Add files using upload-large-folder tool

peer_reviews/supplementary_0_Peer Review File__a1c9b6bf0328e35265d1b23ac7812525f8bbef78b8c2462342fb11c2ba711f2b/supplementary_0_Peer Review File__a1c9b6bf0328e35265d1b23ac7812525f8bbef78b8c2462342fb11c2ba711f2b_det.mmd

@@ -0,0 +1,453 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 505, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[106, 162, 883, 220]]<|/det|>
+Title: On the importance of the electric double layer structure in aqueous electrocatalysis  
+
+<|ref|>image<|/ref|><|det|>[[95, 732, 260, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[270, 732, 880, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 90, 350, 106]]<|/det|>
+<b>REVIEWER COMMENTS</B>  
+
+<|ref|>text<|/ref|><|det|>[[116, 129, 393, 145]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 166, 883, 342]]<|/det|>
+This work aims to demystify the structure of the electrochemical double layer using a combined computational and experimental approach. Through DFT calculations of a silver (111) surface the authors modeled the electrochemical double layer composed of \(\mathrm{Na + }\) , \(\mathrm{F - }\) , and water revealing the dynamic structure and transformations at the electrode/electrolyte interface upon polarization. The authors further performed experimental electrochemistry to understand the role that cations play in the EDL by complexing the sodium cations with a crown ether analog. Using the well characterized CO2 reduction reaction as a platform to study the effects on complexed sodium ions the authors note a logarithmic decrease in current density in concert with increasing chelator concentration. The work presented herein provides a very systematic analysis of the electrochemical double layer structure  
+
+<|ref|>text<|/ref|><|det|>[[115, 363, 166, 378]]<|/det|>
+Notes:  
+
+<|ref|>text<|/ref|><|det|>[[115, 382, 880, 420]]<|/det|>
+- I recommend the authors consider revising the title to be more specific to the impact of the paper. This is a great paper, but it is hard to get excited by the title.  
+
+<|ref|>text<|/ref|><|det|>[[115, 440, 879, 497]]<|/det|>
+- The authors do not provide enough relevant literature as to the recent developments, both experimental and computational, in the field of interfacial electrocatalysis. It is therefore recommended that some of the works listed below be used to motivate the aims of this manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 516, 860, 574]]<|/det|>
+10.1039/C7CP06087D, doi.org/10.1063/1.5124878, doi.org/10.1039/C9EE01341E, doi.org/10.1021/jacs.6b07612, doi.org/10.1021/jacs.7b06765, doi.org/10.1038/s41929-021-00655-5, doi.org/10.1021/acs.jpcc.0c07004.  
+
+<|ref|>text<|/ref|><|det|>[[114, 594, 872, 710]]<|/det|>
+The authors provide a method for modelling the electrochemical double layer through DFT calculations revealing interesting molecular dynamics upon polarization. The experimental work contained herein supports computational findings by tying CO2 electrocatalysis to the EDL structure. This work provides new insights and methods towards elucidating the double layer's structure/property reactions to electrocatalysis. The manuscript is detailed, organized, and insightful. I recommend this work for publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 771, 395, 788]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 811, 640, 828]]<|/det|>
+My review is in the attached file (because I needed to include graphics).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 166, 877, 302]]<|/det|>
+The report is a study using mainly computational methods and a few experiments, to study the origin of the electric double layer. The system is a Ag111 surface with water/Na+ or water/F- solution. Several observed effects are very clearly reproduced and a credible molecular explanation is presented for the capacitance peaks. The paper is well- written and the conclusions are clearly presented. However, there are some things to address before the conclusions can be seen as verified. Since the conclusions fully rely on a single computational model which is non- standard and therefore not well tested, it is crucial that this model is benchmarked. Below are some specific points to address.  
+
+<|ref|>text<|/ref|><|det|>[[113, 322, 877, 440]]<|/det|>
+1. There are some questionable arguments in the molecular origin discussion. It is stated that the cause for F- to adsorb on the surface while Na+ stays further away is due to the smaller hydration energy of anions. In the supporting information the hydration energy for F- is presented as 115-120 kcal/mol, while Na+ has a value of 80-90 kcal/mol. This is precisely opposite to the argument on line 119. It is also stated that the dispersive energy is larger for F- than for Na+, which is likely correct. This should be straight forward to estimate from the Uvdw term.  
+
+<|ref|>text<|/ref|><|det|>[[113, 441, 851, 517]]<|/det|>
+2. In the parametrization of the Buckingham potential for water, only the geometry with the oxygen adsorbed is probed. However, in the simulations at the cathode most water molecules point the hydrogen towards the surface. This geometry should also be probed, and should probably be tested with a charged Ag cluster.  
+
+<|ref|>text<|/ref|><|det|>[[113, 518, 883, 692]]<|/det|>
+3. It is not clear if the TIP3P Ag(111) interaction is balanced. I cannot find any benchmark of that. Especially when the surface is charged and the hydrogen points to the surface, the interaction in the presented model seems very strong so that the cations are even pushed out of the first layer. It could be correct, but it could as well be an artefact from an model that has a too strong interaction between H and Ag. Since TIP3P has an inflated charge distribution to compensate for the lack of anisotropy and other effects, it could lead to the formation of the silver-hydrogen bond formation at the cathode, which in turn seem to completely outcompete the silver cation interaction. I suggest that another water model is tested to see if the electrolyte structure is the same or if it is changed, to avoid the risk of an artefact due to a too simple water model.  
+
+<|ref|>text<|/ref|><|det|>[[113, 693, 877, 809]]<|/det|>
+4. The interpretation that the electric field difference is the determining factor for the difference in activity when crown-ether is added, could be correct but could as well be incorrect. Direct interaction between the oxygen atoms of -COO at the surface could also stabilize the formation of that adduct, and this interaction would also be limited by addition of crown-ether. There are some recent reports that discuss this phenomenon including Nature Catalysis 2021, 4, 654-662 and J. Phys. Chem. C 2020, 124, 41, 22479-22487.  
+
+<|ref|>text<|/ref|><|det|>[[113, 829, 881, 906]]<|/det|>
+Overall I believe that this report could provide very interesting and important insight on the catalyst- solvent interface under working conditions. There are some questions on the reliability of the method that needs to be addressed and some discussion that could be improved, but the key points of the paper are of high interest.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 105, 883, 147]]<|/det|>
+Comments to Reviewer. Thank you for your helpful comments, which are reproduced here in italics. Our responses are in boldface.  
+
+<|ref|>text<|/ref|><|det|>[[115, 169, 211, 186]]<|/det|>
+Reviewer: 1  
+
+<|ref|>text<|/ref|><|det|>[[115, 218, 206, 235]]<|/det|>
+Comments:  
+
+<|ref|>text<|/ref|><|det|>[[112, 251, 875, 586]]<|/det|>
+This work aims to demystify the structure of the electrochemical double layer using a combined computational and experimental approach. Through DFT calculations of a silver (111) surface the authors modeled the electrochemical double layer composed of \(\mathrm{Na}^+\) , \(F^-\) , and water revealing the dynamic structure and transformations at the electrode/electrolyte interface upon polarization. The authors further performed experimental electrochemistry to understand the role that cations play in the EDL by complexing the sodium cations with a crown ether analog. Using the well characterized \(\mathrm{CO}_2\) reduction reaction as a platform to study the effects on complexed sodium ions the authors note a logarithmic decrease in current density in concert with increasing chelator concentration. The work presented herein provides a very systematic analysis of the electrochemical double layer structure  
+
+<|ref|>text<|/ref|><|det|>[[114, 616, 883, 670]]<|/det|>
+We would like to thank the reviewer for the favorable comments and recognition of the significance of our work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 752, 167, 768]]<|/det|>
+Notes:  
+
+<|ref|>text<|/ref|><|det|>[[114, 800, 841, 855]]<|/det|>
+- I recommend the authors consider revising the title to be more specific to the impact of the paper. This is a great paper, but it is hard to get excited by the title.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 103, 884, 266]]<|/det|>
+We would like to thank the reviewer for the thoughtful suggestions regarding the title. Following the recommendations made by the reviewer, we changed the title from "Revealing the significance of the electric double layer structure for electrocatalysis" to "Electric double layer structure in aqueous electrolyte and its electrocatalytic importance" to reflect the impact of the paper more clearly.  
+
+<|ref|>text<|/ref|><|det|>[[113, 343, 838, 469]]<|/det|>
+- The authors do not provide enough relevant literature as to the recent developments, both experimental and computational, in the field of interfacial electrocatalysis. It is therefore recommended that some of the works listed below be used to motivate the aims of this manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 497, 843, 589]]<|/det|>
+10.1039/C7CP06087D, doi.org/10.1063/1.5124878, doi.org/10.1039/C9EE01341E, doi.org/10.1021/jacs.6b07612, doi.org/10.1021/jacs.7b06765, doi.org/10.1038/s41929-021-00655-5, doi.org/10.1021/acs.jpcc.0c07004.  
+
+<|ref|>text<|/ref|><|det|>[[113, 617, 884, 709]]<|/det|>
+We would like to thank the reviewer suggesting these additional references. We agree that all of the recommended references are important and have added corresponding citations on pages 3 and 12 of the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 787, 884, 878]]<|/det|>
+The authors provide a method for modelling the electrochemical double layer through DFT calculations revealing interesting molecular dynamics upon polarization. The experimental work contained herein supports computational findings by tying \(CO_2\) electrocatalysis to the EDL
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 104, 884, 195]]<|/det|>
+structure. This work provides new insights and methods towards elucidating the double layer's structure/property reactions to electrocatalysis. The manuscript is detailed, organized, and insightful. I recommend this work for publication.  
+
+<|ref|>text<|/ref|><|det|>[[112, 223, 884, 351]]<|/det|>
+We appreciate the recognition of the reviewer regarding the importance of the EDL study, which will have a significant impact on the understanding of electrocatalysis. We also would like to thank the reviewer for recommending the publication of our work in Nature Communications.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 152, 886, 526]]<|/det|>
+I have reviewed the manuscript titled "Revealing the significance of the electric double layer structure for electrocatalysis" by Shin et al. In brief, I think the manuscript has interesting scientific insight. However, the authors have entirely neglected previous literature and have made exaggerated claims about the nature of the double- hump capacitance curve (their main focus) as being entirely unexplained in the last 100 years. This lack of honest scholarship on their part, gives me serious pause about the paper. As I will discuss below, this problem has been addressed with many theoretical models in the past, and unlike what is claimed it is not the first time that it is explained. Their work is interesting, and perhaps should be published only and only after it is compared with existing theories and the exaggerations are toned down. I am not quite sure if this is the right journal for it, but that is the editor's call. Similar papers have been published in the more specialized journal regularly (see examples below).  
+
+<|ref|>text<|/ref|><|det|>[[112, 553, 886, 899]]<|/det|>
+We appreciate the efforts made by the reviewer in critiquing our manuscript and providing us with valuable comments. We also would like to thank the reviewer for recognizing the scientific insights that the present work attempted to address. We fully acknowledge and admire the previous research works that have been devoted to electric double layers. Following the thoughtful suggestion provided by the reviewer, we avoided using unnecessary superlative expressions and revised the title to address the impact of our present work clearly. We understand that the major concern of the reviewer is the lack of comparison of our results with previous theoretical works, such as the famous Kornyshev model. As the reviewer noted, the Kornyshev model successfully explains the emergence of camel- shaped differential capacitance [Chem. Rev. 114, 2978- 3036 (2014)], which has been achieved by
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 103, 886, 404]]<|/det|>
+including the ion saturation behavior in the Gouy- Chapman model. Consequently, the results of the Kornyshev model approach those of the Gouy- Chapman model at the dilute electrolyte limit [Chem. Rev. 114, 2978–3036 (2014)]; thus, it cannot be applied to dilute aqueous electrolyte systems, such as those of interest in this study, but more appropriately fits dense Coulomb systems such as ionic- liquid electrolyte systems. We would like to emphasize further the importance of aqueous electrochemistry in electrocatalysis. Electrocatalytic reactions in aqueous media offer an efficient means of converting small molecules (e.g., water, \(\mathrm{H}_2\) , \(\mathrm{O}_2\) , \(\mathrm{CO}_2\) , \(\mathrm{NO_x}\) , etc.) into other chemical species, coupled with renewable electricity storage and release.  
+
+<|ref|>text<|/ref|><|det|>[[112, 432, 886, 627]]<|/det|>
+To the best of our knowledge, there has been no full understanding of the molecular origin of the camel- shaped differential capacitance observed in dilute aqueous electrolyte systems, which we believe warrants the publication of our work in Nature Communications. We further believe that our fundamental understanding of the EDL structure in aqueous electrolytes provides insight into controlling the electrocatalytic reaction that is relevant to modern renewable energy technologies. A more detailed discussion is provided below.  
+
+<|ref|>text<|/ref|><|det|>[[114, 708, 464, 726]]<|/det|>
+Below I list my major critiques of this work:  
+
+<|ref|>text<|/ref|><|det|>[[115, 757, 560, 776]]<|/det|>
+1)- Unwarranted Superlatives and Exaggerated Claims:  
+
+<|ref|>text<|/ref|><|det|>[[115, 808, 728, 826]]<|/det|>
+The paper is replete with unwarranted superlatives. Here are some examples:  
+
+<|ref|>text<|/ref|><|det|>[[113, 858, 620, 876]]<|/det|>
+a)- In the abstract "Unprecedented structural phase transition"
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 105, 884, 230]]<|/det|>
+Two issues here. First, the structural changes and phase transitions near the electrode as a function of potential is reasonably accepted and is not unprecedented. Second, just from a language point of view even if it were not the case, it is not the structural phase transition that is unprecedented, but rather its explanation.  
+
+<|ref|>text<|/ref|><|det|>[[114, 260, 884, 315]]<|/det|>
+I can't go through all of the literature discussing ionic structure and phase transitions near the surface. But as an example here is something from 25 years ago:  
+
+<|ref|>text<|/ref|><|det|>[[114, 344, 884, 398]]<|/det|>
+Blum, L., Dale A. Huckaby, and M. Legault. "Phase transitions at electrode interfaces." Electrochimica acta 41.14 (1996): 2207- 2227.  
+
+<|ref|>text<|/ref|><|det|>[[114, 428, 884, 518]]<|/det|>
+For a modern reference on this topic, please see below: Zhang, Yufan, et al. "Enforced freedom: electric- field- induced declustering of ionic- liquid ions in the electrical double layer." Energy & Environmental Materials 3.3 (2020): 414- 420.  
+
+<|ref|>text<|/ref|><|det|>[[114, 548, 883, 602]]<|/det|>
+This last paper sounds very similar to the claims of the paper under review. Similarly there are many like this in the literature.  
+
+<|ref|>text<|/ref|><|det|>[[113, 632, 885, 899]]<|/det|>
+We agree with the reviewer that the word "unprecedented" was unnecessary and thus have removed it from the manuscript. However, we would like to emphasize that the structural phase transition predicted from our cathodic- polarization simulation is different from the previously suggested ones. For example, the first paper that the reviewer mentioned [Electrochim. Acta, 41, 2207- 2227, 1996] discussed a phase transition in the adsorbate layer of the ions, and the second paper by Kornyshev et al. [Energy Environ. Mater., 3, 414- 420, 2020] described a phase transition from bound cation- anion pairs to free ones in the moderate (or high) ion concentration regime, which cannot be related to the dilute aqueous
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 103, 885, 336]]<|/det|>
+electrolyte case. Based on the comment provided by the reviewer that "I like the simulations, ionic structure change, and especially the Maxwell construction used by the authors. I think it is a decent explanation amongst many that already exist.", we believe that the reviewer fully acknowledges that our work is distinctive from previous works but has a concern regarding our means of expression. We greatly appreciate the thoughtful suggestion and constructive comments provided by the reviewer and have revised our expressions accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[113, 414, 491, 433]]<|/det|>
+b)- Another exaggerated claim is the following:  
+
+<|ref|>text<|/ref|><|det|>[[115, 463, 883, 518]]<|/det|>
+"While the EDL is one of the oldest concepts in electrochemistry1, its significance in controlling electrochemical reactions has been recognized recently2- 7. "  
+
+<|ref|>text<|/ref|><|det|>[[113, 548, 884, 638]]<|/det|>
+References 2- 7 are all published in 2020- 2021. Do the authors mean that no one in the last century ever recognized the importance of EDL for electrochemistry? Or they have something more specific in mind.  
+
+<|ref|>text<|/ref|><|det|>[[113, 668, 883, 723]]<|/det|>
+In continuation of our response in above, we deeply appreciate the constructive comments provided by the reviewer. We clarified the indicated expression as follows:  
+
+<|ref|>text<|/ref|><|det|>[[113, 752, 884, 844]]<|/det|>
+"The EDL is one of the oldest and most fundamental concepts in electrochemistry1,2. As a recent example, the electrochemical carbon dioxide reduction reaction (CO2RR) has been suggested to be controlled by the EDL structure 3- 10,"
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 106, 482, 125]]<|/det|>
+c)- This one is about spectroscopy of the EDL:  
+
+<|ref|>text<|/ref|><|det|>[[115, 155, 884, 245]]<|/det|>
+"Nonetheless, to date, the atomic- level details of the EDL still remain unknown because not only the EDL is spatially concealed between the two bulk phases of solid and liquid, impeding spectroscopic measurements,..."  
+
+<|ref|>text<|/ref|><|det|>[[113, 274, 884, 435]]<|/det|>
+Have the authors not found any optical spectroscopic literature dedicated to the EDL structure that they can so confidently say the spectroscopic measurements are impeded at the EDL? There are hundreds, if not thousands of papers on spectroscopy of the electrode- electrolyte interface, including Raman, and IR vibrational spectroscopy dedicated to understanding ionic structure. A quote like this is inaccurate at best.  
+
+<|ref|>text<|/ref|><|det|>[[113, 463, 884, 590]]<|/det|>
+We agree that the original expressions could have been interpreted as ignoring the importance of spectroscopic dedications given in this field, although that was not our intention. Following the suggestion made by the reviewer, we have revised the indicated text as follows:  
+
+<|ref|>text<|/ref|><|det|>[[113, 617, 884, 742]]<|/det|>
+"Nonetheless, to date, the microscopic structure of the EDL has not been fully resolved not only because the EDL is spatially concealed between the two bulk phases of solid and liquid<sup>11</sup>, but also because the electrochemical signals are highly convoluted by the complex, coupled EDL responses of the multiple components in the electrified interface<sup>12</sup>."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 105, 486, 125]]<|/det|>
+2)- Lack of Comparison with Previous Models:  
+
+<|ref|>text<|/ref|><|det|>[[115, 154, 884, 210]]<|/det|>
+One of the best known models of the double hump capacitance is described by Kornyshev's (and cited heavily in the literature).  
+
+<|ref|>text<|/ref|><|det|>[[114, 239, 884, 364]]<|/det|>
+Kornyshev, Alexei A. "Double- layer in ionic liquids: paradigm change?." (2007): 5545- 5557. Fedorov, Maxim V., Nikolaj Georgi, and Alexei A. Kornyshev. "Double layer in ionic liquids: The nature of the camel shape of capacitance." Electrochemistry Communications 12.2 (2010): 296- 299.  
+
+<|ref|>text<|/ref|><|det|>[[115, 393, 883, 450]]<|/det|>
+Chen, Ming, et al. "On the temperature dependence of the double layer capacitance of ionic liquids." Journal of electroanalytical Chemistry 819 (2018): 347- 358.  
+
+<|ref|>text<|/ref|><|det|>[[113, 478, 884, 640]]<|/det|>
+The figure to the right explains the camel shape of capacitance is from Kornyshev's 2007 paper mentioned above. If this is the case, then how come the authors claim the following in the paper: "We are now thus ready to address the century- year- long question, unresolved since the development of the early EDL theories in the 1900s: what type of molecular- scale change in the EDL structure is responsible for the two humps of the camel- shaped capacitance curve?"21- 30. "  
+
+<|ref|>text<|/ref|><|det|>[[115, 668, 883, 724]]<|/det|>
+Why is the above not referenced and contrasted with their new findings? This sounds very suspicious.  
+
+<|ref|>text<|/ref|><|det|>[[114, 752, 884, 879]]<|/det|>
+The authors could argue against the existing models if they wish, or show similarities and/or differences between the two approaches. However, the authors have circled around citing this well- known (and reasonably successful) model. Kornyshev is cited in passing only [ref 30], without any detailed comparison of the conceptual details. They have not even attempted to compare their
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 104, 884, 230]]<|/det|>
+results with what is known. Avoiding Kornyshev's model is like avoiding an elephant in the room. Just to emphasize, this reviewer is not Kornyshev and has no association with him. I would be very much interested in understanding in what ways Kornyshev's lattice model and this work's phase transition ideas are similar or different.  
+
+<|ref|>text<|/ref|><|det|>[[113, 259, 884, 349]]<|/det|>
+Please note that it is no only Kornyshev who has models for the double hump capacitance. Here is an example from the more recent literature (and see the attached figure explaining the double hump capacitance:  
+
+<|ref|>text<|/ref|><|det|>[[113, 378, 883, 434]]<|/det|>
+Cruz, Carolina, et al. "Electrical double layers close to ionic liquid- solvent demixing." The Journal of Physical Chemistry C 123.3 (2018): 1596- 1601.  
+
+<|ref|>text<|/ref|><|det|>[[112, 463, 886, 904]]<|/det|>
+First, the lack of comparison of our results with those of the Kornyshev model was an oversight on our part. As the reviewer noted below, the contribution of Kornyshev to the EDL field has been so great and his model is highly important in this field. However, we believe that this model (as well as many other models relying on a coarse- grained description of water as a continuum dielectric) has intrinsic limitations in resolving the microscopic origin of capacitance behavior measured in dilute electrolytes. First, the Kornyshev model is a lattice- gas model that does not include explicit treatment of water molecules. Although we do not criticize such a theoretical setting, as it also provides important insight with analytical convenience, this model cannot fully reflect the response of water molecules to the EDL field. The role of water molecules in the EDL becomes important for dilute, that is, water- rich electrolytes. Second, as mentioned previously, the essence of the Kornyshev model is the inclusion of ion saturation behavior in the Gouy- Chapman model, and the emergence of a camel shape is ascribed to ion saturation. Thus, a concentrated electrolyte is required to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 103, 886, 441]]<|/det|>
+manifest a camel shape, and in the dilute limit, the results of this model simply approach those of the Gouy- Chapman model, predicting a U- shaped capacitance curve. Therefore, the lack of explicit water molecules and the requirement of a finite ion concentration make the Kornyshev model suitable for explaining the camel- shaped capacitance measured in the ionic- liquid electrolyte system, rather than that measured in the dilute aqueous electrolyte system. In addition, owing to the lack of explicit water molecular structures in the model, the phase transition concept discussed based on the Kornyshev model is mostly related to the ion structure changes (e.g., from bound ion pairs to free ions [Energy Environ. Mater., 3, 414–420, 2020]), whereas the phase transition predicted from our cathodic charging simulation has direct implications for the orientation response of the water dipoles in the EDL.  
+
+<|ref|>text<|/ref|><|det|>[[113, 468, 885, 523]]<|/det|>
+Following the comment made by the reviewer, we added the following paragraph to compare our results with those of the Kornyshev model on pages 11 and 12 of the revised manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[112, 551, 886, 888]]<|/det|>
+"Notably, some theoretical models have predicted the emergence of a camel shape in the capacitance<sup>34,40–42</sup>; thus, it is useful to compare our approach to the previous model. One of the most recent and elaborate EDL models predicting the camel shape is the Kornyshev model<sup>34,43–47</sup>, which is a lattice- gas model incorporating ion saturation behavior into the Gouy<sup>25</sup>–Chapman<sup>26</sup> theory, where water is coarse- grained as a dielectric. Although our mechanism for the cathodic hump indicates that the key to bistability is orientation polarization of the water molecular dipoles in the EDL, the Kornyshev model ascribes the emergence of the capacitance hump to ion saturation<sup>34</sup>. Thus, a concentrated electrolyte is essential to manifest a camel shape in the Kornyshev model, and in the dilute limit, the results of this model approach those of Gouy<sup>25</sup>–Chapman<sup>26</sup> model. Therefore, the Kornyshev model
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 103, 883, 196]]<|/det|>
+is suitable for explaining the camel- shaped capacitance measured in a dense Coulomb system such as an ionic- liquid electrolyte<sup>34,43–48</sup>, whereas our mechanism explains the camel- shaped capacitance measured in a dilute aqueous electrolyte<sup>15–19</sup>.”  
+
+<|ref|>text<|/ref|><|det|>[[112, 272, 886, 541]]<|/det|>
+As discussed so far, our results are distinct from those of the previous models, which were mostly based on coarse- grained descriptions and/or intended to explain the camel- shape behavior of ionic- liquid systems. In this regard, to the best of our knowledge, no theory has provided molecular- level understanding of the origin of camel- shaped capacitance characteristically measured from a simple system, such as the interface between a planar metal electrode and dilute aqueous electrolyte. However, we agree with the thoughtful comment provided by the reviewer that some expressions, which sound exaggerated, needed to be toned down and clarified. Therefore, we changed the original text reading  
+
+<|ref|>text<|/ref|><|det|>[[113, 568, 884, 688]]<|/det|>
+“We are now thus ready to address the century- year- long question, unresolved since the development of the early EDL theories in the 1900s: what type of molecular- scale change in the EDL structure is responsible for the two humps of the camel- shaped capacitance curve? 21–30”  
+
+<|ref|>text<|/ref|><|det|>[[113, 725, 134, 740]]<|/det|>
+to  
+
+<|ref|>text<|/ref|><|det|>[[113, 772, 884, 898]]<|/det|>
+“Therefore, we are now ready to elucidate the microscopic structural details of the EDL, which have been questioned but not fully resolved since the development of the early EDL theories in the 1900s<sup>2</sup>. In particular, we focus on what type of molecular structural response in the EDL is responsible for the two humps in the camel- shaped capacitance curves that
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 104, 883, 161]]<|/det|>
+have been measured from simple systems, such as the interfaces between planar metal electrodes and dilute aqueous electrolytes<sup>25- 34</sup>."  
+
+<|ref|>text<|/ref|><|det|>[[113, 239, 884, 364]]<|/det|>
+I do like the explanations given the paper under review on the origin of the double hump capacitance. I like the simulations, ionic structure change, and especially the Maxwell construction used by the authors. I think it is a decent explanation amongst many that already exist.  
+
+<|ref|>text<|/ref|><|det|>[[112, 393, 884, 556]]<|/det|>
+We would like to thank the reviewer for recognizing the significance of our work. We sincerely appreciate the efforts made by the reviewer in critiquing our manuscript and providing constructive comments. Thanks to the thoughtful comments provided by the reviewer, our manuscript has been greatly improved and the distinct points of our study have been well clarified.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 105, 884, 195]]<|/det|>
+3)- Relation to \(CO_2\) reduction: The authors efforts in relating this work to \(CO_2\) reduction is appreciated. However, it is done in passing and without much depth. Perhaps it can be done in a separate paper. The explanation in the manuscript is not in depth enough.  
+
+<|ref|>text<|/ref|><|det|>[[112, 224, 886, 593]]<|/det|>
+We appreciate the reviewer for noting this issue. Indeed, after the submission of our manuscript, we became aware of the paper by Koper et al., suggesting the importance of short- range electrostatic interactions between cations and adsorbed \(CO_2\) [Nat. Catal., 4, 654- 662 (2021)]. Thus, we further performed DFT- CES simulations to understand the interaction between the cation and adsorbed \(CO_2\) in a bent form ( \(*CO_2\) ). Our DFT- CES simulation revealed that the cation can coordinate to the \(*CO_2\) and that the coordinating ability of the cation to \(*CO_2\) is weakened when the cation is chelated by the crown ether. Specifically, the coordination number of cation to \(*CO_2\) decreases from 1.0 to 0.3 when the cation is chelated (Figure R1). This finding suggests that the linear dependence of the \(CO_2RR\) activity on the crown ether concentration can also be explained using a mechanism based on a direct cation- \(*CO_2\) interaction.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[152, 110, 890, 496]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 522, 880, 655]]<|/det|>
+<center>Figure R1. Radial distribution function, \(g_{Na - O}(r)\) and its integrated value, \(\int 4\pi r^2 g_{Na - O}(r)dr\) as a function of a distance between \(\mathrm{Na^{+}}\) and O of adsorbed \(\mathrm{CO_2}\) . a, b, \(g_{Na - O}(r)\) for uncomplexed \(\mathrm{Na^{+}}\) (a) and value of \(\int 4\pi r^2 g_{Na - O}(r)dr\) (b). c, d, \(g_{Na - O}(r)\) for 15C5-complexed \(\mathrm{Na^{+}}\) (c) and value of \(\int 4\pi r^2 g_{Na - O}(r)dr\) (d). </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 732, 885, 893]]<|/det|>
+However, as described in the original manuscript, the linear dependence of the \(\mathrm{CO_2RR}\) activity on the crown ether concentration can also be explained in terms of the weakened electric field in the EDL. Thus, we believe that it is more appropriate to address both mechanistic possibilities in the present study than to close the discussion, which will inspire many other researchers to study the mechanism of \(\mathrm{CO_2RR}\) further. Indeed, we are also
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 105, 883, 160]]<|/det|>
+conducting a theoretical- experimental joint study to identify the mechanistic role of cations during \(\mathrm{CO_2RR}\) , which will be presented in a future paper.  
+
+<|ref|>text<|/ref|><|det|>[[113, 190, 883, 245]]<|/det|>
+To address both possibilities based on short- range direct and long- range field- dipole interactions equally, we modified Figure 4c in the revised manuscript as shown in Figure R2.  
+
+<|ref|>image<|/ref|><|det|>[[285, 293, 732, 576]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 612, 884, 701]]<|/det|>
+<center>Figure R2. DFT-CES snapshots showing that uncomplexed \(\mathrm{Na^{+}}\) develops a more direct interaction with the adsorbed \(\mathrm{CO_2}\) than 15C5-complexed \(\mathrm{Na^{+}}\) , forming a compact EDL structure with a stronger field. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 782, 799, 837]]<|/det|>
+We also added Figure R1 into to the supporting information and appended below discussion to page 12 of the revised manuscript:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 103, 886, 370]]<|/det|>
+"Not only the long- range dipole- field interaction, but also the short- range direct interaction of the cation with the adsorbate \(\mathrm{CO_2}\) has been highlighted recently \(^{10,53}\) . Our DFT- CES simulation further revealed that the coordination number of \(\mathrm{Na^{+}}\) to the adsorbed \(\mathrm{CO_2}\) decreases from 1.0 to 0.3 when the cation is complexed with 15C5 (Supplementary Fig. 12). Thus, the decrease in the \(\mathrm{CO_2RR}\) activity can also be explained in terms of the decrease in the coordinating ability of a cation to the adsorbed \(\mathrm{CO_2}\) . In both mechanistic possibilities, our work demonstrates the importance of identifying the EDL structure for controlling the electrocatalytic activity."  
+
+<|ref|>text<|/ref|><|det|>[[113, 448, 884, 575]]<|/det|>
+Minor problems 1)- There are numerous English grammar and style problems in the manuscript. A couple of examples are below. In the abstract: "As based on first principles." "that linearly scales the carbon dioxide reduction activity" 2)- The references section is split by the figures section. 3)- The unit Angstrom is missing in several places.  
+
+<|ref|>text<|/ref|><|det|>[[113, 604, 836, 625]]<|/det|>
+We properly revised the grammatical errors and carefully proof- read the manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 105, 211, 122]]<|/det|>
+Reviewer: 3  
+
+<|ref|>text<|/ref|><|det|>[[112, 153, 885, 420]]<|/det|>
+The report is a study using mainly computational methods and a few experiments, to study the origin of the electric double layer. The system is a Ag111 surface with water/Na+ or water/F- solution. Several observed effects are very clearly reproduced and a credible molecular explanation is presented for the capacitance peaks. The paper is well- written and the conclusions are clearly presented. However, there are some things to address before the conclusions can be seen as verified. Since the conclusions fully rely on a single computational model which is non- standard and therefore not well tested, it is crucial that this model is benchmarked. Below are some specific points to address.  
+
+<|ref|>text<|/ref|><|det|>[[112, 432, 885, 594]]<|/det|>
+We would like to thank the reviewer for the favorable comments and recognition of the significance of our present work. Following the valuable comments provided, we additionally performed an extensive benchmark study of our model, concluding that our findings are not parameter- or model- specific, but rather can be reproduced well in general. A more detailed discussion is provided below.  
+
+<|ref|>text<|/ref|><|det|>[[112, 672, 885, 867]]<|/det|>
+1. There are some questionable arguments in the molecular origin discussion. It is stated that the cause for \(F^{-}\) to adsorb on the surface while \(Na^{+}\) stays further away is due to the smaller hydration energy of anions. In the supporting information the hydration energy for \(F^{-}\) is presented as 115-120 kcal/mol, while \(Na^{+}\) has a value of 80-90 kcal/mol. This is precisely opposite to the argument on line 119. It is also stated that the dispersive energy is larger for \(F^{-}\) than for \(Na^{+}\) , which is likely correct. This should be straight forward to estimate from the Uvdw term.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 103, 885, 300]]<|/det|>
+We agree with the reviewer that the adsorption behavior of \(\mathbf{F}^{- }\) can be ascribed to the fact that its dispersive energy is larger than that of \(\mathrm{Na^{+}}\) . Indeed, the dispersion coefficients (i.e., C6 parameters) and \(\mathrm{F}^{- }\) and \(\mathrm{Na^{+}}\) are 1012 kcal mol \(^{- 1}\mathrm{\AA}^{6}\) and 21 kcal mol \(^{- 1}\mathrm{\AA}^{6}\) , respectively, as determined from real- time time- dependent density functional theory (RT- TDDFT) calculations [J. Chem. Theory Comput. 12, 3603- 3613 (2016)], which were employed in our simulations. Thus, we changed the original text reading  
+
+<|ref|>text<|/ref|><|det|>[[113, 328, 885, 420]]<|/det|>
+"This specific adsorption of anions, proposed by Grahame<sup>24</sup> and demonstrated through various approaches<sup>29,31</sup>, occurs because of the relatively small hydration free energy of anions<sup>32</sup> and their large dispersive attraction toward the electrode<sup>33</sup>."  
+
+<|ref|>text<|/ref|><|det|>[[114, 451, 135, 467]]<|/det|>
+to  
+
+<|ref|>text<|/ref|><|det|>[[113, 497, 884, 589]]<|/det|>
+"This specific adsorption of anions, proposed by Grahame<sup>28</sup> and demonstrated through various approaches<sup>33,35</sup>, occurs because of their large dispersive attraction toward the electrode<sup>36</sup>."  
+
+<|ref|>text<|/ref|><|det|>[[113, 668, 885, 793]]<|/det|>
+2. In the parametrization of the Buckingham potential for water, only the geometry with the oxygen adsorbed is probed. However, in the simulations at the cathode most water molecules point the hydrogen towards the surface. This geometry should also be probed, and should probably be tested with a charged Ag cluster.  
+
+<|ref|>text<|/ref|><|det|>[[113, 822, 884, 878]]<|/det|>
+We would like to thank the reviewer for validating our FF parameters. To respond to the question posed by the reviewer, we refitted our interfacial FF parameters to reproduce both
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 105, 884, 265]]<|/det|>
+the O- and H-head interactions on the Ag surface. The original FF parameters only include Ag- Owater interactions, but new FF parameters include both Ag- Owater and Ag- Hwater interactions using Buckingham potentials (see Table R1), which reproduce the nonlocal vdW-corrected DFT binding energy curves of water to the Ag(111) surface for both O- and H-head orientations (Figure R3).  
+
+<|ref|>table<|/ref|><|det|>[[120, 342, 876, 545]]<|/det|>
+<|ref|>table_caption<|/ref|><|det|>[[147, 551, 874, 570]]<|/det|>
+Table R1. Original and new force-field (FF) parameters for the Buckingham potential.   
+
+<table><tr><td></td><td></td><td>A (kcal mol-1)</td><td>R (Å)</td><td>C6 (kcal mol-1 Å6)</td></tr><tr><td rowspan="2">Original<br>FF<br>parameters</td><td>Ag-Hwater</td><td>0</td><td>1.0</td><td>0</td></tr><tr><td>Ag-Owater</td><td>28888</td><td>0.328</td><td>2009</td></tr><tr><td rowspan="2">New FF<br>parameters</td><td>Ag-Hwater</td><td>3389</td><td>0.343</td><td>355</td></tr><tr><td>Ag-Owater</td><td>17427</td><td>0.348</td><td>1739</td></tr></table>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[153, 110, 963, 528]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[142, 558, 884, 683]]<|/det|>
+<center>Figure R3. DFT-CES benchmark versus QM-level vdW-corrected DFT calculation results. a, b, Binding energy curves for the O-head geometry: binding energy curves (a) and O-head geometry (b). c, d, Binding energy curves for the H-head geometry; binding energy curves (c) and H-head geometry (d). </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 762, 884, 890]]<|/det|>
+Using the new FF parameters, we re- calculated the charging curve (Figure R4a) and differential capacitance curve (Figure R4b). The newly obtained curves show the same characteristic features as the previously obtained curves using the original FF parameters, despite small quantitative changes in the potential values. In addition, the newly obtained
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 103, 885, 265]]<|/det|>
+differential capacitance curve more closely matches the experimental curve in terms of the peak positions and capacitance at the potential at the point of zero charge (EPZC). Thus, we updated all the data in the revised manuscript using the simulation results based on the new FF parameters. The key findings and conclusions remain the same despite some minor quantitative updates.  
+
+<|ref|>image<|/ref|><|det|>[[152, 304, 900, 550]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 580, 884, 636]]<|/det|>
+<center>Figure R4. Comparison of DFT-CES results using original and new FFs. a, Surface charge density, \(\sigma\) , versus electrode potential, \(E\) . b, Differential capacitance, \(C\) , versus \(E\) . </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 715, 885, 876]]<|/det|>
+As requested by the reviewer, we further validated our new FF parameters using a charged Ag cluster. Utilizing neutral and \(- 1\) charged Ag hexamer clusters, we calculated the QM- level binding energy curves of water molecules for O- and H- head orientations. Because of the lack of a nonlocal vdW- corrected functional of vdW- df2, which was used for slab calculations, in the non- periodic DFT codes, we employed the D3 vdW- correction method of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 103, 884, 230]]<|/det|>
+Grimme coupled with the HSE06 functional. As shown in Figure R5, we found that the DFT- CES based on the new FF parameters successfully reproduced the binding energy curves of water molecules to neutral and charged Ag hexamer clusters, both for O- and H- head orientations.  
+
+<|ref|>image<|/ref|><|det|>[[163, 262, 928, 608]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 638, 884, 761]]<|/det|>
+<center>Figure R5. DFT-CES benchmark versus QM-level vdW-corrected DFT calculation results. a, b, c Binding energy curves of water to neutral Ag hexamer (a) and to \(-1\) charged Ag hexamer (b) for O-head geometry (c). d, e, f, Binding energy curves of water to neutral Ag hexamer (d) and to \(-1\) charged Ag hexamer (e) for H-head geometry (f). </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 103, 886, 405]]<|/det|>
+3. It is not clear if the TIP3P Ag(111) interaction is balanced. I cannot find any benchmark of that. Especially when the surface is charged and the hydrogen points to the surface, the interaction in the presented model seems very strong so that the cations are even pushed out of the first layer. It could be correct, but it could as well be an artefact from an model that has a too strong interaction between H and Ag. Since TIP3P has an inflated charge distribution to compensate for the lack of anisotropy and other effects, it could lead to the formation of the silver-hydrogen bond formation at the cathode, which in turn seem to completely outcompete the silver cation interaction. I suggest that another water model is tested to see if the electrolyte structure is the same or if it is changed, to avoid the risk of an artefact due to a too simple water model.  
+
+<|ref|>text<|/ref|><|det|>[[113, 432, 885, 560]]<|/det|>
+We firstly note that the original FF parameters predicted nearly the same binding energy of water to the Ag surface for the H- head configuration but that the separation distance was slightly underestimated (Figure R3). Thus, very strong binding of water to the surface is unlikely to occur even when using the original FF parameters.  
+
+<|ref|>text<|/ref|><|det|>[[113, 587, 884, 712]]<|/det|>
+By using the new FF parameters, which were fitted to reproduce the Ag- H and Ag- O interactions equally, we also found that the first- coordination shell of the cation remained intact even for the highly charged Ag surface case, in agreement with the results obtained using the previous FF parameters (Figure R6).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[191, 115, 830, 522]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 553, 884, 713]]<|/det|>
+<center>Figure R6. DFT-CES simulation results showing the solvation structure of cations and the charge-separation distance, \(d\) , when using the original and new FFs. a, b, Coordination number (CN) of water to the cation (a) and \(d\) versus surface charge density, \(\sigma\) . (b) when using the original FF parameters. c, d, CN of water to the cation (c) and \(d\) versus \(\sigma\) (d) when using the new FF parameters. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 793, 884, 885]]<|/det|>
+To test any possible artefacts due to the use of a simple TIP3P- EW water model, as requested by the reviewer, we employed the TIP4P- EW water model and performed additional calculations. Even after changing the water model from TIP3P- EW to TIP4P- EW, we found
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 104, 883, 160]]<|/det|>
+that the first- coordination shell of the cation remained intact even for the highly charged Ag surface (Figure R7).  
+
+<|ref|>image<|/ref|><|det|>[[191, 201, 830, 399]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 429, 884, 590]]<|/det|>
+<center>Figure R7. DFT-CES simulation results showing the solvation structure of cations and the charge-separation distance, \(d\) , when using the TIP3P-EW and TIP4P-EW water models. a, b, Coordination number (CN) of water to the cation (a) and \(d\) versus surface charge density, \(\sigma\) , (b) when using the TIP3P-EW model (violet) and TIP4P-EW model (red). </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 669, 866, 725]]<|/det|>
+We thus concluded that our finding is not an artefact caused by a specific choice of model parameters, but rather is physically sound.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 103, 886, 300]]<|/det|>
+4. The interpretation that the electric field difference is the determining factor for the difference in activity when crown-ether is added, could be correct but could as well be incorrect. Direct interaction between the oxygen atoms of -COO at the surface could also stabilize the formation of that adduct, and this interaction would also be limited by addition of crown-ether. There are some recent reports that discuss this phenomenon including Nature Catalysis 2021, 4, 654-662 and J. Phys. Chem. C 2020, 124, 41, 22479-22487.  
+
+<|ref|>text<|/ref|><|det|>[[112, 328, 886, 664]]<|/det|>
+We appreciate the reviewer for noting this issue. Indeed, after submitting our manuscript, we became aware of the paper by Koper et al., which suggests the importance of short- range electrostatic interactions between cations and adsorbed \(\mathrm{CO_2}\) [Nat. Catal., 4, 654- 662 (2021)]. Thus, we further performed DFT- CES simulations to understand the interaction between the cation and adsorbed \(\mathrm{CO_2}\) in a bent form \((*\mathrm{CO_2})\) . Our DFT- CES simulation revealed that the cation can coordinate to the \(* \mathrm{CO_2}\) and that the coordinating ability of the cation to \(* \mathrm{CO_2}\) is weakened when the cation is chelated by the crown ether; the coordination number of cation to \(* \mathrm{CO_2}\) decreases from 1.0 to 0.3 when the cation is chelated (Figure R8). This finding suggests that the linear dependence of the \(\mathrm{CO_2RR}\) activity on the crown ether concentration can also be explained using a mechanism based on a direct cation- \(* \mathrm{CO_2}\) interaction.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[153, 110, 900, 500]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 526, 880, 660]]<|/det|>
+<center>Figure R8. Radial distribution function, \(g_{Na - O}(r)\) and its integrated value, \(\int 4\pi r^2 g_{Na - O}(r)dr\) as a function of a distance between \(\mathrm{Na^{+}}\) and O of adsorbed \(\mathrm{CO_2}\) a, b, \(g_{Na - O}(r)\) for uncomplexed \(\mathrm{Na^{+}}\) (a) and value of \(\int 4\pi r^2 g_{Na - O}(r)dr\) (b). c, d, \(g_{Na - O}(r)\) for 15C5-complexed \(\mathrm{Na^{+}}\) (c) and value of \(\int 4\pi r^2 g_{Na - O}(r)dr\) (d). </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 736, 884, 899]]<|/det|>
+However, as described in the original manuscript, the linear dependence of the \(\mathrm{CO_2RR}\) activity on the crown ether concentration can also be explained in terms of the weakened electric field in the EDL. Thus, we believe that it is more appropriate to address both mechanistic possibilities in the present study than to close the discussion, which will inspire many other researchers to study the mechanism of \(\mathrm{CO_2RR}\) further. Indeed, we are also
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 105, 883, 160]]<|/det|>
+conducting a theoretical- experimental joint study to identify the mechanistic role of cations during \(\mathrm{CO_2RR}\) , which will be presented in a future paper.  
+
+<|ref|>text<|/ref|><|det|>[[114, 189, 883, 245]]<|/det|>
+To discuss both possibilities based on short- range direct and long- range field- dipole interactions equally, we modified Figure 4c in the revised manuscript as shown in Figure R9.  
+
+<|ref|>image<|/ref|><|det|>[[285, 291, 732, 576]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 611, 884, 701]]<|/det|>
+<center>Figure R9. DFT-CES snapshots showing that uncomplexed \(\mathrm{Na^{+}}\) develops a more direct interaction with the adsorbed \(\mathrm{CO_2}\) than 15C5-complexed \(\mathrm{Na^{+}}\) , forming a compact EDL structure with a stronger field. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 730, 884, 891]]<|/det|>
+"Not only the long- range dipole- field interaction, but also the short- range direct interaction of the cation with the adsorbate \(\mathrm{CO_2}\) has been highlighted recently \(^{10,53}\) . Our DFT- CES simulation also revealed that the coordination number of \(\mathrm{Na^{+}}\) to the adsorbed \(\mathrm{CO_2}\) decreased from 1.0 to 0.3 when the cation was complexed with 15C5 (Supplementary Fig. 12). Thus, the decrease in the \(\mathrm{CO_2RR}\) activity can also be explained in terms of the decrease in the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 104, 884, 196]]<|/det|>
+coordinating ability of a cation to the adsorbed \(\mathbf{CO}_2\) . In both mechanistic possibilities, our work demonstrates the importance of identifying the EDL structure for controlling the electrocatalytic activity."  
+
+<|ref|>text<|/ref|><|det|>[[114, 274, 870, 400]]<|/det|>
+Overall I believe that this report could provide very interesting and important insight on the catalyst- solvent interface under working conditions. There are some questions on the reliability of the method that needs to be addressed and some discussion that could be improved, but the key points of the paper are of high interest.  
+
+<|ref|>text<|/ref|><|det|>[[113, 428, 884, 556]]<|/det|>
+We would like to thank the reviewer for recognizing the significance of our work. We sincerely appreciate the efforts made by the reviewer in critiquing our manuscript and providing constructive comments. Thanks to the thoughtful comments provided by the reviewer, our manuscript has been greatly improved and clarified.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 90, 362, 106]]<|/det|>
+<b>REVIEWERS' COMMENTS</b>  
+
+<|ref|>text<|/ref|><|det|>[[116, 129, 393, 145]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 167, 854, 204]]<|/det|>
+The authors have addressed most of my major concerns. At this point, their results merits publishing and being shared with the community.  
+
+<|ref|>text<|/ref|><|det|>[[116, 245, 393, 262]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 284, 878, 380]]<|/det|>
+I find the changes and replies to the comments I made (reviewer 3) are very satisfactory. The newly developed model has a much better balance between the H- bound and O- bound conformations, and appears to give even better agreement with the experimental results. The explanation of the F- and Na+ interaction with the surface was also changed to one that made more sense from the data. From my point of view the manuscript is now suitable for publications.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a1ebe21f7f3c67a6280e34d92c1b09fc770427553402a04d944a5aff2adde781/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__a1ebe21f7f3c67a6280e34d92c1b09fc770427553402a04d944a5aff2adde781/supplementary_0_Peer Review File__a1ebe21f7f3c67a6280e34d92c1b09fc770427553402a04d944a5aff2adde781.mmd

@@ -0,0 +1,298 @@
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+Azadmanesh et al. have collected neutron diffraction data on the reduced and oxidized forms of manganese SOD in an attempt to ascertain the protonation states of residues around the active sites in both oxidation states. Neutron diffraction is not frequently used on enzymatic samples, and I consequently find the study to be quite novel on that basis alone.  
+
+The results are highly intriguing and provide insight into how previously studied mutations altered the activity of the enzyme. Unexpectedly, Gln143 appears to be deprotonated in the reduced form of the enzyme. This is highly counter- intuitive since Mn(II)- OH2 groups would normally be more acidic than amides. To their credit, the authors are clearly cognizant of this oddity and provide a clear rationale for why the amide of Gln143 may be more acidic than anticipated.  
+
+My only substantial concern is that the authors' methodology likewise suggests that His 30 may exist as an anionic imidazolate, which is chemically implausible. Could the anionic Gln143 residue likewise be an artifact?  
+
+Once this issue is addressed, the manuscript should be suitable for publication in Nature Communications.  
+
+A. On page 6, the authors describe the interaction between Mn(II) and water as being between a soft acid and a soft base. Although these are softer than Mn(III) and hydroxide, respectively, both are still considered to be hard on the HSAB scale.  
+B. I am concerned that the data also suggest a chemically unrealistic imidazolate anion for His30. I am glad that the authors discount this possibility, but I worry that similarly misleading data may be suggesting an artificial amidate for Gln143.  
+C. I had trouble following the proton and electron relays in Figure 6. An accompanying Chemdraw scheme might elucidate matters.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+In this manuscript, the authors present neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The structures show several very interesting features: The Mn(II) state, shows a OH2 ligand, as expected, but the proton is delivered from the - NH2 group of Gln143, which becomes deprotonated. In one of the two active sites in the protein, Mn(II) becomes 6- coordinate with two OH- ligands. In the oxidized state in one of the two active sites, His30 is a deprotonated imidazolate group. All these are very interesting suggestions and merit publication. However, a few clarifications are needed first.  
