar for \(\mathrm{CO / H_2}\) mixtures and as high temperatures as 506 K. The POLARIS instrument is based on the virtual pressure cell, where we create a \(\sim 30\) micron thick local high
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 82, 883, 377]]<|/det|>
+pressure cushion and utilize grazing incidence of the incoming hard X- rays to provide surface sensitivity, despite high kinetic energy of the photoelectrons<sup>30</sup>. We have used flat Co(0001) and stepped Co(10¯14) single crystal substrates that have been shown previously to turn- over the FT reaction towards mainly methane but also minor fractions of C<sub>2+</sub> hydrocarbon species at close to 1 bar and 500 K<sup>4</sup>. Since the FT reaction is known to be structure sensitive<sup>31- 34</sup> (i.e. a Co stepped crystal, Co(10¯15) gives much higher turn- over than the terrace surface<sup>3</sup>) we have thus directly compared the Co(0001) with the Co(10¯14) surface to elucidate the influence of steps on the reaction. As FT reactions have been demonstrated at the same conditions as in the current study we will denote the experiments as operando. Furthermore, we show the facile appearance and disappearance of C<sub>x</sub>H<sub>y</sub> adsorbates as seen in section II.d as indicators of an operando state. Our complementary operando surface X- ray diffraction experiments yield atomic surface structure information under reaction conditions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[147, 434, 421, 452]]<|/det|>
+## II. Results and Discussion  
+
+<|ref|>text<|/ref|><|det|>[[205, 473, 852, 491]]<|/det|>
+a. Chemical and Structural State of Co Single Crystals and Adsorbates at 1bar  
+
+<|ref|>text<|/ref|><|det|>[[115, 504, 884, 868]]<|/det|>
+First, we address the chemical state of Co: whether it is metallic, oxidic or carbidic in the near surface region. This information would not necessary be observable in bulk sensitive measurements, as this active phase exists only close to the surface (i.e. the first few monolayers). Fig. 1a-c shows the operando Co 2p<sub>3/2</sub>, C 1s, and O 1s signal of the Co(0001) catalyst at a pressure of 1 bar with a reaction mixture of 1:2 CO:H<sub>2</sub> and a temperature of 406 K and 506 K (the higher corresponding to the typical FT high yield conditions) measured in the POLARIS instrument at a photon energy of 4600 eV (for samples and gases description turn to S1, for experimental setup description turn to S2). The Co 2p<sub>3/2</sub> spectrum is composed of a single peak at 778.0eV that shows a completely metallic state with no shoulder at 780.0 eV<sup>35</sup> that would indicate an oxide. A carbide modification of Co would be expected at somewhat higher binding energy compared to the Co<sup>0</sup> metal peak if the shift follows as seen in Ni 2p<sub>3/2</sub> upon carbon dissolution into bulk Ni<sup>14</sup> and the formation of Fe carbides<sup>13</sup>. The peaks in the C 1s region at 285.7 eV and in the O 1s region at 531.7 eV (blue) correspond to adsorbed CO in top position<sup>28</sup>. The C1s feature at 284.1 eV at 506 K is related to hydrocarbon species and exemplifies the reaction intermediates. Other states are only observable at 406 K which will be further
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 880, 127]]<|/det|>
+discussed below. A carbide species would be seen at around \(283.0 \mathrm{eV}\) and oxides at around \(529.3 \mathrm{eV}^{36}\) and none are detected in the spectra.  
+
+<|ref|>text<|/ref|><|det|>[[117, 147, 883, 314]]<|/det|>
+The stepped \(\mathrm{Co}(10\bar{1} 4)\) surface is probed at the same conditions as the \(\mathrm{Co}(0001)\) and the results there of are presented in Fig. 1d- f for the \(\mathrm{Co} 2\mathrm{p}_{3 / 2}\) , C 1s, and O 1s core- levels. Co is metallic during the reaction also on this facet. Adsorbates in the C 1s region, however show a striking difference. Peaks at \(284.1 \mathrm{eV}\) and \(284.7 \mathrm{eV}\) are now observed at \(406 \mathrm{K}\) and as well at \(506 \mathrm{K}\) . Also, there is intensity in the region of \(283.5 \mathrm{eV}\) for both temperatures. Overall the total C coverage is higher than for the flat surface. The O 1s intensity again shows only significant contributions of \(\mathrm{CO}_{\mathrm{top}}\) adsorbates at \(506 \mathrm{K}\) and additionally adsorbed \(\mathrm{H}_2\mathrm{O}\) at \(406 \mathrm{K}\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[125, 88, 866, 617]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 632, 883, 817]]<|/det|>
+<center>Figure 1. Surface state of Co(0001) sample at indicated temperatures in an atmosphere of CO:H2 1:2 at 1 bar studied by hard X-ray photoelectron spectroscopy (HAXPES) utilizing 4600eV photons at \(0.3^{\circ}\) incidence. a shows Co \(2\mathrm{p}_{3 / 2}\) core-level spectra b displays the C 1s region and c depicts the O 1s region. Subplots d, e, f show the same conditions as a, b, c but for a Co(1014) surface. The columns of XPS data have constant scaling of the vertical axis. Subplot g shows a representative figure of the high energy surface X-ray diffraction HESXRD data at 67.4keV of the Co(0001) crystal (full set is shown in SI). The detector is protected by beam stops at the bulk Bragg peak positions. h X-ray structure factor extracted from the 2D diffraction data shown in g at reaction conditions at 496K and 1 bar reaction mixture (1:2 CO:H2), data from the hcp part of the surface used for the fit (orange circles with vertical lines indicating an estimation 10% relative error), data from the fcc part (grey circles), fit result (solid line). </center>  