+
+1. Several mechanisms for MnSOD have been suggested based on QM calculations. They should be discussed.  
+2. Some discussion of previous crystallographic studies of MnSOD should also be mentioned in the introduction.  
+3. The study in ref. 17 actually is an experimental (X-ray) observation of protonated water in the Mn(II) state.  
+4. From an inorganic perspective, it seems quite unlikely that Mn(II) should have a single H2O ligand in one subunit but suddenly two OH- ligands in the other subunit.  
+5. The authors should better explain what they mean by differential protonation and the difference between Fig 5a and 5b.
+
+<--- Page Split --->
+
+6. Fig 5b should show the omit difference density of the proton on Wat2 also.  
+
+7. To put the reported H-bond covalencies in a perspective, the authors should report the covalency of a normal H2O-HOH H-bond, calculated with the same QM method.  
+
+8. The suggested reaction mechanism in 6 is a bit strange as it ignores the substrates.  
+
+9. Na2S2O4 is not sodium hydrosulfite.  
+
+10. The type of refinement is not fully clear. Did you perform joint refinement of the neutron and X-ray data?  
+
+11. I do not think you used COSMO-RS in the QM calculations, only COSMO.  
+
+12. A convergence criterium of 5E-4 a.u. is very floppy. Most QM programs have closer to 1E-6 a.u., i.e. 500 times stricter.  
+
+13. The difference between Figures S3a and b should be clarified.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+The mechanisms of the concerted proton and electron transfers (CEPTs) of human manganese superoxide dismutase (MnSOD) are provided in the paper. Since authors' most important statement, "a novel mechanism is proposed from the direct observation of glutamine deprotonation, the involvement of Tyr and His with altered pKaS, and the three unusual strong- short hydrogen bonds that change with the oxidation state of the metal Mn", is strongly based on the neutron protein crystallography (NPC) results, the accuracy of the NPC results is decisive. Therefore, there are several questions about the NPC results. When these would be solved, we can admit that NPC could contribute to detect the CEPT of MnSOD well.  
+
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases?  
+
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states?  
+
+3) Why the De22 (Q143) cannot be seen in the Figure 2a?  
+
+4) Line 108-116: Since the Fourier map peak of D2 (WAT1) is rather small, it is dangerous to conclude that this does not belong to the De21 (Q143). The authors have concluded that O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1)the difference of pKa between donor and acceptor atoms should be near zero, (2)the hydrogen bond distance is short, and (3)the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria.  
+
+5) Why is the De1(W123) in Figure 2b rather smaller than De1(W123) in Figure 2a?  
+
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the D8 1 is too large and it cannot be seen on the map, isn't it?  
+
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK.
+
+<--- Page Split --->
+
+Places in the manuscript that were changed are highlighted in light grey.  
+
+Reviewer #1 (Remarks to the Author): Review of Manuscript NCOMMS- 20- 43515  
+
+Azadmanesh et al. have collected neutron diffraction data on the reduced and oxidized forms of manganese SOD in an attempt to ascertain the protonation states of residues around the active sites in both oxidation states. Neutron diffraction is not frequently used on enzymatic samples, and I consequently find the study to be quite novel on that basis alone.  
+
+The results are highly intriguing and provide insight into how previously studied mutations altered the activity of the enzyme. Unexpectedly, Gln143 appears to be deprotonated in the reduced form of the enzyme. This is highly counter- intuitive since Mn(II)- OH2 groups would normally be more acidic than amides. To their credit, the authors are clearly cognizant of this oddity and provide a clear rationale for why the amide of Gln143 may be more acidic than anticipated.  
+
+My only substantial concern is that the authors' methodology likewise suggests that His 30 may exist as an anionic imidazolate, which is chemically implausible. Could the anionic Gln143 residue likewise be an artifact?  
+
+Once this issue is addressed, the manuscript should be suitable for publication in Nature Communications.  
+
+A. On page 6, the authors describe the interaction between Mn(II) and water as being between a soft acid and a soft base. Although these are softer than Mn(III) and hydroxide, respectively, both are still considered to be hard on the HSAB scale. The reviewer is correct in that Mn(II) and water are considered hard on the HSAB scale and our wording was technically incorrect. We have rewritten this portion of the manuscript.  
+
+B. I am concerned that the data also suggest a chemically unrealistic imidazolate anion for His30. I am glad that the authors discount this possibility, but I worry that similarly misleading data may be suggesting an artificial amidate for Gln143. We have completely reanalyzed our data concerning His30 and have included more data into the manuscript based on our new observations. Reviewer #1 comments (and reviewer #3 below) helped us to look at the data with a new critical eye. We have refocused this section on the low barrier hydrogen bond that exists between His30 and Tyr166 across the dimer interface in the reduced state and the apparent altered pKas of this pair of amino acids that is probably important to the enzymatic mechanism. We also included a B-value table in the supplement (Table S4) to help the reader evaluate the data for His 30 in the oxidized state, if they are interested.  
+
+The data concerning Gln143 is without question. Gln143 was identified as an amide anion in the apo MnSOD reduced state due to the presence of the neutron scattering length density. While the peak is indeed small relative to other peaks at the same contour it would not be unexpected if there is indeed movement of the proton between WAT1 and Q143. Simply lowering the contour to 2.3 \(\sigma\) from 2.5 \(\sigma\) puts the density size close to the other 2.5 \(\sigma\) peaks. This has been added to the figure. We realize that we are suggesting unlikely chemistry for the Q143- WAT1 interaction and thus sought validation through QM calculations as demonstrated with Figure 3 and careful investigation of the literature. We believe the culmination of the experimental data, the QM calculations, and past studies noted in the manuscript thus strongly favor presence of the amide anion over the neutral amide.  
+
+C. I had trouble following the proton and electron relays in Figure 6. An accompanying Chemdraw scheme might elucidate matters. We have included Chemdraw summaries of the results as well as revised the final suggested mechanism figure into a Chemdraw scheme. We find this to be a great improvement in clarity, thank you for the suggestion.
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+In this manuscript, the authors present neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The structures show several very interesting features: The Mn(II) state, shows a OH2 ligand, as expected, but the proton is delivered from the - NH2 group of Gln143, which becomes deprotonated. In one of the two active sites in the protein, Mn(II) becomes 6- coordinate with two OH- ligands. In the oxidized state in one of the two active sites, His30 is a deprotonated imidazolate group. All these are very interesting suggestions and merit publication. However, a few clarifications are needed first.  
+
+1. Several mechanisms for MnSOD have been suggested based on QM calculations. They should be discussed.  
+
+2. Some discussion of previous crystallographic studies of MnSOD should also be mentioned in the introduction. For suggestions 1 & 2 from reviewer #2, a paragraph has been added into the introduction that discusses previous crystallographic and QM studies.  
+
+3. The study in ref. 17 actually is an experimental (X-ray) observation of protonated water in the Mn(II) state. This is correct and the text of the manuscript has been edited accordingly.  
+
+4. From an inorganic perspective, it seems quite unlikely that Mn(II) should have a single H2O ligand in one subunit but suddenly two OH- ligands in the other subunit. We agree with Reviewer #2 and also consider Mn(II) with two OH- ligands as implausible. However, the respective active site is not strictly two OH- ligands, but rather one strict OH- ligand and another OH- ligand with partial HOH character from its strong short hydrogen bond (1.4 Å) with the proximal amide proton. This is further supported by the intermediary Mn- O(WAT1) bond distance of 2.1 Å compared to the 2.2 Å Mn-O(WAT1) bond distance of the opposing chain where the data indicates H2O as the identity of WAT1. X-ray structures of MnSOD are known to differ in coordination between chains of the asymmetric unit [1-2].  
+
+We also performed DFT geometry optimizations to assess the plausibility of the six- coordinate Mn(II) seen with the neutron data. Indeed, it is stable and has Mn- O(WAT1) bond distance of 2.1 Å. As a further test, replacing the OH ligand with H2O at the sixth coordinate position causes the H2O molecule to disassociate from Mn(II) during DFT geometry optimizations. These details have been added to the manuscript.  
+
+[1] Borgstahl, G. E. O., Pokross, M., Chehab, R., Sekher, A. & Snell, E. H. Cryo- trapping the six- coordinate, distorted- octahedral active site of manganese superoxide dismutase. J. Mol. Biol. 296, 951- 959, doi:10.1006/jmbi.1999.3506 (2000).  
+
+[2] Porta, J., Vahedi- Faridi, A. & Borgstahl, G. E. O. Structural analysis of peroxide- soaked MnSOD crystals reveals side- on binding of peroxide to active- site manganese. J. Mol. Biol. 399, 377- 384, doi:10.1016/j.jmb.2010.04.031 (2010).  
+
+5. The authors should better explain what they mean by differential protonation and the difference between Fig 5a and 5b. We understand the confusion of the figures. We have rewritten the text to be more explicit in the changes of protonation.  
+
+6. Fig 5b should show the omit difference density of the proton on Wat2 also. Good catch, this has been added.  
+
+7. To put the reported H-bond covalencies in a perspective, the authors should report the covalency of a normal H2O–HOH H-bond, calculated with the same QM method. This is a good suggestion and we have implemented this value into the text, \(\sim 90 / 10\) covalence from the same QM method.  
+
+8. The suggested reaction mechanism in 6 is a bit strange as it ignores the substrates. We agree that this is atypical and may be off-putting. However, we do not have experimental data with the substrates (or products) that may change the protonation states of the active site – though it is certainly in the works. The benefit with the present work is the changes of protonation states that occur not as a consequence of the substrate but rather
+
+<--- Page Split --->
+
+the redox state.  
+
+9. Na2S2O4 is not sodium hydrosulfite. Corrected  
+
+10. The type of refinement is not fully clear. Did you perform joint refinement of the neutron and X-ray data? The final stages of refinement were intentionally done separately from the X-ray data due to the lack of isomorphism caused by the known perturbations that X-rays have on solvent structure and metal redox states which are not present with neutrons. X-rays are substantially reducing to metalloproteins and consequentially make it difficult to obtain a totally oxidized structure. Also, X-rays are known to perturb ordered solvent which are of importance for studying MnSOD. The neutron refinement implemented restraints from the refined X-ray structure with the PHENIX refinement software. These details have been clarified in the methods section along with appropriate references.  
+
+11. I do not think you used COSMO-RS in the QM calculations, only COSMO. The reviewer is correct and the methods sections have been corrected. The appropriate reference has also been added.  
+
+12. A convergence criterium of 5E-4 a.u. is very floppy. Most QM programs have closer to 1E-6 a.u., i.e. 500 times stricter. This was an error and in reality the expected value had been used. The orbital gradient convergence criterion was mistakenly interpreted as the energy convergence criterion. The version of the NWChem QM software used indeed has a default energy convergence criterion of 1E-6 that was unaltered during calculations. See images below. This has been clarified in both the methods section and the supplementary methods section.  
+
+![PLACEHOLDER_4_0]
+  
+
+13. The difference between Figures S3a and b should be clarified. Figures S3a-b have been reworked into the main manuscript into Figure 5 and differences have been clarified, specifically the chain identity is identified and has a color assigned to it.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+The mechanisms of the concerted proton and electron transfers (CEPTs) of human manganese superoxide dismutase (MnSOD) are provided in the paper. Since authors' most important statement, "a novel mechanism is proposed from the direct observation of glutamine deprotonation, the involvement of Tyr and His with altered pKaS, and the three unusual strong-short hydrogen bonds that change with the oxidation state of the metal Mn", is strongly based on the neutron protein crystallography (NPC) results, the accuracy of the NPC results is decisive. Therefore, there are several questions about the NPC results. When these would be solved, we can admit that NPC could contribute to detect the CEPT's of MnSOD well.
+
+<--- Page Split --->
+
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases?  
+
+We understand the lack of clarity here and revised the text of the methods section to be explicit about the pH/pD both during crystallization and data collection.  
+
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states?  
+
+We understand how the topic of chain identity may be puzzling to the reader and have reworked the manuscript to specify chain identity. For example, note chain labels in Fig 1a corresponding to the color of the chain.  
+
+There were differences between 2b and 4c. We attribute this to differences in solvent accessibility as noted in supplementary figure S3. Previous MnSOD structures are known to have differences in coordination among the chains [1- 2]. For the oxidized state, Fig. 2a, both subunits looked the same. These have been clarified.  
+
+[1] Borgstahl, G. E. O., Pokross, M., Chehab, R., Sekher, A. & Snell, E. H. Cryo- trapping the six- coordinate, distorted- octahedral active site of manganese superoxide dismutase. J. Mol. Biol. 296, 951- 959, doi:10.1006/jmbi.1999.3506 (2000).  
+
+[2] Porta, J., Vahedi- Faridi, A. & Borgstahl, G. E. O. Structural analysis of peroxide- soaked MnSOD crystals reveals side- on binding of peroxide to active- site manganese. J. Mol. Biol. 399, 377- 384, doi:10.1016/j.jmb.2010.04.031 (2010).  
+
+3) Why the De22 (Q143) cannot be seen in the Figure 2a ? This has been added!  
+
+4) Line 108-116: Since the Fourier map peak of D 2 (WAT1) is rather small, it is dangerous to conclude that this does not belong to the De21 (Q143). The authors have concluded that O(WAT1)-D 2 (WAT1)-Ne2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1)the difference of pKa between donor and acceptor atoms should be near zero, (2)the hydrogen bond distance is short, and (3)the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D 2 (WAT1)-Ne2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria. While the peak is indeed small relative to other peaks at the same contour it would not be unexpected if there is indeed movement of the proton between WAT1 and Q143. Simply lowering the contour to 2.3 \(\sigma\) from 2.5 \(\sigma\) puts the density size close to the other 2.5 \(\sigma\) peaks. This has been added to the figure. Likewise, one would expect a larger peak if the proton belonged to Q143 and acted as a hydrogen bond donor to WAT1. We realize that we are suggesting unlikely chemistry for the Q143-WAT1 interaction and thus sought validation through QM calculations as demonstrated with Figure 3 and careful investigation of the literature. We believe the culmination of the experimental data, the QM calculations, and past studies noted in the manuscript thus strongly favor presence of the amide anion over the neutral amide.  
+
+Investigation of the literature suggests that the term SSHB is sometimes used interchangeably with the term 'low- barrier hydrogen bond' (LBHB) [1] and thus generates confusion. Also, there seems to be different definitions of the terms among studies. To remediate this, we have decided to explicitly define these terms in the manuscript.  
+
+A SSHB is simply a shorter hydrogen bond that indicates greater strength that may have enzymatic implications [2]. For example, the His, Asp, and Ser of catalytic triads have short hydrogen bonds and stark differences of canonical \(\mathrm{pK}_{\mathrm{a}}\) values but use proton transfers.
+
+<--- Page Split --->
+
+A LBHB is a type of SSHB with the added characteristics of (1) the difference of pKa between donor and acceptor atoms should be near zero and (2) the hydrogen atom is located roughly equidistant from the donor and acceptor atoms [3]. This is the same definition used by a recent 2019 Nature publication [4].  
+
+[1] Cleland WW. Low- barrier hydrogen bonds and enzymatic catalysis. Arch Biochem Biophys. 2000 Oct 1;382(1):1- 5. doi: 10.1006/abbi.2000.2011. PMID: 11051090.  
+
+[2] Remer LC., and Jensen JH. Toward a general theory of hydrogen bond:the short, strong hydrogen bond. J. Phys. Chem. A 2000, 104, 40, 9266- 9275. doi:10.1021/jp002726n  
+
+[3] Perrin, C. L. & Nielson, J. B. "Strong" hydrogen bonds in chemistry and biology. Annu. Rev. Phys. Chem. 48, 511- 544 (1997).  
+
+[4] Dai, S., Funk, LM., von Pappenheim, F.R. et al. Low- barrier hydrogen bonds in enzyme cooperativity. Nature 573, 609- 613 (2019). https://doi.org/10.1038/s41586- 019- 1581- 9  
+
+5) Why is the De1(W123) in Figure 2b rather smaller than De1(W123) in Figure 2a? While the neutron data sets from the reduced state (2b) and the oxidized state (2a) are indeed from the same crystal, the strength of the Fourier map peaks should not be expected to be equal. There are notable differences in data collection statistics that may contribute to the difference in Fourier map peaks. The most apparent examples are differences among the unit cell dimensions, resolution, and I/σ(I) as seen in table S5.  
+
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the Dδ 1 is too large and it cannot be seen on the map, isn't it? The section for His30 has been reworked and does not emphasize the possibility of an imidazolate anion. The section was refocused on the LBHB between His30 and Tyr166 across the dimer interface.  
+
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK. We understand this issue of technical terms and wish for the manuscript to be the most technically sound as possible. We have edited the manuscript accordingly.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The authors have adequately addressed my concerns. In my opinion, the manuscript can be published as is.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+This is the revised version of a manuscript describing neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The authors have made strong improvements and the manuscript can now be accepted after two very minor points have been clarified.  
+
+1. The text says that the H-bond length between Asp159 and Wat1 is 2.1 Å, but Figure 2a says 1.9 Å.  
+
+Line 272: "suggests stability of the six-coordinate complex": What six-coordinate complex? What ox state and what ligands?  
+
+## Reviewer #3 (Remarks to the Author):  
+
+I have re- reviewed the revised version mainly according to my previous questions. I found there still were unclear parts as follows: (Italic: comments and new questions)  
+
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases? --- \(\diamond\) Line 443-448: Normally pD is determined as pD is added 0.4 to the measured pH value with a pH meter. According to the revised version, both pH and pD are the same value. Have you adjusted as pD is the same as pH? How have you done it and was it necessary to adjust that pD and pH are the same value?  
+
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states? --- \(\diamond\) I have understood that in one crystallographic asymmetric unit there are two chains, chain A and chain B. Moreover, in the reduced, the protonation structures between chain A and chain B are not the same. How is the case of oxidized state between chain A and chain B? Is it OK to discuss the CEPTs by using together the structure of chain A and chain B, but not using independently only chain A or chain B?  
+
+3) Why the De22 (Q143) cannot be seen in the Figure 2a? --- \(\diamond\) (OK) 
+4) Line 108-116: Since the Fourier map peak of D2 (WAT1) is rather small, it is dangerous to conclude that this does not belong to the De21 (Q143). The authors have concluded that O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1) the difference of pKa between donor and acceptor atoms should be near zero, (2) the hydrogen bond distance is short, and (3) the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria. --- \(\diamond\) (OK)
+
+<--- Page Split --->
+
+5) Why is the De1(W123) in Figure 2b rather smaller than De1(W123) in Figure 2a?---(OK)  
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the Dö 1 is too large and it cannot be seen on the map, isn't it? ---(OK)  
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK. ---(line 85-86) The sentence 'Mn scatters negatively and therefore lacks density' has no sense. Simply, the neutron scattering length of Mn is negative.
+
+<--- Page Split --->
+
+Places in the manuscript that were changed are highlighted in light grey.  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The authors have adequately addressed my concerns. In my opinion, the manuscript can be published as is. We kindly thank the reviewer for their suggestions that have improved the manuscript.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+This is the revised version of a manuscript describing neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The authors have made strong improvements and the manuscript can now be accepted after two very minor points have been clarified.  
+
+1. The text says that the H-bond length between Asp159 and Wat1 is \(2.1 \mathring{\mathrm{A}}\) , but Figure 2a says \(1.9 \mathring{\mathrm{A}}\) . Line 272: "suggests stability of the six-coordinate complex": What six-coordinate complex? What ox state and what ligands?  
+
+We are thankful to the reviewer for the suggestions in clarification! The text has been corrected to \(1.9 \mathring{\mathrm{A}}\) to correspond with Figure 2a. The sentence of line 272 has been clarified:  
+
+>Line 272: Further validation was sought with DFT geometry optimizations of the observed neutron structure, \(\mathrm{Mn}^{2 + }\) SOD with sixth- coordinate \(^2\mathrm{OH}(\mathrm{OL})\) , and suggests stability of the six- coordinate complex while replacing OL with \(\mathrm{H}_2\mathrm{O}\) causes disassociation into five- coordinate \(\mathrm{Mn}^{2 + }\) SOD.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+I have re- reviewed the revised version mainly according to my previous questions. I found there still were unclear parts as follows: (Italic: comments and new questions)  
+
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases?  
+
+---\(\diamond\) Line 443- 448: Normally pD is determined as pD is added 0.4 to the measured pH value with a pH meter. According to the revised version, both pH and pD are the same value. Have you adjusted as pD is the same as pH? How have you done it and was it necessary to adjust that pD and pH are the same value?  
+
+We thank the reviewer for pointing out the lack of clarity. We have added the explicit mention of measured pH values that led to the pD values discussed.  
+
+>Line 443- 445 now reads: Deuterium exchange of crystals was performed by vapor diffusion in capillaries using deuterated solutions at pH 7.4 that is the equivalent pD value of 7.8. The pD value was calculated from \(\mathrm{pD} = \mathrm{pH}_a\) (apparent reading from pH meter) \(+0.4\) .  
+
+>Line 448 indicates that the pD value of 7.8 was obtained by adding 0.4 to the measured pH value.  
+
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states?  
+
+\(\diamond\) I have understood that in one crystallographic asymmetric unit there are two chains, chain A and chain B. Moreover, in the reduced, the protonation structures between chain A and chain B are not the same. How is the case of oxidized state between chain A and chain B? Is it OK to discuss the CEPTs by using together the structure of chain A and chain B, but not using independently only chain A or chain B ?  
+
+The reviewer makes a good point and we have added the following clarifications.
+
+<--- Page Split --->
+
+For the oxidized MnSOD, there is high structural similarity between Chain A and Chain B. The only change in protonation between the chains is the protonation state of His30. When discussing the oxidized structure in the conclusions, we have clarified:  
+
+>Line 367- 371: For the oxidized resting state that is described by both chains of the \(\mathrm{Mn^{3 + }SOD}\) neutron structure (Fig. 6a), the proton bridging His30 and Tyr166 appears to be moving and this suggests the possibility of Tyr166 alternating between an ionized or protonated form and a deprotonated or protonated \(\mathrm{N}^{2 + }(\mathrm{His30})\) . \(\mathrm{N}^{0\mathrm{i}}(\mathrm{His30})\) is also observed to be both deprotonated or protonated (Fig. 5c- d). For simplicity, only one protonation form of His30 and Tyr34 is shown (Fig 6a).  
+
+>Line 375: For the reduced resting state described by chain B of \(\mathrm{Mn^{2 + }SOD}\) (Fig. 6b),  
+
+>Line 379- 380: The third active site state is a six- coordinate \(\mathrm{Mn^{2 + }}\) with \(\mathrm{^3OH}\) bound opposite Asp159 (Fig. 6c) described by chain A of \(\mathrm{Mn^{2 + }SOD}\) .  
+
+Otherwise, we have interpreted the two chains for \(\mathrm{Mn^{3 + }SOD}\) as one structural state. This is the reason why we only included figures of one chain for Fig. 2 and Fig. 4 in the manuscript while placing figures of the other chain in the supplemental. For the figures legends, we have clarified:  
+
+>Legend 2a: Both panels are for chain B. Only one chain is shown due to high structural similarities, see Fig. S1 for chain A.  
+
+>Legend 4a: For the oxidized state, only one chain is shown due to high structural similarities, see Fig. S3 for chain B.  
+
+For the reduced MnSOD, the chains have stark differences, especially the coordination state of \(\mathrm{Mn^{2 + }}\) . Other differences include protonation changes at Tyr34, Gln143, and WAT1. For these reasons, we have interpreted the chains as being in two different states and is the apparent reason discussion of CPETs considers chain A and chain B two independent structures.  
+
+3) Why the \(\mathrm{D\epsilon 22}\) (Q143) cannot be seen in the Figure 2a ? - \(\mathrm{\Delta}\) (OK)  
+
+The reviewer has indicated that this comment was addressed.  
+
+4) Line 108-116: Since the Fourier map peak of D2(WAT1) is rather small, it is dangerous to conclude that this does not belong to the Dε21 (Q143). The authors have concluded that O(WAT1)-D2(WAT1)-Nε2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1)the difference of pKa between donor and acceptor atoms should be near zero, (2)the hydrogen bond distance is short, and (3)the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D2(WAT1)-Nε2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria. - \(\mathrm{\Delta}\) (OK)  
+
+The reviewer has indicated that this comment was addressed.  
+
+5) Why is the Dε1(W123) in Figure 2b rather smaller than Dε1(W123) in Figure 2a? - \(\mathrm{\Delta}\) (OK)  
+
+The reviewer has indicated that this comment was addressed.  
+
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the Dδ 1 is too large and it cannot be seen on the map, isn't it? - \(\mathrm{\Delta}\) (OK)  
+
+The reviewer has indicated that this comment was addressed.  
+
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK.  
+
+- \(\mathrm{\Delta}\) (line 85-86) The sentence 'Mn scatters negatively and therefore lacks density' has no sense. Simply, the neutron scattering length of Mn is negative.
+
+<--- Page Split --->
+
+The reviewer is correct and we thank them for the assistance in technicalities. Line 85- 85 reads "Of note, the neutron scattering length of Mn is negative."
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+## Reviewer #3 (Remarks to the Author):  
+
+The authors have adequately solved my questions. In my opinion, the manuscript can be published as is.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a1ebe21f7f3c67a6280e34d92c1b09fc770427553402a04d944a5aff2adde781/supplementary_0_Peer Review File__a1ebe21f7f3c67a6280e34d92c1b09fc770427553402a04d944a5aff2adde781_det.mmd

@@ -0,0 +1,414 @@
+<|ref|>sub_title<|/ref|><|det|>[[119, 86, 378, 103]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 140, 485, 156]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 169, 878, 227]]<|/det|>
+Azadmanesh et al. have collected neutron diffraction data on the reduced and oxidized forms of manganese SOD in an attempt to ascertain the protonation states of residues around the active sites in both oxidation states. Neutron diffraction is not frequently used on enzymatic samples, and I consequently find the study to be quite novel on that basis alone.  
+
+<|ref|>text<|/ref|><|det|>[[119, 240, 870, 311]]<|/det|>
+The results are highly intriguing and provide insight into how previously studied mutations altered the activity of the enzyme. Unexpectedly, Gln143 appears to be deprotonated in the reduced form of the enzyme. This is highly counter- intuitive since Mn(II)- OH2 groups would normally be more acidic than amides. To their credit, the authors are clearly cognizant of this oddity and provide a clear rationale for why the amide of Gln143 may be more acidic than anticipated.  
+
+<|ref|>text<|/ref|><|det|>[[119, 323, 866, 366]]<|/det|>
+My only substantial concern is that the authors' methodology likewise suggests that His 30 may exist as an anionic imidazolate, which is chemically implausible. Could the anionic Gln143 residue likewise be an artifact?  
+
+<|ref|>text<|/ref|><|det|>[[119, 379, 794, 408]]<|/det|>
+Once this issue is addressed, the manuscript should be suitable for publication in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[117, 421, 875, 536]]<|/det|>
+A. On page 6, the authors describe the interaction between Mn(II) and water as being between a soft acid and a soft base. Although these are softer than Mn(III) and hydroxide, respectively, both are still considered to be hard on the HSAB scale.  
+B. I am concerned that the data also suggest a chemically unrealistic imidazolate anion for His30. I am glad that the authors discount this possibility, but I worry that similarly misleading data may be suggesting an artificial amidate for Gln143.  
+C. I had trouble following the proton and electron relays in Figure 6. An accompanying Chemdraw scheme might elucidate matters.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 624, 485, 640]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 653, 857, 752]]<|/det|>
+In this manuscript, the authors present neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The structures show several very interesting features: The Mn(II) state, shows a OH2 ligand, as expected, but the proton is delivered from the - NH2 group of Gln143, which becomes deprotonated. In one of the two active sites in the protein, Mn(II) becomes 6- coordinate with two OH- ligands. In the oxidized state in one of the two active sites, His30 is a deprotonated imidazolate group. All these are very interesting suggestions and merit publication. However, a few clarifications are needed first.  
+
+<|ref|>text<|/ref|><|det|>[[115, 764, 875, 905]]<|/det|>
+1. Several mechanisms for MnSOD have been suggested based on QM calculations. They should be discussed.  
+2. Some discussion of previous crystallographic studies of MnSOD should also be mentioned in the introduction.  
+3. The study in ref. 17 actually is an experimental (X-ray) observation of protonated water in the Mn(II) state.  
+4. From an inorganic perspective, it seems quite unlikely that Mn(II) should have a single H2O ligand in one subunit but suddenly two OH- ligands in the other subunit.  
+5. The authors should better explain what they mean by differential protonation and the difference between Fig 5a and 5b.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 710, 98]]<|/det|>
+6. Fig 5b should show the omit difference density of the proton on Wat2 also.  
+
+<|ref|>text<|/ref|><|det|>[[118, 98, 805, 126]]<|/det|>
+7. To put the reported H-bond covalencies in a perspective, the authors should report the covalency of a normal H2O-HOH H-bond, calculated with the same QM method.  
+
+<|ref|>text<|/ref|><|det|>[[118, 126, 780, 140]]<|/det|>
+8. The suggested reaction mechanism in 6 is a bit strange as it ignores the substrates.  
+
+<|ref|>text<|/ref|><|det|>[[118, 140, 420, 154]]<|/det|>
+9. Na2S2O4 is not sodium hydrosulfite.  
+
+<|ref|>text<|/ref|><|det|>[[118, 154, 861, 182]]<|/det|>
+10. The type of refinement is not fully clear. Did you perform joint refinement of the neutron and X-ray data?  
+
+<|ref|>text<|/ref|><|det|>[[118, 183, 702, 197]]<|/det|>
+11. I do not think you used COSMO-RS in the QM calculations, only COSMO.  
+
+<|ref|>text<|/ref|><|det|>[[118, 197, 848, 225]]<|/det|>
+12. A convergence criterium of 5E-4 a.u. is very floppy. Most QM programs have closer to 1E-6 a.u., i.e. 500 times stricter.  
+
+<|ref|>text<|/ref|><|det|>[[118, 225, 620, 239]]<|/det|>
+13. The difference between Figures S3a and b should be clarified.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 312, 485, 328]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 341, 875, 456]]<|/det|>
+The mechanisms of the concerted proton and electron transfers (CEPTs) of human manganese superoxide dismutase (MnSOD) are provided in the paper. Since authors' most important statement, "a novel mechanism is proposed from the direct observation of glutamine deprotonation, the involvement of Tyr and His with altered pKaS, and the three unusual strong- short hydrogen bonds that change with the oxidation state of the metal Mn", is strongly based on the neutron protein crystallography (NPC) results, the accuracy of the NPC results is decisive. Therefore, there are several questions about the NPC results. When these would be solved, we can admit that NPC could contribute to detect the CEPT of MnSOD well.  
+
+<|ref|>text<|/ref|><|det|>[[118, 456, 878, 511]]<|/det|>
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases?  
+
+<|ref|>text<|/ref|><|det|>[[118, 511, 878, 567]]<|/det|>
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states?  
+
+<|ref|>text<|/ref|><|det|>[[118, 567, 570, 581]]<|/det|>
+3) Why the De22 (Q143) cannot be seen in the Figure 2a?  
+
+<|ref|>text<|/ref|><|det|>[[118, 581, 880, 696]]<|/det|>
+4) Line 108-116: Since the Fourier map peak of D2 (WAT1) is rather small, it is dangerous to conclude that this does not belong to the De21 (Q143). The authors have concluded that O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1)the difference of pKa between donor and acceptor atoms should be near zero, (2)the hydrogen bond distance is short, and (3)the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria.  
+
+<|ref|>text<|/ref|><|det|>[[118, 696, 760, 710]]<|/det|>
+5) Why is the De1(W123) in Figure 2b rather smaller than De1(W123) in Figure 2a?  
+
+<|ref|>text<|/ref|><|det|>[[118, 710, 878, 765]]<|/det|>
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the D8 1 is too large and it cannot be seen on the map, isn't it?  
+
+<|ref|>text<|/ref|><|det|>[[118, 767, 878, 880]]<|/det|>
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 83, 625, 100]]<|/det|>
+Places in the manuscript that were changed are highlighted in light grey.  
+
+<|ref|>text<|/ref|><|det|>[[56, 110, 410, 143]]<|/det|>
+Reviewer #1 (Remarks to the Author): Review of Manuscript NCOMMS- 20- 43515  
+
+<|ref|>text<|/ref|><|det|>[[56, 161, 933, 230]]<|/det|>
+Azadmanesh et al. have collected neutron diffraction data on the reduced and oxidized forms of manganese SOD in an attempt to ascertain the protonation states of residues around the active sites in both oxidation states. Neutron diffraction is not frequently used on enzymatic samples, and I consequently find the study to be quite novel on that basis alone.  
+
+<|ref|>text<|/ref|><|det|>[[56, 246, 930, 333]]<|/det|>
+The results are highly intriguing and provide insight into how previously studied mutations altered the activity of the enzyme. Unexpectedly, Gln143 appears to be deprotonated in the reduced form of the enzyme. This is highly counter- intuitive since Mn(II)- OH2 groups would normally be more acidic than amides. To their credit, the authors are clearly cognizant of this oddity and provide a clear rationale for why the amide of Gln143 may be more acidic than anticipated.  
+
+<|ref|>text<|/ref|><|det|>[[56, 350, 936, 385]]<|/det|>
+My only substantial concern is that the authors' methodology likewise suggests that His 30 may exist as an anionic imidazolate, which is chemically implausible. Could the anionic Gln143 residue likewise be an artifact?  
+
+<|ref|>text<|/ref|><|det|>[[60, 393, 895, 411]]<|/det|>
+Once this issue is addressed, the manuscript should be suitable for publication in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[56, 428, 933, 497]]<|/det|>
+A. On page 6, the authors describe the interaction between Mn(II) and water as being between a soft acid and a soft base. Although these are softer than Mn(III) and hydroxide, respectively, both are still considered to be hard on the HSAB scale. The reviewer is correct in that Mn(II) and water are considered hard on the HSAB scale and our wording was technically incorrect. We have rewritten this portion of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[56, 505, 939, 660]]<|/det|>
+B. I am concerned that the data also suggest a chemically unrealistic imidazolate anion for His30. I am glad that the authors discount this possibility, but I worry that similarly misleading data may be suggesting an artificial amidate for Gln143. We have completely reanalyzed our data concerning His30 and have included more data into the manuscript based on our new observations. Reviewer #1 comments (and reviewer #3 below) helped us to look at the data with a new critical eye. We have refocused this section on the low barrier hydrogen bond that exists between His30 and Tyr166 across the dimer interface in the reduced state and the apparent altered pKas of this pair of amino acids that is probably important to the enzymatic mechanism. We also included a B-value table in the supplement (Table S4) to help the reader evaluate the data for His 30 in the oxidized state, if they are interested.  
+
+<|ref|>text<|/ref|><|det|>[[56, 670, 940, 824]]<|/det|>
+The data concerning Gln143 is without question. Gln143 was identified as an amide anion in the apo MnSOD reduced state due to the presence of the neutron scattering length density. While the peak is indeed small relative to other peaks at the same contour it would not be unexpected if there is indeed movement of the proton between WAT1 and Q143. Simply lowering the contour to 2.3 \(\sigma\) from 2.5 \(\sigma\) puts the density size close to the other 2.5 \(\sigma\) peaks. This has been added to the figure. We realize that we are suggesting unlikely chemistry for the Q143- WAT1 interaction and thus sought validation through QM calculations as demonstrated with Figure 3 and careful investigation of the literature. We believe the culmination of the experimental data, the QM calculations, and past studies noted in the manuscript thus strongly favor presence of the amide anion over the neutral amide.  
+
+<|ref|>text<|/ref|><|det|>[[56, 833, 933, 903]]<|/det|>
+C. I had trouble following the proton and electron relays in Figure 6. An accompanying Chemdraw scheme might elucidate matters. We have included Chemdraw summaries of the results as well as revised the final suggested mechanism figure into a Chemdraw scheme. We find this to be a great improvement in clarity, thank you for the suggestion.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 100, 360, 116]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[56, 118, 937, 220]]<|/det|>
+In this manuscript, the authors present neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The structures show several very interesting features: The Mn(II) state, shows a OH2 ligand, as expected, but the proton is delivered from the - NH2 group of Gln143, which becomes deprotonated. In one of the two active sites in the protein, Mn(II) becomes 6- coordinate with two OH- ligands. In the oxidized state in one of the two active sites, His30 is a deprotonated imidazolate group. All these are very interesting suggestions and merit publication. However, a few clarifications are needed first.  
+
+<|ref|>text<|/ref|><|det|>[[60, 220, 930, 238]]<|/det|>
+1. Several mechanisms for MnSOD have been suggested based on QM calculations. They should be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[56, 246, 910, 298]]<|/det|>
+2. Some discussion of previous crystallographic studies of MnSOD should also be mentioned in the introduction. For suggestions 1 & 2 from reviewer #2, a paragraph has been added into the introduction that discusses previous crystallographic and QM studies.  
+
+<|ref|>text<|/ref|><|det|>[[56, 308, 922, 342]]<|/det|>
+3. The study in ref. 17 actually is an experimental (X-ray) observation of protonated water in the Mn(II) state. This is correct and the text of the manuscript has been edited accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[56, 351, 934, 489]]<|/det|>
+4. From an inorganic perspective, it seems quite unlikely that Mn(II) should have a single H2O ligand in one subunit but suddenly two OH- ligands in the other subunit. We agree with Reviewer #2 and also consider Mn(II) with two OH- ligands as implausible. However, the respective active site is not strictly two OH- ligands, but rather one strict OH- ligand and another OH- ligand with partial HOH character from its strong short hydrogen bond (1.4 Å) with the proximal amide proton. This is further supported by the intermediary Mn- O(WAT1) bond distance of 2.1 Å compared to the 2.2 Å Mn-O(WAT1) bond distance of the opposing chain where the data indicates H2O as the identity of WAT1. X-ray structures of MnSOD are known to differ in coordination between chains of the asymmetric unit [1-2].  
+
+<|ref|>text<|/ref|><|det|>[[56, 498, 905, 567]]<|/det|>
+We also performed DFT geometry optimizations to assess the plausibility of the six- coordinate Mn(II) seen with the neutron data. Indeed, it is stable and has Mn- O(WAT1) bond distance of 2.1 Å. As a further test, replacing the OH ligand with H2O at the sixth coordinate position causes the H2O molecule to disassociate from Mn(II) during DFT geometry optimizations. These details have been added to the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[56, 576, 916, 628]]<|/det|>
+[1] Borgstahl, G. E. O., Pokross, M., Chehab, R., Sekher, A. & Snell, E. H. Cryo- trapping the six- coordinate, distorted- octahedral active site of manganese superoxide dismutase. J. Mol. Biol. 296, 951- 959, doi:10.1006/jmbi.1999.3506 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[56, 637, 919, 689]]<|/det|>
+[2] Porta, J., Vahedi- Faridi, A. & Borgstahl, G. E. O. Structural analysis of peroxide- soaked MnSOD crystals reveals side- on binding of peroxide to active- site manganese. J. Mol. Biol. 399, 377- 384, doi:10.1016/j.jmb.2010.04.031 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[56, 698, 930, 750]]<|/det|>
+5. The authors should better explain what they mean by differential protonation and the difference between Fig 5a and 5b. We understand the confusion of the figures. We have rewritten the text to be more explicit in the changes of protonation.  
+
+<|ref|>text<|/ref|><|det|>[[58, 758, 927, 776]]<|/det|>
+6. Fig 5b should show the omit difference density of the proton on Wat2 also. Good catch, this has been added.  
+
+<|ref|>text<|/ref|><|det|>[[56, 785, 936, 837]]<|/det|>
+7. To put the reported H-bond covalencies in a perspective, the authors should report the covalency of a normal H2O–HOH H-bond, calculated with the same QM method. This is a good suggestion and we have implemented this value into the text, \(\sim 90 / 10\) covalence from the same QM method.  
+
+<|ref|>text<|/ref|><|det|>[[56, 846, 928, 915]]<|/det|>
+8. The suggested reaction mechanism in 6 is a bit strange as it ignores the substrates. We agree that this is atypical and may be off-putting. However, we do not have experimental data with the substrates (or products) that may change the protonation states of the active site – though it is certainly in the works. The benefit with the present work is the changes of protonation states that occur not as a consequence of the substrate but rather
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 84, 177, 98]]<|/det|>
+the redox state.  
+
+<|ref|>text<|/ref|><|det|>[[56, 100, 455, 117]]<|/det|>
+9. Na2S2O4 is not sodium hydrosulfite. Corrected  
+
+<|ref|>text<|/ref|><|det|>[[55, 125, 934, 264]]<|/det|>
+10. The type of refinement is not fully clear. Did you perform joint refinement of the neutron and X-ray data? The final stages of refinement were intentionally done separately from the X-ray data due to the lack of isomorphism caused by the known perturbations that X-rays have on solvent structure and metal redox states which are not present with neutrons. X-rays are substantially reducing to metalloproteins and consequentially make it difficult to obtain a totally oxidized structure. Also, X-rays are known to perturb ordered solvent which are of importance for studying MnSOD. The neutron refinement implemented restraints from the refined X-ray structure with the PHENIX refinement software. These details have been clarified in the methods section along with appropriate references.  
+
+<|ref|>text<|/ref|><|det|>[[56, 273, 933, 308]]<|/det|>
+11. I do not think you used COSMO-RS in the QM calculations, only COSMO. The reviewer is correct and the methods sections have been corrected. The appropriate reference has also been added.  
+
+<|ref|>text<|/ref|><|det|>[[55, 316, 920, 420]]<|/det|>
+12. A convergence criterium of 5E-4 a.u. is very floppy. Most QM programs have closer to 1E-6 a.u., i.e. 500 times stricter. This was an error and in reality the expected value had been used. The orbital gradient convergence criterion was mistakenly interpreted as the energy convergence criterion. The version of the NWChem QM software used indeed has a default energy convergence criterion of 1E-6 that was unaltered during calculations. See images below. This has been clarified in both the methods section and the supplementary methods section.  
+
+<|ref|>image<|/ref|><|det|>[[164, 427, 815, 644]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[56, 650, 920, 703]]<|/det|>
+13. The difference between Figures S3a and b should be clarified. Figures S3a-b have been reworked into the main manuscript into Figure 5 and differences have been clarified, specifically the chain identity is identified and has a color assigned to it.  
+
+<|ref|>sub_title<|/ref|><|det|>[[56, 730, 361, 746]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[55, 748, 940, 867]]<|/det|>
+The mechanisms of the concerted proton and electron transfers (CEPTs) of human manganese superoxide dismutase (MnSOD) are provided in the paper. Since authors' most important statement, "a novel mechanism is proposed from the direct observation of glutamine deprotonation, the involvement of Tyr and His with altered pKaS, and the three unusual strong-short hydrogen bonds that change with the oxidation state of the metal Mn", is strongly based on the neutron protein crystallography (NPC) results, the accuracy of the NPC results is decisive. Therefore, there are several questions about the NPC results. When these would be solved, we can admit that NPC could contribute to detect the CEPT's of MnSOD well.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 82, 925, 135]]<|/det|>
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases?  
+
+<|ref|>text<|/ref|><|det|>[[56, 143, 937, 178]]<|/det|>
+We understand the lack of clarity here and revised the text of the methods section to be explicit about the pH/pD both during crystallization and data collection.  
+
+<|ref|>text<|/ref|><|det|>[[56, 187, 939, 240]]<|/det|>
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states?  
+
+<|ref|>text<|/ref|><|det|>[[56, 249, 936, 283]]<|/det|>
+We understand how the topic of chain identity may be puzzling to the reader and have reworked the manuscript to specify chain identity. For example, note chain labels in Fig 1a corresponding to the color of the chain.  
+
+<|ref|>text<|/ref|><|det|>[[56, 292, 936, 344]]<|/det|>
+There were differences between 2b and 4c. We attribute this to differences in solvent accessibility as noted in supplementary figure S3. Previous MnSOD structures are known to have differences in coordination among the chains [1- 2]. For the oxidized state, Fig. 2a, both subunits looked the same. These have been clarified.  
+
+<|ref|>text<|/ref|><|det|>[[56, 353, 919, 404]]<|/det|>
+[1] Borgstahl, G. E. O., Pokross, M., Chehab, R., Sekher, A. & Snell, E. H. Cryo- trapping the six- coordinate, distorted- octahedral active site of manganese superoxide dismutase. J. Mol. Biol. 296, 951- 959, doi:10.1006/jmbi.1999.3506 (2000).  
+
+<|ref|>text<|/ref|><|det|>[[56, 413, 920, 465]]<|/det|>
+[2] Porta, J., Vahedi- Faridi, A. & Borgstahl, G. E. O. Structural analysis of peroxide- soaked MnSOD crystals reveals side- on binding of peroxide to active- site manganese. J. Mol. Biol. 399, 377- 384, doi:10.1016/j.jmb.2010.04.031 (2010).  
+
+<|ref|>text<|/ref|><|det|>[[56, 474, 689, 491]]<|/det|>
+3) Why the De22 (Q143) cannot be seen in the Figure 2a ? This has been added!  
+
+<|ref|>text<|/ref|><|det|>[[55, 501, 940, 766]]<|/det|>
+4) Line 108-116: Since the Fourier map peak of D 2 (WAT1) is rather small, it is dangerous to conclude that this does not belong to the De21 (Q143). The authors have concluded that O(WAT1)-D 2 (WAT1)-Ne2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1)the difference of pKa between donor and acceptor atoms should be near zero, (2)the hydrogen bond distance is short, and (3)the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D 2 (WAT1)-Ne2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria. While the peak is indeed small relative to other peaks at the same contour it would not be unexpected if there is indeed movement of the proton between WAT1 and Q143. Simply lowering the contour to 2.3 \(\sigma\) from 2.5 \(\sigma\) puts the density size close to the other 2.5 \(\sigma\) peaks. This has been added to the figure. Likewise, one would expect a larger peak if the proton belonged to Q143 and acted as a hydrogen bond donor to WAT1. We realize that we are suggesting unlikely chemistry for the Q143-WAT1 interaction and thus sought validation through QM calculations as demonstrated with Figure 3 and careful investigation of the literature. We believe the culmination of the experimental data, the QM calculations, and past studies noted in the manuscript thus strongly favor presence of the amide anion over the neutral amide.  
+
+<|ref|>text<|/ref|><|det|>[[56, 775, 916, 844]]<|/det|>
+Investigation of the literature suggests that the term SSHB is sometimes used interchangeably with the term 'low- barrier hydrogen bond' (LBHB) [1] and thus generates confusion. Also, there seems to be different definitions of the terms among studies. To remediate this, we have decided to explicitly define these terms in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[56, 853, 937, 905]]<|/det|>
+A SSHB is simply a shorter hydrogen bond that indicates greater strength that may have enzymatic implications [2]. For example, the His, Asp, and Ser of catalytic triads have short hydrogen bonds and stark differences of canonical \(\mathrm{pK}_{\mathrm{a}}\) values but use proton transfers.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 82, 940, 136]]<|/det|>
+A LBHB is a type of SSHB with the added characteristics of (1) the difference of pKa between donor and acceptor atoms should be near zero and (2) the hydrogen atom is located roughly equidistant from the donor and acceptor atoms [3]. This is the same definition used by a recent 2019 Nature publication [4].  
+
+<|ref|>text<|/ref|><|det|>[[55, 143, 891, 179]]<|/det|>
+[1] Cleland WW. Low- barrier hydrogen bonds and enzymatic catalysis. Arch Biochem Biophys. 2000 Oct 1;382(1):1- 5. doi: 10.1006/abbi.2000.2011. PMID: 11051090.  
+
+<|ref|>text<|/ref|><|det|>[[55, 187, 920, 222]]<|/det|>
+[2] Remer LC., and Jensen JH. Toward a general theory of hydrogen bond:the short, strong hydrogen bond. J. Phys. Chem. A 2000, 104, 40, 9266- 9275. doi:10.1021/jp002726n  
+
+<|ref|>text<|/ref|><|det|>[[55, 231, 866, 266]]<|/det|>
+[3] Perrin, C. L. & Nielson, J. B. "Strong" hydrogen bonds in chemistry and biology. Annu. Rev. Phys. Chem. 48, 511- 544 (1997).  
+
+<|ref|>text<|/ref|><|det|>[[55, 274, 783, 310]]<|/det|>
+[4] Dai, S., Funk, LM., von Pappenheim, F.R. et al. Low- barrier hydrogen bonds in enzyme cooperativity. Nature 573, 609- 613 (2019). https://doi.org/10.1038/s41586- 019- 1581- 9  
+
+<|ref|>text<|/ref|><|det|>[[55, 317, 934, 405]]<|/det|>
+5) Why is the De1(W123) in Figure 2b rather smaller than De1(W123) in Figure 2a? While the neutron data sets from the reduced state (2b) and the oxidized state (2a) are indeed from the same crystal, the strength of the Fourier map peaks should not be expected to be equal. There are notable differences in data collection statistics that may contribute to the difference in Fourier map peaks. The most apparent examples are differences among the unit cell dimensions, resolution, and I/σ(I) as seen in table S5.  
+
+<|ref|>text<|/ref|><|det|>[[55, 412, 939, 519]]<|/det|>
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the Dδ 1 is too large and it cannot be seen on the map, isn't it? The section for His30 has been reworked and does not emphasize the possibility of an imidazolate anion. The section was refocused on the LBHB between His30 and Tyr166 across the dimer interface.  
+
+<|ref|>text<|/ref|><|det|>[[55, 527, 936, 685]]<|/det|>
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK. We understand this issue of technical terms and wish for the manuscript to be the most technically sound as possible. We have edited the manuscript accordingly.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 84, 357, 101]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 138, 448, 152]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 166, 825, 195]]<|/det|>
+The authors have adequately addressed my concerns. In my opinion, the manuscript can be published as is.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 269, 448, 284]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 297, 855, 340]]<|/det|>
+This is the revised version of a manuscript describing neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The authors have made strong improvements and the manuscript can now be accepted after two very minor points have been clarified.  