+
+<|ref|>text<|/ref|><|det|>[[118, 832, 880, 900]]<|/det|>
+As XPS measurements are sensitive to the chemical state we have completed this data set with structure sensitive high energy surface X- ray diffractometry (HESXRD) on a Co(0001) single crystal at 200 mbar and 456 K with 1:2 CO:H2 under flow conditions (further conditions
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 883, 447]]<|/det|>
+are shown in S5). In Fig. 1g we display a maximum intensity per pixel from an angular scan rotating the sample around the surface normal in the range of the Co(0001) (1,0) crystal truncation rod. Full experimental details are given in the supporting information (S4). Our key observation is the appearance of a single surface rod at (1,0) reciprocal lattice units that indicates an unreconstructed hcp surface. A more detailed analysis shows that the behavior at partial pressures of 200mbar and 1 bar of \(\mathrm{CO:H_2}\) 1:2 mixtures the surfaces neither reconstruct (S5), which is in line with previous findings of operando STM observations<sup>4</sup>. No indication for the formation of ordered carbide formation is found. In Fig. 1h we present the X- ray structure factor extracted from the data in g. From the fit we can deduce, that the surface is atomically smooth under all gas mixtures studied with a slight inward relaxation of the topmost layer of \(\sim 0.04 \mathrm{\AA}\) (S2). The fit improves by including CO molecules on the surface, but due to the small data set available, the occupancies and position could not be further refined. The full data set gives also evidence, that a few percent of the surface is fcc (111) terminated. Due to the low number of fcc- terminated sites the contribution from fcc can therefore be neglected for the XPS data analysis.  
+
+<|ref|>text<|/ref|><|det|>[[117, 452, 883, 570]]<|/det|>
+We can thereby conclude that on a \(\mathrm{Co(0001)}\) single crystal at 1 bar and around \(500\mathrm{K}\) during the operation of the FT reaction the Co surface remains fully metallic and retains an ordered, flat surface which exhibits considerable crystal truncation rod signal. Furthermore, there are no signs of a surface carbide or surface oxide indicating that the C 1s and O 1s spectral intensities are related to chemisorbed species.  
+
+<|ref|>sub_title<|/ref|><|det|>[[207, 590, 536, 608]]<|/det|>
+### b. Detailed C 1s Spectral Interpretation  
+
+<|ref|>text<|/ref|><|det|>[[117, 621, 883, 839]]<|/det|>
+When inspecting the XPS spectra from the adsorbate at different conditions we have curve fitted the data into specific components. Since many different conditions in terms of pressure and temperature are measured an assigned spectroscopic component should at least be clearly visible as a peak or strong shoulder in one spectrum. The chemical assignment is based either on experimental spectra obtained from model compounds in ultrahigh vacuum (UHV) on \(\mathrm{Co(0001)^{26,28,37,38}}\) or on density functional theory (DFT) binding energy calculations (S8). The binding energy scale potentially could differ by two tenths of an eV due to recoil effects at high kinetic energies that depends on the bonding strength (for discussion the reader is referred to S6.3), however, we estimate these effects to be negligible.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[210, 90, 785, 547]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 567, 883, 658]]<|/det|>
+<center>Figure 2. Isobar C 1s spectra for Co(0001) at a 500mbar and b 200mbar. In c we show 150mbar on Co(10\bar{1}4). All mixtures are 1:2 CO:H2. Subplot d shows a comparison of 406 K data as function of pressure on the Co(0001) crystal, and subplot e displays a direct comparison of stepped Co(10\bar{1}4) and flat Co(0001) crystals at 406 K. All y-axes have the same scaling. Data has been normalized according to SI Section S6c. The color code is the same as in Fig. 1 b/e. </center>  
+
+<|ref|>text<|/ref|><|det|>[[116, 704, 883, 897]]<|/det|>
+Fig. 2a and 2b shows the C 1s spectra from the reaction of CO and \(\mathrm{H}_{2}\) with a mix ratio of 1:2 at a pressure of 500 mbar and 200 mbar, respectively, at temperatures in the range of 406 K to 523K over Co(0001). The corresponding O 1s spectra are shown in the supplementary material (S6.1). We assign the 283.2 eV (green) feature to chemisorbed C or CH on the surface based on XPS spectra obtained from either decomposition of ethylene on Co(0001) as observed at 283.2 eV \(^{37,38}\) or at 282.8 eV \(^{28}\) and seen in exposure of CO and \(\mathrm{H}_{2}\) at 4 bar followed by evacuation at 283.3 eV \(^{2}\) . The DFT calculation (S8) gives a binding energy of 283.1 eV for adsorbed C, (energy scale corrected against experimental value of adsorbed CO in on- top
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 881, 176]]<|/det|>
+position) whereas adsorbed CH has a somewhat lower value of \(283.0 \mathrm{eV}\) . With only such a small difference in C 1s binding energy between adsorbed C and CH and since there is a variation of the experimental value between \(282.8 - 283.3 \mathrm{eV}\) it is not possible to distinguish the two adsorbates we thus denote this peak C/CH at \(283.2 \mathrm{eV}\) (green).  