+
+<|ref|>text<|/ref|><|det|>[[118, 353, 875, 381]]<|/det|>
+1. The text says that the H-bond length between Asp159 and Wat1 is 2.1 Å, but Figure 2a says 1.9 Å.  
+
+<|ref|>text<|/ref|><|det|>[[118, 381, 864, 410]]<|/det|>
+Line 272: "suggests stability of the six-coordinate complex": What six-coordinate complex? What ox state and what ligands?  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 498, 448, 512]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 526, 860, 555]]<|/det|>
+I have re- reviewed the revised version mainly according to my previous questions. I found there still were unclear parts as follows: (Italic: comments and new questions)  
+
+<|ref|>text<|/ref|><|det|>[[117, 568, 878, 666]]<|/det|>
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases? --- \(\diamond\) Line 443-448: Normally pD is determined as pD is added 0.4 to the measured pH value with a pH meter. According to the revised version, both pH and pD are the same value. Have you adjusted as pD is the same as pH? How have you done it and was it necessary to adjust that pD and pH are the same value?  
+
+<|ref|>text<|/ref|><|det|>[[117, 666, 878, 780]]<|/det|>
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states? --- \(\diamond\) I have understood that in one crystallographic asymmetric unit there are two chains, chain A and chain B. Moreover, in the reduced, the protonation structures between chain A and chain B are not the same. How is the case of oxidized state between chain A and chain B? Is it OK to discuss the CEPTs by using together the structure of chain A and chain B, but not using independently only chain A or chain B?  
+
+<|ref|>text<|/ref|><|det|>[[117, 780, 879, 908]]<|/det|>
+3) Why the De22 (Q143) cannot be seen in the Figure 2a? --- \(\diamond\) (OK) 
+4) Line 108-116: Since the Fourier map peak of D2 (WAT1) is rather small, it is dangerous to conclude that this does not belong to the De21 (Q143). The authors have concluded that O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1) the difference of pKa between donor and acceptor atoms should be near zero, (2) the hydrogen bond distance is short, and (3) the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D2 (WAT1)-Ne2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria. --- \(\diamond\) (OK)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 878, 295]]<|/det|>
+5) Why is the De1(W123) in Figure 2b rather smaller than De1(W123) in Figure 2a?---(OK)  
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the Dö 1 is too large and it cannot be seen on the map, isn't it? ---(OK)  
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK. ---(line 85-86) The sentence 'Mn scatters negatively and therefore lacks density' has no sense. Simply, the neutron scattering length of Mn is negative.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 83, 625, 100]]<|/det|>
+Places in the manuscript that were changed are highlighted in light grey.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 109, 353, 125]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[56, 138, 829, 175]]<|/det|>
+The authors have adequately addressed my concerns. In my opinion, the manuscript can be published as is. We kindly thank the reviewer for their suggestions that have improved the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 203, 353, 219]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[56, 233, 928, 277]]<|/det|>
+This is the revised version of a manuscript describing neutron structures of Mn superoxide dismutase for both the Mn(II) and Mn(III) states. The authors have made strong improvements and the manuscript can now be accepted after two very minor points have been clarified.  
+
+<|ref|>text<|/ref|><|det|>[[56, 290, 900, 336]]<|/det|>
+1. The text says that the H-bond length between Asp159 and Wat1 is \(2.1 \mathring{\mathrm{A}}\) , but Figure 2a says \(1.9 \mathring{\mathrm{A}}\) . Line 272: "suggests stability of the six-coordinate complex": What six-coordinate complex? What ox state and what ligands?  
+
+<|ref|>text<|/ref|><|det|>[[56, 355, 912, 386]]<|/det|>
+We are thankful to the reviewer for the suggestions in clarification! The text has been corrected to \(1.9 \mathring{\mathrm{A}}\) to correspond with Figure 2a. The sentence of line 272 has been clarified:  
+
+<|ref|>text<|/ref|><|det|>[[56, 392, 936, 437]]<|/det|>
+>Line 272: Further validation was sought with DFT geometry optimizations of the observed neutron structure, \(\mathrm{Mn}^{2 + }\) SOD with sixth- coordinate \(^2\mathrm{OH}(\mathrm{OL})\) , and suggests stability of the six- coordinate complex while replacing OL with \(\mathrm{H}_2\mathrm{O}\) causes disassociation into five- coordinate \(\mathrm{Mn}^{2 + }\) SOD.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 466, 353, 481]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[56, 494, 936, 525]]<|/det|>
+I have re- reviewed the revised version mainly according to my previous questions. I found there still were unclear parts as follows: (Italic: comments and new questions)  
+
+<|ref|>text<|/ref|><|det|>[[56, 538, 940, 584]]<|/det|>
+1) When discussing the pKa of amino acids, the pH value of the sample is very important. It seems that crystallization was performed at physiological pH, but it is necessary to describe the pH (or pD) value of the protein solution under the crystallization. What was the actual pH(or pD) in both oxidized and reduced cases?  
+
+<|ref|>text<|/ref|><|det|>[[56, 589, 936, 635]]<|/det|>
+---\(\diamond\) Line 443- 448: Normally pD is determined as pD is added 0.4 to the measured pH value with a pH meter. According to the revised version, both pH and pD are the same value. Have you adjusted as pD is the same as pH? How have you done it and was it necessary to adjust that pD and pH are the same value?  
+
+<|ref|>text<|/ref|><|det|>[[56, 641, 936, 672]]<|/det|>
+We thank the reviewer for pointing out the lack of clarity. We have added the explicit mention of measured pH values that led to the pD values discussed.  
+
+<|ref|>text<|/ref|><|det|>[[56, 678, 875, 723]]<|/det|>
+>Line 443- 445 now reads: Deuterium exchange of crystals was performed by vapor diffusion in capillaries using deuterated solutions at pH 7.4 that is the equivalent pD value of 7.8. The pD value was calculated from \(\mathrm{pD} = \mathrm{pH}_a\) (apparent reading from pH meter) \(+0.4\) .  
+
+<|ref|>text<|/ref|><|det|>[[56, 728, 770, 745]]<|/det|>
+>Line 448 indicates that the pD value of 7.8 was obtained by adding 0.4 to the measured pH value.  
+
+<|ref|>text<|/ref|><|det|>[[56, 764, 930, 810]]<|/det|>
+2) There are two subunits in the crystallographic asymmetric unit. In case of the reduced state, did the structure shown in Figure 2b and the structure shown in Figure 4c coexist? If so, why? And, in case of the oxidized state does only the structure shown in Figure 2a exist? If so, why were there such differences between the two states?  
+
+<|ref|>text<|/ref|><|det|>[[56, 816, 936, 876]]<|/det|>
+\(\diamond\) I have understood that in one crystallographic asymmetric unit there are two chains, chain A and chain B. Moreover, in the reduced, the protonation structures between chain A and chain B are not the same. How is the case of oxidized state between chain A and chain B? Is it OK to discuss the CEPTs by using together the structure of chain A and chain B, but not using independently only chain A or chain B ?  
+
+<|ref|>text<|/ref|><|det|>[[56, 881, 644, 897]]<|/det|>
+The reviewer makes a good point and we have added the following clarifications.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 82, 860, 126]]<|/det|>
+For the oxidized MnSOD, there is high structural similarity between Chain A and Chain B. The only change in protonation between the chains is the protonation state of His30. When discussing the oxidized structure in the conclusions, we have clarified:  
+
+<|ref|>text<|/ref|><|det|>[[56, 132, 936, 207]]<|/det|>
+>Line 367- 371: For the oxidized resting state that is described by both chains of the \(\mathrm{Mn^{3 + }SOD}\) neutron structure (Fig. 6a), the proton bridging His30 and Tyr166 appears to be moving and this suggests the possibility of Tyr166 alternating between an ionized or protonated form and a deprotonated or protonated \(\mathrm{N}^{2 + }(\mathrm{His30})\) . \(\mathrm{N}^{0\mathrm{i}}(\mathrm{His30})\) is also observed to be both deprotonated or protonated (Fig. 5c- d). For simplicity, only one protonation form of His30 and Tyr34 is shown (Fig 6a).  
+
+<|ref|>text<|/ref|><|det|>[[56, 212, 670, 229]]<|/det|>
+>Line 375: For the reduced resting state described by chain B of \(\mathrm{Mn^{2 + }SOD}\) (Fig. 6b),  
+
+<|ref|>text<|/ref|><|det|>[[56, 234, 925, 265]]<|/det|>
+>Line 379- 380: The third active site state is a six- coordinate \(\mathrm{Mn^{2 + }}\) with \(\mathrm{^3OH}\) bound opposite Asp159 (Fig. 6c) described by chain A of \(\mathrm{Mn^{2 + }SOD}\) .  
+
+<|ref|>text<|/ref|><|det|>[[56, 270, 890, 316]]<|/det|>
+Otherwise, we have interpreted the two chains for \(\mathrm{Mn^{3 + }SOD}\) as one structural state. This is the reason why we only included figures of one chain for Fig. 2 and Fig. 4 in the manuscript while placing figures of the other chain in the supplemental. For the figures legends, we have clarified:  
+
+<|ref|>text<|/ref|><|det|>[[56, 321, 936, 352]]<|/det|>
+>Legend 2a: Both panels are for chain B. Only one chain is shown due to high structural similarities, see Fig. S1 for chain A.  
+
+<|ref|>text<|/ref|><|det|>[[56, 357, 914, 374]]<|/det|>
+>Legend 4a: For the oxidized state, only one chain is shown due to high structural similarities, see Fig. S3 for chain B.  
+
+<|ref|>text<|/ref|><|det|>[[56, 378, 925, 439]]<|/det|>
+For the reduced MnSOD, the chains have stark differences, especially the coordination state of \(\mathrm{Mn^{2 + }}\) . Other differences include protonation changes at Tyr34, Gln143, and WAT1. For these reasons, we have interpreted the chains as being in two different states and is the apparent reason discussion of CPETs considers chain A and chain B two independent structures.  
+
+<|ref|>text<|/ref|><|det|>[[56, 460, 548, 476]]<|/det|>
+3) Why the \(\mathrm{D\epsilon 22}\) (Q143) cannot be seen in the Figure 2a ? - \(\mathrm{\Delta}\) (OK)  
+
+<|ref|>text<|/ref|><|det|>[[56, 482, 498, 497]]<|/det|>
+The reviewer has indicated that this comment was addressed.  
+
+<|ref|>text<|/ref|><|det|>[[56, 517, 930, 608]]<|/det|>
+4) Line 108-116: Since the Fourier map peak of D2(WAT1) is rather small, it is dangerous to conclude that this does not belong to the Dε21 (Q143). The authors have concluded that O(WAT1)-D2(WAT1)-Nε2(Gln143) is a short-strong hydrogen bond (SSHB). The criteria whether it is a SSHB or not are that (1)the difference of pKa between donor and acceptor atoms should be near zero, (2)the hydrogen bond distance is short, and (3)the hydrogen atom is located roughly equidistant from the donor and acceptor atoms. O(WAT1)-D2(WAT1)-Nε2(Gln143) is a simple hydrogen bond, isn't it? It is necessary to consider whether the other SSHBs appeared in the paper also fit the above mentioned criteria. - \(\mathrm{\Delta}\) (OK)  
+
+<|ref|>text<|/ref|><|det|>[[56, 613, 496, 628]]<|/det|>
+The reviewer has indicated that this comment was addressed.  
+
+<|ref|>text<|/ref|><|det|>[[56, 635, 737, 652]]<|/det|>
+5) Why is the Dε1(W123) in Figure 2b rather smaller than Dε1(W123) in Figure 2a? - \(\mathrm{\Delta}\) (OK)  
+
+<|ref|>text<|/ref|><|det|>[[56, 657, 496, 672]]<|/det|>
+The reviewer has indicated that this comment was addressed.  
+
+<|ref|>text<|/ref|><|det|>[[56, 679, 937, 748]]<|/det|>
+6) Line 285-286: It is a little bit dangerous to conclude His30 has an imidazolate anion. It is often observed that a histidine on the surface of a protein looks like an imidazolate anion simply because the B-factor of a hydrogen atom is too large and the hydrogen atom cannot be seen on the Fourier map. The B-factor of the Dδ 1 is too large and it cannot be seen on the map, isn't it? - \(\mathrm{\Delta}\) (OK)  
+
+<|ref|>text<|/ref|><|det|>[[56, 753, 496, 768]]<|/det|>
+The reviewer has indicated that this comment was addressed.  
+
+<|ref|>text<|/ref|><|det|>[[56, 774, 936, 878]]<|/det|>
+7) The Authors use the word "nuclear density" for the peak on the Fourier contour map so often. It's a matter of technical terms, and strictly speaking, they should use a "neutron scattering length (NSL) density", but not a "nuclear density". The unit of the neutron scattering power is "neutron scattering length" and the peak on the Fourier contour map obtained by neutron diffraction Bragg reflections should be called as "neutron scattering length (NSL) density", but not "nuclear density". For example, the neutron scattering length of Mn is negative, such as -0.373 x 10-12 cm. So, the peak of Mn on the Fourier contour map must be obtained as negative value. The "negative nuclear density" has physically no sense, but "negative NSL density" is just OK.  
+
+<|ref|>text<|/ref|><|det|>[[56, 884, 901, 915]]<|/det|>
+- \(\mathrm{\Delta}\) (line 85-86) The sentence 'Mn scatters negatively and therefore lacks density' has no sense. Simply, the neutron scattering length of Mn is negative.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 83, 616, 118]]<|/det|>
+The reviewer is correct and we thank them for the assistance in technicalities. Line 85- 85 reads "Of note, the neutron scattering length of Mn is negative."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 84, 377, 101]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 138, 449, 153]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 167, 878, 195]]<|/det|>
+The authors have adequately solved my questions. In my opinion, the manuscript can be published as is.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a211576ac4ea643b0147d034b6c6b5d4b856451c2af07c558428c4642ea7e8c9/images_list.json

@@ -0,0 +1,85 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_13.jpg",
+    "caption": "Fig. A: Re-analysis of panels G, H in Supplementary Fig. 13.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_3.jpg",
+    "caption": "Supplementary Figure 3: Goodness of fit. Panels show posterior predictive checks for fits to \\(Ct = 40\\) (A-D) and \\(Ct = 33\\) (E-H) data, broken down by chicken type and recruitment group. Each panel shows posterior mean (bars) and 95% C.I. (grey error bars) for the numbers of chickens becoming positive (filled) at different stages of the experiment or remaining susceptible (hatched). Black dots and error bars denote data and 95% C.I. computed under",
+    "footnote": [],
+    "bbox": [
+      [
+        120,
+        620,
+        879,
+        840
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_6.jpg",
+    "caption": "Supplementary Fig. 6: Basic reproductive number and AIV persistence. (A) Posterior distribution for the basic reproductive number \\(R_{0}\\) for fits to \\(Ct = 40\\) (red) and \\(Ct = 33\\) (teal). The latter is calculated as the basic reproductive number for a continuous time SEEIRR model with a constant rate of chicken removal \\(\\nu\\) , where \\(\\nu\\) is set to the inverse mean length of stay. Combining previous theoretical results yields the expression \\(R_{0} = \\sum_{b = BR,BY}\\left(\\frac{2\\sigma_{b}}{2\\sigma_{b} + \\nu}\\right)^{2}\\mu +\\nu \\frac{N_{b}}{N}\\left(D'Onofrio, Mathematical Biosciences, 2002; Diekman et Al., Journal of the Royal Society Interface, 2010; Champredon et Al., SIAM Journal on Applied Mathematics, 2018), where \\(N_{b}\\) is the mean number of chickens of type \\(b\\) present in the LBM, and is estimated through simulations, and \\(N = \\sum_{b}N_{b}\\) is the total number of chickens introduced daily. (B) Posterior distribution for the ratio \\(\\beta /\\mu\\) , which represents the basic reproductive number in a closed population \\((\\nu = 0)\\) of constant size \\(N\\) . (C) Posterior probability of AIV persistence as a function of \\(R_{0}\\) . This is measured as the proportion of 2000 simulations where at least one latent or infectious chicken is observed at \\(t = 50\\) days, assuming that all chickens entering the market after \\(t = 20\\) days are susceptible. Results in A,B are based on 5000 posterior samples.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        240,
+        877,
+        388
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_5.jpg",
+    "caption": "Supplementary Figure 5: Relative importance of external introductions and local transmission. (A) Violin plots denote posterior distributions for the fraction of initially susceptible chickens that are infected locally and the fraction of those that become infectious before being sold. (B) Posterior distribution for the fraction of LBM-acquired infections that are caused by other chickens that became infected within the LBM, as opposed to externally-introduced infections. Results are based on 20000 simulations from 2000 independent samples from the posterior distribution. (C) Mean number of new cases \\((R_{t})\\) caused by chickens infected at time \\(t\\) . (D) Mean generation time versus the timing of primary infection \\(t\\) . The generation time is defined as the delay between the infection of a primary and a secondary case. Statistics in C,D were obtained by tracking transmission pairs in an individual-based version of our model. Panels E,F mirror C,D but assume that all chickens entering the market after 30 days are susceptible (i.e. in absence of external introductions). We consider all primary cases infected between \\(t = 45\\) and \\(t = 48\\) days. For externally-introduced infections, the infection time was set to the time of introduction. Daily shipments of chickens are denoted with vertical lines. Results in C-F based on 2000 simulations using the same number of independent samples from the posterior distribution.",
+    "footnote": [],
+    "bbox": [
+      [
+        140,
+        105,
+        861,
+        590
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. B: Summary of estimates of \\(\\theta^{-1}\\) from the literature.",
+    "footnote": [],
+    "bbox": [
+      [
+        235,
+        88,
+        760,
+        365
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_14.jpg",
+    "caption": "Supplementary Figure 14: Synergistic effects of combining interventions. This figure considers a simultaneous reduction of the maximum length of stay of chickens to \\(T_{m}\\) and the probability of prior exposure \\(\\rho_{c,b}\\) to \\((1 - r)\\rho_{c,b}\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        210,
+        312,
+        789,
+        867
+      ]
+    ],
+    "page_idx": 11
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a211576ac4ea643b0147d034b6c6b5d4b856451c2af07c558428c4642ea7e8c9/supplementary_0_Peer Review File__a211576ac4ea643b0147d034b6c6b5d4b856451c2af07c558428c4642ea7e8c9.mmd

@@ -0,0 +1,397 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Modelling the transmission dynamics of H9N2 avian influenza viruses in a live bird market  
+
+![](images/Supplementary_Figure_13.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+This paper analyses live bird market data on the transmission of H9N2 (low pathogenicity) avian influenza, fitting a 13 parameter model to field data from an LBM in Cambodia.  
+
+The paper is broadly speaking clearly written for a modelling heavy paper. It clearly goes through model parameters and discusses in detail the results of interventions from the fitted model. The supplementary information contains considerable detail on the model fit itself and justification for priors. The illustration of fit to simulated data is welcome. On a related point, while the fit to simulated data is helpful, it does show that some parameters are poorly recovered in the posterior. this may not be important in terms of the overall paper conclusions but the authors need to show this. One thing that might help, is to compare the results of interventions for the original simulated parameters, and the results using the recovered posteriors. If key outcomes are insensitive to this difference, it provides some confidence that those differences posteriors are not important. Unfortunately it is at best partial confidence of course (not accounting for differences between the model and the real world processes behind the data of course) but would be an important step.  
+
+Another thing makes the paper difficult to evaluate - nowhere did I see the actual fit of the model to the data; we have some evidence that we have the best fitted model, but we don't have much evidence on how good the best fit is.  
+
+In addition, a discussion of the fundemental drivers of the transmission dynamics (what is R0? R(t)? the generation time of transmission? What proportion of susceptibles are infected in the first, second or third generation)? What are the initial conditions and how much of infection is due to introduction rather than within market transmission. These are important because they help us to understand the "why's" of the interventions.  
+
+A broader point is that the motivation behind the analysis seems to sit a bit between two stools. High prevalence of H9N2 in markets and the issue of control may be important for one of two reasons - first, if it results in substantial zoonotic infection, or if it is important for maintaining the circulation of the virus. However, H9N2, even though it may contribute to the generation of zoonotic forms of flu, is itself not a zoonosis. In this case, the prevalence in the market is less important than the role the market has in the circulation of H9N2 - and this can only be evaluated by considering onward transmission from the market. The authors need to make a better case for why their analysis helps us to understand or control that circulation of virus. This is discussed in the introduction (Paragraph starting line 49) but this at best touches upon the circulation question. There is more detail about trading networks in their other papers (e.g. Moyen et al 2021) that they reference, so it would be useful to contextualise the analysis of this LBM in that study (e.g. is the LBM an important source of birds for mobile traders?).  
+
+A few minor points follow:  
+
+line 41: "especially"  
+
+line 81: this reference is a preprint which is useful to have but it would be better if, the interim has it been submitted for publication and/or accepted.  
+
+line 97: I struggled a bit with the definition of control and intervention chickens. Its clearer when you look at the companion preprint (Kohnle et al) but a bit more here would be useful.  
+
+line 123: this is a striking result - are these all infected in one generation due to a high prevalence of infection of birds entering the market? Or is there another reason?  
+
+line 139: why "tentative"  
+
+line 170 onwards. As discussed, it seems that in the model, environmental contamination only differs from direct transmission by allowing for a longer term persistence once infected birds are
+
+<--- Page Split --->
+
+gone. Otherwise the mechanism of transmission (and impact on transmission) is the same. Kim et al 2018 which the authors cite suggest that there is a strong variability in prevalence across stalls and slaughter areas of the market - is this likely to make a difference here in terms of the way contamination vs direct transmission would work?  
+
+line 220 onwards - are there any references that would help us to understand how difficult it would be to sanitise market? A brief description of what this would entail would be helpful.  
+
+line 227: there is a useful point here in terms of the recovery of infection - and it would therefore be important to gain a better understanding of why - referring to my earlier general point, is it because, for example the R value is very high? Or is it because the incoming infection prevalence is high.  
+
+line 342: That the two scenarios of transmission yield similar result is reassuring but as noted earlier, I suspect this may at least partially be because the dynamics in the model really aren't all that different. While the authors cannot be expected to do everything in a single paper, I think some discussion of the implications of model similarity is worthwhile.  
+
+line 361: "we believe that our main results ... are not affected ..." - why?  
+
+line 381: "with similar conditions" - which are the key conditions that are important?  
+
+S.92 onwards. after all long discussion of the effects of environmental conditions on environmental survival, on line S.103 the authors then reduce the difference to one of temperature only.  
+
+supp fig 7 - as discussed a the beginning of this review, really useful to have this but more details on the parameter sets - were they chosen to be relevant to the posteriors for the real data (e.g. drawn from the posterior distributions)? Also, the mismatch in scenario 5 is indeed striking and it would be good to have confidence it isn't important.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors present an interesting study modelling the transmission dynamics of avian influenza with parameters fit to empirical data from a live bird market. I find the model proposed to be reasonable; the statistical analysis appropriate; and the limitations of the approach to be unambiguously communicated. Generally, I find the manuscript to be uncommonly clearly written.  
+
+Regarding the presentation of the model, I would encourage the authors to briefly state (perhaps any additional analysis is beyond the scope of this manuscript) how they would expect compartment topology to modify the parameter estimates or otherwise invest more time in the model description to motivate why SEEIRR is clearly the most appropriate framework.  
+
+The authors describe several predicted epidemiological differences between exotic broilers and backyard chickens including:  
+
+"From our model's output, we found a shorter latent period in exotic broiler compared to backyard chickens (Fig. 1B,C), lasting an average of 5.3 hours for exotic broiler, and 1 days for backyard chickens."  
+
+and  
+
+"in the case of exotic broilers, most chickens with prior exposure to H9N2 were either infectious or latent, with only a minor proportion of them being immune (Fig. 1E). In contrast, most previously- exposed backyard chickens were immune to H9N2 (Fig. 1F). Our results thus tentatively suggest that prior infection occurs close to marketing age for broilers, whereas in backyard chickens it may occur further in the past, which is consistent with the latter being raised for a longer time compared to broilers."
+
+<--- Page Split --->
+
+I encourage the authors to comment on whether they believe these findings are entirely consistent with variable patterns of prior exposure for these two populations or if the data presented suggest that there may be some unrecognized genetic determinants of susceptibility. Additionally I understand that to support the second passage quoted above, there has been an underlying assumption made that prior infection confers long lasting immunity (relative to the life cycle of a market chicken). If this is correct, I encourage the authors to explicitly state this assumption and perhaps introduce some additional background information regarding the market chicken life cycle so that the unfamiliar reader, like myself, may have some expectation of the maximum number of infections for one individual.  
+
+The authors consider both direct and environmental transmission, writing, "Ienv(t) accumulates due to shedding from infectious chickens and decays progressively at rate \(\Theta\) . Here we consider three values of \(\Theta\) , namely \(\Theta - 1 = 10\) , 3, 1days, corresponding to slow, intermediate and fast decay, respectively. These values are based on actual estimates from the scientific literature and capture a broad range of environmental conditions." I encourage the authors to briefly describe the physical nature of environmental transmission in a live bird market for readers, like myself, who are more familiar with influenza transmission dynamics in the human population which I expect substantially differ.  
+
+I find one of the more striking findings reported in the manuscript to be that while, "reduced length of stay and reduced probability of prior exposure, proved to be only modestly effective. Combining both interventions proved considerably more effective at reducing transmission compared to individual measures" I encourage the authors to clarify in the text if this effect is in fact nonadditive and if so to include some description of the expected mechanism behind the nonadditivity.  
+
+In conclusion, the authors present an interesting and well written study which I believe will be of broad interest.  
+
+Sincerely, Nash Rochman (invited 12/5/23, returned 12/19/23)  
+
+Reviewer #3 (Remarks to the Author):  
+
+The manuscript titled "Modelling the transmission dynamics of H9N2 avian influenza viruses in a live bird market" aims to understand the transmission dynamics of H9N2 avian influenza viruses in live bird markets (LBMs) in Bangladesh. The study uses a mechanistic transmission model, fitted to field data collected in a specific LBM, to simulate the spread of the virus among poultry.  
+
+The study presents significant findings on H9N2 transmission in LBMs, highlighting a rapid infection rate among susceptible chickens, often within a day. The virus's latent period is notably short, particularly in broiler chickens, sometimes as brief as 5.3 hours, facilitating its swift spread. It also uncovers considerable differences in exposure and immunity between exotic broiler and backyard chickens. Its unique approach, combining field data with mathematical modelling, and the robustness of its findings, underscored by sensitivity analyses and Bayesian methods, mark it as particularly relevant for areas where H9N2 is prevalent, guiding interventions to control its spread. Thus, this research offers crucial insights into the epidemiology of H9N2 in LBMs, vital for public health and the poultry industry.  
+
+I thoroughly enjoyed reading this manuscript and believe it is already of high quality. However, I suggest incorporating discussions on several interesting points. Firstly, the study's results are drawn from a single LBM in Bangladesh. It would be beneficial to explore how these findings might apply to other LBMs with varying dynamics, bird populations, and environments. Additionally, a more in- depth examination of environmental factors, like temperature, humidity, and market hygiene, is crucial as they significantly influence H9N2's survival and spread. A comparison with other avian influenza strains could also offer a broader perspective, highlighting differences and
+
+<--- Page Split --->
+
+similarities in transmission dynamics. Finally, it would be interesting to consider whether these findings could be applicable to livestock markets other than those for birds and discuss the generalizability of such high infection rates in group settings, dependent on the virus.
+
+<--- Page Split --->
+
+## General information  
+
+General informationFirst of all, we thank the three anonymous reviewers for their assessments and helpful comments. Please find our detailed replies to these comments below. We also provided a copy of the revised manuscript and supplementary material where all the main changes have been highlighted in red/blue.  
+
+## Reviewer 1  
+
+This paper analyses live bird market data on the transmission of H9N2 (low pathogenicity) avian influenza, fitting a 13 parameter model to field data from an LBM in Cambodia.  
+
+The paper is broadly speaking clearly written for a modelling heavy paper. It clearly goes through model parameters and discusses in detail the results of interventions from the fitted model. The supplementary information contains considerable detail on the model fit itself and justification for priors. The illustration of fit to simulated data is welcome. On a related point, while the fit to simulated data is helpful, it does show that some parameters are poorly recovered in the posterior. this may not be important in terms of the overall paper conclusions but the authors need to show this. One thing that might help, is to compare the results of interventions for the original simulated parameters, and the results using the recovered posteriors. If key outcomes are insensitive to this difference, it provides some confidence that those differences posteriors are not important. Unfortunately it is at best partial confidence of course (not accounting for differences between the model and the real world processes behind the data of course) but would be an important step.  
+
+We are pleased that the reviewer appreciated our work and we thank them for their comments.  
+
+The reviewer raises an important point, and we agree that further work is required to assess to what extent the discrepancies between "true" and recovered parameters affect other key findings. Following the advice from the reviewer, we assessed the effectiveness of individual interventions displayed in Fig. 2 using original parameters and recovered posteriors from the scenarios reported in Supplementary Fig. 7 (now 9). These results are displayed in the new Supplementary Fig. 13 (attached below). Each column in the latter figure represents a different scenario (i.e. simulated dataset), while rows correspond to distinct interventions. We find an excellent agreement between simulations using original (dots) and recovered (bars) parameters in terms of intervention effectiveness in the case of:  
+
+Shorter length of stay (first row) - Shorter length of stay + Control of chickens entering the market (third row) Vaccination (last row)  
+
+However, we also underestimate the effectiveness of reducing past exposure in chickens entering the market in scenarios 2 and 3 (second row, second and third columns). As a
+
+<--- Page Split --->
+
+reminder, the latter intervention reduces the probability that an incoming chicken is not susceptible by a factor \(1 - r\) . Upon further investigation, we discover that such discrepancy disappears completely in scenario 2 and is considerably reduced in scenario 3 after setting the probability \(P_{c,b,X}^{(0)}\) that an incoming bulk (i.e. control group) chicken of type \(b\) is in compartment \(X\) to the corresponding true value. These results are shown in Fig. A below alongside this reply for completeness.  
+
+Overall, these findings suggest that our main results about the effectiveness of modelled interventions are robust to modest discrepancies in posterior estimates. The new supplementary figure is referenced in the Results section:  
+
+"Estimates of intervention effectiveness presented in Fig. 2 are robust to mismatches between inferred and original parameters in simulated scenarios (Supplementary Fig. 13)."  
+
+![](images/Supplementary_Figure_3.jpg)
+  
+
+Supplementary Figure 13: Estimating effectiveness of interventions in artificial scenarios. This figure compares the effectiveness of veterinary public health interventions calculated using recovered (bars) and original (dots) parameters in the 5 artificial scenarios considered in Supplementary Fig. 9 (from left to right). We consider the same interventions presented in Fig. 2 in the main manuscript: shorter length of stay (top row), control of infected chickens entering the LBM (second row) and vaccination (bottom row). The third row corresponds to the simultaneous implementation of the first two interventions. Moderate bias in posterior estimates of model parameters (Supplementary Fig. 9) does not alter our conclusions about the impact of interventions. Panels G,H indicate an underestimation of the impact of controlling incoming infected birds, which we ascribe to overestimating \(\rho_{c,b}\) in artificial scenarios 2 and 3. Results are based on 1000 and 5000 simulations for original and recovered parameters, respectively. In the second case, simulations are evenly split across 500 independent samples from the recovered posterior distribution.
+
+<--- Page Split --->
+![](images/Supplementary_Figure_6.jpg)
+
+<center>Fig. A: Re-analysis of panels G, H in Supplementary Fig. 13. </center>  
+
+Another thing makes the paper difficult to evaluate - nowhere did I see the actual fit of the model to the data; we have some evidence that we have the best fitted model, but we don't have much evidence on how good the best fit is.  
+
+We agree with the reviewer about the importance of showcasing goodness of fit. In presenting our results, posterior predictive checks for the model fitted to Ct=40 data ended up in Supplementary Fig. 4 and 5 (now 7 and 8), while goodness of fit for the model fitted to Ct=33 data was not shown.  
+
+We now display posterior predictive checks for both fits to Ct=40 and Ct=33 data in Supplementary Fig. 3. The latter shows posterior predictions (alongside 95% C.I.) for the counts of chickens testing positive at different time points of the experiment (T0- T4, solid bars) or testing negative throughout the experiment (hatched bars). These predictions are further stratified by chicken type and recruitment group, and are overlaid to data (black dots).  
+
+Supplementary Fig. 3 suggests that the fit of the model to available data is adequate.  
+
+![](images/Supplementary_Figure_5.jpg)
+
+<center>Supplementary Figure 3: Goodness of fit. Panels show posterior predictive checks for fits to \(Ct = 40\) (A-D) and \(Ct = 33\) (E-H) data, broken down by chicken type and recruitment group. Each panel shows posterior mean (bars) and 95% C.I. (grey error bars) for the numbers of chickens becoming positive (filled) at different stages of the experiment or remaining susceptible (hatched). Black dots and error bars denote data and 95% C.I. computed under </center>
+
+<--- Page Split --->
+
+a binomial distribution assumption. We ran 20000 simulations with 5 control and 5 intervention chickens from 2000 independent posterior samples to estimate the probability of turning positive at different times. Expected counts were obtained by multiplying these probabilities by the number of recruited chickens in each category.  
+
+In addition, a discussion of the fundamental drivers of the transmission dynamics (what is R0? R(t)? the generation time of transmission? What proportion of susceptibles are infected in the first, second or third generation)? What are the initial conditions and how much of infection is due to introduction rather than within market transmission. These are important because they help us to understand the "why's" of the interventions.  
+
+We thank the reviewer for this comment. In order to better disentangle the roles of external introductions and local transmission on H9N2 AIV spread, we measured additional epidemiological metrics in our simulations (using posterior samples to construct their posterior distribution).  
+
+First, we estimated the basic reproductive number \(R_{0}\) for this model by considering an effective constant removal rate \(\nu = T_{\text{stay}}^{- 1}\) , where \(T_{\text{stay}}\) is the average length of stay (obtained from the distribution of length of stay in Supplementary Fig. 1). With this simplifying assumption, \(R_{0}\) can be calculated explicitly as:  
+
+\[R_{0} = \sum_{b = BR,BY}\left(\frac{2\sigma_{b}}{2\sigma_{b} + \nu}\right)^{2}\frac{\beta}{\mu + \nu}\frac{N_{b}}{N},\]  
+
+where \(N = \sum_{b = BR,BY}N_{b}\) is the total number of chickens introduced daily and \(\overline{N}_{b}\) is the average number of chickens of type \(b = BR, BY\) present in the LBM. The expression above can be derived by means of the Next- Generation- Matrix for a multi- type \(SE_{n m}\) model (Diekman et Al., Journal of the Royal Society Interface, 2010; Champredon et Al., SIAM Journal on Applied Mathematics, 2018). The term \(\overline{N}_{b}\) reflects the assumption of density- dependent transmission and is calculated through simulations. This term also accounts for daily variations in population size due to trade (D'Onofrio, Mathematical Biosciences, 2002). Please note that the presence of \(N\) is due to our definition of the force of infection, where it appeared as a (constant) scaling factor.  
+
+The posterior distribution for \(R_{0}\) is shown in Supplementary Fig. 6A. We find that this quantity is significantly larger than 1 (between 3.5 and 5 depending on Ct threshold), in agreement with our previous assessment of high spreading potential. It also provides a quantitative justification for the high probability of persistence \((\approx 1)\) shown in the old Supplementary Fig. 6. The latter probability is now shown in Supplementary Fig. 6C against \(R_{0}\) , which we vary through the transmission rate \(\beta\) . As expected, persistence is essentially 0 for \(R_{0} < 1\) , and increases with \(R_{0}\) as \(R_{0} > 1\) . We also provide an estimate of
+
+<--- Page Split --->
+
+the basic reproductive number \(R_{0}(closed) = \beta /\mu\) for a closed population \((\nu = 0)\) with constant size \(N\) . As shown in Supplementary Fig. 6, \(R_{0}(closed)\) is extremely high, thus demonstrating the high infection pressure exerted on marketed chickens. Please note that we present \(R_{0}(closed)\) just for illustrative purposes, as the relevant transmission potential is quantified by \(R_{0}\) . All new references have been added to the list of supplementary references.  
+
+![](images/Figure_unknown_0.jpg)
+
+<center>Supplementary Fig. 6: Basic reproductive number and AIV persistence. (A) Posterior distribution for the basic reproductive number \(R_{0}\) for fits to \(Ct = 40\) (red) and \(Ct = 33\) (teal). The latter is calculated as the basic reproductive number for a continuous time SEEIRR model with a constant rate of chicken removal \(\nu\) , where \(\nu\) is set to the inverse mean length of stay. Combining previous theoretical results yields the expression \(R_{0} = \sum_{b = BR,BY}\left(\frac{2\sigma_{b}}{2\sigma_{b} + \nu}\right)^{2}\mu +\nu \frac{N_{b}}{N}\left(D'Onofrio, Mathematical Biosciences, 2002; Diekman et Al., Journal of the Royal Society Interface, 2010; Champredon et Al., SIAM Journal on Applied Mathematics, 2018), where \(N_{b}\) is the mean number of chickens of type \(b\) present in the LBM, and is estimated through simulations, and \(N = \sum_{b}N_{b}\) is the total number of chickens introduced daily. (B) Posterior distribution for the ratio \(\beta /\mu\) , which represents the basic reproductive number in a closed population \((\nu = 0)\) of constant size \(N\) . (C) Posterior probability of AIV persistence as a function of \(R_{0}\) . This is measured as the proportion of 2000 simulations where at least one latent or infectious chicken is observed at \(t = 50\) days, assuming that all chickens entering the market after \(t = 20\) days are susceptible. Results in A,B are based on 5000 posterior samples. </center>  
+
+Next, we further characterised the relative contributions of introductions and secondary infections within LBMs, the effective reproductive number \(R_{t}\) and the mean generation time \(GT_{t}\) at different times of the day. These results can be found in Supplementary Fig. 5 and are summarised below:  
+
+- Panel A shows the proportions of chickens that enter the market as susceptible (S) and become infected (i.e. sold as E, I and R), or infectious (sold as I or R) before being sold. These proportions are further broken down by chicken type. We find that >70% of initially susceptible chickens are infected before being sold. More than 50% of broiler chickens are also able to become infectious before leaving the market due to their short latent period. In contrast, less than 25% of initially susceptible backyard chickens transition from E to I within the market.
+
+<--- Page Split --->
+
+Panel B shows the proportion of infection events that were generated by chickens infected in the market rather than elsewhere prior to being offered for sale. While the posterior distribution for this quantity is rather wide, it shows that, on average, more than \(50\%\) of infections are caused by secondary, market- acquired cases. Panels C- F show \(R_{t}\) (C,E) and the mean generation time \(GT_{t}\) (D,F). To measure these quantities, we casted our model in an individual- based framework and tracked all transmission pairs; we then followed Liu et Al., PNAS, 2018 and calculated \(R_{t}\) and \(GT_{t}\) as the mean number and mean generation time of infections generated by incident cases at time \(t\) (i.e. we used a "forward- looking" definition for both \(R_{t}\) and \(GT_{t}\) ). Panels C,D are obtained under the same regime of external introductions considered in panels A,B and in the main analysis. Panels E,F, in contrast, correspond to the same scenario evaluated in Supplementary Fig. 6C, where all admitted chickens are susceptible, and all infections are locally acquired. In this case we find that viral circulation is sustained, i.e. the virus is able to circulate without becoming extinct. Panels C shows that the maxima of \(R_{t}\) correspond to the arrival of new chickens in the market (vertical dotted lines). \(R_{t}\) rises in the \(\sim 12\) hours preceding a new shipment due to the sudden increased availability of susceptible hosts entailed by this event. Indeed, \(GT_{t}\) becomes shorter as a new shipment approaches. The spiky shape displayed by \(R_{t}\) within panel C follows from the observation that most infections occur when a new shipment arrives, i.e. when the size of the susceptible population is at its largest and readily infectious birds are admitted into the market. Removing introductions (panels E,F and Supplementary Fig. 6C) produces smoother \(R_{t}\) and \(GT_{t}\) curves that are not affected by the sudden introduction of infectious (I) birds. Panels \(E,F\) further reinforce our interpretation that \(R_{t} > 1\) is driven by the cyclical enhanced availability of susceptible birds.  
+
+Overall, our additional analyses suggest that both external viral introductions and high within- market transmission are important drivers of H9N2 AIV epidemiology, as described in the main manuscript. These findings are mentioned in the Results section (below), and the figure is included in the Supplementary Material. These additional results are particularly relevant in the context of our main analysis, since they complement our previous findings and should hence be part of the supplementary material.  
+
+"The relative importance of external introductions of infected chickens and local transmission is assessed in Supplementary Fig. 5. We find comparable proportions of LBM- acquired infections caused by chickens infected before and after entering the LBM. Moreover, we estimate that within- LBM transmission is high enough ( \(R_{0} = 3.7 - 4.9\) depending on Ct threshold) to ensure long- term persistence of H9N2 AIV within the LBM in the absence of further external introductions (Supplementary Fig. 6)."
+
+<--- Page Split --->
+![](images/Supplementary_Figure_14.jpg)
+
+<center>Supplementary Figure 5: Relative importance of external introductions and local transmission. (A) Violin plots denote posterior distributions for the fraction of initially susceptible chickens that are infected locally and the fraction of those that become infectious before being sold. (B) Posterior distribution for the fraction of LBM-acquired infections that are caused by other chickens that became infected within the LBM, as opposed to externally-introduced infections. Results are based on 20000 simulations from 2000 independent samples from the posterior distribution. (C) Mean number of new cases \((R_{t})\) caused by chickens infected at time \(t\) . (D) Mean generation time versus the timing of primary infection \(t\) . The generation time is defined as the delay between the infection of a primary and a secondary case. Statistics in C,D were obtained by tracking transmission pairs in an individual-based version of our model. Panels E,F mirror C,D but assume that all chickens entering the market after 30 days are susceptible (i.e. in absence of external introductions). We consider all primary cases infected between \(t = 45\) and \(t = 48\) days. For externally-introduced infections, the infection time was set to the time of introduction. Daily shipments of chickens are denoted with vertical lines. Results in C-F based on 2000 simulations using the same number of independent samples from the posterior distribution. </center>  
+
+A broader point is that the motivation behind the analysis seems to sit a bit between two stools. High prevalence of H9N2 in markets and the issue of control may be important for one of two reasons - first, if it results in substantial zoonotic infection, or if it is important for maintaining the circulation of the virus. However, H9N2, even though it
+
+<--- Page Split --->
+
+may contribute to the generation of zoonotic forms of flu, is itself not a zoonosis. In this case, the prevalence in the market is less important than the role the market has in the circulation of H9N2 - and this can only be evaluated by considering onward transmission from the market. The authors need to make a better case for why their analysis helps us to understand or control that circulation of virus. This is discussed in the introduction (Paragraph starting line 49) but this at best touches upon the circulation question. There is more detail about trading networks in their other papers (e.g. Moyen et al 2021) that they reference, so it would be useful to contextualise the analysis of this LBM in that study (e.g. is the LBM an important source of birds for mobile traders?).  
+
+We thank the reviewer for this comment. We strongly agree about the importance of embracing a systemic viewpoint in order to investigate pathogen spread across live poultry markets. In fact, we recently developed a computational model (https://doi.org/10.1371/journal.pcbi.1011375) to model AIV transmission within the entire poultry production and distribution network. While viral movements between LBMs represent a topic of crucial importance in general, Moyen et Al. found limited evidence of inter- LBM poultry movements in Chattogram. Nonetheless, onward spread from a LBM to others and farms could occur through the movements of traders' contaminated equipment and vehicles. We respectfully disagree with the reviewer about the nature of H9N2 AIV which, as stated in the Introduction, can be zoonotic (see also Peacock et Al., Viruses, 2019 and Peacock et Al., J. Virol., 2020). It is then reasonable, in our opinion, to expect that larger prevalence increases the risk of AIV exposure for market workers and customers alike. Moreover, larger prevalence in poultry means more opportunities for viral reassortment and more exportations to other markets and farms.  
+
+A few minor points follow:  
+
+line 41: "especially"  
+
+This has been fixed.  
+
+line 81: this reference is a preprint which is useful to have but it would be better if, the interim has it been submitted for publication and/or accepted.  
+
+We confirm that the referenced preprint has been submitted for publication. However, we can not guarantee that this manuscript will be published/accepted in the short term as both manuscripts were submitted almost simultaneously. We commit to updating this reference as soon as we receive a notification of acceptance, conditional on the timeline of peer- review for the present work.  
+
+line 97: I struggled a bit with the definition of control and intervention chickens. Its clearer when you look at the companion preprint (Kohnle et al) but a bit more here would be useful.  
+
+We have clarified this distinction by adding further details. This sentence now reads:
+
+<--- Page Split --->
+
+"We further distinguished between chickens traded along conventional (control, c) and altered (intervention, i) marketing channels. The latter involved purchasing chickens from farms rather than from traders at the market, thus avoiding intermediate transport and storage steps."  
+
+line 123: this is a striking result - are these all infected in one generation due to a high prevalence of infection of birds entering the market? Or is there another reason?  
+
+This is one of our main results and follows from a combination of high transmission and high prevalence of infection in introduced chickens. The first aspect creates conditions to guarantee viral persistence in the market (see Supplementary Fig. 6), while the second aspect enhances the probability of infection due to the larger initial number of infections after each shipment. We believe that the new Supplementary Fig. 5 discussed above addresses the relative roles of transmission and introductions in an adequate manner. We emphasize that the survival curve shown in Fig. 1A is conditional on a chicken spending at least \(x\) hours in the market. In reality, only \(11.5\%\) of chickens remain in the market for more than 24 hours, the median length of stay being 16 hours (see Supplementary Fig. 1A).  
+
+line 139: why "tentative"  
+
+We removed this word.  
+
+line 170 onwards. As discussed, it seems that in the model, environmental contamination only differs from direct transmission by allowing for a longer term persistence once infected birds are gone. Otherwise the mechanism of transmission (and impact on transmission) is the same. Kim et al 2018 which the authors cite suggest that there is a strong variability in prevalence across stalls and slaughter areas of the market - is this likely to make a difference here in terms of the way contamination vs direct transmission would work?  
+
+The reviewer is right in pointing out that both modes of transmission lead to similar dynamical outcomes. In fact, previous theoretical work (Cortez & Weitz, Am. Nat., 2013 and Benson et Al., PLOS Comp. Biol., 2021) demonstrates that direct transmission models can be seen as the limiting cases of indirect transmission models when the dynamics of environmental contamination is fast (as confirmed by our analysis as well). Some dynamical differences may arise in the case of slow environmental contamination, but Cortez & Weitz show that the specific dynamical signatures depend on the underlying model. On one hand, we believe that such similarity is reassuring, as it suggests that different modelling assumptions lead to similar conclusions about given epidemiological aspects. On the other hand, a realistic accounting of environmental contamination is necessary when addressing specific questions, such as assessing the impact of LBM depopulation and cleaning. Heterogeneities in the handling of meat and/or carcasses and cleaning practices across an LBM may lead to distinct outcomes if transmission is direct or indirect. Realistically, the latter mode should better capture differences in
+
+<--- Page Split --->
+
+prevalence of contamination across different market areas. For example, market stalls located near slaughter areas could be at an increased risk of contamination than more distant ones. Further discrepancies between models may arise in the case of frequent removal of diseased birds (as shown by Benson et Al., PLOS Comp. Biol., 2021), which we did not include in our model due to the low pathogenicity of H9N2 AIV.  
+
+We have now incorporated these points in the Discussion, and included the two references mentioned above. We also replaced one reference (Martin et Al., Preventive Veterinary Medicine, 2011) with a more comprehensive reference that reviews the impact of LBM disinfection and rest days (Offeddu et Al., One Health, 2016). The full paragraph discussing environmental contamination now reads:  
+
+"We considered two alternative modes of transmission, direct and mediated by the environment. Both scenarios were able to explain observed dynamic patterns and yielded similar results in the context of interventions targeting chickens only. Previous theoretical work indeed demonstrated that both modes of transmission lead to similar dynamical outcomes, especially when environmental contamination unfolds on a fast time scale, and that it may be difficult to prefer one or another based solely on prevalence or incidence data (Cortez & Weitz, Am. Nat., 2013; Benson et Al., PLOS Comp. Biol., 2021). This is a reassuring finding as it suggests that some epidemiological conclusions are not affected by precise modelling assumptions. However, further work is needed to identify dynamical signatures of direct and environmental transmission. Nonetheless, incorporating environmental transmission is necessary if the objective is to assess the impact of LBM depopulation and routine cleaning/disinfection, as done in this work. In this case, moderate levels of cleaning were able to curb transmission significantly in simulations, especially with small decay rates, as that corresponds to a slower accumulation of contaminated material. Periodic cleaning/disinfection, usually carried out during rest days, has been shown to reduce AIV burden in Chinese, Hong Kong and Bangladeshi LBMs (Fournié et Al., Journal of the Royal Society Interface, 2011; Offeddu et Al., One Health, 2016; Chowdhury et Al., Emerging Infectious Diseases, 2020). Including environmental transmission may also be more appropriate to capture differences in prevalence of contamination across LBM sections (e.g. stalls and slaughter areas) (Kim et Al., Emerging Infectious Diseases, 2018), and assess how the distance from slaughter areas affects the risk of contamination. Practical difficulties in successfully implementing sanitation in LBMs (Barnett et Al., The Lancet Planetary Health, 2021) further underscore the importance of adopting a multi- pronged approach to reduce the burden of H9N2 AIV in LBMs. Our study also makes the case for the vaccination of poultry intended to be sold in LBMs in Bangladesh."  
+
+line 220 onwards - are there any references that would help us to understand how difficult it would be to sanitise market? A brief description of what this would entail would be helpful.