+
+<|ref|>text<|/ref|><|det|>[[115, 189, 882, 801]]<|/det|>
+The component observed at \(283.5 \mathrm{eV}\) (yellow) we assign to chemisorbed \(\mathrm{CH}_2\) species based on DFT calculations (S8). This binding energy has previously been reported as related to the \(\mathrm{CH}_3\) group in ethylidyne \(^{28}\) but spectra of the chemisorbed ethylidyne molecule also include a peak corresponding C to the group bonded to the surface at \(282.9 \mathrm{eV}\) . Since we observe the \(283.5 \mathrm{eV}\) feature at several conditions without any low binding energy component the \(283.5 \mathrm{eV}\) peak cannot be associated with ethylidyne. With a similar argument, the \(283.5 \mathrm{eV}\) feature cannot be one of the carbons in adsorbed ethylene, where the adsorption site generates two inequivalent C atoms, since then it should be accompanied by a second carbon peak at \(283.9 \mathrm{eV}^{28}\) . Next component, located at \(284.1 \mathrm{eV}\) (light red) we assign to hydrogen- saturated monomers of longer chain hydrocarbons, such as \(\mathrm{- CH}_2\) - and \(\mathrm{- CH}_3\) groups on the surface based on DFT calculations and previous post- analysis experiments \(^{2}\) . While some part of these hydrocarbon chains are in contact or in the proximity of the surface through undersaturated monomers, fully saturated parts are most likely pointing away from the surface and would correspond to the \(284.7 \mathrm{eV}\) (light blue) feature \(^{39}\) . The energy difference between the initial and final states in a photoionization event is much smaller when the hydrocarbon group is directly bonded to the surface allowing for metallic screening of the core- hole state resulting in a lower binding energy for the parts of the hydrocarbon in direct contact with the surface (light red peak) than for the parts pointing away from the surface (light blue) \(^{39}\) . Lastly, a clear peak originating from CO adsorbed in on- top configuration, denoted \(\mathrm{CO}_{\mathrm{top}}\) , is observed at \(285.7 \mathrm{eV}\) (dark blue). We find no indication of CO at other adsorption sites, such as the bridge or hollow sites, as commonly reported in UHV studies at liquid nitrogen temperatures \(^{22}\) or in AP- XPS studies at \(\sim 1000 \mathrm{x}\) lower pressures \(^{40}\) , yet we observe only \(\mathrm{CO}_{\mathrm{top}}\) in our experiments. The O 1s spectra contains mostly a feature associated with adsorbed CO in on- top position (S6.1). No clear indication of any significant amount adsorbed O, OH, CHO, COH or \(\mathrm{CH}_3\mathrm{O}\) species are detected.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 100, 883, 480]]<|/det|>
+Fig. 2a shows the C 1s spectra at a pressure of \(500\mathrm{mbar}\) from \(406\mathrm{K}\) to \(523\mathrm{K}\) on \(\mathrm{Co(0001)}\) . The coverage of the adsorbates has been determined through a specific normalization procedure (details: S6.3). We observe the largest total coverage of carbon containing species at the lowest temperature of \(406\mathrm{K}\) corresponding to \(1.5\mathrm{ML}\) with the \(\mathrm{- CH_2}\) - peak (yellow) clearly dominating the spectrum, but also intensity is observed in the region of non- screened hydrocarbon chains (light blue). A "monolayer" is here defined relative to the surface atoms of the Co substrate. What is clearly noted is that the total coverage decreases with increasing temperature to below monolayer coverage for \(\mathrm{T} > 480\mathrm{K}\) . We observe at \(406\mathrm{K}\) a large amount of the hydrocarbon species with C atoms both bonded to the surface and with \(\mathrm{CH_2}\) and \(\mathrm{CH_3}\) groups away from the surface as well as surface bound \(\mathrm{CH_2}\) groups. As we reach the highest temperature of \(523\mathrm{K}\) there is almost only CO on the surface and some small amount of adsorbed \(\mathrm{C / CH}\) . Chain growth requires a considerable coverage of carbon species which are not fully saturated by hydrogen, which on the (0001) surface occurs below \(485\mathrm{K}\) . This process can consequently occur at the lower temperature where there is a higher coverage of \(\mathrm{CH_2}\) groups and various adsorbed hydrocarbon species.  
+
+<|ref|>text<|/ref|><|det|>[[117, 491, 883, 635]]<|/det|>
+Fig. 2b shows the same trend of temperatures but with a total pressure of \(200\mathrm{mbar}\) on \(\mathrm{Co(0001)}\) . At the lowest temperature we have an almost similar total coverage of \(1.3\mathrm{ML}\) . We notice the same trend where the amount of species decreased with increasing temperature. What is mainly different is that the amount of \(\mathrm{CH_2}\) adsorbed species is now much higher in comparison to hydrocarbon molecules. The chain growth becomes less efficient with lower coverage. Again, the CO coverage \((\sim 1 / 3\mathrm{ML})\) is almost independent of temperature.  