+
+<--- Page Split --->
+
+Chowdhury et Al., Emerg. Infect. Dis., 2020, which is already referenced in the manuscript, provides evidence of benefits of market cleaning even with monthly frequency in Bangladeshi LBMs. However, Paritosh et Al. Transboundary and Emerging Diseases, 2017 found that a package of biosecurity measures revolving around cleaning did not yield differences in detection rates of H5N1 AIV. Barnett et Al., The Lancet Planetary Health, 2021 suggest that the apparent limited effectiveness of LBM sanitation is due to practical implementation challenges, particularly the difficult social, economic and cultural context in which poultry workers operate. We tried to convey this nuance in the Results section when introducing disinfection (we have also included the new references in the main manuscript):  
+
+"Daily, weekly, or even monthly cleaning of poultry stalls, with or without weekly disinfection, was found to be associated with lower detection of AIV environmental contamination (Chowdhury et Al, 2019). However, sanitation is not straightforward to implement in practice (Paritosh et Al., Transboundary and Emerging Diseases, 2017; Barnett et Al., The Lancet Planetary Health, 2021)."  
+
+We believe that practical difficulties in achieving sanitation further underscore the importance of combining multiple interventions, which is one of our main conclusions. This point is now highlighted in the Discussion section as well:  
+
+"Practical difficulties in successfully implementing sanitation in LBMs (Barnett et Al., The Lancet Planetary Health, 2021) further underscore the importance of adopting a multi- pronged approach to reduce the burden of H9N2 AIV in LBMs."  
+
+line 227: there is a useful point here in terms of the recovery of infection - and it would therefore be important to gain a better understanding of why - referring to my earlier general point, is it because, for example the R value is very high? Or is it because the incoming infection prevalence is high.  
+
+We interpret the reviewer's comment as addressing the rapid recovery of infection following depopulation in the case of direct transmission rather than the difference between the latter and environmental transmission. As discussed above (see new Supplementary Fig. 5), fast transmission is due to the large transmission rate, while frequent introductions facilitate a quick return to endemic levels of prevalence. Indeed, if introductions were sporadic (e.g. not every day) it would then take longer to build up prevalence.  
+
+line 342: That the two scenarios of transmission yield similar result is reassuring but as noted earlier, I suspect this may at least partially be because the dynamics in the model really aren't all that different. While the authors cannot be expected to do everything in a single paper, I think some discussion of the implications of model similarity is worthwhile.
+
+<--- Page Split --->
+
+We already addressed this point above.  
+
+line 361: "we believe that our main results ... are not affected ..." - why?  
+
+We expect that including additional chicken types would have a small impact on our estimates since broiler and backyard chickens already represent a large proportion of poultry being traded in the LBM (Supplementary Fig. 1). Moreover, we know from additional field studies that prevalence in other chicken types is not higher than in broilers and backyard chickens (Kim et Al., Emerging Infectious Diseases, 2018). This suggests a smaller contribution of other chicken types to overall transmission compared to broilers.  
+
+line 381: "with similar conditions" - which are the key conditions that are important?  
+
+We slightly rewrote the final paragraph in the Discussion to expand on this point:  
+
+"The ubiquity of similar poultry handling and trading practices suggests that our findings apply to other Bangladeshi LBMs as well. Applications of the model to other LBMs may require calibrating specific LBM features such as the number of chickens being traded and their length of stay."  
+
+S.92 onwards. after all long discussion of the effects of environmental conditions on environmental survival, on line S.103 the authors then reduce the difference to one of temperature only.  
+
+The purpose of this section was to discuss potential factors affecting environmental survival while also presenting a broad range of values for the decay rate \(\Theta\) that were also realistic for this host- pathogen system. The sentence at line S.103 just places the chosen values into the context of Handel at Al.'s findings (reference 2 in the supplementary), which include an explicit, AIV subtype- specific relationship between \(\Theta\) and temperature. We agree that this sentence may read confusing, and decided to remove it in the revised version of the Supplementary Material.  
+
+To further support the appropriateness of our choice of \(\Theta\) , we report various estimates of \(\Theta^{- 1}\) from the literature as a function of water temperature (see Fig. B below). The relationship presented by Handel at Al. is shown in black, while individual red dots are distinct estimates retrieved from a systematic review. Importantly, these estimates were obtained under a variety of environmental conditions beyond temperature. Blue and orange lines correspond to values used in two distinct modelling studies investigating AIV transmission in poultry markets. We prefer to keep Fig. B in this letter as we believe that current references in the Supplementary Material already provide sufficient context.
+
+<--- Page Split --->
+![PLACEHOLDER_17_0]
+
+<center>Fig. B: Summary of estimates of \(\theta^{-1}\) from the literature. </center>  
+
+supp fig 7 - as discussed a the beginning of this review, really useful to have this but more details on the parameter sets - were they chosen to be relevant to the posteriors for the real data (e.g. drawn from the posterior distributions)? Also, the mismatch in scenario 5 is indeed striking and it would be good to have confidence it isn't important.  
+
+These parameter values were not selected from the posterior, although the scales of some of these were consistent with posterior estimates reported in the main manuscript. Our rationale was to verify not only whether we could recover initial parameters under optimal conditions (i.e. in absence of model misspecification), but also differences between chicken types and recruitment groups. This is now explained in the caption of Supplementary Fig. 7 (now 9)  
+
+The mismatch occurring in scenario 5 is evident mostly in the context of the duration of infection \(T_{_I}\) . However, as we describe in the corresponding caption, this is partially intended since scenario 5 uses the same parameter set of scenario 4, but uses a misspecified prior on the total duration of shedding \(T_{_E} + T_{_I}\) . The observed mismatch arises because our data is mostly informative about \(T_{_E}\) and not \(T_{_I}\) , as recovery from infection is not tracked in our experimental setting. Please note that some degree of prior misspecification on \(T_{_E} + T_{_I}\) is present also in scenarios 1- 4, but is not as strong as in scenario 5. We have added a brief sentence in the caption of Supplementary Fig. 7 (now 9) to elucidate this point. We also note that while parameters describing the timing of prior infections \((\lambda , \kappa)\) also show some mismatch, the mean and standard deviation of the corresponding distributions display a better agreement with the original values.
+
+<--- Page Split --->
+
+As discussed above, mismatches between original and recovered parameters did not significantly affect our assessment of effectiveness of interventions (see new Supplementary Fig. 13).  
+
+## Reviewer 2  
+
+The authors present an interesting study modelling the transmission dynamics of avian influenza with parameters fit to empirical data from a live bird market. I find the model proposed to be reasonable; the statistical analysis appropriate; and the limitations of the approach to be unambiguously communicated. Generally, I find the manuscript to be uncommonly clearly written.  
+
+We thank the reviewer for their very positive assessment of our work.  
+
+Regarding the presentation of the model, I would encourage the authors to briefly state (perhaps any additional analysis is beyond the scope of this manuscript) how they would expect compartment topology to modify the parameter estimates or otherwise invest more time in the model description to motivate why SEEIRR is clearly the most appropriate framework.  
+
+We thank the reviewer for their suggestion. We now mention in the model description that our model is a variant of the SEIR model, which is commonly used to investigate AIV dynamics:  
+
+"This model is a variant of the more common SEIR model, which is typically used to investigate AIV dynamics (Fournié et Al., Mathematical Models of Infectious Diseases in Livestock: Concepts and Application to the Spread of Highly Pathogenic Avian Influenza Virus Strain Type H5N1, 2011)."  
+
+We added the new reference to the list of references. Furthermore, we commented on the choice of including two and one latent and infectious compartments, respectively:  
+
+"This is often regarded as a more realistic assumption than an exponential distribution of durations, which is implicit in models with single- staged compartments (Bouma et Al., PLOS Pathogens, 2009). On the other hand, including a single infectious compartment should not affect our results significantly since the main limitation to transmission within LBMs comes from the short length of stay. [...]  
+
+The distinction between \(R^{+}\) and \(R^{- }\) compartments allows to capture the persistence of viral RNA in infected chickens that recently stopped shedding (Griffin D.E., PLoS Pathogens, 2022)."  
+
+The reference to Griffin D.E., PLoS Pathogens, 2022 was also included in the list of references.
+
+<--- Page Split --->
+
+The authors describe several predicted epidemiological differences between exotic broilers and backyard chickens including:  
+
+"From our model's output, we found a shorter latent period in exotic broiler compared to backyard chickens (Fig. 1B,C), lasting an average of 5.3 hours for exotic broiler, and 1 days for backyard chickens."  
+
+and  
+
+"in the case of exotic broilers, most chickens with prior exposure to H9N2 were either infectious or latent, with only a minor proportion of them being immune (Fig. 1E). In contrast, most previously- exposed backyard chickens were immune to H9N2 (Fig. 1F). Our results thus tentatively suggest that prior infection occurs close to marketing age for broilers, whereas in backyard chickens it may occur further in the past, which is consistent with the latter being raised for a longer time compared to broilers."  
+
+I encourage the authors to comment on whether they believe these findings are entirely consistent with variable patterns of prior exposure for these two populations or if the data presented suggest that there may be some unrecognized genetic determinants of susceptibility. Additionally I understand that to support the second passage quoted above, there has been an underlying assumption made that prior infection confers long lasting immunity (relative to the life cycle of a market chicken). If this is correct, I encourage the authors to explicitly state this assumption and perhaps introduce some additional background information regarding the market chicken life cycle so that the unfamiliar reader, like myself, may have some expectation of the maximum number of infections for one individual.  
+
+We thank the reviewer for this comment.  
+
+We have now provided some context regarding the ages at sale of broilers and backyard chickens in the Results section:  
+
+"These findings are consistent with known rearing practices and ages at sale of each chicken type: broilers are selectively bred to grow rapidly, and are sold for meat after just 28- 31 days after hatching (Hennessey et Al., Preventive Veterinary Medicine, 2021). Backyard chickens are instead raised for meat and eggs in rural households and can reach much older ages. For context, backyard chickens in our dataset were aged between 90 and 720 days."  
+
+Our findings are also consistent with a previous study (Das Gupta et Al., Transboundary And Emerging Diseases, 2020) that found a higher seroprevalence of antibodies against H9 among backyard chickens in farms surrounding Chattogram, i.e. where the present LBM is located. While we cannot rule out the existence of unrecognised genetic determinants (which we already mention in the Discussion), our findings suggest that these are not necessary to explain observed differences in incidence in broilers and
+
+<--- Page Split --->
+
+backyard chickens. Nonetheless, the reviewer is right in pointing out that this interpretation rests on the role of pre- existing immunity. Ultimately, further confirmation about either hypothesis (which are not mutually exclusive) will require a better understanding of the role of previous immunity. We have discussed these aspects in multiple points in the Discussion:  
+
+"Our results indicate, however, that differences in earlier exposure to H9N2 AIV are sufficient to explain observed incidence patterns, and are consistent with known ages at sale and levels of H9 seropositivity in broilers and backyard chickens (Das Gupta et Al., 2021). A better understanding of the effectiveness of prior immunity, which we have assumed to be sterilising, will help validating or confuting this interpretation."  
+
+The authors consider both direct and environmental transmission, writing, "lenv(t) accumulates due to shedding from infectious chickens and decays progressively at rate 0. Here we consider three values of 0, namely \(\Theta - 1 = 10,3,1\) days, corresponding to slow, intermediate and fast decay, respectively. These values are based on actual estimates from the scientific literature and capture a broad range of environmental conditions." I encourage the authors to briefly describe the physical nature of environmental transmission in a live bird market for readers, like myself, who are more familiar with influenza transmission dynamics in the human population which I expect substantially differ.  
+
+We thank the reviewer for their comment. We have now described the main sources of environmental contamination in the Results section:  
+
+"Within LBMs, the infection can be transmitted through poultry drinking water, as well as cages and floors that are contaminated by faecal material and during slaughtering. The lack of disinfection and the constant moving of cages and birds further promote poultry exposure to environmental contamination."  
+
+I find one of the more striking findings reported in the manuscript to be that while, "reduced length of stay and reduced probability of prior exposure, proved to be only modestly effective. Combining both interventions proved considerably more effective at reducing transmission compared to individual measures" I encourage the authors to clarify in the text if this effect is in fact nonadditive and if so to include some description of the expected mechanism behind the non- additivity.  
+
+The reviewer raises an interesting point that we address in the new Supplementary Fig. 14 shown below. Our reasoning is as follows: if the effect of each intervention in terms of reducing cumulative daily prevalence was additive, then their combined effectiveness \(E'\) would be given by \(E' = E_1 + E_2\) (this of course holds up if and only if \(E' \leq 1\) ).  
+
+Analogously, if their effects were multiplicative, the combined effectiveness would be \(E' = 1 - (1 - E_1)(1 - E_2)\) . To gauge any signatures of additivity and/or multiplicativity,
+
+<--- Page Split --->
+
+we measured the combined effectiveness from simulations \(E\) (panels A,B), and calculated the difference \(E - E'\) under either expectation for \(E'\) (panels C- F). Interestingly, the positive effects of these interventions do not seem to combine additively or multiplicatively. In particular, the fact that \(\Delta E > 0\) for most configurations, especially for stricter control of importations and shorter length of stay, is suggestive of a synergistic effect of stacking these interventions. This is mentioned in the Result section:  
+
+"Notably, the benefits of combining (i) and (ii) exceed expectations under the assumption that their effects were additive or multiplicative, hence suggesting a synergistic effect of multiple interventions (Supplementary Fig. 14)."  
+
+![PLACEHOLDER_21_0]
+
+<center>Supplementary Figure 14: Synergistic effects of combining interventions. This figure considers a simultaneous reduction of the maximum length of stay of chickens to \(T_{m}\) and the probability of prior exposure \(\rho_{c,b}\) to \((1 - r)\rho_{c,b}\) </center>
+
+<--- Page Split --->
+
+(A,B). The effectiveness of these interventions combined is denoted with \(E\) , and is defined as the reduction in cumulative daily prevalence relative to a scenario with no intervention in place. The latter corresponds to \(T_{m} = 5\) days and \(r = 0\) . The effectiveness of individual interventions is denoted as \(E_{1}(T_{m} < 5\) days and \(r = 0\) ) and \(E_{2}(T_{m} = 5\) days and \(0 < r \leq 1\) ). Panels C-F assess whether the effects of individual interventions add up in a multiplicative (C,D) or additive fashion (E,F). If the effect was purely multiplicative, the expected, combined effectiveness would be \(\bar{E} = 1 - (1 - E_{1})(1 - E_{2})\) . In the purely additive case it would be \(\bar{E} = E_{1} + E_{2}\) . C-F show that \(E > \bar{E}\) for most values of \(T_{m}\) and \(r\) , suggesting a synergistic effect of implementing these interventions simultaneously. Results are based on 2000 simulations from 200 independent samples from the posterior distribution. The first and second columns correspond to the fits to Ct=40 and Ct=33 data, respectively.  
+
+In conclusion, the authors present an interesting and well written study which I believe will be of broad interest.  
+
+Sincerely, Nash Rochman (invited 12/5/23, returned 12/19/23)  
+
+## Reviewer 3  
+
+The manuscript titled "Modelling the transmission dynamics of H9N2 avian influenza viruses in a live bird market" aims to understand the transmission dynamics of H9N2 avian influenza viruses in live bird markets (LBMs) in Bangladesh. The study uses a mechanistic transmission model, fitted to field data collected in a specific LBM, to simulate the spread of the virus among poultry.  
+
+The study presents significant findings on H9N2 transmission in LBMs, highlighting a rapid infection rate among susceptible chickens, often within a day. The virus's latent period is notably short, particularly in broiler chickens, sometimes as brief as 5.3 hours, facilitating its swift spread. It also uncovers considerable differences in exposure and immunity between exotic broiler and backyard chickens. Its unique approach, combining field data with mathematical modelling, and the robustness of its findings, underscored by sensitivity analyses and Bayesian methods, mark it as particularly relevant for areas where H9N2 is prevalent, guiding interventions to control its spread. Thus, This research offers crucial insights into the epidemiology of H9N2 in LBMs, vital for public health and the poultry industry.  
+
+I thoroughly enjoyed reading this manuscript and believe it is already of high quality. However, I suggest incorporating discussions on several interesting points. Firstly, the study's results are drawn from a single LBM in Bangladesh. It would be beneficial to explore how these findings might apply to other LBMs with varying dynamics, bird populations, and environments. Additionally, a more in- depth examination of environmental factors, like temperature, humidity, and market hygiene, is crucial as they significantly influence H9N2's survival and spread. A comparison with other avian influenza strains could also offer a broader perspective, highlighting differences and similarities in transmission dynamics. Finally, it would be interesting to consider whether
+
+<--- Page Split --->
+
+these findings could be applicable to livestock markets other than those for birds and discuss the generalizability of such high infection rates in group settings, dependent on the virus  
+
+We are pleased that the reviewer enjoyed our manuscript and thank them for their comments.  
+
+We have addressed the point on applicability to other LBMs in our reply to Reviewer 1.  
+
+We have now added a sentence to discuss potential applications to other LBMs, live animal markets and host- pathogen systems. There, we also argue for the need to account for the main AIV strain- specific characteristics and environmental drivers of transmission, such as the presence of suitable hosts and those factors that affect AIV survival in the environment (which are described in the Supplementary Methods).  
+
+"The ubiquity of similar poultry handling and trading practices suggests that our findings apply to other Bangladeshi LBMs as well. Applications of the model to other LBMs may require calibrating specific LBM features such as the number of chickens being traded and their length of stay. Applications to other AIV strains, e.g. H5N1, will also require accounting for the presence in the LBM of relevant hosts species (e.g. waterfowl in the case of H5N1 AIV), their specific infection parameters and differential ability to survive in the environment (Handel et Al., PLOS Comp. Biol., 2013). Finally, we note that our modelling framework could be applied to disentangle the contributions of external introductions and local transmission in other types of live animal markets and host- pathogen systems."
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have done an excellent job on their revisions and I have no further comments (and thank them for their respectful disagreement in regards to zoonotic potential - tbh, I think they are in a better position than I to judge this).  
+
+I believe the authors have fully addressed the reviewer comments.  
+
+Sincerely,  Nash Rochman
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a211576ac4ea643b0147d034b6c6b5d4b856451c2af07c558428c4642ea7e8c9/supplementary_0_Peer Review File__a211576ac4ea643b0147d034b6c6b5d4b856451c2af07c558428c4642ea7e8c9_det.mmd

@@ -0,0 +1,533 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[68, 110, 361, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 162, 916, 217]]<|/det|>
+Modelling the transmission dynamics of H9N2 avian influenza viruses in a live bird market  
+
+<|ref|>image<|/ref|><|det|>[[56, 732, 239, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 911, 800]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 84, 295, 97]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[120, 111, 414, 125]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[120, 140, 857, 168]]<|/det|>
+This paper analyses live bird market data on the transmission of H9N2 (low pathogenicity) avian influenza, fitting a 13 parameter model to field data from an LBM in Cambodia.  
+
+<|ref|>text<|/ref|><|det|>[[118, 181, 876, 336]]<|/det|>
+The paper is broadly speaking clearly written for a modelling heavy paper. It clearly goes through model parameters and discusses in detail the results of interventions from the fitted model. The supplementary information contains considerable detail on the model fit itself and justification for priors. The illustration of fit to simulated data is welcome. On a related point, while the fit to simulated data is helpful, it does show that some parameters are poorly recovered in the posterior. this may not be important in terms of the overall paper conclusions but the authors need to show this. One thing that might help, is to compare the results of interventions for the original simulated parameters, and the results using the recovered posteriors. If key outcomes are insensitive to this difference, it provides some confidence that those differences posteriors are not important. Unfortunately it is at best partial confidence of course (not accounting for differences between the model and the real world processes behind the data of course) but would be an important step.  
+
+<|ref|>text<|/ref|><|det|>[[119, 350, 868, 392]]<|/det|>
+Another thing makes the paper difficult to evaluate - nowhere did I see the actual fit of the model to the data; we have some evidence that we have the best fitted model, but we don't have much evidence on how good the best fit is.  
+
+<|ref|>text<|/ref|><|det|>[[118, 405, 875, 476]]<|/det|>
+In addition, a discussion of the fundemental drivers of the transmission dynamics (what is R0? R(t)? the generation time of transmission? What proportion of susceptibles are infected in the first, second or third generation)? What are the initial conditions and how much of infection is due to introduction rather than within market transmission. These are important because they help us to understand the "why's" of the interventions.  
+
+<|ref|>text<|/ref|><|det|>[[117, 490, 876, 658]]<|/det|>
+A broader point is that the motivation behind the analysis seems to sit a bit between two stools. High prevalence of H9N2 in markets and the issue of control may be important for one of two reasons - first, if it results in substantial zoonotic infection, or if it is important for maintaining the circulation of the virus. However, H9N2, even though it may contribute to the generation of zoonotic forms of flu, is itself not a zoonosis. In this case, the prevalence in the market is less important than the role the market has in the circulation of H9N2 - and this can only be evaluated by considering onward transmission from the market. The authors need to make a better case for why their analysis helps us to understand or control that circulation of virus. This is discussed in the introduction (Paragraph starting line 49) but this at best touches upon the circulation question. There is more detail about trading networks in their other papers (e.g. Moyen et al 2021) that they reference, so it would be useful to contextualise the analysis of this LBM in that study (e.g. is the LBM an important source of birds for mobile traders?).  
+
+<|ref|>text<|/ref|><|det|>[[119, 672, 318, 686]]<|/det|>
+A few minor points follow:  
+
+<|ref|>text<|/ref|><|det|>[[119, 700, 270, 714]]<|/det|>
+line 41: "especially"  
+
+<|ref|>text<|/ref|><|det|>[[118, 728, 875, 756]]<|/det|>
+line 81: this reference is a preprint which is useful to have but it would be better if, the interim has it been submitted for publication and/or accepted.  
+
+<|ref|>text<|/ref|><|det|>[[118, 770, 863, 798]]<|/det|>
+line 97: I struggled a bit with the definition of control and intervention chickens. Its clearer when you look at the companion preprint (Kohnle et al) but a bit more here would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[118, 812, 872, 840]]<|/det|>
+line 123: this is a striking result - are these all infected in one generation due to a high prevalence of infection of birds entering the market? Or is there another reason?  
+
+<|ref|>text<|/ref|><|det|>[[119, 854, 310, 868]]<|/det|>
+line 139: why "tentative"  
+
+<|ref|>text<|/ref|><|det|>[[118, 882, 857, 910]]<|/det|>
+line 170 onwards. As discussed, it seems that in the model, environmental contamination only differs from direct transmission by allowing for a longer term persistence once infected birds are
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 872, 140]]<|/det|>
+gone. Otherwise the mechanism of transmission (and impact on transmission) is the same. Kim et al 2018 which the authors cite suggest that there is a strong variability in prevalence across stalls and slaughter areas of the market - is this likely to make a difference here in terms of the way contamination vs direct transmission would work?  
+
+<|ref|>text<|/ref|><|det|>[[118, 153, 875, 183]]<|/det|>
+line 220 onwards - are there any references that would help us to understand how difficult it would be to sanitise market? A brief description of what this would entail would be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[118, 196, 866, 253]]<|/det|>
+line 227: there is a useful point here in terms of the recovery of infection - and it would therefore be important to gain a better understanding of why - referring to my earlier general point, is it because, for example the R value is very high? Or is it because the incoming infection prevalence is high.  
+
+<|ref|>text<|/ref|><|det|>[[118, 265, 866, 323]]<|/det|>
+line 342: That the two scenarios of transmission yield similar result is reassuring but as noted earlier, I suspect this may at least partially be because the dynamics in the model really aren't all that different. While the authors cannot be expected to do everything in a single paper, I think some discussion of the implications of model similarity is worthwhile.  
+
+<|ref|>text<|/ref|><|det|>[[118, 336, 680, 350]]<|/det|>
+line 361: "we believe that our main results ... are not affected ..." - why?  
+
+<|ref|>text<|/ref|><|det|>[[118, 364, 766, 379]]<|/det|>
+line 381: "with similar conditions" - which are the key conditions that are important?  
+
+<|ref|>text<|/ref|><|det|>[[118, 392, 874, 421]]<|/det|>
+S.92 onwards. after all long discussion of the effects of environmental conditions on environmental survival, on line S.103 the authors then reduce the difference to one of temperature only.  
+
+<|ref|>text<|/ref|><|det|>[[118, 434, 868, 491]]<|/det|>
+supp fig 7 - as discussed a the beginning of this review, really useful to have this but more details on the parameter sets - were they chosen to be relevant to the posteriors for the real data (e.g. drawn from the posterior distributions)? Also, the mismatch in scenario 5 is indeed striking and it would be good to have confidence it isn't important.  
+
+<|ref|>text<|/ref|><|det|>[[120, 545, 413, 559]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 574, 868, 631]]<|/det|>
+The authors present an interesting study modelling the transmission dynamics of avian influenza with parameters fit to empirical data from a live bird market. I find the model proposed to be reasonable; the statistical analysis appropriate; and the limitations of the approach to be unambiguously communicated. Generally, I find the manuscript to be uncommonly clearly written.  
+
+<|ref|>text<|/ref|><|det|>[[118, 644, 861, 701]]<|/det|>
+Regarding the presentation of the model, I would encourage the authors to briefly state (perhaps any additional analysis is beyond the scope of this manuscript) how they would expect compartment topology to modify the parameter estimates or otherwise invest more time in the model description to motivate why SEEIRR is clearly the most appropriate framework.  
+
+<|ref|>text<|/ref|><|det|>[[118, 714, 841, 743]]<|/det|>
+The authors describe several predicted epidemiological differences between exotic broilers and backyard chickens including:  
+
+<|ref|>text<|/ref|><|det|>[[118, 756, 870, 799]]<|/det|>
+"From our model's output, we found a shorter latent period in exotic broiler compared to backyard chickens (Fig. 1B,C), lasting an average of 5.3 hours for exotic broiler, and 1 days for backyard chickens."  
+
+<|ref|>text<|/ref|><|det|>[[118, 800, 149, 811]]<|/det|>
+and  
+
+<|ref|>text<|/ref|><|det|>[[118, 813, 875, 897]]<|/det|>
+"in the case of exotic broilers, most chickens with prior exposure to H9N2 were either infectious or latent, with only a minor proportion of them being immune (Fig. 1E). In contrast, most previously- exposed backyard chickens were immune to H9N2 (Fig. 1F). Our results thus tentatively suggest that prior infection occurs close to marketing age for broilers, whereas in backyard chickens it may occur further in the past, which is consistent with the latter being raised for a longer time compared to broilers."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 876, 210]]<|/det|>
+I encourage the authors to comment on whether they believe these findings are entirely consistent with variable patterns of prior exposure for these two populations or if the data presented suggest that there may be some unrecognized genetic determinants of susceptibility. Additionally I understand that to support the second passage quoted above, there has been an underlying assumption made that prior infection confers long lasting immunity (relative to the life cycle of a market chicken). If this is correct, I encourage the authors to explicitly state this assumption and perhaps introduce some additional background information regarding the market chicken life cycle so that the unfamiliar reader, like myself, may have some expectation of the maximum number of infections for one individual.  
+
+<|ref|>text<|/ref|><|det|>[[118, 224, 878, 336]]<|/det|>
+The authors consider both direct and environmental transmission, writing, "Ienv(t) accumulates due to shedding from infectious chickens and decays progressively at rate \(\Theta\) . Here we consider three values of \(\Theta\) , namely \(\Theta - 1 = 10\) , 3, 1days, corresponding to slow, intermediate and fast decay, respectively. These values are based on actual estimates from the scientific literature and capture a broad range of environmental conditions." I encourage the authors to briefly describe the physical nature of environmental transmission in a live bird market for readers, like myself, who are more familiar with influenza transmission dynamics in the human population which I expect substantially differ.  
+
+<|ref|>text<|/ref|><|det|>[[118, 350, 868, 435]]<|/det|>
+I find one of the more striking findings reported in the manuscript to be that while, "reduced length of stay and reduced probability of prior exposure, proved to be only modestly effective. Combining both interventions proved considerably more effective at reducing transmission compared to individual measures" I encourage the authors to clarify in the text if this effect is in fact nonadditive and if so to include some description of the expected mechanism behind the nonadditivity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 448, 861, 476]]<|/det|>
+In conclusion, the authors present an interesting and well written study which I believe will be of broad interest.  
+
+<|ref|>text<|/ref|><|det|>[[118, 492, 523, 519]]<|/det|>
+Sincerely, Nash Rochman (invited 12/5/23, returned 12/19/23)  
+
+<|ref|>text<|/ref|><|det|>[[120, 574, 413, 588]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 602, 878, 658]]<|/det|>
+The manuscript titled "Modelling the transmission dynamics of H9N2 avian influenza viruses in a live bird market" aims to understand the transmission dynamics of H9N2 avian influenza viruses in live bird markets (LBMs) in Bangladesh. The study uses a mechanistic transmission model, fitted to field data collected in a specific LBM, to simulate the spread of the virus among poultry.  
+
+<|ref|>text<|/ref|><|det|>[[118, 672, 876, 800]]<|/det|>
+The study presents significant findings on H9N2 transmission in LBMs, highlighting a rapid infection rate among susceptible chickens, often within a day. The virus's latent period is notably short, particularly in broiler chickens, sometimes as brief as 5.3 hours, facilitating its swift spread. It also uncovers considerable differences in exposure and immunity between exotic broiler and backyard chickens. Its unique approach, combining field data with mathematical modelling, and the robustness of its findings, underscored by sensitivity analyses and Bayesian methods, mark it as particularly relevant for areas where H9N2 is prevalent, guiding interventions to control its spread. Thus, this research offers crucial insights into the epidemiology of H9N2 in LBMs, vital for public health and the poultry industry.  
+
+<|ref|>text<|/ref|><|det|>[[118, 813, 875, 910]]<|/det|>
+I thoroughly enjoyed reading this manuscript and believe it is already of high quality. However, I suggest incorporating discussions on several interesting points. Firstly, the study's results are drawn from a single LBM in Bangladesh. It would be beneficial to explore how these findings might apply to other LBMs with varying dynamics, bird populations, and environments. Additionally, a more in- depth examination of environmental factors, like temperature, humidity, and market hygiene, is crucial as they significantly influence H9N2's survival and spread. A comparison with other avian influenza strains could also offer a broader perspective, highlighting differences and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 843, 127]]<|/det|>
+similarities in transmission dynamics. Finally, it would be interesting to consider whether these findings could be applicable to livestock markets other than those for birds and discuss the generalizability of such high infection rates in group settings, dependent on the virus.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 84, 295, 100]]<|/det|>
+## General information  
+
+<|ref|>text<|/ref|><|det|>[[118, 117, 852, 188]]<|/det|>
+General informationFirst of all, we thank the three anonymous reviewers for their assessments and helpful comments. Please find our detailed replies to these comments below. We also provided a copy of the revised manuscript and supplementary material where all the main changes have been highlighted in red/blue.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 205, 214, 222]]<|/det|>
+## Reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[118, 241, 870, 298]]<|/det|>
+This paper analyses live bird market data on the transmission of H9N2 (low pathogenicity) avian influenza, fitting a 13 parameter model to field data from an LBM in Cambodia.  
+
+<|ref|>text<|/ref|><|det|>[[116, 315, 870, 560]]<|/det|>
+The paper is broadly speaking clearly written for a modelling heavy paper. It clearly goes through model parameters and discusses in detail the results of interventions from the fitted model. The supplementary information contains considerable detail on the model fit itself and justification for priors. The illustration of fit to simulated data is welcome. On a related point, while the fit to simulated data is helpful, it does show that some parameters are poorly recovered in the posterior. this may not be important in terms of the overall paper conclusions but the authors need to show this. One thing that might help, is to compare the results of interventions for the original simulated parameters, and the results using the recovered posteriors. If key outcomes are insensitive to this difference, it provides some confidence that those differences posteriors are not important. Unfortunately it is at best partial confidence of course (not accounting for differences between the model and the real world processes behind the data of course) but would be an important step.  
+
+<|ref|>text<|/ref|><|det|>[[118, 578, 797, 612]]<|/det|>
+We are pleased that the reviewer appreciated our work and we thank them for their comments.  
+
+<|ref|>text<|/ref|><|det|>[[116, 629, 877, 785]]<|/det|>
+The reviewer raises an important point, and we agree that further work is required to assess to what extent the discrepancies between "true" and recovered parameters affect other key findings. Following the advice from the reviewer, we assessed the effectiveness of individual interventions displayed in Fig. 2 using original parameters and recovered posteriors from the scenarios reported in Supplementary Fig. 7 (now 9). These results are displayed in the new Supplementary Fig. 13 (attached below). Each column in the latter figure represents a different scenario (i.e. simulated dataset), while rows correspond to distinct interventions. We find an excellent agreement between simulations using original (dots) and recovered (bars) parameters in terms of intervention effectiveness in the case of:  
+
+<|ref|>text<|/ref|><|det|>[[147, 802, 788, 855]]<|/det|>
+Shorter length of stay (first row) - Shorter length of stay + Control of chickens entering the market (third row) Vaccination (last row)  
+
+<|ref|>text<|/ref|><|det|>[[116, 872, 844, 907]]<|/det|>
+However, we also underestimate the effectiveness of reducing past exposure in chickens entering the market in scenarios 2 and 3 (second row, second and third columns). As a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 869, 200]]<|/det|>
+reminder, the latter intervention reduces the probability that an incoming chicken is not susceptible by a factor \(1 - r\) . Upon further investigation, we discover that such discrepancy disappears completely in scenario 2 and is considerably reduced in scenario 3 after setting the probability \(P_{c,b,X}^{(0)}\) that an incoming bulk (i.e. control group) chicken of type \(b\) is in compartment \(X\) to the corresponding true value. These results are shown in Fig. A below alongside this reply for completeness.  
+
+<|ref|>text<|/ref|><|det|>[[118, 216, 843, 269]]<|/det|>
+Overall, these findings suggest that our main results about the effectiveness of modelled interventions are robust to modest discrepancies in posterior estimates. The new supplementary figure is referenced in the Results section:  
+
+<|ref|>text<|/ref|><|det|>[[118, 285, 861, 320]]<|/det|>
+"Estimates of intervention effectiveness presented in Fig. 2 are robust to mismatches between inferred and original parameters in simulated scenarios (Supplementary Fig. 13)."  
+
+<|ref|>image<|/ref|><|det|>[[118, 337, 880, 685]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[116, 688, 870, 850]]<|/det|>
+Supplementary Figure 13: Estimating effectiveness of interventions in artificial scenarios. This figure compares the effectiveness of veterinary public health interventions calculated using recovered (bars) and original (dots) parameters in the 5 artificial scenarios considered in Supplementary Fig. 9 (from left to right). We consider the same interventions presented in Fig. 2 in the main manuscript: shorter length of stay (top row), control of infected chickens entering the LBM (second row) and vaccination (bottom row). The third row corresponds to the simultaneous implementation of the first two interventions. Moderate bias in posterior estimates of model parameters (Supplementary Fig. 9) does not alter our conclusions about the impact of interventions. Panels G,H indicate an underestimation of the impact of controlling incoming infected birds, which we ascribe to overestimating \(\rho_{c,b}\) in artificial scenarios 2 and 3. Results are based on 1000 and 5000 simulations for original and recovered parameters, respectively. In the second case, simulations are evenly split across 500 independent samples from the recovered posterior distribution.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[264, 88, 734, 243]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[280, 249, 713, 265]]<|/det|>
+<center>Fig. A: Re-analysis of panels G, H in Supplementary Fig. 13. </center>  
+
+<|ref|>text<|/ref|><|det|>[[118, 280, 878, 337]]<|/det|>
+Another thing makes the paper difficult to evaluate - nowhere did I see the actual fit of the model to the data; we have some evidence that we have the best fitted model, but we don't have much evidence on how good the best fit is.  
+
+<|ref|>text<|/ref|><|det|>[[118, 355, 839, 431]]<|/det|>
+We agree with the reviewer about the importance of showcasing goodness of fit. In presenting our results, posterior predictive checks for the model fitted to Ct=40 data ended up in Supplementary Fig. 4 and 5 (now 7 and 8), while goodness of fit for the model fitted to Ct=33 data was not shown.  
+
+<|ref|>text<|/ref|><|det|>[[117, 448, 874, 563]]<|/det|>
+We now display posterior predictive checks for both fits to Ct=40 and Ct=33 data in Supplementary Fig. 3. The latter shows posterior predictions (alongside 95% C.I.) for the counts of chickens testing positive at different time points of the experiment (T0- T4, solid bars) or testing negative throughout the experiment (hatched bars). These predictions are further stratified by chicken type and recruitment group, and are overlaid to data (black dots).  
+
+<|ref|>text<|/ref|><|det|>[[116, 582, 844, 601]]<|/det|>
+Supplementary Fig. 3 suggests that the fit of the model to available data is adequate.  
+
+<|ref|>image<|/ref|><|det|>[[120, 620, 879, 840]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 845, 878, 905]]<|/det|>
+<center>Supplementary Figure 3: Goodness of fit. Panels show posterior predictive checks for fits to \(Ct = 40\) (A-D) and \(Ct = 33\) (E-H) data, broken down by chicken type and recruitment group. Each panel shows posterior mean (bars) and 95% C.I. (grey error bars) for the numbers of chickens becoming positive (filled) at different stages of the experiment or remaining susceptible (hatched). Black dots and error bars denote data and 95% C.I. computed under </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 81, 877, 128]]<|/det|>
+a binomial distribution assumption. We ran 20000 simulations with 5 control and 5 intervention chickens from 2000 independent posterior samples to estimate the probability of turning positive at different times. Expected counts were obtained by multiplying these probabilities by the number of recruited chickens in each category.  
+
+<|ref|>text<|/ref|><|det|>[[117, 145, 872, 241]]<|/det|>
+In addition, a discussion of the fundamental drivers of the transmission dynamics (what is R0? R(t)? the generation time of transmission? What proportion of susceptibles are infected in the first, second or third generation)? What are the initial conditions and how much of infection is due to introduction rather than within market transmission. These are important because they help us to understand the "why's" of the interventions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 259, 821, 335]]<|/det|>
+We thank the reviewer for this comment. In order to better disentangle the roles of external introductions and local transmission on H9N2 AIV spread, we measured additional epidemiological metrics in our simulations (using posterior samples to construct their posterior distribution).  
+
+<|ref|>text<|/ref|><|det|>[[117, 352, 852, 446]]<|/det|>
+First, we estimated the basic reproductive number \(R_{0}\) for this model by considering an effective constant removal rate \(\nu = T_{\text{stay}}^{- 1}\) , where \(T_{\text{stay}}\) is the average length of stay (obtained from the distribution of length of stay in Supplementary Fig. 1). With this simplifying assumption, \(R_{0}\) can be calculated explicitly as:  
+
+<|ref|>equation<|/ref|><|det|>[[366, 470, 629, 512]]<|/det|>
+\[R_{0} = \sum_{b = BR,BY}\left(\frac{2\sigma_{b}}{2\sigma_{b} + \nu}\right)^{2}\frac{\beta}{\mu + \nu}\frac{N_{b}}{N},\]  
+
+<|ref|>text<|/ref|><|det|>[[115, 541, 855, 742]]<|/det|>
+where \(N = \sum_{b = BR,BY}N_{b}\) is the total number of chickens introduced daily and \(\overline{N}_{b}\) is the average number of chickens of type \(b = BR, BY\) present in the LBM. The expression above can be derived by means of the Next- Generation- Matrix for a multi- type \(SE_{n m}\) model (Diekman et Al., Journal of the Royal Society Interface, 2010; Champredon et Al., SIAM Journal on Applied Mathematics, 2018). The term \(\overline{N}_{b}\) reflects the assumption of density- dependent transmission and is calculated through simulations. This term also accounts for daily variations in population size due to trade (D'Onofrio, Mathematical Biosciences, 2002). Please note that the presence of \(N\) is due to our definition of the force of infection, where it appeared as a (constant) scaling factor.  
+
+<|ref|>text<|/ref|><|det|>[[117, 759, 876, 905]]<|/det|>
+The posterior distribution for \(R_{0}\) is shown in Supplementary Fig. 6A. We find that this quantity is significantly larger than 1 (between 3.5 and 5 depending on Ct threshold), in agreement with our previous assessment of high spreading potential. It also provides a quantitative justification for the high probability of persistence \((\approx 1)\) shown in the old Supplementary Fig. 6. The latter probability is now shown in Supplementary Fig. 6C against \(R_{0}\) , which we vary through the transmission rate \(\beta\) . As expected, persistence is essentially 0 for \(R_{0} < 1\) , and increases with \(R_{0}\) as \(R_{0} > 1\) . We also provide an estimate of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 81, 856, 220]]<|/det|>
+the basic reproductive number \(R_{0}(closed) = \beta /\mu\) for a closed population \((\nu = 0)\) with constant size \(N\) . As shown in Supplementary Fig. 6, \(R_{0}(closed)\) is extremely high, thus demonstrating the high infection pressure exerted on marketed chickens. Please note that we present \(R_{0}(closed)\) just for illustrative purposes, as the relevant transmission potential is quantified by \(R_{0}\) . All new references have been added to the list of supplementary references.  
+
+<|ref|>image<|/ref|><|det|>[[123, 240, 877, 388]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 394, 870, 630]]<|/det|>
+<center>Supplementary Fig. 6: Basic reproductive number and AIV persistence. (A) Posterior distribution for the basic reproductive number \(R_{0}\) for fits to \(Ct = 40\) (red) and \(Ct = 33\) (teal). The latter is calculated as the basic reproductive number for a continuous time SEEIRR model with a constant rate of chicken removal \(\nu\) , where \(\nu\) is set to the inverse mean length of stay. Combining previous theoretical results yields the expression \(R_{0} = \sum_{b = BR,BY}\left(\frac{2\sigma_{b}}{2\sigma_{b} + \nu}\right)^{2}\mu +\nu \frac{N_{b}}{N}\left(D'Onofrio, Mathematical Biosciences, 2002; Diekman et Al., Journal of the Royal Society Interface, 2010; Champredon et Al., SIAM Journal on Applied Mathematics, 2018), where \(N_{b}\) is the mean number of chickens of type \(b\) present in the LBM, and is estimated through simulations, and \(N = \sum_{b}N_{b}\) is the total number of chickens introduced daily. (B) Posterior distribution for the ratio \(\beta /\mu\) , which represents the basic reproductive number in a closed population \((\nu = 0)\) of constant size \(N\) . (C) Posterior probability of AIV persistence as a function of \(R_{0}\) . This is measured as the proportion of 2000 simulations where at least one latent or infectious chicken is observed at \(t = 50\) days, assuming that all chickens entering the market after \(t = 20\) days are susceptible. Results in A,B are based on 5000 posterior samples. </center>  
+
+<|ref|>text<|/ref|><|det|>[[117, 646, 868, 732]]<|/det|>
+Next, we further characterised the relative contributions of introductions and secondary infections within LBMs, the effective reproductive number \(R_{t}\) and the mean generation time \(GT_{t}\) at different times of the day. These results can be found in Supplementary Fig. 5 and are summarised below:  
+
+<|ref|>text<|/ref|><|det|>[[145, 751, 869, 884]]<|/det|>
+- Panel A shows the proportions of chickens that enter the market as susceptible (S) and become infected (i.e. sold as E, I and R), or infectious (sold as I or R) before being sold. These proportions are further broken down by chicken type. We find that >70% of initially susceptible chickens are infected before being sold. More than 50% of broiler chickens are also able to become infectious before leaving the market due to their short latent period. In contrast, less than 25% of initially susceptible backyard chickens transition from E to I within the market.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[145, 80, 880, 622]]<|/det|>
+Panel B shows the proportion of infection events that were generated by chickens infected in the market rather than elsewhere prior to being offered for sale. While the posterior distribution for this quantity is rather wide, it shows that, on average, more than \(50\%\) of infections are caused by secondary, market- acquired cases. Panels C- F show \(R_{t}\) (C,E) and the mean generation time \(GT_{t}\) (D,F). To measure these quantities, we casted our model in an individual- based framework and tracked all transmission pairs; we then followed Liu et Al., PNAS, 2018 and calculated \(R_{t}\) and \(GT_{t}\) as the mean number and mean generation time of infections generated by incident cases at time \(t\) (i.e. we used a "forward- looking" definition for both \(R_{t}\) and \(GT_{t}\) ). Panels C,D are obtained under the same regime of external introductions considered in panels A,B and in the main analysis. Panels E,F, in contrast, correspond to the same scenario evaluated in Supplementary Fig. 6C, where all admitted chickens are susceptible, and all infections are locally acquired. In this case we find that viral circulation is sustained, i.e. the virus is able to circulate without becoming extinct. Panels C shows that the maxima of \(R_{t}\) correspond to the arrival of new chickens in the market (vertical dotted lines). \(R_{t}\) rises in the \(\sim 12\) hours preceding a new shipment due to the sudden increased availability of susceptible hosts entailed by this event. Indeed, \(GT_{t}\) becomes shorter as a new shipment approaches. The spiky shape displayed by \(R_{t}\) within panel C follows from the observation that most infections occur when a new shipment arrives, i.e. when the size of the susceptible population is at its largest and readily infectious birds are admitted into the market. Removing introductions (panels E,F and Supplementary Fig. 6C) produces smoother \(R_{t}\) and \(GT_{t}\) curves that are not affected by the sudden introduction of infectious (I) birds. Panels \(E,F\) further reinforce our interpretation that \(R_{t} > 1\) is driven by the cyclical enhanced availability of susceptible birds.  
+
+<|ref|>text<|/ref|><|det|>[[116, 643, 861, 777]]<|/det|>
+Overall, our additional analyses suggest that both external viral introductions and high within- market transmission are important drivers of H9N2 AIV epidemiology, as described in the main manuscript. These findings are mentioned in the Results section (below), and the figure is included in the Supplementary Material. These additional results are particularly relevant in the context of our main analysis, since they complement our previous findings and should hence be part of the supplementary material.  
+
+<|ref|>text<|/ref|><|det|>[[116, 794, 861, 915]]<|/det|>
+"The relative importance of external introductions of infected chickens and local transmission is assessed in Supplementary Fig. 5. We find comparable proportions of LBM- acquired infections caused by chickens infected before and after entering the LBM. Moreover, we estimate that within- LBM transmission is high enough ( \(R_{0} = 3.7 - 4.9\) depending on Ct threshold) to ensure long- term persistence of H9N2 AIV within the LBM in the absence of further external introductions (Supplementary Fig. 6)."
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[140, 105, 861, 590]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 595, 875, 787]]<|/det|>
+<center>Supplementary Figure 5: Relative importance of external introductions and local transmission. (A) Violin plots denote posterior distributions for the fraction of initially susceptible chickens that are infected locally and the fraction of those that become infectious before being sold. (B) Posterior distribution for the fraction of LBM-acquired infections that are caused by other chickens that became infected within the LBM, as opposed to externally-introduced infections. Results are based on 20000 simulations from 2000 independent samples from the posterior distribution. (C) Mean number of new cases \((R_{t})\) caused by chickens infected at time \(t\) . (D) Mean generation time versus the timing of primary infection \(t\) . The generation time is defined as the delay between the infection of a primary and a secondary case. Statistics in C,D were obtained by tracking transmission pairs in an individual-based version of our model. Panels E,F mirror C,D but assume that all chickens entering the market after 30 days are susceptible (i.e. in absence of external introductions). We consider all primary cases infected between \(t = 45\) and \(t = 48\) days. For externally-introduced infections, the infection time was set to the time of introduction. Daily shipments of chickens are denoted with vertical lines. Results in C-F based on 2000 simulations using the same number of independent samples from the posterior distribution. </center>  
+
+<|ref|>text<|/ref|><|det|>[[118, 819, 874, 896]]<|/det|>
+A broader point is that the motivation behind the analysis seems to sit a bit between two stools. High prevalence of H9N2 in markets and the issue of control may be important for one of two reasons - first, if it results in substantial zoonotic infection, or if it is important for maintaining the circulation of the virus. However, H9N2, even though it
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 877, 252]]<|/det|>
+may contribute to the generation of zoonotic forms of flu, is itself not a zoonosis. In this case, the prevalence in the market is less important than the role the market has in the circulation of H9N2 - and this can only be evaluated by considering onward transmission from the market. The authors need to make a better case for why their analysis helps us to understand or control that circulation of virus. This is discussed in the introduction (Paragraph starting line 49) but this at best touches upon the circulation question. There is more detail about trading networks in their other papers (e.g. Moyen et al 2021) that they reference, so it would be useful to contextualise the analysis of this LBM in that study (e.g. is the LBM an important source of birds for mobile traders?).  
+
+<|ref|>text<|/ref|><|det|>[[116, 270, 878, 512]]<|/det|>
+We thank the reviewer for this comment. We strongly agree about the importance of embracing a systemic viewpoint in order to investigate pathogen spread across live poultry markets. In fact, we recently developed a computational model (https://doi.org/10.1371/journal.pcbi.1011375) to model AIV transmission within the entire poultry production and distribution network. While viral movements between LBMs represent a topic of crucial importance in general, Moyen et Al. found limited evidence of inter- LBM poultry movements in Chattogram. Nonetheless, onward spread from a LBM to others and farms could occur through the movements of traders' contaminated equipment and vehicles. We respectfully disagree with the reviewer about the nature of H9N2 AIV which, as stated in the Introduction, can be zoonotic (see also Peacock et Al., Viruses, 2019 and Peacock et Al., J. Virol., 2020). It is then reasonable, in our opinion, to expect that larger prevalence increases the risk of AIV exposure for market workers and customers alike. Moreover, larger prevalence in poultry means more opportunities for viral reassortment and more exportations to other markets and farms.  
+
+<|ref|>text<|/ref|><|det|>[[119, 530, 344, 547]]<|/det|>
+A few minor points follow:  
+
+<|ref|>text<|/ref|><|det|>[[119, 565, 283, 583]]<|/det|>
+line 41: "especially"  
+
+<|ref|>text<|/ref|><|det|>[[119, 603, 290, 619]]<|/det|>
+This has been fixed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 638, 850, 676]]<|/det|>
+line 81: this reference is a preprint which is useful to have but it would be better if, the interim has it been submitted for publication and/or accepted.  