+
+<|ref|>text<|/ref|><|det|>[[117, 648, 882, 765]]<|/det|>
+Fig. 2c shows the trend with the stepped \(\mathrm{Co(1014)}\) surface at total pressure of \(150\mathrm{mbar}\) . In general, we again observe an almost constant CO coverage but an increase in the hydrocarbon content and decrease of \(\mathrm{CH_2}\) adsorbed species indicating more efficient chain growth at steps compared to terraces. The total coverage at the lowest temperature of \(425\mathrm{K}\) is \(1.7\mathrm{ML}\) and compares well with our previous observations on the \(\mathrm{Co(0001)}\) surface.  
+
+<|ref|>text<|/ref|><|det|>[[117, 778, 882, 897]]<|/det|>
+Fig. 2d shows a comparison of spectra at different pressure on the \(\mathrm{Co(0001)}\) surface at the FT temperature of \(406\mathrm{K}\) . We observe an increase in the total coverage of adsorbed hydrocarbon species on the surface going to 1 bar, however, the relative distribution of different molecular fragments is somewhat similar at this low temperature. We can relate that the production of hydrocarbon at this temperature is limited by the desorption of products and only
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 880, 127]]<|/det|>
+becomes more efficient to a smaller degree with the increase in pressure since the surface is blocking its active sites due to kinetic hindrance.  
+
+<|ref|>text<|/ref|><|det|>[[117, 139, 883, 356]]<|/det|>
+Fig. 2e shows a comparison of the C 1s spectra at the temperature of \(506\mathrm{K}\) between the flat \(\mathrm{Co(0001)}\) and the stepped \(\mathrm{Co(10\bar{1}4)}\) surfaces. At this temperature the catalyst is expected to be active for the FT reaction. There is a striking increase in the hydrocarbon content with the presence of steps. It is interesting to note that also the production of methane and minor hydrocarbon species increased by almost an order of magnitude between flat and stepped Co surfaces in a recent STM reactor study. This increased reactivity was attributed to the lowering of the energy barriers for a rate limiting CO dissociation, and the increased hydrocarbon presence at all examined temperatures on the stepped crystal is fully consistent with this hypothesis.  
+
+<|ref|>text<|/ref|><|det|>[[116, 368, 883, 783]]<|/det|>
+Our data supports the view that CO dissociates most efficiently on the steps through a direct dissociation route and not the hydrogen assisted mechanism. The direct route hypothesis is strengthened by not observing any CHO species on the surface that would be visible at 285.0 eV and \(530.1\mathrm{eV}\) (S6). Although a weak component at the C 1s position could possibly be overlapping with adsorbed CO and hydrocarbon species, but there is no appreciable intensity at the low binding energy in the O 1s spectra. Furthermore, ultrafast measurements using X- ray lasers have demonstrated that the CHO species could only exist in an extremely short lived transient regime with a life time of only a few picoseconds and could never build up any appreciable coverage during steady- state reaction conditions. We do not observe any significant \(\mathrm{CH_2O}\) species with a calculated binding energy of \(286.5\mathrm{eV}\) and \(530.2\mathrm{eV}\) or \(\mathrm{CH_3O}\) species at \(286.5\mathrm{eV}\) and \(531.2\mathrm{eV}\) (S6) pointing to that non- dissociated CO does not significantly contribute to the chain growth. Finally, as the coverage of adsorbed hydrocarbon species with more than two attached hydrogens per carbon is quite high the hydrogenation termination step leading to desorption would also be rate limiting. We therefore predict that both CO dissociation and the final hydrogenation leading to desorption is rate limiting under the current conditions on the stepped surface. On the flat surface the different hydrocarbon hydrogenation steps seem to be limiting as well, indicating an overall less active surface.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[207, 85, 649, 102]]<|/det|>
+d. Dynamics upon Changes in Reactant Composition  
+
+<|ref|>image<|/ref|><|det|>[[168, 140, 820, 655]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 666, 883, 776]]<|/det|>
+<center>Figure 3. Dynamic study of a CO addition and removal experiment. a 2D time-resolved spectrum of the C 1s region at 406 K (b for 506 K) total pressure under CO flow is 200mbar with a 1:2 CO:H₂ mixture. In the beginning and end the sample is subjected to a H₂ flow alone. The lines on the right are extracted from a and b at indicated times and resemble these reaction conditions from bottom to top: pure H₂, first 6 minutes under CO:H₂ mixture, last 6min under CO:H₂ mixture, only H₂ after removal of CO from the mixture (red) and at the end of the experiment after at least 15min in H₂(light blue). </center>  
+
+<|ref|>text<|/ref|><|det|>[[117, 785, 881, 904]]<|/det|>
+Fig. 3 depicts the time dependence of the C 1s spectra related to the FT reaction of the Co(0001) surface by applying and removing CO while keeping a constant H₂ flow on the sample at 200 mbar total pressure. We performed experiments at 406 K (panel a with line extracts shown in b) and at 506 K (panel c with line extracts shown in d). At 406 K the sample surface is covered with a tiny amount of species at 284.5eV initially. Upon exposure with CO
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 82, 883, 373]]<|/det|>
+there immediately appears intensity in the \(\mathrm{CH}_2\) and \(\mathrm{CO_{top}}\) regions (283.5 and \(285.7\mathrm{eV}\) respectively) indicating that some CO is dissociated and hydrogenated. Over an interval of approximately \(30\mathrm{min}\) there is a continuous growth of the \(284.5\mathrm{eV}\) state indicating the appearance of longer chain hydrocarbons due to chain growth. Upon removal of the CO in the reaction mixture this component remains for a certain time while \(\mathrm{CO_{top}}\) and \(\mathrm{CH}_2\) are reacted away within the time resolution of this experiment. Continuing in this configuration, the \(284.5\mathrm{eV}\) hydrocarbon peak intensity reduces, indicating facile reaction and departure from the surface aided by the presence of \(\mathrm{H}_2\) , which supports our operando claim. We are here observing the rate limitation of the final hydrogenation step that removes the hydrocarbon species on the surface. A competing reaction to the hydrocarbon chain growth is the fast \(\mathrm{CH}_2\) hydrogenation into methane, which also explains the swift vanishing of the \(\mathrm{CH}_2\) contribution upon CO gas removal from reaction mixture<sup>26</sup>.  
+
+<|ref|>text<|/ref|><|det|>[[116, 385, 884, 600]]<|/det|>
+At \(506\mathrm{K}\) , we observe mainly low \(\mathrm{CO_{top}}\) surface coverage and only to a negligible degree \(\mathrm{C_xH_y}\) species as compared to the \(406\mathrm{K}\) experiment during reaction conditions, which is in line with the trends in the static measurements shown in Fig. 2. After CO is removed from the reaction mixture the intermediate hydrocarbons desorb or react away and the corresponding peak diminishes to baseline intensity. The dynamic response is much faster at the higher temperature. From these temperatures we derive that the surface is highly dynamic (i.e. turning over and in operando) and that changes in the conditions needs time to establish a steady state. Moreover, we observe that the rate limiting step changes from the formation of carbon chains at lower temperatures to the dissociation of CO at higher temperatures.  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 634, 312, 653]]<|/det|>
+## III. Conclusion  
+
+<|ref|>text<|/ref|><|det|>[[116, 673, 883, 891]]<|/det|>
+We have studied the two Co single crystal surfaces of (0001) and \((10\bar{1} 4)\) using operando XPS at almost 100 times higher pressures than traditional NAPXPS and can directly probe adsorbates on the surface during the reaction. The C 1s and O 1s spectra shows only adsorbed species even at pressures close to 1 bar and the Co \(2\mathrm{p}_{3 / 2}\) spectra have no sign of an oxide or a carbide component. The surface X- ray diffraction results on Co(0001) demonstrate that the surface stays atomically smooth under reaction conditions. Thereby, there is no indication of any chemical or structural changes of the Co substrate surface region as the reaction proceeds. Our observations tip the scales in the discussion regarding the nature of the CO dissociation towards the direct or often denoted carbide mechanism since no sign of hydrogen assisted
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 82, 883, 375]]<|/det|>
+dissociation in terms of CHO detected species in the C or O 1s spectra. Furthermore, the chain growth involves only hydrocarbon species, since undissociated CO participation should show up as detected \(\mathrm{CH}_2\mathrm{O}\) spectral components, which we did not observe. There are also no ethylidene or adsorbed ethylene intermediates detected pointing to simple chain growth of - \(\mathrm{CH}_x\) - species resulting in an increasing amount of hydrocarbon species with groups both bonded directly to the surface but also pointing away towards the gas phase. The increasing appearance of hydrocarbon species at \(406\mathrm{K}\) as well as on the stepped surface where CO dissociation is more facile shows that the final termination step in terms of hydrogenation is also rate- limiting. Finally, our observation of the Co based Fischer- Tropsch reaction is highly dynamic meaning that the involved species (despite a potentially long residence time) show changing adsorbate compositions as a direct consequence of changes in the reactant mixtures. The time for the delay is strongly temperature dependent and are on the tens of minutes scale.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 84, 345, 118]]<|/det|>
+## ASSOCIATED CONTENT Supporting Information.  