+
+<|ref|>text<|/ref|><|det|>[[118, 695, 878, 789]]<|/det|>
+We confirm that the referenced preprint has been submitted for publication. However, we can not guarantee that this manuscript will be published/accepted in the short term as both manuscripts were submitted almost simultaneously. We commit to updating this reference as soon as we receive a notification of acceptance, conditional on the timeline of peer- review for the present work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 807, 875, 863]]<|/det|>
+line 97: I struggled a bit with the definition of control and intervention chickens. Its clearer when you look at the companion preprint (Kohnle et al) but a bit more here would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[116, 881, 805, 898]]<|/det|>
+We have clarified this distinction by adding further details. This sentence now reads:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 848, 170]]<|/det|>
+"We further distinguished between chickens traded along conventional (control, c) and altered (intervention, i) marketing channels. The latter involved purchasing chickens from farms rather than from traders at the market, thus avoiding intermediate transport and storage steps."  
+
+<|ref|>text<|/ref|><|det|>[[118, 186, 848, 224]]<|/det|>
+line 123: this is a striking result - are these all infected in one generation due to a high prevalence of infection of birds entering the market? Or is there another reason?  
+
+<|ref|>text<|/ref|><|det|>[[117, 241, 877, 399]]<|/det|>
+This is one of our main results and follows from a combination of high transmission and high prevalence of infection in introduced chickens. The first aspect creates conditions to guarantee viral persistence in the market (see Supplementary Fig. 6), while the second aspect enhances the probability of infection due to the larger initial number of infections after each shipment. We believe that the new Supplementary Fig. 5 discussed above addresses the relative roles of transmission and introductions in an adequate manner. We emphasize that the survival curve shown in Fig. 1A is conditional on a chicken spending at least \(x\) hours in the market. In reality, only \(11.5\%\) of chickens remain in the market for more than 24 hours, the median length of stay being 16 hours (see Supplementary Fig. 1A).  
+
+<|ref|>text<|/ref|><|det|>[[118, 415, 321, 432]]<|/det|>
+line 139: why "tentative"  
+
+<|ref|>text<|/ref|><|det|>[[118, 453, 314, 470]]<|/det|>
+We removed this word.  
+
+<|ref|>text<|/ref|><|det|>[[117, 487, 880, 620]]<|/det|>
+line 170 onwards. As discussed, it seems that in the model, environmental contamination only differs from direct transmission by allowing for a longer term persistence once infected birds are gone. Otherwise the mechanism of transmission (and impact on transmission) is the same. Kim et al 2018 which the authors cite suggest that there is a strong variability in prevalence across stalls and slaughter areas of the market - is this likely to make a difference here in terms of the way contamination vs direct transmission would work?  
+
+<|ref|>text<|/ref|><|det|>[[116, 639, 880, 903]]<|/det|>
+The reviewer is right in pointing out that both modes of transmission lead to similar dynamical outcomes. In fact, previous theoretical work (Cortez & Weitz, Am. Nat., 2013 and Benson et Al., PLOS Comp. Biol., 2021) demonstrates that direct transmission models can be seen as the limiting cases of indirect transmission models when the dynamics of environmental contamination is fast (as confirmed by our analysis as well). Some dynamical differences may arise in the case of slow environmental contamination, but Cortez & Weitz show that the specific dynamical signatures depend on the underlying model. On one hand, we believe that such similarity is reassuring, as it suggests that different modelling assumptions lead to similar conclusions about given epidemiological aspects. On the other hand, a realistic accounting of environmental contamination is necessary when addressing specific questions, such as assessing the impact of LBM depopulation and cleaning. Heterogeneities in the handling of meat and/or carcasses and cleaning practices across an LBM may lead to distinct outcomes if transmission is direct or indirect. Realistically, the latter mode should better capture differences in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 878, 177]]<|/det|>
+prevalence of contamination across different market areas. For example, market stalls located near slaughter areas could be at an increased risk of contamination than more distant ones. Further discrepancies between models may arise in the case of frequent removal of diseased birds (as shown by Benson et Al., PLOS Comp. Biol., 2021), which we did not include in our model due to the low pathogenicity of H9N2 AIV.  
+
+<|ref|>text<|/ref|><|det|>[[118, 196, 879, 290]]<|/det|>
+We have now incorporated these points in the Discussion, and included the two references mentioned above. We also replaced one reference (Martin et Al., Preventive Veterinary Medicine, 2011) with a more comprehensive reference that reviews the impact of LBM disinfection and rest days (Offeddu et Al., One Health, 2016). The full paragraph discussing environmental contamination now reads:  
+
+<|ref|>text<|/ref|><|det|>[[115, 308, 878, 802]]<|/det|>
+"We considered two alternative modes of transmission, direct and mediated by the environment. Both scenarios were able to explain observed dynamic patterns and yielded similar results in the context of interventions targeting chickens only. Previous theoretical work indeed demonstrated that both modes of transmission lead to similar dynamical outcomes, especially when environmental contamination unfolds on a fast time scale, and that it may be difficult to prefer one or another based solely on prevalence or incidence data (Cortez & Weitz, Am. Nat., 2013; Benson et Al., PLOS Comp. Biol., 2021). This is a reassuring finding as it suggests that some epidemiological conclusions are not affected by precise modelling assumptions. However, further work is needed to identify dynamical signatures of direct and environmental transmission. Nonetheless, incorporating environmental transmission is necessary if the objective is to assess the impact of LBM depopulation and routine cleaning/disinfection, as done in this work. In this case, moderate levels of cleaning were able to curb transmission significantly in simulations, especially with small decay rates, as that corresponds to a slower accumulation of contaminated material. Periodic cleaning/disinfection, usually carried out during rest days, has been shown to reduce AIV burden in Chinese, Hong Kong and Bangladeshi LBMs (Fournié et Al., Journal of the Royal Society Interface, 2011; Offeddu et Al., One Health, 2016; Chowdhury et Al., Emerging Infectious Diseases, 2020). Including environmental transmission may also be more appropriate to capture differences in prevalence of contamination across LBM sections (e.g. stalls and slaughter areas) (Kim et Al., Emerging Infectious Diseases, 2018), and assess how the distance from slaughter areas affects the risk of contamination. Practical difficulties in successfully implementing sanitation in LBMs (Barnett et Al., The Lancet Planetary Health, 2021) further underscore the importance of adopting a multi- pronged approach to reduce the burden of H9N2 AIV in LBMs. Our study also makes the case for the vaccination of poultry intended to be sold in LBMs in Bangladesh."  
+
+<|ref|>text<|/ref|><|det|>[[118, 819, 827, 875]]<|/det|>
+line 220 onwards - are there any references that would help us to understand how difficult it would be to sanitise market? A brief description of what this would entail would be helpful.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 82, 878, 272]]<|/det|>
+Chowdhury et Al., Emerg. Infect. Dis., 2020, which is already referenced in the manuscript, provides evidence of benefits of market cleaning even with monthly frequency in Bangladeshi LBMs. However, Paritosh et Al. Transboundary and Emerging Diseases, 2017 found that a package of biosecurity measures revolving around cleaning did not yield differences in detection rates of H5N1 AIV. Barnett et Al., The Lancet Planetary Health, 2021 suggest that the apparent limited effectiveness of LBM sanitation is due to practical implementation challenges, particularly the difficult social, economic and cultural context in which poultry workers operate. We tried to convey this nuance in the Results section when introducing disinfection (we have also included the new references in the main manuscript):  
+
+<|ref|>text<|/ref|><|det|>[[118, 290, 832, 385]]<|/det|>
+"Daily, weekly, or even monthly cleaning of poultry stalls, with or without weekly disinfection, was found to be associated with lower detection of AIV environmental contamination (Chowdhury et Al, 2019). However, sanitation is not straightforward to implement in practice (Paritosh et Al., Transboundary and Emerging Diseases, 2017; Barnett et Al., The Lancet Planetary Health, 2021)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 404, 858, 460]]<|/det|>
+We believe that practical difficulties in achieving sanitation further underscore the importance of combining multiple interventions, which is one of our main conclusions. This point is now highlighted in the Discussion section as well:  
+
+<|ref|>text<|/ref|><|det|>[[118, 479, 864, 535]]<|/det|>
+"Practical difficulties in successfully implementing sanitation in LBMs (Barnett et Al., The Lancet Planetary Health, 2021) further underscore the importance of adopting a multi- pronged approach to reduce the burden of H9N2 AIV in LBMs."  
+
+<|ref|>text<|/ref|><|det|>[[118, 552, 856, 628]]<|/det|>
+line 227: there is a useful point here in terms of the recovery of infection - and it would therefore be important to gain a better understanding of why - referring to my earlier general point, is it because, for example the R value is very high? Or is it because the incoming infection prevalence is high.  
+
+<|ref|>text<|/ref|><|det|>[[117, 647, 876, 779]]<|/det|>
+We interpret the reviewer's comment as addressing the rapid recovery of infection following depopulation in the case of direct transmission rather than the difference between the latter and environmental transmission. As discussed above (see new Supplementary Fig. 5), fast transmission is due to the large transmission rate, while frequent introductions facilitate a quick return to endemic levels of prevalence. Indeed, if introductions were sporadic (e.g. not every day) it would then take longer to build up prevalence.  
+
+<|ref|>text<|/ref|><|det|>[[117, 796, 870, 890]]<|/det|>
+line 342: That the two scenarios of transmission yield similar result is reassuring but as noted earlier, I suspect this may at least partially be because the dynamics in the model really aren't all that different. While the authors cannot be expected to do everything in a single paper, I think some discussion of the implications of model similarity is worthwhile.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 454, 100]]<|/det|>
+We already addressed this point above.  
+
+<|ref|>text<|/ref|><|det|>[[118, 120, 718, 138]]<|/det|>
+line 361: "we believe that our main results ... are not affected ..." - why?  
+
+<|ref|>text<|/ref|><|det|>[[117, 158, 877, 289]]<|/det|>
+We expect that including additional chicken types would have a small impact on our estimates since broiler and backyard chickens already represent a large proportion of poultry being traded in the LBM (Supplementary Fig. 1). Moreover, we know from additional field studies that prevalence in other chicken types is not higher than in broilers and backyard chickens (Kim et Al., Emerging Infectious Diseases, 2018). This suggests a smaller contribution of other chicken types to overall transmission compared to broilers.  
+
+<|ref|>text<|/ref|><|det|>[[118, 309, 830, 327]]<|/det|>
+line 381: "with similar conditions" - which are the key conditions that are important?  
+
+<|ref|>text<|/ref|><|det|>[[118, 345, 775, 363]]<|/det|>
+We slightly rewrote the final paragraph in the Discussion to expand on this point:  
+
+<|ref|>text<|/ref|><|det|>[[118, 380, 877, 455]]<|/det|>
+"The ubiquity of similar poultry handling and trading practices suggests that our findings apply to other Bangladeshi LBMs as well. Applications of the model to other LBMs may require calibrating specific LBM features such as the number of chickens being traded and their length of stay."  
+
+<|ref|>text<|/ref|><|det|>[[118, 472, 848, 529]]<|/det|>
+S.92 onwards. after all long discussion of the effects of environmental conditions on environmental survival, on line S.103 the authors then reduce the difference to one of temperature only.  
+
+<|ref|>text<|/ref|><|det|>[[117, 548, 878, 680]]<|/det|>
+The purpose of this section was to discuss potential factors affecting environmental survival while also presenting a broad range of values for the decay rate \(\Theta\) that were also realistic for this host- pathogen system. The sentence at line S.103 just places the chosen values into the context of Handel at Al.'s findings (reference 2 in the supplementary), which include an explicit, AIV subtype- specific relationship between \(\Theta\) and temperature. We agree that this sentence may read confusing, and decided to remove it in the revised version of the Supplementary Material.  
+
+<|ref|>text<|/ref|><|det|>[[117, 699, 873, 855]]<|/det|>
+To further support the appropriateness of our choice of \(\Theta\) , we report various estimates of \(\Theta^{- 1}\) from the literature as a function of water temperature (see Fig. B below). The relationship presented by Handel at Al. is shown in black, while individual red dots are distinct estimates retrieved from a systematic review. Importantly, these estimates were obtained under a variety of environmental conditions beyond temperature. Blue and orange lines correspond to values used in two distinct modelling studies investigating AIV transmission in poultry markets. We prefer to keep Fig. B in this letter as we believe that current references in the Supplementary Material already provide sufficient context.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[235, 88, 760, 365]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[309, 376, 684, 393]]<|/det|>
+<center>Fig. B: Summary of estimates of \(\theta^{-1}\) from the literature. </center>  
+
+<|ref|>text<|/ref|><|det|>[[118, 406, 857, 483]]<|/det|>
+supp fig 7 - as discussed a the beginning of this review, really useful to have this but more details on the parameter sets - were they chosen to be relevant to the posteriors for the real data (e.g. drawn from the posterior distributions)? Also, the mismatch in scenario 5 is indeed striking and it would be good to have confidence it isn't important.  
+
+<|ref|>text<|/ref|><|det|>[[118, 500, 878, 614]]<|/det|>
+These parameter values were not selected from the posterior, although the scales of some of these were consistent with posterior estimates reported in the main manuscript. Our rationale was to verify not only whether we could recover initial parameters under optimal conditions (i.e. in absence of model misspecification), but also differences between chicken types and recruitment groups. This is now explained in the caption of Supplementary Fig. 7 (now 9)  
+
+<|ref|>text<|/ref|><|det|>[[115, 631, 877, 864]]<|/det|>
+The mismatch occurring in scenario 5 is evident mostly in the context of the duration of infection \(T_{_I}\) . However, as we describe in the corresponding caption, this is partially intended since scenario 5 uses the same parameter set of scenario 4, but uses a misspecified prior on the total duration of shedding \(T_{_E} + T_{_I}\) . The observed mismatch arises because our data is mostly informative about \(T_{_E}\) and not \(T_{_I}\) , as recovery from infection is not tracked in our experimental setting. Please note that some degree of prior misspecification on \(T_{_E} + T_{_I}\) is present also in scenarios 1- 4, but is not as strong as in scenario 5. We have added a brief sentence in the caption of Supplementary Fig. 7 (now 9) to elucidate this point. We also note that while parameters describing the timing of prior infections \((\lambda , \kappa)\) also show some mismatch, the mean and standard deviation of the corresponding distributions display a better agreement with the original values.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 848, 140]]<|/det|>
+As discussed above, mismatches between original and recovered parameters did not significantly affect our assessment of effectiveness of interventions (see new Supplementary Fig. 13).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 158, 216, 175]]<|/det|>
+## Reviewer 2  
+
+<|ref|>text<|/ref|><|det|>[[118, 194, 876, 290]]<|/det|>
+The authors present an interesting study modelling the transmission dynamics of avian influenza with parameters fit to empirical data from a live bird market. I find the model proposed to be reasonable; the statistical analysis appropriate; and the limitations of the approach to be unambiguously communicated. Generally, I find the manuscript to be uncommonly clearly written.  
+
+<|ref|>text<|/ref|><|det|>[[118, 308, 707, 326]]<|/det|>
+We thank the reviewer for their very positive assessment of our work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 343, 879, 438]]<|/det|>
+Regarding the presentation of the model, I would encourage the authors to briefly state (perhaps any additional analysis is beyond the scope of this manuscript) how they would expect compartment topology to modify the parameter estimates or otherwise invest more time in the model description to motivate why SEEIRR is clearly the most appropriate framework.  
+
+<|ref|>text<|/ref|><|det|>[[118, 456, 877, 513]]<|/det|>
+We thank the reviewer for their suggestion. We now mention in the model description that our model is a variant of the SEIR model, which is commonly used to investigate AIV dynamics:  
+
+<|ref|>text<|/ref|><|det|>[[118, 531, 860, 608]]<|/det|>
+"This model is a variant of the more common SEIR model, which is typically used to investigate AIV dynamics (Fournié et Al., Mathematical Models of Infectious Diseases in Livestock: Concepts and Application to the Spread of Highly Pathogenic Avian Influenza Virus Strain Type H5N1, 2011)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 626, 855, 664]]<|/det|>
+We added the new reference to the list of references. Furthermore, we commented on the choice of including two and one latent and infectious compartments, respectively:  
+
+<|ref|>text<|/ref|><|det|>[[117, 681, 867, 796]]<|/det|>
+"This is often regarded as a more realistic assumption than an exponential distribution of durations, which is implicit in models with single- staged compartments (Bouma et Al., PLOS Pathogens, 2009). On the other hand, including a single infectious compartment should not affect our results significantly since the main limitation to transmission within LBMs comes from the short length of stay. [...]  
+
+<|ref|>text<|/ref|><|det|>[[118, 799, 844, 857]]<|/det|>
+The distinction between \(R^{+}\) and \(R^{- }\) compartments allows to capture the persistence of viral RNA in infected chickens that recently stopped shedding (Griffin D.E., PLoS Pathogens, 2022)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 875, 815, 912]]<|/det|>
+The reference to Griffin D.E., PLoS Pathogens, 2022 was also included in the list of references.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 102, 830, 139]]<|/det|>
+The authors describe several predicted epidemiological differences between exotic broilers and backyard chickens including:  
+
+<|ref|>text<|/ref|><|det|>[[118, 156, 870, 212]]<|/det|>
+"From our model's output, we found a shorter latent period in exotic broiler compared to backyard chickens (Fig. 1B,C), lasting an average of 5.3 hours for exotic broiler, and 1 days for backyard chickens."  
+
+<|ref|>text<|/ref|><|det|>[[118, 215, 152, 229]]<|/det|>
+and  
+
+<|ref|>text<|/ref|><|det|>[[118, 232, 878, 345]]<|/det|>
+"in the case of exotic broilers, most chickens with prior exposure to H9N2 were either infectious or latent, with only a minor proportion of them being immune (Fig. 1E). In contrast, most previously- exposed backyard chickens were immune to H9N2 (Fig. 1F). Our results thus tentatively suggest that prior infection occurs close to marketing age for broilers, whereas in backyard chickens it may occur further in the past, which is consistent with the latter being raised for a longer time compared to broilers."  
+
+<|ref|>text<|/ref|><|det|>[[117, 362, 864, 550]]<|/det|>
+I encourage the authors to comment on whether they believe these findings are entirely consistent with variable patterns of prior exposure for these two populations or if the data presented suggest that there may be some unrecognized genetic determinants of susceptibility. Additionally I understand that to support the second passage quoted above, there has been an underlying assumption made that prior infection confers long lasting immunity (relative to the life cycle of a market chicken). If this is correct, I encourage the authors to explicitly state this assumption and perhaps introduce some additional background information regarding the market chicken life cycle so that the unfamiliar reader, like myself, may have some expectation of the maximum number of infections for one individual.  
+
+<|ref|>text<|/ref|><|det|>[[118, 570, 464, 587]]<|/det|>
+We thank the reviewer for this comment.  
+
+<|ref|>text<|/ref|><|det|>[[118, 607, 872, 644]]<|/det|>
+We have now provided some context regarding the ages at sale of broilers and backyard chickens in the Results section:  
+
+<|ref|>text<|/ref|><|det|>[[117, 663, 876, 777]]<|/det|>
+"These findings are consistent with known rearing practices and ages at sale of each chicken type: broilers are selectively bred to grow rapidly, and are sold for meat after just 28- 31 days after hatching (Hennessey et Al., Preventive Veterinary Medicine, 2021). Backyard chickens are instead raised for meat and eggs in rural households and can reach much older ages. For context, backyard chickens in our dataset were aged between 90 and 720 days."  
+
+<|ref|>text<|/ref|><|det|>[[117, 796, 866, 909]]<|/det|>
+Our findings are also consistent with a previous study (Das Gupta et Al., Transboundary And Emerging Diseases, 2020) that found a higher seroprevalence of antibodies against H9 among backyard chickens in farms surrounding Chattogram, i.e. where the present LBM is located. While we cannot rule out the existence of unrecognised genetic determinants (which we already mention in the Discussion), our findings suggest that these are not necessary to explain observed differences in incidence in broilers and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 866, 177]]<|/det|>
+backyard chickens. Nonetheless, the reviewer is right in pointing out that this interpretation rests on the role of pre- existing immunity. Ultimately, further confirmation about either hypothesis (which are not mutually exclusive) will require a better understanding of the role of previous immunity. We have discussed these aspects in multiple points in the Discussion:  
+
+<|ref|>text<|/ref|><|det|>[[118, 196, 861, 290]]<|/det|>
+"Our results indicate, however, that differences in earlier exposure to H9N2 AIV are sufficient to explain observed incidence patterns, and are consistent with known ages at sale and levels of H9 seropositivity in broilers and backyard chickens (Das Gupta et Al., 2021). A better understanding of the effectiveness of prior immunity, which we have assumed to be sterilising, will help validating or confuting this interpretation."  
+
+<|ref|>text<|/ref|><|det|>[[117, 309, 874, 477]]<|/det|>
+The authors consider both direct and environmental transmission, writing, "lenv(t) accumulates due to shedding from infectious chickens and decays progressively at rate 0. Here we consider three values of 0, namely \(\Theta - 1 = 10,3,1\) days, corresponding to slow, intermediate and fast decay, respectively. These values are based on actual estimates from the scientific literature and capture a broad range of environmental conditions." I encourage the authors to briefly describe the physical nature of environmental transmission in a live bird market for readers, like myself, who are more familiar with influenza transmission dynamics in the human population which I expect substantially differ.  
+
+<|ref|>text<|/ref|><|det|>[[118, 497, 857, 534]]<|/det|>
+We thank the reviewer for their comment. We have now described the main sources of environmental contamination in the Results section:  
+
+<|ref|>text<|/ref|><|det|>[[118, 553, 857, 628]]<|/det|>
+"Within LBMs, the infection can be transmitted through poultry drinking water, as well as cages and floors that are contaminated by faecal material and during slaughtering. The lack of disinfection and the constant moving of cages and birds further promote poultry exposure to environmental contamination."  
+
+<|ref|>text<|/ref|><|det|>[[117, 647, 875, 760]]<|/det|>
+I find one of the more striking findings reported in the manuscript to be that while, "reduced length of stay and reduced probability of prior exposure, proved to be only modestly effective. Combining both interventions proved considerably more effective at reducing transmission compared to individual measures" I encourage the authors to clarify in the text if this effect is in fact nonadditive and if so to include some description of the expected mechanism behind the non- additivity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 780, 874, 857]]<|/det|>
+The reviewer raises an interesting point that we address in the new Supplementary Fig. 14 shown below. Our reasoning is as follows: if the effect of each intervention in terms of reducing cumulative daily prevalence was additive, then their combined effectiveness \(E'\) would be given by \(E' = E_1 + E_2\) (this of course holds up if and only if \(E' \leq 1\) ).  
+
+<|ref|>text<|/ref|><|det|>[[118, 860, 875, 898]]<|/det|>
+Analogously, if their effects were multiplicative, the combined effectiveness would be \(E' = 1 - (1 - E_1)(1 - E_2)\) . To gauge any signatures of additivity and/or multiplicativity,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 870, 215]]<|/det|>
+we measured the combined effectiveness from simulations \(E\) (panels A,B), and calculated the difference \(E - E'\) under either expectation for \(E'\) (panels C- F). Interestingly, the positive effects of these interventions do not seem to combine additively or multiplicatively. In particular, the fact that \(\Delta E > 0\) for most configurations, especially for stricter control of importations and shorter length of stay, is suggestive of a synergistic effect of stacking these interventions. This is mentioned in the Result section:  
+
+<|ref|>text<|/ref|><|det|>[[118, 233, 865, 291]]<|/det|>
+"Notably, the benefits of combining (i) and (ii) exceed expectations under the assumption that their effects were additive or multiplicative, hence suggesting a synergistic effect of multiple interventions (Supplementary Fig. 14)."  
+
+<|ref|>image<|/ref|><|det|>[[210, 312, 789, 867]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[115, 875, 864, 909]]<|/det|>
+<center>Supplementary Figure 14: Synergistic effects of combining interventions. This figure considers a simultaneous reduction of the maximum length of stay of chickens to \(T_{m}\) and the probability of prior exposure \(\rho_{c,b}\) to \((1 - r)\rho_{c,b}\) </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 81, 866, 247]]<|/det|>
+(A,B). The effectiveness of these interventions combined is denoted with \(E\) , and is defined as the reduction in cumulative daily prevalence relative to a scenario with no intervention in place. The latter corresponds to \(T_{m} = 5\) days and \(r = 0\) . The effectiveness of individual interventions is denoted as \(E_{1}(T_{m} < 5\) days and \(r = 0\) ) and \(E_{2}(T_{m} = 5\) days and \(0 < r \leq 1\) ). Panels C-F assess whether the effects of individual interventions add up in a multiplicative (C,D) or additive fashion (E,F). If the effect was purely multiplicative, the expected, combined effectiveness would be \(\bar{E} = 1 - (1 - E_{1})(1 - E_{2})\) . In the purely additive case it would be \(\bar{E} = E_{1} + E_{2}\) . C-F show that \(E > \bar{E}\) for most values of \(T_{m}\) and \(r\) , suggesting a synergistic effect of implementing these interventions simultaneously. Results are based on 2000 simulations from 200 independent samples from the posterior distribution. The first and second columns correspond to the fits to Ct=40 and Ct=33 data, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[118, 262, 857, 300]]<|/det|>
+In conclusion, the authors present an interesting and well written study which I believe will be of broad interest.  
+
+<|ref|>text<|/ref|><|det|>[[118, 317, 568, 355]]<|/det|>
+Sincerely, Nash Rochman (invited 12/5/23, returned 12/19/23)  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 372, 216, 390]]<|/det|>
+## Reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[118, 408, 847, 504]]<|/det|>
+The manuscript titled "Modelling the transmission dynamics of H9N2 avian influenza viruses in a live bird market" aims to understand the transmission dynamics of H9N2 avian influenza viruses in live bird markets (LBMs) in Bangladesh. The study uses a mechanistic transmission model, fitted to field data collected in a specific LBM, to simulate the spread of the virus among poultry.  
+
+<|ref|>text<|/ref|><|det|>[[117, 520, 875, 710]]<|/det|>
+The study presents significant findings on H9N2 transmission in LBMs, highlighting a rapid infection rate among susceptible chickens, often within a day. The virus's latent period is notably short, particularly in broiler chickens, sometimes as brief as 5.3 hours, facilitating its swift spread. It also uncovers considerable differences in exposure and immunity between exotic broiler and backyard chickens. Its unique approach, combining field data with mathematical modelling, and the robustness of its findings, underscored by sensitivity analyses and Bayesian methods, mark it as particularly relevant for areas where H9N2 is prevalent, guiding interventions to control its spread. Thus, This research offers crucial insights into the epidemiology of H9N2 in LBMs, vital for public health and the poultry industry.  
+
+<|ref|>text<|/ref|><|det|>[[117, 726, 875, 896]]<|/det|>
+I thoroughly enjoyed reading this manuscript and believe it is already of high quality. However, I suggest incorporating discussions on several interesting points. Firstly, the study's results are drawn from a single LBM in Bangladesh. It would be beneficial to explore how these findings might apply to other LBMs with varying dynamics, bird populations, and environments. Additionally, a more in- depth examination of environmental factors, like temperature, humidity, and market hygiene, is crucial as they significantly influence H9N2's survival and spread. A comparison with other avian influenza strains could also offer a broader perspective, highlighting differences and similarities in transmission dynamics. Finally, it would be interesting to consider whether
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 864, 139]]<|/det|>
+these findings could be applicable to livestock markets other than those for birds and discuss the generalizability of such high infection rates in group settings, dependent on the virus  
+
+<|ref|>text<|/ref|><|det|>[[118, 159, 816, 196]]<|/det|>
+We are pleased that the reviewer enjoyed our manuscript and thank them for their comments.  
+
+<|ref|>text<|/ref|><|det|>[[118, 215, 848, 234]]<|/det|>
+We have addressed the point on applicability to other LBMs in our reply to Reviewer 1.  
+
+<|ref|>text<|/ref|><|det|>[[118, 253, 857, 347]]<|/det|>
+We have now added a sentence to discuss potential applications to other LBMs, live animal markets and host- pathogen systems. There, we also argue for the need to account for the main AIV strain- specific characteristics and environmental drivers of transmission, such as the presence of suitable hosts and those factors that affect AIV survival in the environment (which are described in the Supplementary Methods).  
+
+<|ref|>text<|/ref|><|det|>[[116, 365, 877, 555]]<|/det|>
+"The ubiquity of similar poultry handling and trading practices suggests that our findings apply to other Bangladeshi LBMs as well. Applications of the model to other LBMs may require calibrating specific LBM features such as the number of chickens being traded and their length of stay. Applications to other AIV strains, e.g. H5N1, will also require accounting for the presence in the LBM of relevant hosts species (e.g. waterfowl in the case of H5N1 AIV), their specific infection parameters and differential ability to survive in the environment (Handel et Al., PLOS Comp. Biol., 2013). Finally, we note that our modelling framework could be applied to disentangle the contributions of external introductions and local transmission in other types of live animal markets and host- pathogen systems."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 84, 310, 98]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[119, 111, 415, 126]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 140, 878, 183]]<|/det|>
+The authors have done an excellent job on their revisions and I have no further comments (and thank them for their respectful disagreement in regards to zoonotic potential - tbh, I think they are in a better position than I to judge this).  
+
+<|ref|>text<|/ref|><|det|>[[119, 250, 629, 264]]<|/det|>
+I believe the authors have fully addressed the reviewer comments.  
+
+<|ref|>text<|/ref|><|det|>[[118, 279, 234, 306]]<|/det|>
+Sincerely,  Nash Rochman
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a27a9071773bd5448442fddc2e7d62ff8da38764cfd258c9ac0e24d0109e8f59/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__a27a9071773bd5448442fddc2e7d62ff8da38764cfd258c9ac0e24d0109e8f59/supplementary_0_Peer Review File__a27a9071773bd5448442fddc2e7d62ff8da38764cfd258c9ac0e24d0109e8f59.mmd

@@ -0,0 +1,388 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Afadin couples RAS GTPases to the polarity rheostat Scribble  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+Manuscript title: Afadin couples RAS GTPases to the polarity rheostat Scribble  
+
+By: Marilyn Goudreault, Valérie Gagné, Chang Hwa Jo, Swati Singh, Ryan C. Killoran,  
+
+Anne- Claude Gingras, and Matthew J. Smith  
+
+Comments: This is an extraordinary effort to characterize AFDN interaction network, using proximity labelling- MS approach. Authors were looking into both isoforms of AFDN as well. They found that AFDN binds to its previously unknown interactor, SCRIB, adhesion/polarity tumor suppressor protein. Both AFDN and SCRIB were found to be linked with RAS- induced cell growth and invasion. Next, they were looking into specific binding domains, involved in AFDN- SCRIBs complex and the molecular mechanism of this interaction. They found that that interaction is mediated through the first PDZ domain of SCRIB and the AFDN FHA domain. Further, the knockout experiments reveal the important role of AFDN- SCRIB interaction in MAPK and PI3K activation, and cell motility.  
+
+This data provide insight into new roles of AFDN, as not- so- well characterized RAS effector. I appreciate the quality of results, data and figures, and recommend this manuscript for publication.  
+
+Reviewer #2 (Remarks to the Author):  
+
+In the manuscript by Goudreault et al. entitled 'Afadin couples RAS GTPases to the polarity rheostat protein Scribble', the investigators employ proximity- based proteomics using afadin (AFDN) isoforms and identify the polarity protein SCRIB/Scribble as the top hit. Additional support for this interaction is obtained using in vitro binding assays and in- cell pull down experiments in which the authors find that the first PDZ domain of SCRIB and the AFDN FHA domain directly associate. AFDN has previously been shown to be an effector of the RAS and RAP GTPases and binds through its RA domains. Consistent with these earlier observations, the investigators find that KRAS co- localizes with AFDN to facilitate formation of the AFDN- SCRIB complex. Knockout of AFDN or SCRIB in MCF7 epithelial cells disrupts growth dependent MAPK and PI3K activation and inhibits cell motility. Results from these studies provide a new link between AFDN, SCRIB and RAS- mediated MAPK signaling.
+
+<--- Page Split --->
+
+1) The authors state that 'We observed in Fig. 2F that only the N-SCRIB, but not C-SCRIB construct, is able to pulldown the AFDN-FHA fragment, though the C-SCRIB contains 3 additional PDZ domains.' Have the 4 PDZ domains been compared to assess possible sequence determinants (e.g. specificity for the first PDZ domain)? Also, as the PDZ domains may interact or be involved in autoinhibitory interactions, have the investigators probed the individual PDZ domains or the 4 PDZ domains together?  
+
+2) Have the investigators compared binding of full length AFDN and SCRIB proteins with the individual domains (FHA/PDZ1) by in vitro assays? This comparison would be helpful in assessing whether the FHA and PDZ1 domains are sufficient for binding of AFDN and SCRIB, especially since the binding appears weak (~22 uM). Although full-length AFDN is compared with AFDN delta FHA (Figure 2I), a faint band is discernible for the FHA deleted protein suggesting that other regions of the protein may be involved in the interaction. A similar comparison is also not made for the SCRIB protein (Figure 2J).  
+
+3) In Fig. 2C and 2D, IP with AFDN results in a strong SCRIB band with ectopic expression. Interestingly, this interaction looks to be retained with endogenous protein in panel 2D. However, the level of endogenous SCRIB protein pulled down looks to be much less than the GFP-tagged version despite high levels of endogenous expression. Is there a hypothesis for why this is observed?  
+
+4) The authors state 'To map the interaction site on the SCRIB PDZ1 domain we used NMR spectroscopy. BMRBid 11207 was used to assign 71% of the peaks in a 1H/15N-HSQC of 15N-labelled SCRIB PDZ1.' It would be helpful state the conditions (pH, salt, etc) for published assignments in comparison to the NMR data collected, as the assignments are key for chemical shift mapping of the binding interaction. Please also indicate how well spectra overlay with previously determined assignments. Also, in Figure 3E, the full HSQC spectrum should be shown (with zoom as appropriate). A docked model of the interaction with the AFDN FHA domain could be generated using the NMR data but not included.  
+
+5) It is unclear whether there is a difference in the construct used in figure 3D (371-532) and Figure 3F (371-531). If they are the same then a correction is required. If they are different then the rationale for using two constructs with a difference in 1 amino acid is needed. Also, Figure 3D and 3F could be combined.  
+
+6) In Fig 3C, the minus phosphatase control for the pulldown is not shown.  
+
+7) Additional bands are observed in Fig. 3D, 3F for experiments with purified protein. Is this due to degradation or other protein contaminants? Some discussion is needed here.
+
+<--- Page Split --->
+
+8) In Fig. 4, it is stated that densitometry analysis of western blots showed an 8-fold increase in SCRIB protein pulled down with AFDN/KRAS G12V co-expression. However, this data does not appear to be included in the submission.  
+
+9) In Fig. 4D, authors show that increased SCRIB precipitation is observed with AFDN in the presence of KRAS G12V. AFDN is also shown to be immunoprecipitated with KRAS and related RAP GTPases through the RA domains. Interestingly, in panel 4D, some RAP GTPase appears to be pulled down with just FLAG-AFDN and EGFP-vector, but the same is not observed for KRAS G12V here. What is the source of the disconnect between panel 4C and 4D with regard to AFDN/KRAS G12V?  
+
+10) Similarly in Fig. 4E, a FLAG pulldown control absent of FLAG-AFDN shows modest levels of KRAS G12V immunoprecipitated. Is this due to residual SCRIB interaction with KRAS or an artifact using a FLAG-antibody?  
+
+11) Fig. 4F and 4G require reworking for presentation purposes. The Z-stacks are not very legible given the intent of showing merged co-localization of KRAS:AFDN:SCRIB. Though the figure panel is intended to show increased localization of KRAS G12V vs. KRAS WT to AFDN:SCRIB, this point is not currently made and it is unclear whether this is due to data interpretation or simply representation of the data. The investigators may want to consider alternative color mapping for better contrast of co-localization points, or a larger zoom/view for readers to see the merged fluorescence channels. Additionally, differentiation between SCRIB/AFDN merge vs. SCRIB/AFDN/KRAS merges could be useful for interpretation of the two separate complexes.  
+
+12) The same concern exists with Fig. 5E. The authors should provide a better representation for the immunofluorescence localization and internalization of AFDN. They state that expression of KRAS G12V in the SCRIB KO line induces internalization of AFDN in the figure legends, but this is not addressed in the Results proper. Rather, in Results, the authors state that AFDN is retained at sites of cell-cell contact.  
+
+13) Similar concerns also exist with Fig. 6A and 6B regarding representation of the data. In 6A, it appears that SCRIB may be less recruited to cell contacts with the FHA deletion of AFDN as compared to WT, though authors state the opposite. The trend the authors state, is more clear in the reverse scenario with AFDN and the SCRIB PDZ truncation in that there is less AFDN association at the cell contacts. Perhaps an alternative quantification representation for the co-localization of AFDN and SCRIB will better illustrate the authors' point/clarify the data.  
+
+14) Minor comment. In the Fig. 7 legend consider adding \((A - C)\) at beginning of legend for panels \(A - C\) and \((D - E)\) to maintain parallel structure.
+
+<--- Page Split --->
+
+15) The authors state that a signaling "defect" for the ERK MAPK/PI3K-AKT signaling cascade is induced with KO of AFDN or SCRIB. They may want to consider rewording, as the initial increase in temporal signaling indicates that signaling is not defective, but rather altered. Have the authors examined this temporal signaling effect of AFDN/SCRIB KO in the context of an activated KRAS such as KRAS G12V?  
+
+16) Statistics for pERK and pAKT is lacking to show a difference between "WT"/AFDN or SCRIB KO?  
+
+17) In Fig. 7l, consider a directional quantification of the leading-edge Golgi stain or a better zoomed-in insert to emphasize the loss of directionality in SCRIB/AFDN KO cells.  
+
+18) Mechanistic insight into how the AFSN/SCRIB effects the RAS-mediated pERK or pAKT activation would aid the discussion.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The manuscript "Afadin couples RAS GTPases to the polarity rheostat Scribble" identifies a novel proximity interaction between Afadin and Scribble, using BioID coupled to mass spectrometry. A series of carefully crafted IP experiments identified that the PDZ domain of Scribble interacts with FHA domain of AFDN. Furthermore, the authors characterised the interaction between the domains using different techniques to reveal a model of how the two domains bind to each other. Since, AFDN binds to several RAS GTPases, the authors determined and characterised the interaction of activated forms of the GTPases with AFDN-SCRIB complex. Using CRISPR/Cas9 gene editing, the authors created a suite of single and double KO cell lines of AFDN and SCRIB. Using these cells lines, the authors further support a KRAS-AFDN-SCRIB complex formation and the requirement of a direct FHA-PDZ1 interaction for proper localisation of AFDN-SCRIB at cell contacts. Finally, the authors show that the loss of either AFDN or SCRIB disrupts ERK and AKT activation kinetics and cell motility in growth factor-dependent manner.  
+
+The manuscript is well written and experiments are performed with nice controls. The observations and conclusions drawn in this manuscript would help advance the field in a significant way. The data presented in the manuscript provides ample support to the conclusions drawn. I have few suggestions that would improve the manuscript:
+
+<--- Page Split --->
+
+1: BirA\* is a large tag and can cause a significant level of mislocalisation when expressed in cells. Since the BioID results show some proteins from other compartments like ER/Golgi, mitochondria, etc, it would be nice if the authors can determine subcellular localisation of the AFDN- BirA\* tagged fusion proteins (both isoforms).  
+
+2: Since the authors have created multiple CRISPR/Cas9 gene edited cell lines and assessed the precise nature of the edit by Sanger sequencing (line: 693), it would be recommended to show the precise edits and how they impact expression of that gene (i.e. Introduce STOP codon, impact splicing, destroy START codon, frame- shift, etc).  
+
+Reviewer #4 (Remarks to the Author):  
+
+Afadin (AFDN), a regulator of cell- cell contacts, has long been recognized as an effector of RAS and related small GTPases but little is known about the interaction or its functional consequences. AFDN is unusual among the dozen or so effectors of RAS because it possesses two tandem RA domains. Goudreault et al. set out to explore the AFDN interactome by proximity labeling and here report a comprehensive characterization of the interaction with their most robust hit, SCRIB, a tumor suppressor and polarity protein that possesses four tandem PDZ domains. They show that AFDN and SCRIB associate via a non- canonical interaction of the first PDZ domain with the forkhead associated (FHA) domain of AFDN and that the interaction is enhanced by GTP- bound KRAS that forms a ternary complex with the two polarity proteins. Conversely, they show that silencing AFDN or SCRIB changes the kinetics of growth- factor stimulated ERK and AKT signaling in MCF7 epithelial cells.  
+
+The manuscript is exceedingly well written. The authors walk the reader through not only the experiments and results but also the thinking behind them. The turbolD proximity screen is well done, appropriately controlled, and clearly reported. The biochemical validation of the interaction of AFDN and SCRIB is outstanding, particularly the rigor applied to confirming the non- canonical nature of the interaction between the AFDN FHA domain with the first PDZ domain of SCRIB. The promiscuity of the AFDN RA domain(s) for RAS family proteins relative to the specificity of the RAF1 RBD is well demonstrated. These results are clear, novel, and of interest to cell biologists and are certainly worthy of reporting.
+
+<--- Page Split --->
+
+The weakest part of the study is the overinterpretation of the immunofluorescent localizations and co- localizations of KRAS, AFDN and SCRIB.  
+
+Fig. 2A. Here AFDN and SCRIB, both epitope- tagged, are overexpressed in a HeLa cell that is processed for immunofluorescence. It is not stated if the GFP- AFDN is imaged with the intrinsic fluorescence of GFP or if an anti- GFP antibody is employed along with the anti- FLAG antibody. No control proteins are employed nor is a control with first antibody omitted shown. The localization is indeterminant and uninformative. A single cell is shown such that one cannot determine if this represents the predominant fluorescent pattern. Since tagged, ectopically expressed proteins are used, including one tagged with GFP, it is not clear why the authors did not use mCherry- SCRIB such that they could colocalize the two proteins in live cells, which allow for more precise subcellular localization free of fixation and permeabilization artifacts.  
+
+Fig. 4F,G. In describing these micrographs the authors state on p. 9 that "tagged, wild- type KRAS does not significantly alter AFDN or SCRIB localization and does not co- localize with these proteins (Figure 4F and S4B). ... In contrast, expression of KRAS- G12V markedly disrupted cell- cell contacts and was noticeably co- localized with endogenous AFDN and SCRIB, as determined by z- plane projections (Figure 4G)." In the discussion on p. 13 the authors write that "we show that RAS- G12V is co- localized with AFDN and SCRIB at sites of cell contact, while wild- type RAS is distributed more generally across the plasma membrane." The data do not support these conclusions. First, not shown is any disruption of cell- cell contacts in cells expressing KRAS- G12V. Three GFP- KRAS4B- G12V expressing cells are shown with three levels of expression and perhaps different z- planes. Two of these three maintain robust cell- cell contacts as determined by morphology and SCRIB and AFDN staining of areas of cell contact (anti- ZO- 1 staining would be a way to look at this without imaging the experimental proteins themselves). Not shown are any of the detached cells to which the authors refer as having lost cell- cell contacts as a function of oncogenic KRAS. Second, and more important, the data show that both WT- KRAS and KRAS- G12V decorate the basolateral membrane. Indeed, both the WT and mutant KRAS decorate the entire plasma membrane (PM) as is well established in a vast literature. As expected for cell adhesion proteins imaged in confluent epithelial cells, SCRIB and AFDN decorate primarily the basolateral membrane. The conclusion that these proteins are colocalized to a greater extent with KRAS- G12V that with WT KRAS is not supported by the data shown and contrary to a vast literature on KRAS localization. Some of the problem is semantic; the concept of co- localization is somewhat ambiguous. There is co- localization on the PM in some regions of the cell but not others. This should not necessarily be interpreted as one protein pulling another to a region of PM since the same pattern would be observed if the localizations are true but unrelated to the direct interactions of the proteins. The three proteins colocalize at the basolateral membrane but not the apical membrane and this is not affected by the GTP- binding state of KRAS. Were this localization of mutant KRAS to differ from that of WT it would be contrary to a vast literature on KRAS localization that in total demonstrates that the subcellular localization driven by the prenylated HVR is not affected by the GTP- binding state. Current paradigms of RAS signaling hold the PM localization of KRAS is constitutive and it is effectors that are drawn to RAS (e.g. translocation of RAF) not vice versa. Are the authors arguing that in this case the converse is true and that the localization of KRAS is driven by that of its effector?
+
+<--- Page Split --->
+
+Fig. 5D. The altered localization of endogenous SCRIB as a consequence of silencing AFDN is described on p. 10 as "dispersed throughout the cytoplasm." But unlike the distribution of GFP that is homogeneous and clearly cytosolic, that of SCRIB is punctate consistent with a vesicular localization and should be described as such. Co- localization with Texas- red transferrin would allow an assessment as to whether these are endosomes. Caution must be taken in what is used for permeabilization of the cells (here \(0.05\%\) Tween- 20) as this can alter the appearance of vesicles. It would be wise to also try \(0.1\%\) saponin.  
+
+Despite binding of several RAS family small GTPases, in their colocalization studies the authors restricted their analysis to KRAS4B. This is unfortunate. It would be informative to also study a RAS- related binding partner that is not normally localized exclusively to the PM. RAP2 fits the bill as the authors show strong binding to AFDN and this small GTPase has been localized to endomembrane (PMID: 1923507 and 19061864). It would also be interesting to determine if RHEB interacts with AFDN since this RAS family small GTPase is expressed on lysosomes.  
+
+Minor points:  
+
+Fig. 2B. This figure sets up indirect immunofluorescence (iIF) staining of endogenous AFDN and SCRIB, which is used extensively throughout the paper. The specificity of antibodies used for iIF must always be validated by knockdown of the protein of interest, which is accomplished in Fig. 5 and this should be added to the legend of Fig. 2B.  
+
+Fig. 4D. The authors write on p. 7 that with an \(n = 5\) they saw an 8- fold enhancement of SCRIB co- IP with AFDN upon expression of KRAS4B- G12V, but they do not report results for RAP1B or RAP2C, which appear to also induce some enhancement, albeit to a lesser extent (KRAS4B>RAP2C> RAP1B). It would be informative to report the results for each of the interacting GTPases.  
+
+Fig. S6B. It is very difficult to see the AFDN staining.  
+
+Fig. 6C,D. These z projections are convincing that true colocalization of AFDN and SCRIB requires PDZ1 and FHA, but would be even more so if they were subjected to analysis with Pearson's coefficient.  
+
+The authors refer to KRAS throughout but they mean KRAS4B. They do not study KRAS4A. Since these splice variants differ only in their HVRs that direct subcellular trafficking this should be acknowledged.
+
+<--- Page Split --->
+
+To be a true effector of a small GTPase, three conditions must be met. The effector must bind directly to the GTPase, the binding must depend on GTP- loading of the GTPase, and the binding must in some way change the conformation or activity of the effector. RAF and HK1 meet all of these criteria but the third has been lacking for AFDN. In Fig. 4D the authors establish for the first time a change in the properties of AFDN induced by KRAS, confirming that AFDN is a bone fide effector. This should be discussed.  
+
+Because AFDN is unique in possessing tandem RA domains that, in principal, could bind two GTPases the authors have a unique opportunity to ask if either or both are required for the effect of KRAS seen in Fig. 4D. Indeed, it would be interesting to determine if RAS binds to one and RAP2 to the other RA domain.  
+
+The change in kinetics of ERK and AKT activation downstream of EGF signaling upon silencing AFDN or SCRIB shown in Fig. 7D,E is interesting but the authors do not comment on possible mechanisms. Interestingly they parallel the differential effects of NGF versus EGF in PC12 cells where only the former induces sustained ERK activation (PMID 7834738).
+
+<--- Page Split --->
+
+This revised manuscript addresses all Referee comments with either experimental data or further clarification, as required. We have added the following results to the manuscript:  
+
+- Purified SCRIB PDZ domains 2 and 3-4, measured binding to the AFDN FHA domain (no binding demonstrates specificity of the AFDN domain for SCRIB PDZ1)- Determined GTPase specificity of AFDN RA domains 1 and 2 alone, to contrast what was observed with the RA1-2 construct (GTPase binding is avidity driven by the tandem domains, particularly KRAS)- Derived suitable conditions for our phosphatase treated co-IP of AFDN and SCRIB, and included the untreated control- Demonstrated that AFDN and RAS GTPases co-localize in HeLa cells, which do not express detectable levels of endogenous AFDN, substantiating AFDN as a RAS effector- Obtained new images for KRAS expressed in MCF7 cells to demonstrate how cells expressing activated KRAS detach from the monolayer- Used RAB5, RAB7 and RAB11 GTPases to resolve whether the punctate SCRIB pattern observed in AFDN KO MCF7 cells is localized to endosomes- Demonstrated co-localization of AFDN with activated RAP1B and RAP2C GTPases  
+
+Specific comments to each reviewer follow.  
+
+## Reviewer #1  
+
+Comments: This is an extraordinary effort to characterize AFDN interaction network, using proximity labelling- MS approach. Authors were looking into both isoforms of AFDN as well. They found that AFDN binds to its previously unknown interactor, SCRIB, adhesion/polarity tumor suppressor protein. Both AFDN and SCRIB were found to be linked with RAS- induced cell growth and invasion. Next, they were looking into specific binding domains, involved in AFDN- SCRIBS complex and the molecular mechanism of this interaction. They found that that interaction is mediated through the first PDZ domain of SCRIB and the AFDN FHA domain. Further, the knockout experiments reveal the important role of AFDN- SCRIB interaction in MAPK and PI3K activation, and cell motility.  
+
+This data provide insight into new roles of AFDN, as not- so- well characterized RAS effector. I appreciate the quality of results, data and figures, and recommend this manuscript for publication.  
+
+We thank the reviewer for their positive comments and appreciation for our work.  
+
+## Reviewer #2  
+
+In the manuscript by Goudreault et al. entitled 'Afadin couples RAS GTPases to the polarity rheostat protein Scribble', the investigators employ proximity- based proteomics using afadin (AFDN) isoforms and identify the polarity protein SCRIB/Scribble as the top hit. Additional support for this interaction is obtained using in vitro binding assays and in- cell pull down experiments in which the authors find that the first PDZ domain of SCRIB and the AFDN FHA domain directly associate. AFDN has previously been shown to be an effector of the RAS and RAP GTPases and binds through its RA domains. Consistent with these earlier observations, the investigators find that KRAS co- localizes with AFDN to facilitate formation of the AFDN- SCRIB
+
+<--- Page Split --->
+
+complex. Knockout of AFDN or SCRIB in MCF7 epithelial cells disrupts growth dependent MAPK and PI3K activation and inhibits cell motility. Results from these studies provide a new link between AFDN, SCRIB and RAS- mediated MAPK signaling.  
+
+We thank the reviewer for their thoughtful suggestions and have attempted to address all points raised with comments, or by acquiring new experimental data.  
+
+1) The authors state that 'We observed in Fig. 2F that only the N-SCRIB, but not C-SCRIB construct, is able to pulldown the AFDN-FHA fragment, though the C-SCRIB contains 3 additional PDZ domains.' Have the 4 PDZ domains been compared to assess possible sequence determinants (e.g. specificity for the first PDZ domain)? Also, as the PDZ domains may interact or be involved in autoinhibitory interactions, have the investigators probed the individual PDZ domains or the 4 PDZ domains together?  
+
+There are significant literature describing efforts to determine PDZ domain binding specificity. To date, it has not been possible to establish binding specificity of individual domains based on sequence alone. The most valuable approaches have been large scale, non- biased screens of PDZ binding to peptide libraries and many include the PDZ domains of SCRIB \(^{1 - 3}\) . Indeed, several recent papers have resolved SCRIB PDZ 1- 4 specificity for individual substrates, though resolving specificity determinants within the domains themselves has required structural elucidation \(^{4 - 6}\) . As we have been unable to successfully crystallize the PDZ1- FHA complex, and most library screens to date have focused on C- terminal PDZ motifs, a sequence- based prediction of why PDZ1 binds to AFDN while PDZ2, 3, and 4 domains of SCRIB do not is currently not feasible. To satisfy the reviewers request and demonstrate that our elucidated interaction is specific to PDZ1, we generated expression vectors for SCRIB PDZ2 and the PDZ3- 4 supramodule \(^{7}\) . We expressed and purified these domains to homogeneity and used ITC to measure their binding affinity for the AFDN FHA domain (new Supplementary Fig. 3d). We observed no binding between the FHA domain and SCRIB PDZ2, while the PDZ3- 4 module showed only extremely weak binding on the order of \(\sim 100 \mathrm{mM}\) . Thus, the AFDN- SCRIB interaction is highly specific for the first PDZ of SCRIB, though the determinants of this specificity must be elucidated with future structural characterization of this complex.  
+
+1. Zhang, Y. et al. Convergent and divergent ligand specificity among PDZ domains of the LAP and zonula occludens (ZO) families. Journal of Biological Chemistry 281, 22299-22311 (2006).  
+2. Ivarsson, Y. et al. Large-scale interaction profiling of PDZ domains through proteomic peptide-phage display using human and viral phage peptidomes. Proceedings of the National Academy of Sciences of the United States of America 111, 2542-7 (2014).  
+3. Mu, Y., Cai, P., Hu, S., Ma, S. & Gao, Y. Characterization of diverse internal binding specificities of PDZ domains by yeast two-hybrid screening of a special peptide library. PLoS ONE 9, e88286 (2014).  
+4. Lim, K. Y. B., Gödde, N. J., Humbert, P. O. & Kvansakul, M. Structural basis for the differential interaction of Scribble PDZ domains with the guanine nucleotide exchange factor β-PIX. Journal of Biological Chemistry 292, 20425-20436 (2017).
+
+<--- Page Split --->
+
+5. Caria, C. et al. Structural analysis of phosphorylation-associated interactions of MCC to Scribble PDZ domains. FEBS Journal 286, 4910-4925 (2019).  
+
+6. How, J. Y., Stephens, R. K., Lim, K. Y. B., Humbert, P. O. & Kvansakul, M. Structural basis of the human Scribble–Vang12 association in health and disease. Biochemical Journal 478, 1321–1332 (2021).  
+
+7. Ren, J. et al. Interdomain interface-mediated target recognition by the Scribble PDZ34 supramodule. Biochemical Journal 468, 133–144 (2015).  
+
+2) Have the investigators compared binding of full length AFDN and SCRIB proteins with the individual domains (FHA/PDZ1) by in vitro assays? This comparison would be helpful in assessing whether the FHA and PDZ1 domains are sufficient for binding of AFDN and SCRIB, especially since the binding appears weak (~22 uM). Although full-length AFDN is compared with AFDN delta FHA (Figure 2I), a faint band is discernible for the FHA deleted protein suggesting that other regions of the protein may be involved in the interaction. A similar comparison is also not made for the SCRIB protein (Figure 2J).  
+
+While we would be happy to measure in vitro binding of full length SCRIB and AFDN, this is an extremely challenging experiment. Both proteins are \(\sim 200 \mathrm{kDa}\) in size and cannot be expressed and purified from bacteria. We have attempted to purify these proteins using a mammalian cell expression system, but AFDN in particular has a propensity to aggregate in vitro as a full length protein. While we anticipate purifying and studying this complex in the future, it is beyond the scope of this manuscript and will require many years of work. I do believe the reviewer is correct that there are likely additional contact points outside PDZ1-FHA. The two proteins co-IP under highly stringent conditions (even in RIPA buffer), indicating the measured in vitro affinity between the PDZ1 and FHA domains (5- 15 \(\mu \mathrm{M}\) depending on NaCl concentration) is perhaps insufficient to rationalize the overall robust binding of the two proteins from cell lysate. It is possible that the AFDN oligomers we observe following its purification from mammalian cells are genuine, perhaps supporting a model whereby multiple AFDN proteins can strongly co- precipitate SCRIB through multiple FHA-PDZ1 interactions of moderate affinity.  
+
+3) In Fig. 2C and 2D, IP with AFDN results in a strong SCRIB band with ectopic expression. Interestingly, this interaction looks to be retained with endogenous protein in panel 2D. However, the level of endogenous SCRIB protein pulled down looks to be much less than the GFP-tagged version despite high levels of endogenous expression. Is there a hypothesis for why this is observed?  
+
+There are numerous explanations for why co-immunoprecipitation of endogenous proteins may not re- capitulate the strong complex observed by ectopic expression. A primary factor is the binding of antibodies to the proteins directly, rather than to a recombinant tag. In this case, the AFDN polyclonal antibodies were raised against a "synthetic peptide corresponding to Mouse Afadin aa 1800 to the C-terminus (abcam)". While this should not disrupt the PDZ1-FHA interaction, it is difficult to predict if antibody binding to the C-terminus of AFDN may disrupt SCRIB binding until we have a structure of the full proteins in complex. Indeed, related to the previous question, we now believe that AFDN oligomerization is mediated through its C-terminal
+
+<--- Page Split --->
+
+region and it is possible that disruption of this higher order structure diminishes the amount of SCRIB bound in an IP. Moreover, it is conceivable that a fraction of endogenous SCRIB/AFDN in these cells are post- translationally modified in a manner that may disrupt their interaction, or that a population of the endogenous proteins are engaged in alternative complexes that are competitive with the AFDN- SCRIB module. Altogether, we were happy to observe the complex could be validated with endogenous proteins and future work will resolve the detailed mechanisms by which it is regulated in cells.  
+
+4) The authors state 'To map the interaction site on the SCRIB PDZ1 domain we used NMR spectroscopy. BMRBid 11207 was used to assign 71% of the peaks in a 1H/15N-HSQC of 15N-labelled SCRIB PDZ1.' It would be helpful state the conditions (pH, salt, etc) for published assignments in comparison to the NMR data collected, as the assignments are key for chemical shift mapping of the binding interaction. Please also indicate how well spectra overlay with previously determined assignments. Also, in Figure 3E, the full HSQC spectrum should be shown (with zoom as appropriate). A docked model of the interaction with the AFDN FHA domain could be generated using the NMR data but not included.  
+
+To address the first question, the assignment of SCRIB PDZ1 in the BMRB was deposited by RIKEN and there is no accompanying manuscript, but the buffer conditions in the PDB are:  
+
+20mM d- Tris- HCl; 100mM NaCl; 1mM d- DTT; 0.02% NaN3  
+
+This is nearly identical to the conditions used to acquire our own PDZ1 spectra:  
+
+20 mM Tris (pH 7.5), 100 mM NaCl, 1 mM DTT  
+
+To further indicate how well the previously assigned spectra overlays with our own data, we have added the full spectra superimposed with projected peaks from the BMRB assignment (new Supplementary Fig. 4a). This clearly demonstrates how similar our experimental data is to the previously derived assignment, and how we were able to unambiguously assign 70% of the observed peaks. The second point raised here is to include a full HSQC of the AFDN FHA domain in Fig. 3e. We do not believe a full HSQC would offer any additional information, and the space constraints make it difficult to fit complete spectra in Fig. 3. The panels shown demonstrate how the PDZ1 domain induces chemical shift perturbations in residues located in the C- terminal extension of the FHA domain. There are no chemical shifts from these residues in the broader spectrum, so we have chosen to leave this figure as originally presented. Finally, with regards to modeling we have used HADDOCK to generate models of the PDZ1- FHA interaction, aided by chemical shift perturbations used as restraints. While the models are interesting, they are not experimentally corroborated and do not provide any additional value to the current work. We prefer to wait for a proper structural elucidation of the full complex rather than speculate on a predicted model.  
+
+5) It is unclear whether there is a difference in the construct used in figure 3D (371-532) and Figure 3F (371-531). If they are the same then a correction is required. If they are different then the rationale for using two constructs with a difference in 1 amino acid is needed. Also, Figure 3D and 3F could be combined.
+
+<--- Page Split --->
+
+These are the same construct (371- 532) and the labelling has been corrected.  
+
+6) In Fig 3C, the minus phosphatase control for the pulldown is not shown.  
+
+We have redone this experiment and now include a minus phosphatase control in Fig. 3c.  
+
+7) Additional bands are observed in Fig. 3D, 3F for experiments with purified protein. Is this due to degradation or other protein contaminants? Some discussion is needed here.  
+
+The proteins used in these binding assays are GST- tagged and bound to glutathione beads. The smaller bands beneath our proteins of interest are typical degradation bands observed with GST fusion proteins, particularly following incubation with the partner protein (here, purified PDZ1) on a nutator for \(>1\) hour followed by multiple washes.  
+
+8) In Fig. 4, it is stated that densitometry analysis of western blots showed an 8-fold increase in SCRIB protein pulled down with AFDN/KRAS G12V co-expression. However, this data does not appear to be included in the submission.  
+
+We have added quantitation of SCRIB co- immunoprecipitation with AFDN in the new Fig. 4g. The plot shows SCRIB levels when co- expressed with KRAS- G12V, as well as RAP1B- G12V and RAP2C- G12V. Ratios are in comparison to SCRIB levels co- precipitated when AFDN is expressed with wild- type KRAS.  
+
+9) In Fig. 4D, authors show that increased SCRIB precipitation is observed with AFDN in the presence of KRAS G12V. AFDN is also shown to be immunoprecipitated with KRAS and related RAP GTPases through the RA domains. Interestingly, in panel 4D, some RAP GTPase appears to be pulled down with just FLAG-AFDN and EGFP-vector, but the same is not observed for KRAS G12V here. What is the source of the disconnect between panel 4C and 4D with regard to AFDN/KRAS G12V?  
+
+(Previous Fig. 4d is now Fig. 4e) We show multiple times that KRAS- G12V can be precipitated from cell lysate by the AFDN RA domains (Figs 2f and 4c/d) or that KRAS co- precipitates with full length AFDN (Figs 4a/b, 5c and Supplementary Fig. 5a). We have also previously solved a crystal structure of the AFDN RA1 bound to RAS- GTP and characterized this complex by ITC (Nat. Comm., 2017). These data clearly demonstrate that KRAS- G12V binds AFDN, though a strong band is not visible in lane 9 of Fig. 4e. This is likely because this is a triple transient transfection (3 plasmids) and the levels of GTPase expressed are relatively low. This might be improved by generating series of stable cell lines expressing the GTPases, but the objective of this experiment was to determine the relative levels of SCRIB present in AFDN co- IPs when the activated GTPases are present. As it worked well for this, it did not seem necessary to further resolve the KRAS- AFDN interaction when it is studied extensively in other experiments.
+
+<--- Page Split --->
+
+10) Similarly in Fig. 4E, a FLAG pulldown control absent of FLAG-AFDN shows modest levels of KRAS G12V immunoprecipitated. Is this due to residual SCRIB interaction with KRAS or an artifact using a FLAG-antibody?  
+
+(Previous Fig. 4e is now Fig. 4f) These are merely background bands common when working with RAS GTPases. The proteins are unstable when nucleotide free or when prenylated in the absence of lipid membrane and it is common to have background. As in the comment above, this is a triple transient transfection and the expression level of the GTPases is low (we must expose for \(\sim 2\) minutes to observe strong bands). The objective of this experiment was to resolve whether SCRIBAPDZ1 alters the interaction between AFDN and activated KRAS, and the experiment clearly demonstrates that it does not.  
+
+11) Fig. 4F and 4G require reworking for presentation purposes. The Z-stacks are not very legible given the intent of showing merged co-localization of KRAS:AFDN:SCRIB. Though the figure panel is intended to show increased localization of KRAS G12V vs. KRAS WT to AFDN:SCRIB, this point is not currently made and it is unclear whether this is due to data interpretation or simply representation of the data. The investigators may want to consider alternative color mapping for better contrast of co-localization points, or a larger zoom/view for readers to see the merged fluorescence channels. Additionally, differentiation between SCRIB/AFDN merge vs. SCRIB/AFDN/KRAS merges could be useful for interpretation of the two separate complexes.  
+
+We have taken multiple approaches to address the points raised here. First, we enlarged the z- stack panels in these images to make the expression patterns more legible (now Fig. 5a/b). More importantly, we reconsidered whether MCF7 cells were an appropriate system to demonstrate co- localization of RAS GTPases with AFDN, as both proteins are constitutively membrane localized in polarized epithelial cells. This made it very challenging to demonstrate co- localization driven by RAS activation. Instead, we went back to the HeLa cell system as this line does not express detectable levels of endogenous AFDN and does not form adherens junctions. The new Fig. 5c shows how expression of activated KRAS, and not wild- type KRAS, recruits exogenously expressed AFDN to the cell membrane. We hope the reviewer agrees this is a more robust demonstration that AFDN co- localizes with activated RAS in cells and that it should be considered a RAS effector.  
+
+12) The same concern exists with Fig. 5E. The authors should provide a better representation for the immunofluorescence localization and internalization of AFDN. They state that expression of KRAS G12V in the SCRIB KO line induces internalization of AFDN in the figure legends, but this is not addressed in the Results proper. Rather, in Results, the authors state that AFDN is retained at sites of cell-cell contact.  
+
+In response to this point, and a related point from Reviewer 4, we have added numerous images of immunostained AFDN in KRAS- expressing MCF7 cells (the new Supplementary Fig. 5c/d). These images show that AFDN is no longer visible at cell- cell contacts in the majority of cells expressing KRAS- G12V, while it is retained at cell contacts in those expressing wild- type KRAS. Indeed, most cells expressing KRAS- G12V are no longer associated with adjacent cells and empty
+
+<--- Page Split --->
+
+space is clearly observable surrounding these cells. The results are appropriately discussed in the Results section.  
+
+13) Similar concerns also exist with Fig. 6A and 6B regarding representation of the data. In 6A, it appears that SCRIB may be less recruited to cell contacts with the FHA deletion of AFDN as compared to WT, though authors state the opposite. The trend the authors state, is more clear in the reverse scenario with AFDN and the SCRIB PDZ truncation in that there is less AFDN association at the cell contacts. Perhaps an alternative quantification representation for the colocalization of AFDN and SCRIB will better illustrate the authors' point/clarify the data.  
+
+We have calculated Pearson's coefficients to quantify the levels of endogenous SCRIB colocalized with EGFP- tagged wild- type AFDN or the \(\Delta \mathrm{FHA}\) variant (in AFDN KO cells), or endogenous AFDN with EGFP- tagged wild- type SCRIB or its \(\Delta \mathrm{PDZ1}\) variant (in SCRIB KO cells). These calculations were performed on \(n\geq 8\) images similar to those presented in Fig. 7c/d. The results are presented in the new Supplementary Fig. 7c. This quantitation resolved a significant difference in the co- localization of endogenous SCRIB/AFDN with the respective wild- type rescues compared with domain deletions, and is now presented in the text.  
+
+14) Minor comment. In the Fig. 7 legend consider adding \((A - C)\) at beginning of legend for panels \(A - C\) and \((D - E)\) to maintain parallel structure.  
+
+This was corrected in the figure legend.  
+
+15) The authors state that a signaling "defect" for the ERK MAPK/PI3K-AKT signaling cascade is induced with KO of AFDN or SCRIB. They may want to consider rewording, as the initial increase in temporal signaling indicates that signaling is not defective, but rather altered. Have the authors examined this temporal signaling effect of AFDN/SCRIB KO in the context of an activated KRAS such as KRAS G12V?  
+
+This may be semantics, but 'defect' is defined as "an imperfection or abnormality that impairs quality, function, or utility". We would argue that the disrupted kinetics of AKT and ERK activation demonstrated here represent a genuine abnormality in response to EGF stimulation and can thus be appropriately described as a "defect". See also the response to Reviewer 4, point 12.  
+
+16) Statistics for pERK and pAKT is lacking to show a difference between "WT"/AFDN or SCRIB KO?  
+
+I believe the reviewer is referring to Fig. 8d/e (previously Fig. 7d/e), which graphs quantitation of pERK and pAKT in the parental cell line vs the AFDN and SCRIB KO lines. This is a time course of EGF stimulation, and to my knowledge there is no established statistical approach that will determine a p- value across the entire series. A 2- way ANOVA can be used, but this would treat the time points randomly rather than sequential. If the course resulted in a linear response, we could solve this with regression, but the response is not linear. We can use multiple independent t- tests (perhaps focused on the time points of interest), but this disregards that the series are
+
+<--- Page Split --->
+
+sequential and therefore does not provide value. We have left the qualitative graphs in Fig. 8d/e without p- values for now, but if the reviewer has insights to how such non- linear time course data can be tested statistically, we are happy to perform the analysis.  
+
+17) In Fig. 7I, consider a directional quantification of the leading-edge Golgi stain or a better zoomed-in insert to emphasize the loss of directionality in SCRIB/AFDN KO cells.  
+
+We have added a zoomed image of cells at the leading edge to Fig. 8i (previously Fig. 7i).  
+
+18) Mechanistic insight into how the AFSN/SCRIB effects the RAS-mediated pERK or pAKT activation would aid the discussion.  
+
+Please see our answer to Reviewer 4, point 12.  
+
+## Reviewer #3  
+
+The manuscript "Afadin couples RAS GTPases to the polarity rheostat Scribble" identifies a novel proximity interaction between Afadin and Scribble, using BioID coupled to mass spectrometry. A series of carefully crafted IP experiments identified that the PDZ domain of Scribble interacts with FHA domain of AFDN. Furthermore, the authors characterised the interaction between the domains using different techniques to reveal a model of how the two domains bind to each other. Since, AFDN binds to several RAS GTPases, the authors determined and characterised the interaction of activated forms of the GTPases with AFDN- SCRIB complex. Using CRISPR/Cas9 gene editing, the authors created a suite of single and double KO cell lines of AFDN and SCRIB. Using these cells lines, the authors further support a KRAS-AFDN- SCRIB complex formation and the requirement of a direct FHA- PDZ1 interaction for proper localisation of AFDN- SCRIB at cell contacts. Finally, the authors show that the loss of either AFDN or SCRIB disrupts ERK and AKT activation kinetics and cell motility in growth factor- dependent manner. The manuscript is well written and experiments are performed with nice controls. The observations and conclusions drawn in this manuscript would help advance the field in a significant way. The data presented in the manuscript provides ample support to the conclusions drawn. I have few suggestions that would improve the manuscript:  
+
+1) BirA\* is a large tag and can cause a significant level of mislocalisation when expressed in cells. Since the BioID results show some proteins from other compartments like ER/Golgi, mitochondria, etc, it would be nice if the authors can determine subcellular localisation of the AFDN-BirA\* tagged fusion proteins (both isoforms).  
+
+We thank the reviewer for their helpful comments and will address the points raised here. A distinct advantage of proximity- based proteomics is the insight provided to subcellular localization. The results presented in Fig. 1 (and accompanying Table 1) make clear that BirA\*- AFDN is properly behaved, as the majority of identified preys are plasma membrane proteins involved in cell adhesion (including most previously known AFDN interactors). The reviewer is correct that the
+
+<--- Page Split --->
+
+BirA\* tag is large, and to satisfy the request we imaged our BirA\*- FLAG-AFDN iso1 construct in the MCF7 cell line:  
+
+![PLACEHOLDER_17_0]
+  
+
+This confocal image shows that BirA\*- FLAG- AFDN is prominently localized at cell- cell contacts, as expected. We include here staining with Streptavidin- 488 following 2 hours of incubation with biotin. This shows the majority of biotinylated proteins are also at the plasma membrane, though AFDN and Streptavidin- 488 signals are also visible in the interior of these cells (Strep- 488 also characteristically marks mitochondria in the surrounding cells). We are confident based on these images and the derived proteomic data that BirA\*- AFDN appropriately localizes in cells.  
+
+2) Since the authors have created multiple CRISPR/Cas9 gene edited cell lines and assessed the precise nature of the edit by Sanger sequencing (line: 693), it would be recommended to show the precise edits and how they impact expression of that gene (i.e. Introduce STOP codon, impact splicing, destroy START codon, frame-shift, etc).  
+
+We have added a table to the Results section (under the subheading Generation of AFDN and SCRIB CRISPR KO cells and Rescues) that describes sequencing of the KO cell lines in detail.  
+
+## Reviewer #4  
+
+Afadin (AFDN), a regulator of cell- cell contacts, has long been recognized as an effector of RAS and related small GTPases but little is known about the interaction or its functional consequences. AFDN is unusual among the dozen or so effectors of RAS because it possesses two tandem RA domains. Goudreault et al. set out to explore the AFDN interactome by proximity labeling and here report a comprehensive characterization of the interaction with their most robust hit, SCRIB, a tumor suppressor and polarity protein that possesses four tandem PDZ domains. They show that AFDN and SCRIB associate via a non- canonical interaction of the first PDZ domain with the forkhead associated (FHA) domain of AFDN and that the interaction is enhanced by GTP- bound KRAS that forms a ternary complex with the two polarity proteins. Conversely, they show that silencing AFDN or SCRIB changes the kinetics of growth- factor stimulated ERK and AKT signaling in MCF7 epithelial cells. The manuscript is exceedingly well written. The authors walk the reader through not only the experiments and results but also the thinking behind them. The
+
+<--- Page Split --->
+
+turboID proximity screen is well done, appropriately controlled, and clearly reported. The biochemical validation of the interaction of AFDN and SCRIB is outstanding, particularly the rigor applied to confirming the non- canonical nature of the interaction between the AFDN FHA domain with the first PDZ domain of SCRIB. The promiscuity of the AFDN RA domain(s) for RAS family proteins relative to the specificity of the RAF1 RBD is well demonstrated. These results are clear, novel, and of interest to cell biologists and are certainly worthy of reporting. The weakest part of the study is the overinterpretation of the immunofluorescent localizations and colocalizations of KRAS, AFDN and SCRIB.  
+
+We thank the reviewer for evaluating our manuscript and will address the concerns here, including with the addition of new experimental data.  
+
+1) Fig. 2A. Here AFDN and SCRIB, both epitope-tagged, are overexpressed in a HeLa cell that is processed for immunofluorescence. It is not stated if the GFP-AFDN is imaged with the intrinsic fluorescence of GFP or if an anti-GFP antibody is employed along with the anti-FLAG antibody. No control proteins are employed nor is a control with first antibody omitted shown. The localization is indeterminant and uninformative. A single cell is shown such that one cannot determine if this represents the predominant fluorescent pattern. Since tagged, ectopically expressed proteins are used, including one tagged with GFP, it is not clear why the authors did not use mCherry-SCRIB such that they could colocalize the two proteins in live cells, which allow for more precise subcellular localization free of fixation and permeabilization artifacts.  
+
+In Fig. 2a we show the localization of ectopically tagged AFDN and SCRIB in HeLa cells (as this was the cell line from which our proteomics data was collected) before moving on to MCF7 cells. The reviewer states the localization is 'indeterminant and uninformative', but this is indeed the diffuse localization pattern we observe from these proteins in this epithelial line, which lack defined cell- cell contacts and apical- basal polarity. Levels of endogenous AFDN in HeLa cells were too low to detect by immunofluorescence. Endogenous SCRIB is expressed, can be detected, and displays a diffuse/punctate localization when immunostained that is highly reminiscent of SCRIB in AFDN KO MCF7 cells:  
+
+![PLACEHOLDER_18_0]
+  
+
+Others have observed this pattern in different cell types<sup>1</sup>. To satisfy the reviewers request, we co- expressed EGFP- AFDN and Cherry- SCRIB in these cells (the new Supplementary Fig. 1d), or appropriate controls, to avoid fixation/permeabilization effects. We again observe broad staining in the cytoplasm for both AFDN and SCRIB, with some weak concentration at the cell cortex. In
+
+<--- Page Split --->
+
+the following section we describe how this was used to improve our understanding of RAS- mediated recruitment of AFDN to the plasma membrane, something that was more difficult in the polarized MCF7 cell line where AFDN and SCRIB are characteristically membrane- proximal even in the absence of activated GTPases.  
+
+1. Anastas, J. N. et al. A protein complex of SCRIB, NOS1AP and VANGL1 regulates cell polarity and migration, and is associated with breast cancer progression. Oncogene 31, 3696-3708 (2012).  
+
+2) Fig. 4F,G. In describing these micrographs the authors state on p. 9 that "tagged, wild-type KRAS does not significantly alter AFDN or SCRIB localization and does not co-localize with these proteins (Figure 4F and S4B). ... In contrast, expression of KRAS-G12V markedly disrupted cell-cell contacts and was noticeably co-localized with endogenous AFDN and SCRIB, as determined by z-plane projections (Figure 4G)." In the discussion on p. 13 the authors write that "we show that RAS-G12V is co-localized with AFDN and SCRIB at sites of cell contact, while wild-type RAS is distributed more generally across the plasma membrane." The data do not support these conclusions. First, not shown is any disruption of cell-cell contacts in cells expressing KRAS-G12V. Three GFP-KRAS4B-G12V expressing cells are shown with three levels of expression and perhaps different z-planes. Two of these three maintain robust cell-cell contacts as determined by morphology and SCRIB and AFDN staining of areas of cell contact (anti-ZO-1 staining would be a way to look at this without imaging the experimental proteins themselves). Not shown are any of the detached cells to which the authors refer as having lost cell-cell contacts as a function of oncogenic KRAS. Second, and more important, the data show that both WT-KRAS and KRAS-G12V decorate the basolateral membrane. Indeed, both the WT and mutant KRAS decorate the entire plasma membrane (PM) as is well established in a vast literature. As expected for cell adhesion proteins imaged in confluent epithelial cells, SCRIB and AFDN decorate primarily the basolateral membrane. The conclusion that these proteins are colocalized to a greater extent with KRAS-G12V that with WT KRAS is not supported by the data shown and contrary to a vast literature on KRAS localization. Some of the problem is semantic; the concept of co-localization is somewhat ambiguous. There is co-localization on the PM in some regions of the cell but not others. This should not necessarily be interpreted as one protein pulling another to a region of PM since the same pattern would be observed if the localizations are true but unrelated to the direct interactions of the proteins. The three proteins colocalize at the basolateral membrane but not the apical membrane and this is not affected by the GTP-binding state of KRAS. Were this localization of mutant KRAS to differ from that of WT it would be contrary to a vast literature on KRAS localization that in total demonstrates that the subcellular localization driven by the prenylated HVR is not affected by the GTP-binding state. Current paradigms of RAS signaling hold the PM localization of KRAS is constitutive and it is effectors that are drawn to RAS (e.g. translocation of RAF) not vice versa. Are the authors arguing that in this case the converse is true and that the localization of KRAS is driven by that of its effector?  
+
+It has been difficult to discern specific co- localization between activated KRAS with AFDN in any polarized epithelial cell line we have worked with. There are several reasons for this: 1) both AFDN and KRAS are constitutively membrane localized in these lines independent of each other, as stated by the reviewer; 2) cells expressing activated KRAS detach from adjacent cells; and 3) live cell imaging has proven ineffectual for numerous reasons, primarily due to the unpredictable nature of KRAS- G12V expressing cells but also the requirement for high resolution z- planes and
+
+<--- Page Split --->
+
+the tendency of overexpressed AFDN to accumulate in the cytoplasm. To address the reviewers' points and validate that KRAS and AFDN co- localize in cells in a GTP- dependent manner we have taken several approaches. First, we have added numerous images of immunostained AFDN in KRAS- expressing MCF7 cells (the new Supplementary Fig. 5c/d). These images show that AFDN is no longer visible at cell- cell contacts in cells expressing KRAS- G12V, while it is retained at cell contacts in those expressing wild- type KRAS. The images demonstrate that cells expressing KRAS- G12V are no longer associated with adjacent cells, with empty space clearly observable in the surrounding space (i.e. detached). This addresses the reviewers first point above. Secondly, we reconsidered whether polarized epithelial cells with well- defined cell- cell contacts were the most suitable system to demonstrate co- localization between RAS GTPases and AFDN. As discussed above, HeLa cells lack discernable levels of endogenous AFDN, as well as apical- basal polarity and do not form adherens junctions. We exploited this to examine whether RAS could specifically recruit exogenous AFDN to the plasma membrane in a GTP- dependent manner. The new Fig. 5c shows how expression of activated KRAS, and not wild- type KRAS, co- localizes with AFDN at the plasma membrane. Indeed, most of the AFDN pool is recruited from the cytoplasm with a fraction remaining in the perinuclear region. We hope the reviewer agrees this is more robust demonstration that AFDN should be considered a RAS effector.  
+
+3) Fig. 5D. The altered localization of endogenous SCRIB as a consequence of silencing AFDN is described on p. 10 as "dispersed throughout the cytoplasm." But unlike the distribution of GFP that is homogeneous and clearly cytosolic, that of SCRIB is punctate consistent with a vesicular localization and should be described as such. Co-localization with Texas-red transferrin would allow an assessment as to whether these are endosomes. Caution must be taken in what is used for permeabilization of the cells (here 0.05% Tween-20) as this can alter the appearance of vesicles. It would be wise to also try 0.1% saponin.  
+
+We agree that the localization of SCRIB in AFDN KO MCF7 cells is more appropriately described as punctate and have changed the description in our revised manuscript. To assess whether the observed pattern is consistent with a localization to endosomes we expressed EGFP- tagged RAB5A, RAB7A or RAB11A in these cells followed by immunostaining for endogenous SCRIB. These report on early, late or recycling endosomes, respectively, and the new Supplementary Fig. 6c reveals that SCRIB does not co- localize with any of these markers. As noted above, others have observed a similar punctate pattern of endogenous SCRIB in different cell lines (and we see this in HeLa cells, which do not express significant levels of AFDN), but this experiment suggests these are not endosomes. Future delineation will hopefully shed light on where SCRIB is localized in the absence of AFDN.  
+
+4) Despite binding of several RAS family small GTPases, in their colocalization studies the authors restricted their analysis to KRAS4B. This is unfortunate. It would be informative to also study a RAS-related binding partner that is not normally localized exclusively to the PM. RAP2 fits the bill as the authors show strong binding to AFDN and this small GTPase has been localized to endomembrane (PMID: 1923507 and 19061864). It would also be interesting to determine if RHEB interacts with AFDN since this RAS family small GTPase is expressed on lysosomes.
+
+<--- Page Split --->
+
+To address this point we have studied co- localization of AFDN with the small GTPases RAP1B and RAP2C (the new Fig. 5c). We found that imaging these proteins in MCF7 cells came with many of the same caveats presented by AFDN- KRAS, and therefore used the HeLa cell line to determine if these protein partners co- localize (as described for KRAS above). RAP1B and RAP2C were the most prominent members of this family precipitated by the AFDN RA domains. Consistent with previous data, we observed RAP1B is distributed prominently throughout the cytoplasm with a fraction at the plasma membrane \(^{1,2}\) , while RAP2C is localized to the plasma membrane and on endomembranes \(^{3,4}\) . In both cases, we observed their complete co- localization with exogenous AFDN. Indeed, RAP2C- G12V stimulated recruitment of AFDN to the membrane in a manner very similar to that observed with KRAS- G12V.  
+
+1. Wilson, J. M., Prokop, J. W., Lorimer, E., Ntanti, E. & Williams, C. L. Differences in the Phosphorylation-Dependent Regulation of Prenylation of Rap1A and Rap1B. Journal of Molecular Biology 428, 4929-4945 (2016).  
+
+2. Ntanti, E. et al. An Adenosine-Mediated Signaling Pathway Suppresses Prenylation of the GTPase Rap1B and Promotes Cell Scattering. Sci. Signal. 6, (2013).  
+
+3. Duncan, E. D., Han, K.-J., Trout, M. A. & Prekeris, R. Ubiquitylation by Rab40b/Cul5 regulates Rap2 localization and activity during cell migration. Journal of Cell Biology 221, e202107114 (2022).  
+
+4. Meng, Z. et al. RAP2 mediates mechanoresponses of the Hippo pathway. Nature 560, 655-660 (2018).  
+
+## Minor points:  
+
+5) Fig. 2B. This figure sets up indirect immunofluorescence (iIF) staining of endogenous AFDN and SCRIB, which is used extensively throughout the paper. The specificity of antibodies used for iIF must always be validated by knockdown of the protein of interest, which is accomplished in Fig. 5 and this should be added to the legend of Fig. 2B.  
+
+A statement was added to the figure legend.  
+
+6) Fig. 4D. The authors write on p. 7 that with an \(n = 5\) they saw an 8-fold enhancement of SCRIB co-IP with AFDN upon expression of KRAS4B-G12V, but they do not report results for RAP1B or RAP2C, which appear to also induce some enhancement, albeit to a lesser extent (KRAS4B>>RAP2C> RAP1B). It would be informative to report the results for each of the interacting GTPases.  
+
+We have added quantitation of SCRIB co- immunoprecipitation with AFDN in the new Fig. 4g. The plot shows SCRIB levels when co- expressed with KRAS- G12V, as well as RAP1B- G12V and RAP2C- G12V. Ratios are in comparison to SCRIB levels co- precipitated when AFDN is expressed with wild- type KRAS. While expression of KRAS- G12V results in an 8- fold increase in SCRIB association with AFDN, the levels are indeed lower for RAP2C (2.6- fold increase) and RAP1B (1.7- fold increase).
+
+<--- Page Split --->
+
+7) Fig. S6B. It is very difficult to see the AFDN staining.  
+
+We have increased intensity of the AFDN signal in these images (now Supplementary Fig. 7b).  
+
+8) Fig. 6C,D. These \(z\) projections are convincing that true colocalization of AFDN and SCRIB requires PDZ1 and FHA, but would be even more so if they were subjected to analysis with Pearson's coefficient.  
+
+We have calculated Pearson's coefficients to quantify the levels of endogenous SCRIB colocalized with EGFP- tagged wild- type AFDN or the \(\Delta\) FHA variant (in AFDN KO cells), or endogenous AFDN with EGFP- tagged wild- type SCRIB or its \(\Delta\) PDZ1 variant (in SCRIB KO cells). These calculations were performed on \(n \geq 8\) images similar to those presented in Fig. 7c/d. The results are presented in the new Supplementary Fig. 7c. This quantitation resolved a significant difference in the co- localization of endogenous SCRIB/AFDN with the respective wild- type rescues compared with domain deletions. The analysis has more noise in the AFDN KO cells due to the punctate nature of SCRIB localization, but there is clearly a deficiency in SCRIB colocalization with the AFDN/AFHA rescue compared to wild- type.  
+
+9) The authors refer to KRAS throughout but they mean KRAS4B. They do not study KRAS4A. Since these splice variants differ only in their HVRs that direct subcellular trafficking this should be acknowledged.  
+
+We have acknowledged this in the Results section (page 5) and in the Methods section (Plasmid Constructs and Antibodies).  
+
+10) To be a true effector of a small GTPase, three conditions must be met. The effector must bind directly to the GTPase, the binding must depend on GTP-loading of the GTPase, and the binding must in some way change the conformation or activity of the effector. RAF and HK1 meet all of these criteria but the third has been lacking for AFDN. In Fig. 4D the authors establish for the first time a change in the properties of AFDN induced by KRAS, confirming that AFDN is a bone fide effector. This should be discussed.  
+
+The point is well taken, and we have reinforced this view in the Discussion (bottom of the first paragraph). It is still not completely clear how RAS activates most effectors, including RAF from a completely mechanistic perspective, though cryo- EM and modelling data are making significant progress with this. It does appear that GTPase binding to AFDN (particularly KRAS) increases its complex with SCRIB, and while only a structure of the full proteins complexed with RAS will reveal the mechanism behind this 'activation', it does appear to satisfy the third condition for being a true effector.  
+
+11) Because AFDN is unique in possessing tandem RA domains that, in principal, could bind two GTPases the authors have a unique opportunity to ask if either or both are required for the effect of KRAS seen in Fig. 4D. Indeed, it would be interesting to determine if RAS binds to one and RAP2 to the other RA domain.
+
+<--- Page Split --->
+
+To address this point we generated bacterial expression constructs to allow purification of AFDN RA1 or RA2 alone. We reveal the specificity of the two individual RA domains for RAS and RAP GTPases, in contrast to the RA1/2 protein, in the new Fig. 4d. The first RA domain of AFDN demonstrated a very similar binding profile to the dual RA1/2 construct, but the levels of GTPase precipitated were considerably lower than with the tandem domains, KRAS in particular. The RA2 binding profile is more restricted, with only KRAS, RAP2B and RAP2C showing significant interaction. Overall, the results suggest that the tandem domains do provide an avidity for most of these GTPases which drives tighter binding than with either domain in isolation. RAP1B seems an exception, whereby binding is generated predominantly through RA1. These data are now described in the Results section.  
+
+12) The change in kinetics of ERK and AKT activation downstream of EGF signaling upon silencing AFDN or SCRIB shown in Fig. 7D,E is interesting but the authors do not comment on possible mechanisms. Interestingly they parallel the differential effects of NGF versus EGF in PC12 cells where only the former induces sustained ERK activation (PMID 7834738).  
+
+Indeed, the duration of ERK activation has long been recognized as a foremost feature of signalling through the MAPK pathway, and potentially as a determinant of proliferation vs differentiation outcomes. Interestingly, the sustained activation of MAPK signalling induced by NGF (pointed out by the reviewer) is ostensibly dependent on RAP1<sup>1</sup>, making the AFDN association with RAS and RAP of particular relevance to this phenomenon. As this will be a focus of future work and we do not currently have clear mechanistic data elucidating how loss of AFDN disrupts ERK activation, we have added only a short speculation to the Discussion section (paragraph 4). It seems clear that AFDN association with RAS at the plasma membrane will compete with binding of other effectors to activated RAS, including the RAF kinases, and this is the most likely explanation for the observed signalling defect. It is also probable that competition between activated RAS and RAP GTPases for the AFDN RA domains will further impact effector binding to the individual GTPases themselves. For SCRIB, it is more difficult to reason how loss of expression induces nearly the identical defect as loss of AFDN. It is possible that the RAS-AFDN-SCRIB module is more stable than the RAS-AFDN module alone, and our data supports this. Thus, SCRIB could be an important determinant of competition between AFDN and RAF effectors for activated RAS.  
+
+1. York, R. D. et al. Rap1 mediates sustained MAP kinase activation induced by nerve growth factor. Nature 392, 622-626 (1998).
+
+<--- Page Split --->
+
+REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors significantly approved the quality of their manuscript.  
+
+Reviewer #4 (Remarks to the Author):  
+
+The authors have addressed my comments and those of three other reviewers with new data, a revised manuscript and a well- argued rebuttal. They have done an outstanding job and the revised manuscript is significantly improved. This interesting, comprehensive, and significant paper is suitable for publication in Nature Communications.  
+
+I have two minor comments not intended to diminish the outstanding work presented but for future analyses should the authors carry this work forward. The authors write in their rebuttal: "To satisfy the reviewers request, we coexpressed EGFP- AFDN and Cherry- SCRIB in these cells (the new Supplementary Fig. 1d), or appropriate controls, to avoid fixation/permeabilization effects." However, the authors did not eliminate fixation/permeabilization effects because rather than image these cells alive, as would have been preferable, they fixed the cells, apparently to allow Hoechst staining. But nuclear localization with Hoechst adds little if anything to the analysis and would be far outweighed by the power or live cell imaging.  
+
+One other caveat along these lines, the authors refer to the localizations of proteins like GFP alone and GFP- AFADIN in cells not expressing oncogenic KRAS as "cytoplasmic." But the cytoplasm includes both cytosol and all membrane bound organelles (endomembrane). Live cell imaging allows one to clearly visualize the cytosol (organelles appear to be negatively imaged), which is an important descriptive term for molecules like AFADIN that translocate not from endomembrane to plasma membrane but rather from cytosol to plasma membrane.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a27a9071773bd5448442fddc2e7d62ff8da38764cfd258c9ac0e24d0109e8f59/supplementary_0_Peer Review File__a27a9071773bd5448442fddc2e7d62ff8da38764cfd258c9ac0e24d0109e8f59_det.mmd

@@ -0,0 +1,523 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[70, 110, 363, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 155, 901, 183]]<|/det|>
+Afadin couples RAS GTPases to the polarity rheostat Scribble  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 781]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 912, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 145, 393, 161]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 679, 220]]<|/det|>
+Manuscript title: Afadin couples RAS GTPases to the polarity rheostat Scribble  
+
+<|ref|>text<|/ref|><|det|>[[115, 231, 727, 248]]<|/det|>
+By: Marilyn Goudreault, Valérie Gagné, Chang Hwa Jo, Swati Singh, Ryan C. Killoran,  
+
+<|ref|>text<|/ref|><|det|>[[115, 260, 440, 276]]<|/det|>
+Anne- Claude Gingras, and Matthew J. Smith  
+
+<|ref|>text<|/ref|><|det|>[[115, 317, 877, 461]]<|/det|>
+Comments: This is an extraordinary effort to characterize AFDN interaction network, using proximity labelling- MS approach. Authors were looking into both isoforms of AFDN as well. They found that AFDN binds to its previously unknown interactor, SCRIB, adhesion/polarity tumor suppressor protein. Both AFDN and SCRIB were found to be linked with RAS- induced cell growth and invasion. Next, they were looking into specific binding domains, involved in AFDN- SCRIBs complex and the molecular mechanism of this interaction. They found that that interaction is mediated through the first PDZ domain of SCRIB and the AFDN FHA domain. Further, the knockout experiments reveal the important role of AFDN- SCRIB interaction in MAPK and PI3K activation, and cell motility.  
+
+<|ref|>text<|/ref|><|det|>[[115, 473, 872, 508]]<|/det|>
+This data provide insight into new roles of AFDN, as not- so- well characterized RAS effector. I appreciate the quality of results, data and figures, and recommend this manuscript for publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 605, 393, 621]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 661, 883, 844]]<|/det|>
+In the manuscript by Goudreault et al. entitled 'Afadin couples RAS GTPases to the polarity rheostat protein Scribble', the investigators employ proximity- based proteomics using afadin (AFDN) isoforms and identify the polarity protein SCRIB/Scribble as the top hit. Additional support for this interaction is obtained using in vitro binding assays and in- cell pull down experiments in which the authors find that the first PDZ domain of SCRIB and the AFDN FHA domain directly associate. AFDN has previously been shown to be an effector of the RAS and RAP GTPases and binds through its RA domains. Consistent with these earlier observations, the investigators find that KRAS co- localizes with AFDN to facilitate formation of the AFDN- SCRIB complex. Knockout of AFDN or SCRIB in MCF7 epithelial cells disrupts growth dependent MAPK and PI3K activation and inhibits cell motility. Results from these studies provide a new link between AFDN, SCRIB and RAS- mediated MAPK signaling.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 879, 180]]<|/det|>
+1) The authors state that 'We observed in Fig. 2F that only the N-SCRIB, but not C-SCRIB construct, is able to pulldown the AFDN-FHA fragment, though the C-SCRIB contains 3 additional PDZ domains.' Have the 4 PDZ domains been compared to assess possible sequence determinants (e.g. specificity for the first PDZ domain)? Also, as the PDZ domains may interact or be involved in autoinhibitory interactions, have the investigators probed the individual PDZ domains or the 4 PDZ domains together?  
+
+<|ref|>text<|/ref|><|det|>[[114, 218, 875, 329]]<|/det|>
+2) Have the investigators compared binding of full length AFDN and SCRIB proteins with the individual domains (FHA/PDZ1) by in vitro assays? This comparison would be helpful in assessing whether the FHA and PDZ1 domains are sufficient for binding of AFDN and SCRIB, especially since the binding appears weak (~22 uM). Although full-length AFDN is compared with AFDN delta FHA (Figure 2I), a faint band is discernible for the FHA deleted protein suggesting that other regions of the protein may be involved in the interaction. A similar comparison is also not made for the SCRIB protein (Figure 2J).  
+
+<|ref|>text<|/ref|><|det|>[[114, 367, 874, 439]]<|/det|>
+3) In Fig. 2C and 2D, IP with AFDN results in a strong SCRIB band with ectopic expression. Interestingly, this interaction looks to be retained with endogenous protein in panel 2D. However, the level of endogenous SCRIB protein pulled down looks to be much less than the GFP-tagged version despite high levels of endogenous expression. Is there a hypothesis for why this is observed?  
+
+<|ref|>text<|/ref|><|det|>[[114, 478, 877, 606]]<|/det|>
+4) The authors state 'To map the interaction site on the SCRIB PDZ1 domain we used NMR spectroscopy. BMRBid 11207 was used to assign 71% of the peaks in a 1H/15N-HSQC of 15N-labelled SCRIB PDZ1.' It would be helpful state the conditions (pH, salt, etc) for published assignments in comparison to the NMR data collected, as the assignments are key for chemical shift mapping of the binding interaction. Please also indicate how well spectra overlay with previously determined assignments. Also, in Figure 3E, the full HSQC spectrum should be shown (with zoom as appropriate). A docked model of the interaction with the AFDN FHA domain could be generated using the NMR data but not included.  
+
+<|ref|>text<|/ref|><|det|>[[114, 644, 870, 716]]<|/det|>
+5) It is unclear whether there is a difference in the construct used in figure 3D (371-532) and Figure 3F (371-531). If they are the same then a correction is required. If they are different then the rationale for using two constructs with a difference in 1 amino acid is needed. Also, Figure 3D and 3F could be combined.  
+
+<|ref|>text<|/ref|><|det|>[[114, 756, 660, 774]]<|/det|>
+6) In Fig 3C, the minus phosphatase control for the pulldown is not shown.  
+
+<|ref|>text<|/ref|><|det|>[[114, 813, 836, 848]]<|/det|>
+7) Additional bands are observed in Fig. 3D, 3F for experiments with purified protein. Is this due to degradation or other protein contaminants? Some discussion is needed here.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 857, 143]]<|/det|>
+8) In Fig. 4, it is stated that densitometry analysis of western blots showed an 8-fold increase in SCRIB protein pulled down with AFDN/KRAS G12V co-expression. However, this data does not appear to be included in the submission.  
+
+<|ref|>text<|/ref|><|det|>[[115, 182, 877, 273]]<|/det|>
+9) In Fig. 4D, authors show that increased SCRIB precipitation is observed with AFDN in the presence of KRAS G12V. AFDN is also shown to be immunoprecipitated with KRAS and related RAP GTPases through the RA domains. Interestingly, in panel 4D, some RAP GTPase appears to be pulled down with just FLAG-AFDN and EGFP-vector, but the same is not observed for KRAS G12V here. What is the source of the disconnect between panel 4C and 4D with regard to AFDN/KRAS G12V?  
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 843, 366]]<|/det|>
+10) Similarly in Fig. 4E, a FLAG pulldown control absent of FLAG-AFDN shows modest levels of KRAS G12V immunoprecipitated. Is this due to residual SCRIB interaction with KRAS or an artifact using a FLAG-antibody?  
+
+<|ref|>text<|/ref|><|det|>[[115, 405, 869, 550]]<|/det|>
+11) Fig. 4F and 4G require reworking for presentation purposes. The Z-stacks are not very legible given the intent of showing merged co-localization of KRAS:AFDN:SCRIB. Though the figure panel is intended to show increased localization of KRAS G12V vs. KRAS WT to AFDN:SCRIB, this point is not currently made and it is unclear whether this is due to data interpretation or simply representation of the data. The investigators may want to consider alternative color mapping for better contrast of co-localization points, or a larger zoom/view for readers to see the merged fluorescence channels. Additionally, differentiation between SCRIB/AFDN merge vs. SCRIB/AFDN/KRAS merges could be useful for interpretation of the two separate complexes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 590, 878, 662]]<|/det|>
+12) The same concern exists with Fig. 5E. The authors should provide a better representation for the immunofluorescence localization and internalization of AFDN. They state that expression of KRAS G12V in the SCRIB KO line induces internalization of AFDN in the figure legends, but this is not addressed in the Results proper. Rather, in Results, the authors state that AFDN is retained at sites of cell-cell contact.  