+
+<|ref|>text<|/ref|><|det|>[[118, 133, 881, 217]]<|/det|>
+PDF file contains details of (i) sample preparation and assessment, (ii) the beam damage assessment, (iii) sample heating and surface temperature measurements, (iv) XPS data corrections. It also contains more data sets for the readers' reference.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 264, 351, 281]]<|/det|>
+## AUTHOR INFORMATION  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 297, 327, 315]]<|/det|>
+## Corresponding Authors  
+
+<|ref|>text<|/ref|><|det|>[[118, 329, 501, 380]]<|/det|>
+\*E- mail for P.L.: patrick.loemker@fysik.su.se  \*E- Mail for A.N.: andersn@fysik.su.se  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 395, 310, 412]]<|/det|>
+## Author Contributions  
+
+<|ref|>text<|/ref|><|det|>[[115, 426, 883, 642]]<|/det|>
+P.L. with input from A.N. planned the experiments at Petra III. P.L., D.D., M.S., C.M.G., J.G., H.- Y.W., A.H., R.R. A.S., C.S., A.N., P.A. participated in the XPS experimental work, while Z.H., A.S., H.N., V.V., J.- C.S., R.G., S.T. and L.J. participated in the SXRD measurements. P.L. extracted and plotted the SXRD data and A.S. fitted it. G.L.S.R. performed theoretical calculations. P.L. and M.S. developed the data analysis software and P.L. did the data analysis. P.L. and AN wrote the manuscript. All authors contributed to the literature research, result discussion and manuscript improvement.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 658, 170, 674]]<|/det|>
+## Notes  
+
+<|ref|>text<|/ref|><|det|>[[118, 690, 548, 708]]<|/det|>
+The authors declare no competing financial interests.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 755, 346, 772]]<|/det|>
+## ACKNOWLEDGEMENTS  
+
+<|ref|>text<|/ref|><|det|>[[117, 787, 883, 905]]<|/det|>
+The work was supported by the Swedish Research Council (Vetenskapsrådet, VR, project 2017- 00559 and project 2013- 8823), the Knut & Alice Wallenberg (KAW, grant nr. 2016.0042) foundation as well as the Swedish Foundation for strategic research (Stiftelsen för Strategisk Forskning, SSF, Proj. ITM 17- 0034). The research leading to this result has also
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 883, 333]]<|/det|>
+been supported by the project CALIPSO plus under Grant Agreement 730872 from the EU Framework Program for Research and Innovation HORIZON 2020. The experimental part of this research was carried out at P22 and P21 beamlines at DESY, a member of the Helmholtz Association (HGF). Beamtime was given for in house research proposals. The DFT calculations were performed using resources provided by the Swedish National Infrastructure for Computing (SNIC) at the HPC2N center. The authors would like to acknowledge the help of the P22 beamline engineer Katrin Ederer, and the Technical Division at Stockholm University.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 84, 215, 101]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[115, 113, 881, 165]]<|/det|>
+1. Dry, M. E. The Fischer–Tropsch process: 1950–2000. Catal. Today 71, 227–241 (2002).  
+
+<|ref|>text<|/ref|><|det|>[[115, 179, 881, 233]]<|/det|>
+2. Jin, Y. et al. Elementary Surface Reactions on Co(0001) under Fischer–Tropsch Synthesis Conditions. J. Phys. Chem. C 121, 21535–21540 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[115, 245, 881, 333]]<|/det|>
+3. Golder, K. M. & Wintterlin, J. In Situ/Operando STM of the Fischer–Tropsch Synthesis on a Co(10f15) Surface—A Study to Bridge the Materials Gap between Single-Crystal Models and Supported Catalysts. ACS Catal. 12, 7199–7209 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 344, 881, 430]]<|/det|>
+4. Boller, B., Durner, K. M. & Wintterlin, J. The active sites of a working Fischer–Tropsch catalyst revealed by operando scanning tunnelling microscopy. Nat. Catal. 2, 1027 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 442, 881, 529]]<|/det|>
+5. Ernst, B., Bensaddik, A., Hilaire, L., Chaumette, P. & Kiennemann, A. Study on a cobalt silica catalyst during reduction and Fischer–Tropsch reaction: In situ EXAFS compared to XPS and XRD. Catal. Today 39, 329–341 (1998).  
+
+<|ref|>text<|/ref|><|det|>[[115, 541, 880, 594]]<|/det|>
+6. Cats, K. H. et al. X-ray nanoscopy of cobalt Fischer–Tropsch catalysts at work. Chem. Commun. 49, 4622–4624 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[115, 606, 880, 659]]<|/det|>
+7. Price, S. W. T. et al. Chemical imaging of Fischer-Tropsch catalysts under operating conditions. Sci. Adv. 3, e1602838 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[115, 672, 881, 758]]<|/det|>
+8. Ronning, M. et al. Combined XRD and XANES studies of a Re-promoted Co/γ-Al2O3 catalyst at Fischer–Tropsch synthesis conditions. Catal. Today 155, 289–295 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[115, 771, 880, 824]]<|/det|>
+9. Rochet, A. et al. In situ and operando structural characterisation of a Fischer–Tropsch supported cobalt catalyst. Catal. Today 171, 186–191 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 880, 889]]<|/det|>
+10. Huffman, G. P. et al. In-Situ XAFS Investigation of K-Promoted Co Catalysts. J. Catal. 151, 17–25 (1995).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 881, 168]]<|/det|>
+11. Melaet, G. et al. Evidence of Highly Active Cobalt Oxide Catalyst for the Fischer–Tropsch Synthesis and CO₂ Hydrogenation. J. Am. Chem. Soc. 136, 2260–2263 (2014).  
+
+<|ref|>text<|/ref|><|det|>[[118, 181, 880, 235]]<|/det|>
+12. Zhong, L. et al. Cobalt carbide nanoprisms for direct production of lower olefins from syngas. Nature 538, 84–87 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 880, 301]]<|/det|>
+13. Shipilin, M. et al. In Situ Surface-Sensitive Investigation of Multiple Carbon Phases on Fe(110) in the Fischer–Tropsch Synthesis. ACS Catal. 12, 7609–7621 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[118, 313, 881, 397]]<|/det|>
+14. Degerman, D. et al. State of the Surface During CO Hydrogenation over Ni(111) and Ni(211) Probed by Operando X-ray Photoelectron Spectroscopy. J. Phys. Chem. C acs.jpcc.2c07650 (2023) doi:10.1021/acs.jpcc.2c07650.  