+
+<|ref|>text<|/ref|><|det|>[[115, 701, 879, 810]]<|/det|>
+13) Similar concerns also exist with Fig. 6A and 6B regarding representation of the data. In 6A, it appears that SCRIB may be less recruited to cell contacts with the FHA deletion of AFDN as compared to WT, though authors state the opposite. The trend the authors state, is more clear in the reverse scenario with AFDN and the SCRIB PDZ truncation in that there is less AFDN association at the cell contacts. Perhaps an alternative quantification representation for the co-localization of AFDN and SCRIB will better illustrate the authors' point/clarify the data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 850, 870, 885]]<|/det|>
+14) Minor comment. In the Fig. 7 legend consider adding \((A - C)\) at beginning of legend for panels \(A - C\) and \((D - E)\) to maintain parallel structure.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 117, 870, 191]]<|/det|>
+15) The authors state that a signaling "defect" for the ERK MAPK/PI3K-AKT signaling cascade is induced with KO of AFDN or SCRIB. They may want to consider rewording, as the initial increase in temporal signaling indicates that signaling is not defective, but rather altered. Have the authors examined this temporal signaling effect of AFDN/SCRIB KO in the context of an activated KRAS such as KRAS G12V?  
+
+<|ref|>text<|/ref|><|det|>[[115, 228, 835, 247]]<|/det|>
+16) Statistics for pERK and pAKT is lacking to show a difference between "WT"/AFDN or SCRIB KO?  
+
+<|ref|>text<|/ref|><|det|>[[115, 285, 870, 321]]<|/det|>
+17) In Fig. 7l, consider a directional quantification of the leading-edge Golgi stain or a better zoomed-in insert to emphasize the loss of directionality in SCRIB/AFDN KO cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 360, 845, 396]]<|/det|>
+18) Mechanistic insight into how the AFSN/SCRIB effects the RAS-mediated pERK or pAKT activation would aid the discussion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 493, 393, 509]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 548, 878, 750]]<|/det|>
+The manuscript "Afadin couples RAS GTPases to the polarity rheostat Scribble" identifies a novel proximity interaction between Afadin and Scribble, using BioID coupled to mass spectrometry. A series of carefully crafted IP experiments identified that the PDZ domain of Scribble interacts with FHA domain of AFDN. Furthermore, the authors characterised the interaction between the domains using different techniques to reveal a model of how the two domains bind to each other. Since, AFDN binds to several RAS GTPases, the authors determined and characterised the interaction of activated forms of the GTPases with AFDN-SCRIB complex. Using CRISPR/Cas9 gene editing, the authors created a suite of single and double KO cell lines of AFDN and SCRIB. Using these cells lines, the authors further support a KRAS-AFDN-SCRIB complex formation and the requirement of a direct FHA-PDZ1 interaction for proper localisation of AFDN-SCRIB at cell contacts. Finally, the authors show that the loss of either AFDN or SCRIB disrupts ERK and AKT activation kinetics and cell motility in growth factor-dependent manner.  
+
+<|ref|>text<|/ref|><|det|>[[115, 788, 876, 861]]<|/det|>
+The manuscript is well written and experiments are performed with nice controls. The observations and conclusions drawn in this manuscript would help advance the field in a significant way. The data presented in the manuscript provides ample support to the conclusions drawn. I have few suggestions that would improve the manuscript:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 863, 161]]<|/det|>
+1: BirA\* is a large tag and can cause a significant level of mislocalisation when expressed in cells. Since the BioID results show some proteins from other compartments like ER/Golgi, mitochondria, etc, it would be nice if the authors can determine subcellular localisation of the AFDN- BirA\* tagged fusion proteins (both isoforms).  
+
+<|ref|>text<|/ref|><|det|>[[115, 201, 878, 273]]<|/det|>
+2: Since the authors have created multiple CRISPR/Cas9 gene edited cell lines and assessed the precise nature of the edit by Sanger sequencing (line: 693), it would be recommended to show the precise edits and how they impact expression of that gene (i.e. Introduce STOP codon, impact splicing, destroy START codon, frame- shift, etc).  
+
+<|ref|>text<|/ref|><|det|>[[116, 398, 393, 414]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 454, 879, 637]]<|/det|>
+Afadin (AFDN), a regulator of cell- cell contacts, has long been recognized as an effector of RAS and related small GTPases but little is known about the interaction or its functional consequences. AFDN is unusual among the dozen or so effectors of RAS because it possesses two tandem RA domains. Goudreault et al. set out to explore the AFDN interactome by proximity labeling and here report a comprehensive characterization of the interaction with their most robust hit, SCRIB, a tumor suppressor and polarity protein that possesses four tandem PDZ domains. They show that AFDN and SCRIB associate via a non- canonical interaction of the first PDZ domain with the forkhead associated (FHA) domain of AFDN and that the interaction is enhanced by GTP- bound KRAS that forms a ternary complex with the two polarity proteins. Conversely, they show that silencing AFDN or SCRIB changes the kinetics of growth- factor stimulated ERK and AKT signaling in MCF7 epithelial cells.  
+
+<|ref|>text<|/ref|><|det|>[[114, 675, 884, 821]]<|/det|>
+The manuscript is exceedingly well written. The authors walk the reader through not only the experiments and results but also the thinking behind them. The turbolD proximity screen is well done, appropriately controlled, and clearly reported. The biochemical validation of the interaction of AFDN and SCRIB is outstanding, particularly the rigor applied to confirming the non- canonical nature of the interaction between the AFDN FHA domain with the first PDZ domain of SCRIB. The promiscuity of the AFDN RA domain(s) for RAS family proteins relative to the specificity of the RAF1 RBD is well demonstrated. These results are clear, novel, and of interest to cell biologists and are certainly worthy of reporting.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 867, 125]]<|/det|>
+The weakest part of the study is the overinterpretation of the immunofluorescent localizations and co- localizations of KRAS, AFDN and SCRIB.  
+
+<|ref|>text<|/ref|><|det|>[[114, 164, 878, 328]]<|/det|>
+Fig. 2A. Here AFDN and SCRIB, both epitope- tagged, are overexpressed in a HeLa cell that is processed for immunofluorescence. It is not stated if the GFP- AFDN is imaged with the intrinsic fluorescence of GFP or if an anti- GFP antibody is employed along with the anti- FLAG antibody. No control proteins are employed nor is a control with first antibody omitted shown. The localization is indeterminant and uninformative. A single cell is shown such that one cannot determine if this represents the predominant fluorescent pattern. Since tagged, ectopically expressed proteins are used, including one tagged with GFP, it is not clear why the authors did not use mCherry- SCRIB such that they could colocalize the two proteins in live cells, which allow for more precise subcellular localization free of fixation and permeabilization artifacts.  
+
+<|ref|>text<|/ref|><|det|>[[113, 364, 880, 899]]<|/det|>
+Fig. 4F,G. In describing these micrographs the authors state on p. 9 that "tagged, wild- type KRAS does not significantly alter AFDN or SCRIB localization and does not co- localize with these proteins (Figure 4F and S4B). ... In contrast, expression of KRAS- G12V markedly disrupted cell- cell contacts and was noticeably co- localized with endogenous AFDN and SCRIB, as determined by z- plane projections (Figure 4G)." In the discussion on p. 13 the authors write that "we show that RAS- G12V is co- localized with AFDN and SCRIB at sites of cell contact, while wild- type RAS is distributed more generally across the plasma membrane." The data do not support these conclusions. First, not shown is any disruption of cell- cell contacts in cells expressing KRAS- G12V. Three GFP- KRAS4B- G12V expressing cells are shown with three levels of expression and perhaps different z- planes. Two of these three maintain robust cell- cell contacts as determined by morphology and SCRIB and AFDN staining of areas of cell contact (anti- ZO- 1 staining would be a way to look at this without imaging the experimental proteins themselves). Not shown are any of the detached cells to which the authors refer as having lost cell- cell contacts as a function of oncogenic KRAS. Second, and more important, the data show that both WT- KRAS and KRAS- G12V decorate the basolateral membrane. Indeed, both the WT and mutant KRAS decorate the entire plasma membrane (PM) as is well established in a vast literature. As expected for cell adhesion proteins imaged in confluent epithelial cells, SCRIB and AFDN decorate primarily the basolateral membrane. The conclusion that these proteins are colocalized to a greater extent with KRAS- G12V that with WT KRAS is not supported by the data shown and contrary to a vast literature on KRAS localization. Some of the problem is semantic; the concept of co- localization is somewhat ambiguous. There is co- localization on the PM in some regions of the cell but not others. This should not necessarily be interpreted as one protein pulling another to a region of PM since the same pattern would be observed if the localizations are true but unrelated to the direct interactions of the proteins. The three proteins colocalize at the basolateral membrane but not the apical membrane and this is not affected by the GTP- binding state of KRAS. Were this localization of mutant KRAS to differ from that of WT it would be contrary to a vast literature on KRAS localization that in total demonstrates that the subcellular localization driven by the prenylated HVR is not affected by the GTP- binding state. Current paradigms of RAS signaling hold the PM localization of KRAS is constitutive and it is effectors that are drawn to RAS (e.g. translocation of RAF) not vice versa. Are the authors arguing that in this case the converse is true and that the localization of KRAS is driven by that of its effector?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 117, 879, 245]]<|/det|>
+Fig. 5D. The altered localization of endogenous SCRIB as a consequence of silencing AFDN is described on p. 10 as "dispersed throughout the cytoplasm." But unlike the distribution of GFP that is homogeneous and clearly cytosolic, that of SCRIB is punctate consistent with a vesicular localization and should be described as such. Co- localization with Texas- red transferrin would allow an assessment as to whether these are endosomes. Caution must be taken in what is used for permeabilization of the cells (here \(0.05\%\) Tween- 20) as this can alter the appearance of vesicles. It would be wise to also try \(0.1\%\) saponin.  
+
+<|ref|>text<|/ref|><|det|>[[114, 283, 881, 393]]<|/det|>
+Despite binding of several RAS family small GTPases, in their colocalization studies the authors restricted their analysis to KRAS4B. This is unfortunate. It would be informative to also study a RAS- related binding partner that is not normally localized exclusively to the PM. RAP2 fits the bill as the authors show strong binding to AFDN and this small GTPase has been localized to endomembrane (PMID: 1923507 and 19061864). It would also be interesting to determine if RHEB interacts with AFDN since this RAS family small GTPase is expressed on lysosomes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 433, 215, 448]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 488, 879, 561]]<|/det|>
+Fig. 2B. This figure sets up indirect immunofluorescence (iIF) staining of endogenous AFDN and SCRIB, which is used extensively throughout the paper. The specificity of antibodies used for iIF must always be validated by knockdown of the protein of interest, which is accomplished in Fig. 5 and this should be added to the legend of Fig. 2B.  
+
+<|ref|>text<|/ref|><|det|>[[115, 600, 877, 672]]<|/det|>
+Fig. 4D. The authors write on p. 7 that with an \(n = 5\) they saw an 8- fold enhancement of SCRIB co- IP with AFDN upon expression of KRAS4B- G12V, but they do not report results for RAP1B or RAP2C, which appear to also induce some enhancement, albeit to a lesser extent (KRAS4B>RAP2C> RAP1B). It would be informative to report the results for each of the interacting GTPases.  
+
+<|ref|>text<|/ref|><|det|>[[115, 712, 491, 729]]<|/det|>
+Fig. S6B. It is very difficult to see the AFDN staining.  
+
+<|ref|>text<|/ref|><|det|>[[115, 768, 864, 804]]<|/det|>
+Fig. 6C,D. These z projections are convincing that true colocalization of AFDN and SCRIB requires PDZ1 and FHA, but would be even more so if they were subjected to analysis with Pearson's coefficient.  
+
+<|ref|>text<|/ref|><|det|>[[115, 844, 861, 880]]<|/det|>
+The authors refer to KRAS throughout but they mean KRAS4B. They do not study KRAS4A. Since these splice variants differ only in their HVRs that direct subcellular trafficking this should be acknowledged.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 117, 883, 208]]<|/det|>
+To be a true effector of a small GTPase, three conditions must be met. The effector must bind directly to the GTPase, the binding must depend on GTP- loading of the GTPase, and the binding must in some way change the conformation or activity of the effector. RAF and HK1 meet all of these criteria but the third has been lacking for AFDN. In Fig. 4D the authors establish for the first time a change in the properties of AFDN induced by KRAS, confirming that AFDN is a bone fide effector. This should be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[114, 247, 881, 303]]<|/det|>
+Because AFDN is unique in possessing tandem RA domains that, in principal, could bind two GTPases the authors have a unique opportunity to ask if either or both are required for the effect of KRAS seen in Fig. 4D. Indeed, it would be interesting to determine if RAS binds to one and RAP2 to the other RA domain.  
+
+<|ref|>text<|/ref|><|det|>[[114, 312, 875, 386]]<|/det|>
+The change in kinetics of ERK and AKT activation downstream of EGF signaling upon silencing AFDN or SCRIB shown in Fig. 7D,E is interesting but the authors do not comment on possible mechanisms. Interestingly they parallel the differential effects of NGF versus EGF in PC12 cells where only the former induces sustained ERK activation (PMID 7834738).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 883, 125]]<|/det|>
+This revised manuscript addresses all Referee comments with either experimental data or further clarification, as required. We have added the following results to the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[120, 134, 884, 380]]<|/det|>
+- Purified SCRIB PDZ domains 2 and 3-4, measured binding to the AFDN FHA domain (no binding demonstrates specificity of the AFDN domain for SCRIB PDZ1)- Determined GTPase specificity of AFDN RA domains 1 and 2 alone, to contrast what was observed with the RA1-2 construct (GTPase binding is avidity driven by the tandem domains, particularly KRAS)- Derived suitable conditions for our phosphatase treated co-IP of AFDN and SCRIB, and included the untreated control- Demonstrated that AFDN and RAS GTPases co-localize in HeLa cells, which do not express detectable levels of endogenous AFDN, substantiating AFDN as a RAS effector- Obtained new images for KRAS expressed in MCF7 cells to demonstrate how cells expressing activated KRAS detach from the monolayer- Used RAB5, RAB7 and RAB11 GTPases to resolve whether the punctate SCRIB pattern observed in AFDN KO MCF7 cells is localized to endosomes- Demonstrated co-localization of AFDN with activated RAP1B and RAP2C GTPases  
+
+<|ref|>text<|/ref|><|det|>[[115, 389, 461, 406]]<|/det|>
+Specific comments to each reviewer follow.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 446, 219, 464]]<|/det|>
+## Reviewer #1  
+
+<|ref|>text<|/ref|><|det|>[[115, 473, 883, 630]]<|/det|>
+Comments: This is an extraordinary effort to characterize AFDN interaction network, using proximity labelling- MS approach. Authors were looking into both isoforms of AFDN as well. They found that AFDN binds to its previously unknown interactor, SCRIB, adhesion/polarity tumor suppressor protein. Both AFDN and SCRIB were found to be linked with RAS- induced cell growth and invasion. Next, they were looking into specific binding domains, involved in AFDN- SCRIBS complex and the molecular mechanism of this interaction. They found that that interaction is mediated through the first PDZ domain of SCRIB and the AFDN FHA domain. Further, the knockout experiments reveal the important role of AFDN- SCRIB interaction in MAPK and PI3K activation, and cell motility.  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 883, 675]]<|/det|>
+This data provide insight into new roles of AFDN, as not- so- well characterized RAS effector. I appreciate the quality of results, data and figures, and recommend this manuscript for publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 684, 759, 702]]<|/det|>
+We thank the reviewer for their positive comments and appreciation for our work.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 741, 220, 758]]<|/det|>
+## Reviewer #2  
+
+<|ref|>text<|/ref|><|det|>[[115, 767, 883, 907]]<|/det|>
+In the manuscript by Goudreault et al. entitled 'Afadin couples RAS GTPases to the polarity rheostat protein Scribble', the investigators employ proximity- based proteomics using afadin (AFDN) isoforms and identify the polarity protein SCRIB/Scribble as the top hit. Additional support for this interaction is obtained using in vitro binding assays and in- cell pull down experiments in which the authors find that the first PDZ domain of SCRIB and the AFDN FHA domain directly associate. AFDN has previously been shown to be an effector of the RAS and RAP GTPases and binds through its RA domains. Consistent with these earlier observations, the investigators find that KRAS co- localizes with AFDN to facilitate formation of the AFDN- SCRIB
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 883, 143]]<|/det|>
+complex. Knockout of AFDN or SCRIB in MCF7 epithelial cells disrupts growth dependent MAPK and PI3K activation and inhibits cell motility. Results from these studies provide a new link between AFDN, SCRIB and RAS- mediated MAPK signaling.  
+
+<|ref|>text<|/ref|><|det|>[[115, 153, 883, 188]]<|/det|>
+We thank the reviewer for their thoughtful suggestions and have attempted to address all points raised with comments, or by acquiring new experimental data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 224, 883, 329]]<|/det|>
+1) The authors state that 'We observed in Fig. 2F that only the N-SCRIB, but not C-SCRIB construct, is able to pulldown the AFDN-FHA fragment, though the C-SCRIB contains 3 additional PDZ domains.' Have the 4 PDZ domains been compared to assess possible sequence determinants (e.g. specificity for the first PDZ domain)? Also, as the PDZ domains may interact or be involved in autoinhibitory interactions, have the investigators probed the individual PDZ domains or the 4 PDZ domains together?  
+
+<|ref|>text<|/ref|><|det|>[[114, 339, 883, 636]]<|/det|>
+There are significant literature describing efforts to determine PDZ domain binding specificity. To date, it has not been possible to establish binding specificity of individual domains based on sequence alone. The most valuable approaches have been large scale, non- biased screens of PDZ binding to peptide libraries and many include the PDZ domains of SCRIB \(^{1 - 3}\) . Indeed, several recent papers have resolved SCRIB PDZ 1- 4 specificity for individual substrates, though resolving specificity determinants within the domains themselves has required structural elucidation \(^{4 - 6}\) . As we have been unable to successfully crystallize the PDZ1- FHA complex, and most library screens to date have focused on C- terminal PDZ motifs, a sequence- based prediction of why PDZ1 binds to AFDN while PDZ2, 3, and 4 domains of SCRIB do not is currently not feasible. To satisfy the reviewers request and demonstrate that our elucidated interaction is specific to PDZ1, we generated expression vectors for SCRIB PDZ2 and the PDZ3- 4 supramodule \(^{7}\) . We expressed and purified these domains to homogeneity and used ITC to measure their binding affinity for the AFDN FHA domain (new Supplementary Fig. 3d). We observed no binding between the FHA domain and SCRIB PDZ2, while the PDZ3- 4 module showed only extremely weak binding on the order of \(\sim 100 \mathrm{mM}\) . Thus, the AFDN- SCRIB interaction is highly specific for the first PDZ of SCRIB, though the determinants of this specificity must be elucidated with future structural characterization of this complex.  
+
+<|ref|>text<|/ref|><|det|>[[112, 645, 884, 870]]<|/det|>
+1. Zhang, Y. et al. Convergent and divergent ligand specificity among PDZ domains of the LAP and zonula occludens (ZO) families. Journal of Biological Chemistry 281, 22299-22311 (2006).  
+2. Ivarsson, Y. et al. Large-scale interaction profiling of PDZ domains through proteomic peptide-phage display using human and viral phage peptidomes. Proceedings of the National Academy of Sciences of the United States of America 111, 2542-7 (2014).  
+3. Mu, Y., Cai, P., Hu, S., Ma, S. & Gao, Y. Characterization of diverse internal binding specificities of PDZ domains by yeast two-hybrid screening of a special peptide library. PLoS ONE 9, e88286 (2014).  
+4. Lim, K. Y. B., Gödde, N. J., Humbert, P. O. & Kvansakul, M. Structural basis for the differential interaction of Scribble PDZ domains with the guanine nucleotide exchange factor β-PIX. Journal of Biological Chemistry 292, 20425-20436 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 90, 883, 125]]<|/det|>
+5. Caria, C. et al. Structural analysis of phosphorylation-associated interactions of MCC to Scribble PDZ domains. FEBS Journal 286, 4910-4925 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[115, 134, 882, 188]]<|/det|>
+6. How, J. Y., Stephens, R. K., Lim, K. Y. B., Humbert, P. O. & Kvansakul, M. Structural basis of the human Scribble–Vang12 association in health and disease. Biochemical Journal 478, 1321–1332 (2021).  
+
+<|ref|>text<|/ref|><|det|>[[115, 197, 882, 233]]<|/det|>
+7. Ren, J. et al. Interdomain interface-mediated target recognition by the Scribble PDZ34 supramodule. Biochemical Journal 468, 133–144 (2015).  
+
+<|ref|>text<|/ref|><|det|>[[115, 269, 882, 392]]<|/det|>
+2) Have the investigators compared binding of full length AFDN and SCRIB proteins with the individual domains (FHA/PDZ1) by in vitro assays? This comparison would be helpful in assessing whether the FHA and PDZ1 domains are sufficient for binding of AFDN and SCRIB, especially since the binding appears weak (~22 uM). Although full-length AFDN is compared with AFDN delta FHA (Figure 2I), a faint band is discernible for the FHA deleted protein suggesting that other regions of the protein may be involved in the interaction. A similar comparison is also not made for the SCRIB protein (Figure 2J).  
+
+<|ref|>text<|/ref|><|det|>[[115, 400, 882, 627]]<|/det|>
+While we would be happy to measure in vitro binding of full length SCRIB and AFDN, this is an extremely challenging experiment. Both proteins are \(\sim 200 \mathrm{kDa}\) in size and cannot be expressed and purified from bacteria. We have attempted to purify these proteins using a mammalian cell expression system, but AFDN in particular has a propensity to aggregate in vitro as a full length protein. While we anticipate purifying and studying this complex in the future, it is beyond the scope of this manuscript and will require many years of work. I do believe the reviewer is correct that there are likely additional contact points outside PDZ1-FHA. The two proteins co-IP under highly stringent conditions (even in RIPA buffer), indicating the measured in vitro affinity between the PDZ1 and FHA domains (5- 15 \(\mu \mathrm{M}\) depending on NaCl concentration) is perhaps insufficient to rationalize the overall robust binding of the two proteins from cell lysate. It is possible that the AFDN oligomers we observe following its purification from mammalian cells are genuine, perhaps supporting a model whereby multiple AFDN proteins can strongly co- precipitate SCRIB through multiple FHA-PDZ1 interactions of moderate affinity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 665, 882, 752]]<|/det|>
+3) In Fig. 2C and 2D, IP with AFDN results in a strong SCRIB band with ectopic expression. Interestingly, this interaction looks to be retained with endogenous protein in panel 2D. However, the level of endogenous SCRIB protein pulled down looks to be much less than the GFP-tagged version despite high levels of endogenous expression. Is there a hypothesis for why this is observed?  
+
+<|ref|>text<|/ref|><|det|>[[115, 762, 882, 903]]<|/det|>
+There are numerous explanations for why co-immunoprecipitation of endogenous proteins may not re- capitulate the strong complex observed by ectopic expression. A primary factor is the binding of antibodies to the proteins directly, rather than to a recombinant tag. In this case, the AFDN polyclonal antibodies were raised against a "synthetic peptide corresponding to Mouse Afadin aa 1800 to the C-terminus (abcam)". While this should not disrupt the PDZ1-FHA interaction, it is difficult to predict if antibody binding to the C-terminus of AFDN may disrupt SCRIB binding until we have a structure of the full proteins in complex. Indeed, related to the previous question, we now believe that AFDN oligomerization is mediated through its C-terminal
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 212]]<|/det|>
+region and it is possible that disruption of this higher order structure diminishes the amount of SCRIB bound in an IP. Moreover, it is conceivable that a fraction of endogenous SCRIB/AFDN in these cells are post- translationally modified in a manner that may disrupt their interaction, or that a population of the endogenous proteins are engaged in alternative complexes that are competitive with the AFDN- SCRIB module. Altogether, we were happy to observe the complex could be validated with endogenous proteins and future work will resolve the detailed mechanisms by which it is regulated in cells.  
+
+<|ref|>text<|/ref|><|det|>[[114, 248, 883, 389]]<|/det|>
+4) The authors state 'To map the interaction site on the SCRIB PDZ1 domain we used NMR spectroscopy. BMRBid 11207 was used to assign 71% of the peaks in a 1H/15N-HSQC of 15N-labelled SCRIB PDZ1.' It would be helpful state the conditions (pH, salt, etc) for published assignments in comparison to the NMR data collected, as the assignments are key for chemical shift mapping of the binding interaction. Please also indicate how well spectra overlay with previously determined assignments. Also, in Figure 3E, the full HSQC spectrum should be shown (with zoom as appropriate). A docked model of the interaction with the AFDN FHA domain could be generated using the NMR data but not included.  
+
+<|ref|>text<|/ref|><|det|>[[115, 398, 882, 434]]<|/det|>
+To address the first question, the assignment of SCRIB PDZ1 in the BMRB was deposited by RIKEN and there is no accompanying manuscript, but the buffer conditions in the PDB are:  
+
+<|ref|>text<|/ref|><|det|>[[115, 443, 605, 461]]<|/det|>
+20mM d- Tris- HCl; 100mM NaCl; 1mM d- DTT; 0.02% NaN3  
+
+<|ref|>text<|/ref|><|det|>[[115, 471, 742, 489]]<|/det|>
+This is nearly identical to the conditions used to acquire our own PDZ1 spectra:  
+
+<|ref|>text<|/ref|><|det|>[[115, 498, 506, 516]]<|/det|>
+20 mM Tris (pH 7.5), 100 mM NaCl, 1 mM DTT  
+
+<|ref|>text<|/ref|><|det|>[[114, 525, 882, 788]]<|/det|>
+To further indicate how well the previously assigned spectra overlays with our own data, we have added the full spectra superimposed with projected peaks from the BMRB assignment (new Supplementary Fig. 4a). This clearly demonstrates how similar our experimental data is to the previously derived assignment, and how we were able to unambiguously assign 70% of the observed peaks. The second point raised here is to include a full HSQC of the AFDN FHA domain in Fig. 3e. We do not believe a full HSQC would offer any additional information, and the space constraints make it difficult to fit complete spectra in Fig. 3. The panels shown demonstrate how the PDZ1 domain induces chemical shift perturbations in residues located in the C- terminal extension of the FHA domain. There are no chemical shifts from these residues in the broader spectrum, so we have chosen to leave this figure as originally presented. Finally, with regards to modeling we have used HADDOCK to generate models of the PDZ1- FHA interaction, aided by chemical shift perturbations used as restraints. While the models are interesting, they are not experimentally corroborated and do not provide any additional value to the current work. We prefer to wait for a proper structural elucidation of the full complex rather than speculate on a predicted model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 824, 882, 894]]<|/det|>
+5) It is unclear whether there is a difference in the construct used in figure 3D (371-532) and Figure 3F (371-531). If they are the same then a correction is required. If they are different then the rationale for using two constructs with a difference in 1 amino acid is needed. Also, Figure 3D and 3F could be combined.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 716, 108]]<|/det|>
+These are the same construct (371- 532) and the labelling has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[115, 145, 706, 164]]<|/det|>
+6) In Fig 3C, the minus phosphatase control for the pulldown is not shown.  
+
+<|ref|>text<|/ref|><|det|>[[115, 172, 817, 191]]<|/det|>
+We have redone this experiment and now include a minus phosphatase control in Fig. 3c.  
+
+<|ref|>text<|/ref|><|det|>[[115, 226, 883, 263]]<|/det|>
+7) Additional bands are observed in Fig. 3D, 3F for experiments with purified protein. Is this due to degradation or other protein contaminants? Some discussion is needed here.  
+
+<|ref|>text<|/ref|><|det|>[[115, 271, 882, 342]]<|/det|>
+The proteins used in these binding assays are GST- tagged and bound to glutathione beads. The smaller bands beneath our proteins of interest are typical degradation bands observed with GST fusion proteins, particularly following incubation with the partner protein (here, purified PDZ1) on a nutator for \(>1\) hour followed by multiple washes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 378, 882, 432]]<|/det|>
+8) In Fig. 4, it is stated that densitometry analysis of western blots showed an 8-fold increase in SCRIB protein pulled down with AFDN/KRAS G12V co-expression. However, this data does not appear to be included in the submission.  
+
+<|ref|>text<|/ref|><|det|>[[115, 440, 882, 512]]<|/det|>
+We have added quantitation of SCRIB co- immunoprecipitation with AFDN in the new Fig. 4g. The plot shows SCRIB levels when co- expressed with KRAS- G12V, as well as RAP1B- G12V and RAP2C- G12V. Ratios are in comparison to SCRIB levels co- precipitated when AFDN is expressed with wild- type KRAS.  
+
+<|ref|>text<|/ref|><|det|>[[115, 547, 882, 653]]<|/det|>
+9) In Fig. 4D, authors show that increased SCRIB precipitation is observed with AFDN in the presence of KRAS G12V. AFDN is also shown to be immunoprecipitated with KRAS and related RAP GTPases through the RA domains. Interestingly, in panel 4D, some RAP GTPase appears to be pulled down with just FLAG-AFDN and EGFP-vector, but the same is not observed for KRAS G12V here. What is the source of the disconnect between panel 4C and 4D with regard to AFDN/KRAS G12V?  
+
+<|ref|>text<|/ref|><|det|>[[115, 663, 882, 855]]<|/det|>
+(Previous Fig. 4d is now Fig. 4e) We show multiple times that KRAS- G12V can be precipitated from cell lysate by the AFDN RA domains (Figs 2f and 4c/d) or that KRAS co- precipitates with full length AFDN (Figs 4a/b, 5c and Supplementary Fig. 5a). We have also previously solved a crystal structure of the AFDN RA1 bound to RAS- GTP and characterized this complex by ITC (Nat. Comm., 2017). These data clearly demonstrate that KRAS- G12V binds AFDN, though a strong band is not visible in lane 9 of Fig. 4e. This is likely because this is a triple transient transfection (3 plasmids) and the levels of GTPase expressed are relatively low. This might be improved by generating series of stable cell lines expressing the GTPases, but the objective of this experiment was to determine the relative levels of SCRIB present in AFDN co- IPs when the activated GTPases are present. As it worked well for this, it did not seem necessary to further resolve the KRAS- AFDN interaction when it is studied extensively in other experiments.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 883, 143]]<|/det|>
+10) Similarly in Fig. 4E, a FLAG pulldown control absent of FLAG-AFDN shows modest levels of KRAS G12V immunoprecipitated. Is this due to residual SCRIB interaction with KRAS or an artifact using a FLAG-antibody?  
+
+<|ref|>text<|/ref|><|det|>[[115, 152, 882, 275]]<|/det|>
+(Previous Fig. 4e is now Fig. 4f) These are merely background bands common when working with RAS GTPases. The proteins are unstable when nucleotide free or when prenylated in the absence of lipid membrane and it is common to have background. As in the comment above, this is a triple transient transfection and the expression level of the GTPases is low (we must expose for \(\sim 2\) minutes to observe strong bands). The objective of this experiment was to resolve whether SCRIBAPDZ1 alters the interaction between AFDN and activated KRAS, and the experiment clearly demonstrates that it does not.  
+
+<|ref|>text<|/ref|><|det|>[[115, 311, 882, 451]]<|/det|>
+11) Fig. 4F and 4G require reworking for presentation purposes. The Z-stacks are not very legible given the intent of showing merged co-localization of KRAS:AFDN:SCRIB. Though the figure panel is intended to show increased localization of KRAS G12V vs. KRAS WT to AFDN:SCRIB, this point is not currently made and it is unclear whether this is due to data interpretation or simply representation of the data. The investigators may want to consider alternative color mapping for better contrast of co-localization points, or a larger zoom/view for readers to see the merged fluorescence channels. Additionally, differentiation between SCRIB/AFDN merge vs. SCRIB/AFDN/KRAS merges could be useful for interpretation of the two separate complexes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 882, 653]]<|/det|>
+We have taken multiple approaches to address the points raised here. First, we enlarged the z- stack panels in these images to make the expression patterns more legible (now Fig. 5a/b). More importantly, we reconsidered whether MCF7 cells were an appropriate system to demonstrate co- localization of RAS GTPases with AFDN, as both proteins are constitutively membrane localized in polarized epithelial cells. This made it very challenging to demonstrate co- localization driven by RAS activation. Instead, we went back to the HeLa cell system as this line does not express detectable levels of endogenous AFDN and does not form adherens junctions. The new Fig. 5c shows how expression of activated KRAS, and not wild- type KRAS, recruits exogenously expressed AFDN to the cell membrane. We hope the reviewer agrees this is a more robust demonstration that AFDN co- localizes with activated RAS in cells and that it should be considered a RAS effector.  
+
+<|ref|>text<|/ref|><|det|>[[115, 690, 882, 778]]<|/det|>
+12) The same concern exists with Fig. 5E. The authors should provide a better representation for the immunofluorescence localization and internalization of AFDN. They state that expression of KRAS G12V in the SCRIB KO line induces internalization of AFDN in the figure legends, but this is not addressed in the Results proper. Rather, in Results, the authors state that AFDN is retained at sites of cell-cell contact.  
+
+<|ref|>text<|/ref|><|det|>[[115, 787, 882, 874]]<|/det|>
+In response to this point, and a related point from Reviewer 4, we have added numerous images of immunostained AFDN in KRAS- expressing MCF7 cells (the new Supplementary Fig. 5c/d). These images show that AFDN is no longer visible at cell- cell contacts in the majority of cells expressing KRAS- G12V, while it is retained at cell contacts in those expressing wild- type KRAS. Indeed, most cells expressing KRAS- G12V are no longer associated with adjacent cells and empty
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 882, 125]]<|/det|>
+space is clearly observable surrounding these cells. The results are appropriately discussed in the Results section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 161, 882, 268]]<|/det|>
+13) Similar concerns also exist with Fig. 6A and 6B regarding representation of the data. In 6A, it appears that SCRIB may be less recruited to cell contacts with the FHA deletion of AFDN as compared to WT, though authors state the opposite. The trend the authors state, is more clear in the reverse scenario with AFDN and the SCRIB PDZ truncation in that there is less AFDN association at the cell contacts. Perhaps an alternative quantification representation for the colocalization of AFDN and SCRIB will better illustrate the authors' point/clarify the data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 276, 882, 399]]<|/det|>
+We have calculated Pearson's coefficients to quantify the levels of endogenous SCRIB colocalized with EGFP- tagged wild- type AFDN or the \(\Delta \mathrm{FHA}\) variant (in AFDN KO cells), or endogenous AFDN with EGFP- tagged wild- type SCRIB or its \(\Delta \mathrm{PDZ1}\) variant (in SCRIB KO cells). These calculations were performed on \(n\geq 8\) images similar to those presented in Fig. 7c/d. The results are presented in the new Supplementary Fig. 7c. This quantitation resolved a significant difference in the co- localization of endogenous SCRIB/AFDN with the respective wild- type rescues compared with domain deletions, and is now presented in the text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 435, 882, 471]]<|/det|>
+14) Minor comment. In the Fig. 7 legend consider adding \((A - C)\) at beginning of legend for panels \(A - C\) and \((D - E)\) to maintain parallel structure.  
+
+<|ref|>text<|/ref|><|det|>[[116, 481, 427, 499]]<|/det|>
+This was corrected in the figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 535, 882, 623]]<|/det|>
+15) The authors state that a signaling "defect" for the ERK MAPK/PI3K-AKT signaling cascade is induced with KO of AFDN or SCRIB. They may want to consider rewording, as the initial increase in temporal signaling indicates that signaling is not defective, but rather altered. Have the authors examined this temporal signaling effect of AFDN/SCRIB KO in the context of an activated KRAS such as KRAS G12V?  
+
+<|ref|>text<|/ref|><|det|>[[115, 633, 882, 703]]<|/det|>
+This may be semantics, but 'defect' is defined as "an imperfection or abnormality that impairs quality, function, or utility". We would argue that the disrupted kinetics of AKT and ERK activation demonstrated here represent a genuine abnormality in response to EGF stimulation and can thus be appropriately described as a "defect". See also the response to Reviewer 4, point 12.  
+
+<|ref|>text<|/ref|><|det|>[[115, 740, 882, 774]]<|/det|>
+16) Statistics for pERK and pAKT is lacking to show a difference between "WT"/AFDN or SCRIB KO?  
+
+<|ref|>text<|/ref|><|det|>[[115, 785, 882, 907]]<|/det|>
+I believe the reviewer is referring to Fig. 8d/e (previously Fig. 7d/e), which graphs quantitation of pERK and pAKT in the parental cell line vs the AFDN and SCRIB KO lines. This is a time course of EGF stimulation, and to my knowledge there is no established statistical approach that will determine a p- value across the entire series. A 2- way ANOVA can be used, but this would treat the time points randomly rather than sequential. If the course resulted in a linear response, we could solve this with regression, but the response is not linear. We can use multiple independent t- tests (perhaps focused on the time points of interest), but this disregards that the series are
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 882, 142]]<|/det|>
+sequential and therefore does not provide value. We have left the qualitative graphs in Fig. 8d/e without p- values for now, but if the reviewer has insights to how such non- linear time course data can be tested statistically, we are happy to perform the analysis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 179, 882, 215]]<|/det|>
+17) In Fig. 7I, consider a directional quantification of the leading-edge Golgi stain or a better zoomed-in insert to emphasize the loss of directionality in SCRIB/AFDN KO cells.  
+
+<|ref|>text<|/ref|><|det|>[[115, 224, 825, 243]]<|/det|>
+We have added a zoomed image of cells at the leading edge to Fig. 8i (previously Fig. 7i).  
+
+<|ref|>text<|/ref|><|det|>[[115, 278, 882, 314]]<|/det|>
+18) Mechanistic insight into how the AFSN/SCRIB effects the RAS-mediated pERK or pAKT activation would aid the discussion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 324, 481, 341]]<|/det|>
+Please see our answer to Reviewer 4, point 12.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 406, 220, 423]]<|/det|>
+## Reviewer #3  
+
+<|ref|>text<|/ref|><|det|>[[114, 433, 882, 714]]<|/det|>
+The manuscript "Afadin couples RAS GTPases to the polarity rheostat Scribble" identifies a novel proximity interaction between Afadin and Scribble, using BioID coupled to mass spectrometry. A series of carefully crafted IP experiments identified that the PDZ domain of Scribble interacts with FHA domain of AFDN. Furthermore, the authors characterised the interaction between the domains using different techniques to reveal a model of how the two domains bind to each other. Since, AFDN binds to several RAS GTPases, the authors determined and characterised the interaction of activated forms of the GTPases with AFDN- SCRIB complex. Using CRISPR/Cas9 gene editing, the authors created a suite of single and double KO cell lines of AFDN and SCRIB. Using these cells lines, the authors further support a KRAS-AFDN- SCRIB complex formation and the requirement of a direct FHA- PDZ1 interaction for proper localisation of AFDN- SCRIB at cell contacts. Finally, the authors show that the loss of either AFDN or SCRIB disrupts ERK and AKT activation kinetics and cell motility in growth factor- dependent manner. The manuscript is well written and experiments are performed with nice controls. The observations and conclusions drawn in this manuscript would help advance the field in a significant way. The data presented in the manuscript provides ample support to the conclusions drawn. I have few suggestions that would improve the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[115, 722, 881, 793]]<|/det|>
+1) BirA\* is a large tag and can cause a significant level of mislocalisation when expressed in cells. Since the BioID results show some proteins from other compartments like ER/Golgi, mitochondria, etc, it would be nice if the authors can determine subcellular localisation of the AFDN-BirA\* tagged fusion proteins (both isoforms).  
+
+<|ref|>text<|/ref|><|det|>[[115, 802, 882, 890]]<|/det|>
+We thank the reviewer for their helpful comments and will address the points raised here. A distinct advantage of proximity- based proteomics is the insight provided to subcellular localization. The results presented in Fig. 1 (and accompanying Table 1) make clear that BirA\*- AFDN is properly behaved, as the majority of identified preys are plasma membrane proteins involved in cell adhesion (including most previously known AFDN interactors). The reviewer is correct that the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 90, 883, 125]]<|/det|>
+BirA\* tag is large, and to satisfy the request we imaged our BirA\*- FLAG-AFDN iso1 construct in the MCF7 cell line:  
+
+<|ref|>image<|/ref|><|det|>[[270, 133, 725, 352]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[114, 363, 883, 470]]<|/det|>
+This confocal image shows that BirA\*- FLAG- AFDN is prominently localized at cell- cell contacts, as expected. We include here staining with Streptavidin- 488 following 2 hours of incubation with biotin. This shows the majority of biotinylated proteins are also at the plasma membrane, though AFDN and Streptavidin- 488 signals are also visible in the interior of these cells (Strep- 488 also characteristically marks mitochondria in the surrounding cells). We are confident based on these images and the derived proteomic data that BirA\*- AFDN appropriately localizes in cells.  
+
+<|ref|>text<|/ref|><|det|>[[114, 505, 883, 576]]<|/det|>
+2) Since the authors have created multiple CRISPR/Cas9 gene edited cell lines and assessed the precise nature of the edit by Sanger sequencing (line: 693), it would be recommended to show the precise edits and how they impact expression of that gene (i.e. Introduce STOP codon, impact splicing, destroy START codon, frame-shift, etc).  
+
+<|ref|>text<|/ref|><|det|>[[115, 585, 883, 621]]<|/det|>
+We have added a table to the Results section (under the subheading Generation of AFDN and SCRIB CRISPR KO cells and Rescues) that describes sequencing of the KO cell lines in detail.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 658, 220, 675]]<|/det|>
+## Reviewer #4  
+
+<|ref|>text<|/ref|><|det|>[[114, 685, 883, 895]]<|/det|>
+Afadin (AFDN), a regulator of cell- cell contacts, has long been recognized as an effector of RAS and related small GTPases but little is known about the interaction or its functional consequences. AFDN is unusual among the dozen or so effectors of RAS because it possesses two tandem RA domains. Goudreault et al. set out to explore the AFDN interactome by proximity labeling and here report a comprehensive characterization of the interaction with their most robust hit, SCRIB, a tumor suppressor and polarity protein that possesses four tandem PDZ domains. They show that AFDN and SCRIB associate via a non- canonical interaction of the first PDZ domain with the forkhead associated (FHA) domain of AFDN and that the interaction is enhanced by GTP- bound KRAS that forms a ternary complex with the two polarity proteins. Conversely, they show that silencing AFDN or SCRIB changes the kinetics of growth- factor stimulated ERK and AKT signaling in MCF7 epithelial cells. The manuscript is exceedingly well written. The authors walk the reader through not only the experiments and results but also the thinking behind them. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 882, 230]]<|/det|>
+turboID proximity screen is well done, appropriately controlled, and clearly reported. The biochemical validation of the interaction of AFDN and SCRIB is outstanding, particularly the rigor applied to confirming the non- canonical nature of the interaction between the AFDN FHA domain with the first PDZ domain of SCRIB. The promiscuity of the AFDN RA domain(s) for RAS family proteins relative to the specificity of the RAF1 RBD is well demonstrated. These results are clear, novel, and of interest to cell biologists and are certainly worthy of reporting. The weakest part of the study is the overinterpretation of the immunofluorescent localizations and colocalizations of KRAS, AFDN and SCRIB.  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 882, 275]]<|/det|>
+We thank the reviewer for evaluating our manuscript and will address the concerns here, including with the addition of new experimental data.  
+
+<|ref|>text<|/ref|><|det|>[[114, 310, 882, 469]]<|/det|>
+1) Fig. 2A. Here AFDN and SCRIB, both epitope-tagged, are overexpressed in a HeLa cell that is processed for immunofluorescence. It is not stated if the GFP-AFDN is imaged with the intrinsic fluorescence of GFP or if an anti-GFP antibody is employed along with the anti-FLAG antibody. No control proteins are employed nor is a control with first antibody omitted shown. The localization is indeterminant and uninformative. A single cell is shown such that one cannot determine if this represents the predominant fluorescent pattern. Since tagged, ectopically expressed proteins are used, including one tagged with GFP, it is not clear why the authors did not use mCherry-SCRIB such that they could colocalize the two proteins in live cells, which allow for more precise subcellular localization free of fixation and permeabilization artifacts.  
+
+<|ref|>text<|/ref|><|det|>[[114, 478, 882, 618]]<|/det|>
+In Fig. 2a we show the localization of ectopically tagged AFDN and SCRIB in HeLa cells (as this was the cell line from which our proteomics data was collected) before moving on to MCF7 cells. The reviewer states the localization is 'indeterminant and uninformative', but this is indeed the diffuse localization pattern we observe from these proteins in this epithelial line, which lack defined cell- cell contacts and apical- basal polarity. Levels of endogenous AFDN in HeLa cells were too low to detect by immunofluorescence. Endogenous SCRIB is expressed, can be detected, and displays a diffuse/punctate localization when immunostained that is highly reminiscent of SCRIB in AFDN KO MCF7 cells:  
+
+<|ref|>image<|/ref|><|det|>[[370, 626, 627, 826]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 882, 907]]<|/det|>
+Others have observed this pattern in different cell types<sup>1</sup>. To satisfy the reviewers request, we co- expressed EGFP- AFDN and Cherry- SCRIB in these cells (the new Supplementary Fig. 1d), or appropriate controls, to avoid fixation/permeabilization effects. We again observe broad staining in the cytoplasm for both AFDN and SCRIB, with some weak concentration at the cell cortex. In
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 882, 160]]<|/det|>
+the following section we describe how this was used to improve our understanding of RAS- mediated recruitment of AFDN to the plasma membrane, something that was more difficult in the polarized MCF7 cell line where AFDN and SCRIB are characteristically membrane- proximal even in the absence of activated GTPases.  
+
+<|ref|>text<|/ref|><|det|>[[115, 169, 882, 205]]<|/det|>
+1. Anastas, J. N. et al. A protein complex of SCRIB, NOS1AP and VANGL1 regulates cell polarity and migration, and is associated with breast cancer progression. Oncogene 31, 3696-3708 (2012).  
+
+<|ref|>text<|/ref|><|det|>[[113, 240, 883, 784]]<|/det|>
+2) Fig. 4F,G. In describing these micrographs the authors state on p. 9 that "tagged, wild-type KRAS does not significantly alter AFDN or SCRIB localization and does not co-localize with these proteins (Figure 4F and S4B). ... In contrast, expression of KRAS-G12V markedly disrupted cell-cell contacts and was noticeably co-localized with endogenous AFDN and SCRIB, as determined by z-plane projections (Figure 4G)." In the discussion on p. 13 the authors write that "we show that RAS-G12V is co-localized with AFDN and SCRIB at sites of cell contact, while wild-type RAS is distributed more generally across the plasma membrane." The data do not support these conclusions. First, not shown is any disruption of cell-cell contacts in cells expressing KRAS-G12V. Three GFP-KRAS4B-G12V expressing cells are shown with three levels of expression and perhaps different z-planes. Two of these three maintain robust cell-cell contacts as determined by morphology and SCRIB and AFDN staining of areas of cell contact (anti-ZO-1 staining would be a way to look at this without imaging the experimental proteins themselves). Not shown are any of the detached cells to which the authors refer as having lost cell-cell contacts as a function of oncogenic KRAS. Second, and more important, the data show that both WT-KRAS and KRAS-G12V decorate the basolateral membrane. Indeed, both the WT and mutant KRAS decorate the entire plasma membrane (PM) as is well established in a vast literature. As expected for cell adhesion proteins imaged in confluent epithelial cells, SCRIB and AFDN decorate primarily the basolateral membrane. The conclusion that these proteins are colocalized to a greater extent with KRAS-G12V that with WT KRAS is not supported by the data shown and contrary to a vast literature on KRAS localization. Some of the problem is semantic; the concept of co-localization is somewhat ambiguous. There is co-localization on the PM in some regions of the cell but not others. This should not necessarily be interpreted as one protein pulling another to a region of PM since the same pattern would be observed if the localizations are true but unrelated to the direct interactions of the proteins. The three proteins colocalize at the basolateral membrane but not the apical membrane and this is not affected by the GTP-binding state of KRAS. Were this localization of mutant KRAS to differ from that of WT it would be contrary to a vast literature on KRAS localization that in total demonstrates that the subcellular localization driven by the prenylated HVR is not affected by the GTP-binding state. Current paradigms of RAS signaling hold the PM localization of KRAS is constitutive and it is effectors that are drawn to RAS (e.g. translocation of RAF) not vice versa. Are the authors arguing that in this case the converse is true and that the localization of KRAS is driven by that of its effector?  
+
+<|ref|>text<|/ref|><|det|>[[115, 791, 882, 898]]<|/det|>
+It has been difficult to discern specific co- localization between activated KRAS with AFDN in any polarized epithelial cell line we have worked with. There are several reasons for this: 1) both AFDN and KRAS are constitutively membrane localized in these lines independent of each other, as stated by the reviewer; 2) cells expressing activated KRAS detach from adjacent cells; and 3) live cell imaging has proven ineffectual for numerous reasons, primarily due to the unpredictable nature of KRAS- G12V expressing cells but also the requirement for high resolution z- planes and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 882, 387]]<|/det|>
+the tendency of overexpressed AFDN to accumulate in the cytoplasm. To address the reviewers' points and validate that KRAS and AFDN co- localize in cells in a GTP- dependent manner we have taken several approaches. First, we have added numerous images of immunostained AFDN in KRAS- expressing MCF7 cells (the new Supplementary Fig. 5c/d). These images show that AFDN is no longer visible at cell- cell contacts in cells expressing KRAS- G12V, while it is retained at cell contacts in those expressing wild- type KRAS. The images demonstrate that cells expressing KRAS- G12V are no longer associated with adjacent cells, with empty space clearly observable in the surrounding space (i.e. detached). This addresses the reviewers first point above. Secondly, we reconsidered whether polarized epithelial cells with well- defined cell- cell contacts were the most suitable system to demonstrate co- localization between RAS GTPases and AFDN. As discussed above, HeLa cells lack discernable levels of endogenous AFDN, as well as apical- basal polarity and do not form adherens junctions. We exploited this to examine whether RAS could specifically recruit exogenous AFDN to the plasma membrane in a GTP- dependent manner. The new Fig. 5c shows how expression of activated KRAS, and not wild- type KRAS, co- localizes with AFDN at the plasma membrane. Indeed, most of the AFDN pool is recruited from the cytoplasm with a fraction remaining in the perinuclear region. We hope the reviewer agrees this is more robust demonstration that AFDN should be considered a RAS effector.  