+
+<|ref|>text<|/ref|><|det|>[[118, 410, 883, 529]]<|/det|>
+15. Dry, M. E. The Fischer-Tropsch (FT) Synthesis Processes. in Handbook of Heterogeneous Catalysis (eds. Ertl, G., Knözinger, H., Schüth, F. & Weitkamp, J.) hetcat0150 (Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, Germany, 2008). doi:10.1002/9783527610044. hetcat0150.  
+
+<|ref|>text<|/ref|><|det|>[[118, 542, 881, 595]]<|/det|>
+16. Shetty, S. & Santen, R. A. van. Hydrogen induced CO activation on open Ru and Co surfaces. Phys. Chem. Chem. Phys. 12, 6330–6332 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[118, 608, 881, 692]]<|/det|>
+17. Petersen, M. A., van den Berg, J.-A., Ciobica, I. M. & van Helden, P. Revisiting CO Activation on Co Catalysts: Impact of Step and Kink Sites from DFT. ACS Catal. 7, 1984–1992 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[118, 705, 881, 789]]<|/det|>
+18. Storsæter, S., Chen, D. & Holmen, A. Microkinetic modelling of the formation of C1 and C2 products in the Fischer–Tropsch synthesis over cobalt catalysts. Surf. Sci. 600, 2051–2063 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[118, 803, 881, 888]]<|/det|>
+19. Inderwildi, O. R., Jenkins, S. J. & King, D. A. Fischer–Tropsch Mechanism Revisited: Alternative Pathways for the Production of Higher Hydrocarbons from Synthesis Gas. J. Phys. Chem. C 112, 1305–1307 (2008).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 882, 168]]<|/det|>
+20. Banerjee, A., van Bavel, A. P., Kuipers, H. P. C. E. & Saeys, M. CO Activation on Realistic Cobalt Surfaces: Kinetic Role of Hydrogen. ACS Catal. 7, 5289–5293 (2017).  
+
+<|ref|>text<|/ref|><|det|>[[115, 180, 882, 268]]<|/det|>
+21. Yang, J., Tveten, E. Z., Chen, D. & Holmen, A. Understanding the Effect of Cobalt Particle Size on Fischer–Tropsch Synthesis: Surface Species and Mechanistic Studies by SSITKA and Kinetic Isotope Effect. Langmuir 26, 16558–16567 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[115, 279, 882, 397]]<|/det|>
+22. Weststrate, C. J., van de Loosdrecht, J. & Niemantsverdriet, J. W. Spectroscopic insights into cobalt-catalyzed Fischer-Tropsch synthesis: A review of the carbon monoxide interaction with single crystalline surfaces of cobalt. J. Catal. 342, 1–16 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 409, 881, 464]]<|/det|>
+23. Wu, L., Liu, H., Reese, J. M. & Zhang, Y. Non-equilibrium dynamics of dense gas under tight confinement. J. Fluid Mech. 794, 252–266 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 476, 881, 561]]<|/det|>
+24. Oosterbeek, H. Bridging the pressure and material gap in heterogeneous catalysis: cobalt Fischer–Tropsch catalysts from surface science to industrial application. Phys. Chem. Chem. Phys. 9, 3570–3576 (2007).  
+
+<|ref|>text<|/ref|><|det|>[[115, 573, 881, 659]]<|/det|>
+25. Navarro, V., van Spronsen, M. A. & Frenken, J. W. M. In situ observation of self-assembled hydrocarbon Fischer–Tropsch products on a cobalt catalyst. Nat. Chem. 8, 929–934 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 671, 881, 757]]<|/det|>
+26. Weststrate, C. J., van Helden, P. & Niemantsverdriet, J. W. Reflections on the Fischer-Tropsch synthesis: Mechanistic issues from a surface science perspective. Catal. Today 275, 100–110 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 770, 881, 824]]<|/det|>
+27. Weststrate, C. J. & Niemantsverdriet, J. W. Understanding FTS selectivity: the crucial role of surface hydrogen. Faraday Discuss. 197, 101–116 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 881, 167]]<|/det|>
+28. Weststrate, C. J. (Kees-J. et al. Mechanistic insight into carbon-carbon bond formation on cobalt under simulated Fischer-Tropsch synthesis conditions. Nat. Commun. 11, 750 (2020).  
+
+<|ref|>text<|/ref|><|det|>[[115, 180, 881, 234]]<|/det|>
+29. Geerlings, I. J. C., Zonneyville, M. C. & De Groot, C. P. M. The Fischer-Tropsch reaction on a cobalt (0001) single crystal. Catal. Lett. 5, 309-314 (1990).  
+
+<|ref|>text<|/ref|><|det|>[[115, 246, 881, 331]]<|/det|>
+30. Amann, P. et al. A high-pressure x-ray photoelectron spectroscopy instrument for studies of industrially relevant catalytic reactions at pressures of several bars. Rev. Sci. Instrum. 90, 103102 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 344, 881, 430]]<|/det|>
+31. Bezemer, G. L. et al. Cobalt Particle Size Effects in the Fischer-Tropsch Reaction Studied with Carbon Nanofiber Supported Catalysts. J. Am. Chem. Soc. 128, 3956-3964 (2006).  
+
+<|ref|>text<|/ref|><|det|>[[115, 443, 881, 562]]<|/det|>
+32. Prieto, G., Martínez, A., Concepción, P. & Moreno-Tost, R. Cobalt particle size effects in Fischer-Tropsch synthesis: structural and in situ spectroscopic characterisation on reverse micelle-synthesised Co/ITQ-2 model catalysts. J. Catal. 266, 129-144 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[115, 574, 881, 628]]<|/det|>
+33. den Breejen, J. P. et al. On the Origin of the Cobalt Particle Size Effects in Fischer-Tropsch Catalysis. J. Am. Chem. Soc. 131, 7197-7203 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 881, 728]]<|/det|>
+34. Xiong, H., Motchelaho, M. A. M., Moyo, M., Jewell, L. L. & Coville, N. J. Correlating the preparation and performance of cobalt catalysts supported on carbon nanotubes and carbon spheres in the Fischer-Tropsch synthesis. J. Catal. 278, 26-40 (2011).  
+
+<|ref|>text<|/ref|><|det|>[[115, 739, 881, 825]]<|/det|>
+35. Biesinger, M. C. et al. Resolving surface chemical states in XPS analysis of first row transition metals, oxides and hydroxides: Cr, Mn, Fe, Co and Ni. Appl. Surf. Sci. 257, 2717-2730 (2011).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 881, 168]]<|/det|>
+36. Weststrate, C. J., van Helden, P., van de Loosdrecht, J. & Niemantsverdriet, J. W. Elementary steps in Fischer–Tropsch synthesis: CO bond scission, CO oxidation and surface carbiding on Co(0001). Surf. Sci. 648, 60–66 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 180, 881, 267]]<|/det|>
+37. Xu, L. et al. A Photoemission Study of Ethylene Decomposition on a Co(0001) Surface: Formation of Different Types of Carbon Species. J. Phys. Chem. C 116, 4167–4174 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[115, 279, 881, 366]]<|/det|>
+38. Xu, L. et al. Transformation of Carbon Monomers and Dimers to Graphene Islands on Co(0001): Thermodynamics and Kinetics. J. Phys. Chem. C 117, 2952–2958 (2013).  
+
+<|ref|>text<|/ref|><|det|>[[115, 377, 881, 431]]<|/det|>
+39. Weiss, K. et al. XPS and XAS investigation of condensed and adsorbed n-octane on a Cu(110) surface. J. Electron Spectrosc. Relat. Phenom. 128, 179–191 (2003).  
+
+<|ref|>text<|/ref|><|det|>[[115, 443, 881, 530]]<|/det|>
+40. Chai, P. et al. Near Ambient-Pressure X-ray Photoelectron Spectroscopy Study of CO Activation and Hydrogenation on Co(0001). J. Phys. Chem. C 125, 22223–22230 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 880, 596]]<|/det|>
+41. LaRue, J. et al. Real-Time Elucidation of Catalytic Pathways in CO Hydrogenation on Ru. J. Phys. Chem. Lett. 8, 3820–3825 (2017).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[43, 42, 312, 70]]<|/det|>
+## Supplementary Files  
+
+<|ref|>text<|/ref|><|det|>[[42, 93, 768, 112]]<|/det|>
+This is a list of supplementary files associated with this preprint. Click to download.  
+
+<|ref|>text<|/ref|><|det|>[[60, 130, 475, 150]]<|/det|>
+- CoFTsynthesisSupplementaryMaterials.docx
+
+<--- Page Split --->

preprint/preprint__b46d8d3df6b02d19a7c5175f1d79036d28add0aaedacb640efb371097fda6156/images_list.json

@@ -0,0 +1,40 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 Exploration of the CuAg alloy-based selector. a, Schematic illustrations of a metal-filament-based selector under different applied voltage ( \\(V_{\\mathrm{a}}\\) ). b, Representative current-voltage ( \\(I - V\\) ) characteristics of a Cu/Ag metal-filament-based selector, the ON/OFF ratio corresponds to the current variation at the read voltage ( \\(V_{\\mathrm{read}}\\) ) and half-read voltage ( \\(V_{\\mathrm{read}} / 2\\) ). c, \\(I - V\\) characteristics of \\(\\mathrm{Ag / Al_2O_3 / Ag}\\) , \\(\\mathrm{Cu / Al_2O_3 / Cu}\\) and \\(\\mathrm{CuAg / Al_2O_3 / CuAg}\\) selectors before annealing. d, Surface morphologies of \\(\\mathrm{Ag / Al_2O_3 / Ag}\\) , \\(\\mathrm{Cu / Al_2O_3 / Cu}\\) and \\(\\mathrm{CuAg / Al_2O_3 / CuAg}\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        117,
+        348,
+        876,
+        714
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 The \\(64 \\times 64\\) 1S1R array. a, Schematic illustration of the integrated 1S1R devices. b,",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 On the validation of selector-based LIF neuron. a, Schematic diagram of a biological neuron. b, Circuit model of a LIF neuron, the relationship between the \\(I(t)\\) and the \\(u(t)\\) is",
+    "footnote": [],
+    "bbox": [
+      [
+        115,
+        485,
+        875,
+        835
+      ]
+    ],
+    "page_idx": 12
+  }
+]