+
+<|ref|>text<|/ref|><|det|>[[114, 422, 882, 547]]<|/det|>
+3) Fig. 5D. The altered localization of endogenous SCRIB as a consequence of silencing AFDN is described on p. 10 as "dispersed throughout the cytoplasm." But unlike the distribution of GFP that is homogeneous and clearly cytosolic, that of SCRIB is punctate consistent with a vesicular localization and should be described as such. Co-localization with Texas-red transferrin would allow an assessment as to whether these are endosomes. Caution must be taken in what is used for permeabilization of the cells (here 0.05% Tween-20) as this can alter the appearance of vesicles. It would be wise to also try 0.1% saponin.  
+
+<|ref|>text<|/ref|><|det|>[[114, 554, 882, 730]]<|/det|>
+We agree that the localization of SCRIB in AFDN KO MCF7 cells is more appropriately described as punctate and have changed the description in our revised manuscript. To assess whether the observed pattern is consistent with a localization to endosomes we expressed EGFP- tagged RAB5A, RAB7A or RAB11A in these cells followed by immunostaining for endogenous SCRIB. These report on early, late or recycling endosomes, respectively, and the new Supplementary Fig. 6c reveals that SCRIB does not co- localize with any of these markers. As noted above, others have observed a similar punctate pattern of endogenous SCRIB in different cell lines (and we see this in HeLa cells, which do not express significant levels of AFDN), but this experiment suggests these are not endosomes. Future delineation will hopefully shed light on where SCRIB is localized in the absence of AFDN.  
+
+<|ref|>text<|/ref|><|det|>[[114, 766, 883, 872]]<|/det|>
+4) Despite binding of several RAS family small GTPases, in their colocalization studies the authors restricted their analysis to KRAS4B. This is unfortunate. It would be informative to also study a RAS-related binding partner that is not normally localized exclusively to the PM. RAP2 fits the bill as the authors show strong binding to AFDN and this small GTPase has been localized to endomembrane (PMID: 1923507 and 19061864). It would also be interesting to determine if RHEB interacts with AFDN since this RAS family small GTPase is expressed on lysosomes.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 882, 265]]<|/det|>
+To address this point we have studied co- localization of AFDN with the small GTPases RAP1B and RAP2C (the new Fig. 5c). We found that imaging these proteins in MCF7 cells came with many of the same caveats presented by AFDN- KRAS, and therefore used the HeLa cell line to determine if these protein partners co- localize (as described for KRAS above). RAP1B and RAP2C were the most prominent members of this family precipitated by the AFDN RA domains. Consistent with previous data, we observed RAP1B is distributed prominently throughout the cytoplasm with a fraction at the plasma membrane \(^{1,2}\) , while RAP2C is localized to the plasma membrane and on endomembranes \(^{3,4}\) . In both cases, we observed their complete co- localization with exogenous AFDN. Indeed, RAP2C- G12V stimulated recruitment of AFDN to the membrane in a manner very similar to that observed with KRAS- G12V.  
+
+<|ref|>text<|/ref|><|det|>[[115, 273, 883, 327]]<|/det|>
+1. Wilson, J. M., Prokop, J. W., Lorimer, E., Ntanti, E. & Williams, C. L. Differences in the Phosphorylation-Dependent Regulation of Prenylation of Rap1A and Rap1B. Journal of Molecular Biology 428, 4929-4945 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[115, 336, 882, 373]]<|/det|>
+2. Ntanti, E. et al. An Adenosine-Mediated Signaling Pathway Suppresses Prenylation of the GTPase Rap1B and Promotes Cell Scattering. Sci. Signal. 6, (2013).  
+
+<|ref|>text<|/ref|><|det|>[[115, 380, 882, 434]]<|/det|>
+3. Duncan, E. D., Han, K.-J., Trout, M. A. & Prekeris, R. Ubiquitylation by Rab40b/Cul5 regulates Rap2 localization and activity during cell migration. Journal of Cell Biology 221, e202107114 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 443, 882, 479]]<|/det|>
+4. Meng, Z. et al. RAP2 mediates mechanoresponses of the Hippo pathway. Nature 560, 655-660 (2018).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 517, 224, 534]]<|/det|>
+## Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 543, 882, 614]]<|/det|>
+5) Fig. 2B. This figure sets up indirect immunofluorescence (iIF) staining of endogenous AFDN and SCRIB, which is used extensively throughout the paper. The specificity of antibodies used for iIF must always be validated by knockdown of the protein of interest, which is accomplished in Fig. 5 and this should be added to the legend of Fig. 2B.  
+
+<|ref|>text<|/ref|><|det|>[[115, 623, 460, 641]]<|/det|>
+A statement was added to the figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 678, 882, 766]]<|/det|>
+6) Fig. 4D. The authors write on p. 7 that with an \(n = 5\) they saw an 8-fold enhancement of SCRIB co-IP with AFDN upon expression of KRAS4B-G12V, but they do not report results for RAP1B or RAP2C, which appear to also induce some enhancement, albeit to a lesser extent (KRAS4B>>RAP2C> RAP1B). It would be informative to report the results for each of the interacting GTPases.  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 882, 880]]<|/det|>
+We have added quantitation of SCRIB co- immunoprecipitation with AFDN in the new Fig. 4g. The plot shows SCRIB levels when co- expressed with KRAS- G12V, as well as RAP1B- G12V and RAP2C- G12V. Ratios are in comparison to SCRIB levels co- precipitated when AFDN is expressed with wild- type KRAS. While expression of KRAS- G12V results in an 8- fold increase in SCRIB association with AFDN, the levels are indeed lower for RAP2C (2.6- fold increase) and RAP1B (1.7- fold increase).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 558, 108]]<|/det|>
+7) Fig. S6B. It is very difficult to see the AFDN staining.  
+
+<|ref|>text<|/ref|><|det|>[[115, 117, 875, 137]]<|/det|>
+We have increased intensity of the AFDN signal in these images (now Supplementary Fig. 7b).  
+
+<|ref|>text<|/ref|><|det|>[[115, 172, 882, 225]]<|/det|>
+8) Fig. 6C,D. These \(z\) projections are convincing that true colocalization of AFDN and SCRIB requires PDZ1 and FHA, but would be even more so if they were subjected to analysis with Pearson's coefficient.  
+
+<|ref|>text<|/ref|><|det|>[[115, 234, 882, 392]]<|/det|>
+We have calculated Pearson's coefficients to quantify the levels of endogenous SCRIB colocalized with EGFP- tagged wild- type AFDN or the \(\Delta\) FHA variant (in AFDN KO cells), or endogenous AFDN with EGFP- tagged wild- type SCRIB or its \(\Delta\) PDZ1 variant (in SCRIB KO cells). These calculations were performed on \(n \geq 8\) images similar to those presented in Fig. 7c/d. The results are presented in the new Supplementary Fig. 7c. This quantitation resolved a significant difference in the co- localization of endogenous SCRIB/AFDN with the respective wild- type rescues compared with domain deletions. The analysis has more noise in the AFDN KO cells due to the punctate nature of SCRIB localization, but there is clearly a deficiency in SCRIB colocalization with the AFDN/AFHA rescue compared to wild- type.  
+
+<|ref|>text<|/ref|><|det|>[[115, 428, 882, 481]]<|/det|>
+9) The authors refer to KRAS throughout but they mean KRAS4B. They do not study KRAS4A. Since these splice variants differ only in their HVRs that direct subcellular trafficking this should be acknowledged.  
+
+<|ref|>text<|/ref|><|det|>[[115, 490, 882, 526]]<|/det|>
+We have acknowledged this in the Results section (page 5) and in the Methods section (Plasmid Constructs and Antibodies).  
+
+<|ref|>text<|/ref|><|det|>[[115, 562, 882, 667]]<|/det|>
+10) To be a true effector of a small GTPase, three conditions must be met. The effector must bind directly to the GTPase, the binding must depend on GTP-loading of the GTPase, and the binding must in some way change the conformation or activity of the effector. RAF and HK1 meet all of these criteria but the third has been lacking for AFDN. In Fig. 4D the authors establish for the first time a change in the properties of AFDN induced by KRAS, confirming that AFDN is a bone fide effector. This should be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 677, 882, 800]]<|/det|>
+The point is well taken, and we have reinforced this view in the Discussion (bottom of the first paragraph). It is still not completely clear how RAS activates most effectors, including RAF from a completely mechanistic perspective, though cryo- EM and modelling data are making significant progress with this. It does appear that GTPase binding to AFDN (particularly KRAS) increases its complex with SCRIB, and while only a structure of the full proteins complexed with RAS will reveal the mechanism behind this 'activation', it does appear to satisfy the third condition for being a true effector.  
+
+<|ref|>text<|/ref|><|det|>[[115, 837, 882, 907]]<|/det|>
+11) Because AFDN is unique in possessing tandem RA domains that, in principal, could bind two GTPases the authors have a unique opportunity to ask if either or both are required for the effect of KRAS seen in Fig. 4D. Indeed, it would be interesting to determine if RAS binds to one and RAP2 to the other RA domain.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 882, 265]]<|/det|>
+To address this point we generated bacterial expression constructs to allow purification of AFDN RA1 or RA2 alone. We reveal the specificity of the two individual RA domains for RAS and RAP GTPases, in contrast to the RA1/2 protein, in the new Fig. 4d. The first RA domain of AFDN demonstrated a very similar binding profile to the dual RA1/2 construct, but the levels of GTPase precipitated were considerably lower than with the tandem domains, KRAS in particular. The RA2 binding profile is more restricted, with only KRAS, RAP2B and RAP2C showing significant interaction. Overall, the results suggest that the tandem domains do provide an avidity for most of these GTPases which drives tighter binding than with either domain in isolation. RAP1B seems an exception, whereby binding is generated predominantly through RA1. These data are now described in the Results section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 301, 882, 372]]<|/det|>
+12) The change in kinetics of ERK and AKT activation downstream of EGF signaling upon silencing AFDN or SCRIB shown in Fig. 7D,E is interesting but the authors do not comment on possible mechanisms. Interestingly they parallel the differential effects of NGF versus EGF in PC12 cells where only the former induces sustained ERK activation (PMID 7834738).  
+
+<|ref|>text<|/ref|><|det|>[[115, 380, 882, 643]]<|/det|>
+Indeed, the duration of ERK activation has long been recognized as a foremost feature of signalling through the MAPK pathway, and potentially as a determinant of proliferation vs differentiation outcomes. Interestingly, the sustained activation of MAPK signalling induced by NGF (pointed out by the reviewer) is ostensibly dependent on RAP1<sup>1</sup>, making the AFDN association with RAS and RAP of particular relevance to this phenomenon. As this will be a focus of future work and we do not currently have clear mechanistic data elucidating how loss of AFDN disrupts ERK activation, we have added only a short speculation to the Discussion section (paragraph 4). It seems clear that AFDN association with RAS at the plasma membrane will compete with binding of other effectors to activated RAS, including the RAF kinases, and this is the most likely explanation for the observed signalling defect. It is also probable that competition between activated RAS and RAP GTPases for the AFDN RA domains will further impact effector binding to the individual GTPases themselves. For SCRIB, it is more difficult to reason how loss of expression induces nearly the identical defect as loss of AFDN. It is possible that the RAS-AFDN-SCRIB module is more stable than the RAS-AFDN module alone, and our data supports this. Thus, SCRIB could be an important determinant of competition between AFDN and RAF effectors for activated RAS.  
+
+<|ref|>text<|/ref|><|det|>[[115, 652, 882, 688]]<|/det|>
+1. York, R. D. et al. Rap1 mediates sustained MAP kinase activation induced by nerve growth factor. Nature 392, 622-626 (1998).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 300, 106]]<|/det|>
+REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 147, 393, 163]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 204, 600, 220]]<|/det|>
+The authors significantly approved the quality of their manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 290, 393, 305]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 881, 416]]<|/det|>
+The authors have addressed my comments and those of three other reviewers with new data, a revised manuscript and a well- argued rebuttal. They have done an outstanding job and the revised manuscript is significantly improved. This interesting, comprehensive, and significant paper is suitable for publication in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[114, 456, 880, 602]]<|/det|>
+I have two minor comments not intended to diminish the outstanding work presented but for future analyses should the authors carry this work forward. The authors write in their rebuttal: "To satisfy the reviewers request, we coexpressed EGFP- AFDN and Cherry- SCRIB in these cells (the new Supplementary Fig. 1d), or appropriate controls, to avoid fixation/permeabilization effects." However, the authors did not eliminate fixation/permeabilization effects because rather than image these cells alive, as would have been preferable, they fixed the cells, apparently to allow Hoechst staining. But nuclear localization with Hoechst adds little if anything to the analysis and would be far outweighed by the power or live cell imaging.  
+
+<|ref|>text<|/ref|><|det|>[[114, 641, 880, 749]]<|/det|>
+One other caveat along these lines, the authors refer to the localizations of proteins like GFP alone and GFP- AFADIN in cells not expressing oncogenic KRAS as "cytoplasmic." But the cytoplasm includes both cytosol and all membrane bound organelles (endomembrane). Live cell imaging allows one to clearly visualize the cytosol (organelles appear to be negatively imaged), which is an important descriptive term for molecules like AFADIN that translocate not from endomembrane to plasma membrane but rather from cytosol to plasma membrane.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a29cbea41c68e3e071c4195dab8528dd223479c9c2dfab21918b644a3ec30b56/images_list.json

@@ -0,0 +1,295 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure S13. Venn diagram comparing SG-proximal RNAs from arsenite-stressed HEK293T and U-2 OS cells.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure S14. Box plot comparing the translation efficiencies (left), AU content (middle) and transcript length (right) between HEK-specific, cell type-independent and U-2 OS-specific mRNAs. ns, not significant \\((p > 0.05)\\) ; \\(*p< 0.05\\) ; \\(**p< 0.01\\) ; \\(***p< 0.0001\\) (Mann-Whitney test). Translation efficiencies are counted from a previous study (Sidrauski, et al., Elife, 2015). AU content and transcript length features are referenced from Ensembl website.",
+    "footnote": [],
+    "bbox": [
+      [
+        160,
+        261,
+        816,
+        457
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure S5. Western blot analysis of elF2α phosphorylation (p-elF2α) in G3BP1-miniSOG HEK293T cells after CAP-seq labeling. The intensity of p-elF2α is normalized with respect to \\(\\alpha\\) -Tubulin.",
+    "footnote": [],
+    "bbox": [
+      [
+        344,
+        499,
+        636,
+        768
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure for Reviewers: Immunofluorescence microscopy of U-2 OS cells expressing G3BP1-miniSOG (green). Cells were treated with \\(0.5 \\mathrm{mM}\\) sodium arsenite for \\(60 \\mathrm{min}\\) and labeled by \\(10 \\mathrm{mM}\\) PA under \\(15 \\mathrm{min}\\) blue light illumination. THTPA and BTTAa represents CuAAC reaction with THPTA ( \\(100 \\mu \\mathrm{M}\\) \\(\\mathrm{N}_3\\) -PEG3-biotin, \\(2 \\mathrm{mM}\\) THPTA, \\(0.5 \\mathrm{mM}\\) CuSO4 and \\(5 \\mathrm{mM}\\) sodium ascorbate) or BTTAa ( \\(50 \\mu \\mathrm{M}\\) \\(\\mathrm{N}_3\\) -PEG3-biotin, \\(2 \\mathrm{mM}\\) CuSO4, \\(1 \\mathrm{mM}\\) BTTAa and \\(0.5 \\mathrm{mg / mL}\\) sodium ascorbate), respectively. Biotinylated signals are shown in magenta. Scale bars, \\(10 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        290,
+        85,
+        722,
+        265
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Figure S23. Comparing the proportion of mRNA with different \\(\\mathsf{m}^6\\mathsf{A}\\) density ( \\(\\mathsf{m}^6\\mathsf{A}\\) sites per kilobase) in basal-specific, pre-existing and de novo datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        312,
+        88,
+        694,
+        275
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Figure S26. Comparing the proportion of mRNA with different \\(\\mathsf{m}^6\\mathsf{A}\\) density ( \\(\\mathsf{m}^6\\mathsf{A}\\) sites per kilobase) between 4 clusters.",
+    "footnote": [],
+    "bbox": [
+      [
+        308,
+        355,
+        686,
+        531
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1B.jpg",
+    "caption": "Figure 1B Immunofluorescence microscopy of HEK293T cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Cells were treated with \\(0.5 \\mathrm{mM}\\) sodium arsenite for 60 min and labeled by \\(10 \\mathrm{mM}\\) PA under 15 min blue light illumination. Endogenous SG marker G3BP2 and biotinylated signal are shown in cyan and magenta, respectively. Scale bars, \\(10 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        264,
+        435,
+        730,
+        751
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2A.jpg",
+    "caption": "Figure 2A Immunofluorescence microscopy of U-2 OS cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Cells were treated with 0.5 mM sodium arsenite for 60 min and labeled by 10 mM PA under 15 min blue light illumination. Endogenous SG marker G3BP2 and biotinylated signal are shown in cyan and magenta, respectively. Scale bars, 10 μm.",
+    "footnote": [],
+    "bbox": [
+      [
+        280,
+        95,
+        716,
+        390
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3A.jpg",
+    "caption": "Figure 3A Immunofluorescence microscopy of HEK293T cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Cells were treated with 0.4 M sorbitol for 150 min and labeled by 10 mM PA under 15 min blue light illumination. Endogenous SG marker G3BP2 and biotinylated signal are shown in cyan and magenta, respectively. Scale bars, 10 μm.",
+    "footnote": [],
+    "bbox": [
+      [
+        330,
+        515,
+        663,
+        794
+      ]
+    ],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Figure S3. Co-localization of G3BP1-miniSOG and untargeted-miniSOG with SG. (A) Immunofluorescence microscopy of HEK293T cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Stressed cells were treated with \\(0.5 \\text{mM}\\) sodium arsenite for 60 min. Endogenous SG marker G3BP2 or TIA1 signal are shown in cyan. Scale bars, \\(10 \\mu \\text{m}\\) . (B) Co-localization of G3BP1-miniSOG and untargeted-miniSOG with the SG marker G3BP2. Quantification was performed with a provided script written in MATLAB. ( \\(n = 30\\) from three biological replicates).",
+    "footnote": [],
+    "bbox": [
+      [
+        180,
+        92,
+        828,
+        283
+      ]
+    ],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Figure S1. Western blot detection of G3BP1 overexpression in HEK293T cells stably expressing G3BP1-miniSOG. elF2α was set as loading control.",
+    "footnote": [],
+    "bbox": [
+      [
+        383,
+        484,
+        645,
+        606
+      ]
+    ],
+    "page_idx": 27
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Figure S8. Western blot detection of G3BP1 overexpression in U-2 OS cells stably expressing G3BP1-miniSOG. elF2α was set as loading control.",
+    "footnote": [],
+    "bbox": [
+      [
+        323,
+        672,
+        660,
+        833
+      ]
+    ],
+    "page_idx": 28
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Figure S2. Western blot analysis of elF2α phosphorylation (p- elF2α) in G3BP1-miniSOG HEK293T cells in unstressed (basal), stressed (T0), and recovery (T1 and T3) stages. The intensity of elF2α phosphorylation are normalized with elF2α expression. Error bars represent standard deviation.",
+    "footnote": [],
+    "bbox": [
+      [
+        171,
+        91,
+        825,
+        217
+      ]
+    ],
+    "page_idx": 28
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1E.jpg",
+    "caption": "Figure 1E smFISH of SG-proximal (GAS1, BMS1, APLP2, USP7) and SG-excluded mRNAs (CCNL2, PCBP2). Scale bars, 5 μm. (F) Quantitation of the ratio of smFISH within SG vs. whole cell ( \\(n = 20\\) cells from three biological replicates).",
+    "footnote": [],
+    "bbox": [
+      [
+        370,
+        95,
+        627,
+        485
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2E.jpg",
+    "caption": "Figure 2E smFISH validation of CAP-seq uniquely captured SG-proximal mRNA (SMC1A) and SG-excluded mRNA (SRRM2). Scale bars, \\(5 \\mu \\mathrm{m}\\) . (F) Quantitation of the ratio of smFISH within SG vs. whole cell \\((n = 20\\) cells from three biological replicates).",
+    "footnote": [],
+    "bbox": [
+      [
+        410,
+        95,
+        593,
+        444
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Figure S11. smFISH validation of CAP-seq defined SG-excluded mRNAs (PLXNB2, PKD1, MT-ND4) in arsenite-stressed U-2 OS cells. Scale bars, \\(5 \\mu \\mathrm{m}\\) . Quantitation of the ratio of smFISH within SG vs. whole cell \\((n = 20\\) cells from three biological replicates) is shown on the right.",
+    "footnote": [],
+    "bbox": [
+      [
+        152,
+        548,
+        845,
+        710
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Figure S19. Box plot comparing the translation efficiencies (left) and transcript length (right) in",
+    "footnote": [],
+    "bbox": [
+      [
+        202,
+        604,
+        790,
+        881
+      ]
+    ],
+    "page_idx": 31
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_12.jpg",
+    "caption": "Figure S2. Western blot analysis of elF2α phosphorylation (p- elF2α) in G3BP1-miniSOG HEK293T cells in unstressed (basal), stressed (T0), and recovery (T1 and T3) stages. The intensity of elF2α phosphorylation are normalized with elF2α expression. Error bars represent standard deviation.",
+    "footnote": [],
+    "bbox": [
+      [
+        155,
+        362,
+        857,
+        500
+      ]
+    ],
+    "page_idx": 33
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_13.jpg",
+    "caption": "Figure S23. Comparing the proportion of mRNA with different \\(\\mathsf{m}^6\\mathsf{A}\\) density ( \\(\\mathsf{m}^6\\mathsf{A}\\) sites per kilobase) in basal-specific, pre-existing and de novo datasets.",
+    "footnote": [],
+    "bbox": [
+      [
+        315,
+        258,
+        696,
+        448
+      ]
+    ],
+    "page_idx": 34
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_14.jpg",
+    "caption": "Figure S26. Comparing the proportion of mRNA with different \\(\\mathsf{m}^6\\mathsf{A}\\) density ( \\(\\mathsf{m}^6\\mathsf{A}\\) sites per kilobase) between 4 clusters.",
+    "footnote": [],
+    "bbox": [
+      [
+        305,
+        524,
+        692,
+        704
+      ]
+    ],
+    "page_idx": 37
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a29cbea41c68e3e071c4195dab8528dd223479c9c2dfab21918b644a3ec30b56/supplementary_0_Peer Review File__a29cbea41c68e3e071c4195dab8528dd223479c9c2dfab21918b644a3ec30b56.mmd

@@ -0,0 +1,672 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Profiling stress- triggered RNA condensation with photocatalytic proximity labeling  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## Reviewer #1 (Remarks to the Author):  
+
+Stress granule (SG) is known for buffering proteomic stress, and its defective function is linked to various neuronal disorders. Identifying the definitive composition of SG is crucial for understanding its related physiology and developing therapy for SG- related diseases. Since SG is considered a liquid- liquid condensate, proximity labeling is the preferred method for identifying its contents because conventional purification methods cannot perfectly isolate it without contaminants. Apex, BioID, and TurboID have nicely revealed the protein components of SG. SG also contains various RNA molecules, and it is believed that the specific RNA and protein interaction supports the formation of SG. Therefore, RNA components in SG are of great interest, and various studies have identified this RNA information with fractionation and Co- IP methods. This method has revealed that RNA molecules can be dynamically trapped in SG under various stress. Recently, Dr. Peng Zou's group developed an efficient proximal RNA labeling method such as APEX- Aniline labeling (Zhou et al. Angew Chem Int Ed, 2019) and Cap- seq (Wang et al. Nat Chem. Biol. 2019) to identify SG- localized RNA information, which has lower labeling efficiency than that of proteins. The proximal RNA labeling method, developed by Zou group, is also utilized in the recent work of cell surface- localized RNA mapping by Bertozzi's group (Flynn et al. Cell, 2021). Thus, I believe that this work of identifying SG- localized RNA using Cap- seq by the Zou group will be a valuable resource that can be extensively exploited in the SG biology community after further comparison with previous findings. After these corrections and additions during the revision process, I recommend publishing this work in Nature Communications.  
+
+## Major points:  
+
+1. There are several important points that the authors need to address in their study. Firstly, as there are previously published studies on SG-enriched RNA using various techniques, the authors need to compare their data with these previous findings. For example, previous approaches did not mention enrichment of AU-rich RNA. If these RNA species are not temporal residents of SG, their absence in previous studies raises questions about whether the molecular information provided by these species is reliable.  
+
+2. Secondly, the authors should extensively compare their Cap-seq data with previous SG-RNA profiling studies, such as APEX-seq data (PMID: 31442426). While the labeling mechanism (conjugation to exposed guanosine) is similar to each other, there may be similarities and differences between the two datasets that the authors need to identify.  
+
+3. Thirdly, many SG biologists believe that SG proteome is highly cell-type specific. For example, TDP-43, a well-known SG component protein in neurons, is not found in SG proteomics in Hela cells (PMID: 29395067). The authors should investigate whether the current Cap-seq findings of resident RNA molecules such as AU-rich RNA can be considered general components in other cell lines or cell-type specific findings in Hek293 cells. If the authors conduct additional Cap-seq experiments with other cell lines, they can provide more insight into the generalism of their findings. Utilizing neural-originated cell lines for another round of Cap-seq experiment can make this work more valuable as many known SG
+
+<--- Page Split --->
+
+related diseases are neurological disorders, such as ALS.  
+
+Minor points:  
+
+1. The authors should clarify why they used 10mM PA and whether the increased concentration is required due to the Core-Shell structure of SG, which might hinder probe penetration. The authors should also measure or mention any toxicity or stress caused by probe treatment in the Discussion section.  
+
+2. Cu-click protocol used in the immunofluorescence and RNA isolation experiments appear to be slightly different (e.g., copper conc. THPTA vs BTTAA), and the authors should provide a brief comment explaining this experimental difference.  
+
+3. The authors need to address the limitation of their analysis method, specifically the possibility of indirect identification of RNA. If this indirect issue is present in this study, it needs to be clearly mentioned in the Discussion section.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The manuscript, "Profiling stress- triggered RNA condensation with photocatalytic proximity labeling" by Ren et al., contributes a novel approach to profile the transcripts that are enriched in stress granules called CAP- seq. The CAP- seq approach employs a miniSOG fusion protein to label and subsequently purify and identify the RNAs that are in proximity to G3BP1- miniSOG in stressed and unstressed conditions. The study demonstrates that there are similarities and differences in the RNAs that are differentially enriched in stress granules depending on cell type, stress type, and time over a stress time- course experiment. The primary novelty of this study is in defining the G3BP1- associated transcriptome during the recovery from stress, which is an important contribution to the field. However, major weaknesses in the data presented and the interpretation of the data must be addressed for this manuscript to be considered for publication. For this reason, I do not recommend publication of this manuscript in Nature Communications. Specific areas to be addressed are listed below.  
+
+Major comments:  
+
+1. The observations that RNA length, AU rich elements, and m6A elements are associated with RNA enrichment in stress granules are not novel and have already been shown by several groups and published in articles including Khong et al., Mol Cell 2017; Namkoong et al., Mol Cell 2018; Van Treeck et al., PNAS 2018; Ries et al., 2019 Nature; and Matheny et al., Mol Cell Biol 2019. Therefore, these observations should be used to bolster the validity of the novel CAP-seq approach rather than portrayed as novel findings throughout the manuscript.  
+
+2. Several key manuscripts are not discussed or included in citations in the introduction of the
+
+<--- Page Split --->
+
+manuscript. These include sources listed in comment 1, in addition to the study by Khong et al., Nat Commun 2022 which describes a limited effect of m6A on mRNA recruitment to stress granules. Additionally, Moon et al., Nat Cell Biol 2019 demonstrated that translating mRNAs can transiently colocalize with stress granules, which should be included with the Mateju et al., 2020 Cell citation.  
+
+3. G3BP1 is an RNA binding protein, therefore it is expected that mRNAs that bind to G3BP1 will be present in the unstressed and stressed G3BP1-miniSOG datasets. The statement, "It appears that a subpopulation of mRNAs remain proximal to G3BP1 in the cytoplasm despite apparent SG disassembly" implies that these detected RNAs are within stress granule cores, but no experimental evidence is presented to support that statement.  
+
+4. The m6A enrichment data shown in Figure 4H and 5G must be normalized to total transcript length to be meaningful.  
+
+5. The details of the CAP-seq approach must be clarified to indicate the limitations of the assay. The following should be addressed to give confidence that the method is resulting in specific labeling of RNAs in stress granules:  
+
+a. The distance between the miniSOG protein and the RNA target that allows labeling should be stated. 
+b. The degree of G3BP1-miniSOG localization to stress granules and the degree of the control miniSOG in stress granules must be quantified and reported, especially considering that miniSOG is present in stress granules. 
+c. The definition of 'post-versus pre-enrichment of RNA" should be given. Is this equivalent to an 'input' sample in an RNA-protein co-immunoprecipitation experiment? 
+d. The rationale for pooling the three negative controls (post-versus pre-enrichment, PA-omitted, and untargeted miniSOG) should be clarified and results should be shown for each individual treatment as they are controlling for different aspects of the approach. 
+e. It is unclear whether this same approach of pooling all three negative control datasets is used for the unstressed HEK293T cell and the recovery from stress datasets.  
+
+6. Poor data quality throughout the manuscript makes the study results difficult to interpret. In particular:  
+
+a. Unstressed cells must be shown in Figure 1B, Figure 2A, Figure 3A, Figure S2, 
+b. Statistical analyses must be done for all Venn diagrams to enable a meaningful interpretation to be made (Figure 1D, Figure 2C, Figure 2D, Figure 3C, Figure 3D, Figure S5, Figure S12, Figure S16; Figure S19, Figure S20) 
+c. Loading controls and results from experimental replicates must be included for all western blots (Figure S1, Figure S6, Figure S18) 
+d. Some smFISH images do not appear to have detectable RNA and/or stress granules in the cytoplasm, making them uninterpretable (Figure 1E CCNL2, Figure 2E SRRM2, Figure S8 PKD1) 
+e. The results of independent experimental replicates for the smFISH data (e.g., Figure 1F) should be reported rather than the number of individual cells quantified.
+
+<--- Page Split --->
+
+7. The stress granule transcriptome generated by Khong et al. 2017 Mol Cell is treated throughout the manuscript as if all the RNAs within it are enriched in stress granules, however, this dataset contains all detectable RNAs that are in stress granules at varying degrees. This is an important distinction because CAP-seq results throughout the paper are not generally showing different results from previously published data in Khong et al., 2017.  
+
+8. Several statements in the text are misrepresenting the presented data:  
+
+The statement "No significant differences in TE and transcript length were found among mRNAs in stress-independent, sorbitol-specific and arsenite-specific datasets" is not supported by the data shown in Figure S14, which shows significant differences in TE and transcript length in the arsenite-specific dataset.  
+
+The statement "At 3 hr post-stress, the elF2a phosphorylation appears fully restored to its basal level" is not supported by the data in Figure S18 which shows P- elF2 is 2- 3x higher at 3 hr compared to the basal unstressed condition.  
+
+Minor comments:  
+
+References to Parker et al., and Jeffrey (Jaffrey) et al., should be replaced with the citation in proper format.  
+
+The observation that mitochondrial gene encoded RNAs are depleted from stress granules using CAP- seq is not novel as it is stated in Khong et al., Mol Cell 2017 that mitochondrial RNAs present in their transcriptome dataset may be contaminants introduced during purification.  
+
+The source of the data used to create translation efficiency and transcript length plots throughout should be defined in the figure legends.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In this manuscript, Ren and colleagues use the RNA proximity labeling approach CAP- seq to identify and quantify transcripts localized to stress granules (SG). They use this technique to profile how the SG transcriptome changes in response to different stresses (arsenite vs. sorbitol) and how its RNA content dynamically changes upon SG disassembly. Generally, the conclusions in the manuscript are supported by the data. I have only a few comments that may improve the manuscript.  
+
+## MAJOR COMMENTS  
+
+1. Again and again, the authors find that SG-proximal RNAs are (1) long, (2) AU-rich, and (3) poorly
+
+<--- Page Split --->
+
+translated. They ascribe function to all of these characteristics and note that they may be involved in localizing RNAs to SG. However, across the transcriptome, all 3 of these characteristics are correlated with each other. That is, RNAs that are long also tend to be AU- rich, and vice versa (Marin et al Yeast 2003; Lopez et al Frontiers in Genetics 2021, and others). It could be then that just one of these characteristics is important and the others come along for the ride through these correlations but are not actually functional themselves in terms of getting RNAs to SGs. This point should at least be noted.  
+
+2. Related to point 1, I particularly disagree with the statement that "RNA binding proteins generally prefer AU-rich motifs". Yes, many RBPs do bind AU-rich sequences, but many also do not. Characterizing the entirety of the RNA binding proteome in this way is not accurate.  
+
+3. Any time that membership in a group is tested across experiments (e.g. Venn diagrams comparing SG-enriched RNAs across stresses or cell types, etc.), p-values should be calculated for the overlap. Without some kind of test, it's difficult for me to know whether or not the overlap observed is more or less than expected. This is important as observing a greater overlap than expected would lend additional confidence to the results.  
+
+## MINOR COMMENTS  
+
+1. Where did the translational efficiency data come from? I may have missed it, but it was not obvious to me. This should be made clearer.  
+
+2. "Transcription factors binding to AU-rich elements [have] been found in SG." Are you sure you mean transcription factors? Those don't normally bind RNA.  
+
+3. "However, the purification procedure is prone to contamination and loss of weakly associated material, thus causing high false positive rate." The loss of material that in truth really is associated with SG would result in false negatives, not false positives.
+
+<--- Page Split --->
+
+We thank all three reviewers for their thorough and thoughtful comments to help us improve this manuscript. In the revised manuscript, we have provided additional experimental data on the evaluation of miniSOG expression levels and labeling- induced toxicity. We have also provided additional statistical analysis of the colocalization between G3BP1- miniSOG and stress granules and compared our CAP- seq datasets between HEK293T and U- 2 OS cells to highlight cell- type specific RNA enrichment. As suggested by the reviewers, we have added discussions to compare our findings with previous reports, especially regarding the RNA features associated with SG enrichment. Please see our point- by- point responses below. In the revised manuscript, these changes have been marked in red.  
+
+Reviewer #1 (Remarks to the Author):  
+
+Stress granule (SG) is known for buffering proteomic stress, and its defective function is linked to various neuronal disorders. Identifying the definitive composition of SG is crucial for understanding its related physiology and developing therapy for SG- related diseases. Since SG is considered a liquid- liquid condensate, proximity labeling is the preferred method for identifying its contents because conventional purification methods cannot perfectly isolate it without contaminants. Apex, BiolD, and TurboID have nicely revealed the protein components of SG. SG also contains various RNA molecules, and it is believed that the specific RNA and protein interaction supports the formation of SG. Therefore, RNA components in SG are of great interest, and various studies have identified this RNA information with fractionation and Co- IP methods. This method has revealed that RNA molecules can be dynamically trapped in SG under various stress. Recently, Dr. Peng Zou's group developed an efficient proximal RNA labeling method such as APEX- Aniline labeling (Zhou et al. Angew Chem Int Ed, 2019) and Cap- seq (Wang et al. Nat Chem. Biol. 2019) to identify SG- localized RNA information, which has lower labeling efficiency than that of proteins. The proximal RNA labeling method, developed by Zou group, is also utilized in the recent work of cell surface- localized RNA mapping by Bertozzi's group (Flynn et al. Cell, 2021). Thus, I believe that this work of identifying SG- localized RNA using Cap- seq by the Zou group will be a valuable resource that can be extensively exploited in the SG biology community after further comparison with previous findings. After these corrections and additions during the revision process, I recommend publishing this work in Nature Communications.  
+
+## Major points:
+
+<--- Page Split --->
+
+1. There are several important points that the authors need to address in their study. Firstly, as there are previously published studies on SG-enriched RNA using various techniques, the authors need to compare their data with these previous findings. For example, previous approaches did not mention enrichment of AU-rich RNA. If these RNA species are not temporal residents of SG, their absence in previous studies raises questions about whether the molecular information provided by these species.  
+
+Response: We thank the reviewer for the advice on comparing our data with previous findings regarding the features of SG transcriptome. In the previous submission, we only compared out CAP-seq dataset with the \(\mathsf{SG}_{\mathsf{coreRNA}}\) dataset derived from arsenite stressed U- 2OS (by Parker and co- workers). In the revised manuscript, we have provided additional comparisons with several other published SG- enriched RNA datasets, as suggested by both reviewers #1 and #2.  
+
+In the work by Namkoong et al. and co- workers (PMID: 29576526), RNA granules were purified from NIH- 3T3 cells stressed with thapsigargin, heat shock, or sodium arsenite. The authors found that these stress- induced RNA granules tend to enrich transcripts with longer length and more AU- elements. Notably, 181 out of 457 transcripts enriched in our CAP- seq dataset are also enriched in their arsenite treated NIH- 3T3 dataset (Clusters 1 and 2 in Figure 7C in their paper, 2920 transcripts in total), and both our studies identified the same trend of length distributions in SG- enriched RNAs. However, the purification- based methods may have lower SG- specificity, as the purified RNA granules might contain components from other cytosolic membrane- less organelle such as processing bodies.  
+
+The correlation between RNA length and SG- enrichment have also been investigated by Van Treeck et al. (PMID: 29483269) and Matheny et al. (PMID: 31591142). Van Treeck et al. analyzed the components of liquid- liquid phase separation (LLPS) droplets formed by yeast RNAs in vitro. They found similar RNA components in these droplets as those in the SGs in yeast, and droplet- enriched RNAs contained longer transcripts than droplet- depleted RNAs. Matheny et al. analyzed the RNA components of purified RNA granules from arsenite- treated U- 2 OS cells, where they found impaired translation efficiency and longer transcript length.  
+
+More recently, Kleiner and co- workers applied proximity- dependent RNA editing technique (TRIBE- ID) to profile G3BP1- associated RNAs in HEK293T cells challenged with sodium arsenite stress (PMID: 37349582). Their data revealed that G3BP1- associated RNAs are positively correlated with their transcript length and negatively correlated with translation efficiency, which is consistent with our findings. Among the 745 G3BP1- associated RNAs identified in their dataset, 106 were also enriched in our CAP- seq SG RNA datasets derived from arsenite stressed HEK293T.  
+
+All of the above references have been added to the revised manuscripts:
+
+<--- Page Split --->
+
+Page 9, lines 12 - 13: Previous studies have revealed a negative correlation between mRNA association with SG and translational efficiency (TE) (NamKoong, et al., Mol. Cell, 2017; Khong, et al., Mol. Cell, 2017; Padron, et al., Mol. Cell, 2019; Matheny, et al., Mol. Cell Biol., 2019)  
+
+Page 9, lines 15 - 19: We also observed longer transcript length and higher AU content in SG- proximal versus SG- excluded mRNAs (Figure S9, Table S2B), which follows a similar trend as previous reports that \(\mathrm{SG}_{\mathrm{coreRNA}}\) contained less GC content. (Khong, et al., Mol. Cell, 2017). The above analysis shows that data acquired with CAP- seq are generally in agreement with prior knowledge of SG transcriptome.  
+
+Page 19, line 24 - page 20, line 8: "For example, mRNAs SRRM2, PLXNB2, PKD1 and MT- ND4, were previously enriched from purified SG (Khong, et al., Mol. Cell, 2017) but were found to be SG- excluded by CAP- seq. Our quantitative smFISH imaging analysis shows less than \(10\%\) co- localization of these RNAs with SG, thus supporting the CAP- seq dataset. We also compared our SG- proximal datasets with RNAs enriched in granules isolated from arsenite- stressed NIH- 3T3 cells (NamKoong, et al., Mol. Cell, 2017), revealing an overlap of 181 transcripts. Notably, both studies identified a positive correlation between transcript length and SG- enrichment. Such correlation has also been reported by previous research on liquid- liquid phase separation of yeast RNAs in vitro (Van Treeck, et al., Proc. Natl. Acad. Sci. U.S.A., 2018) and RNA granules purified from arsenite- stressed U- 2 OS cells (Matheny, et al., Mol. Cell Biol., 2019)."  
+
+Page 20, lines 19 - 25: "More recently, Kleiner and co- workers applied proximity- dependent RNA editing technique (TRIBE- ID) to profile G3BP1- associated RNAs in HEK293T cells challenged with sodium arsenite stress (Seo & Kleiner, Nat. Chem. Biol., 2023). Among the 745 G3BP1- associated RNAs identified in their dataset, 106 were enriched in our CAP- seq SG RNA datasets derived from arsenite stressed HEK293T (p value \(< 4 \times 10^{- 100}\) ). Notably, their data revealed that G3BP1- associated RNAs are positively correlated with their transcript length and negatively correlated with translation efficiency, which is consistent with our findings."  
+
+Regarding our statement on AU- rich RNAs, it is noteworthy that our definition of AU content is essentially the opposite to the definition of GC content in previously published reports, i.e. \(\mathrm{AU\%} = 100\% - \mathrm{GC\%}\) . Thus, AU- rich means low GC content. The correlation between SG enrichment and GC content has been reported in previous studies. For example, Khong et al. (PMID: 29129640) purified the SG cores from arsenite- stressed U- 2OS cells and found less GC content in the RNA components. Therefore, our finding of AU- rich RNAs in SG- proximal transcriptome does not contradict with previous reports. We have clarified this point in the Methods section. In
+
+<--- Page Split --->
+
+the revised manuscript, we have added a comment on AU- rich and GC- poor in the main text:  
+
+Page 9, lines 15 - 18: We also observed longer transcript length and higher AU content in SG- proximal versus SG- excluded mRNAs (Figure S9, Table S2B), which follows a similar trend as previous reports that SGcoreRNA contained less GC content. (Khong, et al., Mol. Cell, 2017)  
+
+2. Secondly, the authors should extensively compare their Cap-seq data with previous SG-RNA profiling studies, such as APEX-seq data (PMID: 31442426). While the labeling mechanism (conjugation to exposed guanosine) is similar to each other, there may be similarities and differences between the two datasets that the authors need to identify.  
+
+Response: We agree with the reviewer that it would be informative to compare our CAP-seq dataset with APEX-seq, especially when considering their similar labeling mechanisms. In the work by Padron, et al. (PMID 31442426), as recommended by the reviewer), the translation factor eIF4A was used as the bait to fuse with APEX2, and SG formation was induced by heat-shock in HEK293T cells. Unfortunately, the authors did not provide a list of enriched SG-proximal transcripts in the published paper, so we started with re-analyzing their raw RNA-seq data of enriched and input samples. By applying the same cutoff as in our analysis (log2FC > 0,3 and padj < 0.05), differential expression analysis of the APEX-seq dataset reveals 394 transcripts as significant enriched. Among these, only 18 transcripts overlap with our CAP-seq dataset of 457 transcripts. The lack of overlap may be attributed to the differences in our choices of baits (G3BP1 vs. eIF4A), stress conditions (sodium arsenite vs. heat shock), and duration of stress (1 hr vs. 20 min). Nevertheless, both our current study and the APEX-seq study identified a strong correlation between SG-proximal RNA labeling and transcript length, suggesting that longer RNAs are more likely localized to the SG. In the revised manuscript, we have added the above comparison and discussions:  
+
+Page 20, lines 9 - 18: "We also compared our CAP-seq dataset with a previously reported SG transcriptome profiling with APEX-seq, in which the translation factor eIF4A was used as the bait and SG formation was induced by heat- shock for 20 min in HEK293T cells (Padron, et al., Mol. Cell, 2019). By applying the same cutoff as in our analysis (log2FC > 0.3 and padj < 0.05), differential expression analysis of the APEX-seq dataset reveals 394 transcripts as significant enriched. Among these, only 18 transcripts overlap with our CAP-seq dataset of 457 transcripts, which may be attributed to the differences in the choices of baits, stress conditions, and durations of stress. Nevertheless, both datasets identified a strong correlation between SG- proximal RNA labeling and transcript length, suggesting that longer RNAs are more
+
+<--- Page Split --->
+
+likely localized to the SG."  
+
+3. Thirdly, many SG biologists believe that SG proteome is highly cell-type specific. For example, TDP-43, a well-known SG component protein in neurons, is not found in SG proteomics in Hela cells (PMID: 29395067). The authors should investigate whether the current Cap-seq findings of resident RNA molecules such as AU-rich RNA can be considered general components in other cell lines or cell-type specific findings in Hek293 cells. If the authors conduct additional Cap-seq experiments with other cell lines, they can provide more insight into the generalism of their findings. Utilizing neural-originated cell lines for another round of Cap-seq experiment can make this work more valuable as many known SG-related diseases are neurological disorders, such as ALS.  
+
+Response: We thank the reviewer for the advice on comparing CAP- seq findings in different cell lines. As the reviewer pointed out, SG proteome is highly cell- type specific. Previous proteomic profiling experiments identified 123 and 411 SG proteins in arsenite- stressed HEK293T (Markmiller et al., Cell, 2018; PMID: 29373831) and U- 2 OS cells (Jain et al., Cell, 2016; PMID: 26777405), respectively. Among these, only 82 proteins were enriched in both datasets. It is therefore expected that the SG transcriptome might also be cell- type specific. In our previous submission, we have performed CAP- seq in both HEK293T and U- 2 OS cells but did not compare these datasets. In the revised manuscript, we have provided additional data analysis by comparing these datasets in a head- to- head manner. Such comparisons are particularly meaningful as they employ the same labeling method with the same bait protein. Our CAP- seq analysis identified a total of 457 and 1135 transcripts from arsenite- stressed HEK293T and U- 2 OS cells, respectively. Among these, 261 transcripts are shared by both datasets (i.e. cell type- independent). The remaining 196 transcripts in the HEK293T SG CAP- seq dataset and 874 transcripts in the U- 2OS SG CAP- seq dataset are defined as HEK293T- specific and U- 2OS specific, respectively.  
+
+We also compared the RNA features of each category. Overall, we did not observe substantial differences in the translation efficiency, AU content, and transcript length. U- 2 OS- specific CAP- seq RNAs tend to have higher AU content and shorter CDS length. The trend is quite modest yet statistically significant. These additional results are presented in Figure S13 - S14 in the revised manuscript.
+
+<--- Page Split --->
+![](images/Figure_unknown_1.jpg)
+
+<center>Figure S13. Venn diagram comparing SG-proximal RNAs from arsenite-stressed HEK293T and U-2 OS cells. </center>  
+
+![](images/Figure_unknown_2.jpg)
+
+<center>Figure S14. Box plot comparing the translation efficiencies (left), AU content (middle) and transcript length (right) between HEK-specific, cell type-independent and U-2 OS-specific mRNAs. ns, not significant \((p > 0.05)\) ; \(*p< 0.05\) ; \(**p< 0.01\) ; \(***p< 0.0001\) (Mann-Whitney test). Translation efficiencies are counted from a previous study (Sidrauski, et al., Elife, 2015). AU content and transcript length features are referenced from Ensembl website. </center>  
+
+We thus conclude that while the SG RNA compositions may vary substantially in different cell lines, presumably due to differences in the SG protein components that are responsible for recruiting these RNAs, the overall SG RNA features are quite similar, suggesting a conserved mechanism of RNA sorting into the SG. In the revised manuscript, we have added the following comparisons  
+
+Page 9, line 21 - page 10, line 5: "Notably, the molecular composition of SG is known to be cell- type specific. For example, TDP- 43, the well- known SG protein marker in neuron, is not found in the SG proteome in Hela cells. In addition, proteomic profiling experiments identified 123 and 411 SG proteins in arsenite- stressed HEK293T and U- 2 OS cells, by APEX proximity labeling or purification of SG, respectively. Among these, only 82 proteins were enriched in both datasets. It is therefore expected that the SG transcriptome might also be cell- type specific. We compared the 457 and 1135 SG- proximal transcripts from arsenite- stressed HEK293T and U- 2 OS cells, respectively. Among these, 261 transcripts are shared by both datasets (i.e. cell type- independent). The remaining 196 transcripts in the HEK293T
+
+<--- Page Split --->
+
+SG CAP- seq dataset and 874 transcripts in the U- 2 OS SG CAP- seq dataset are defined as HEK293T- specific and U- 2 OS- specific, respectively (Figure S13).  
+
+We compared the RNA features of cell type- specific and cell type- independent categories, but did not observe substantial differences in the translation efficiency, AU content, and transcript length. U- 2 OS- specific CAP- seq RNAs tend to have higher AU content and shorter CDS length (Figure S14). The trend is quite modest yet statistically significant. The above comparisons indicate that while the SG RNA compositions may vary substantially across different cell lines, presumably due to differences in the SG protein components that are responsible for recruiting these RNAs, the overall SG RNA features are quite similar, suggesting a conserved mechanism of RNA sorting into the SG."  
+
+We also thank the reviewer for suggesting CAP- seq experiments in neural- originated cells. Projects of SG CAP- seq in cultured cortical neurons are already underway in our laboratory. Due to difficulties associated with culturing primary cells and labeling in these delicate samples, these projects involve engineering a better version of CAP- seq labeling with directed evolution, which, in our opinion, are beyond the scope of the current study. In the revised manuscript, we have provided additional discussions on the potential limitation of the current CAP- seq method and future directions that aims to improve its efficiency and compatibility with delicate samples such as primary neuron culture.  
+
+## Minor points:  
+
+1. The authors should clarify why they used 10mM PA and whether the increased concentration is required due to the Core-Shell structure of SG, which might hinder probe penetration. The authors should also measure or mention any toxicity or stress caused by probe treatment in the Discussion section.  
+
+Response: The choice of 10 mM of PA is based on our previously published CAP- seq method (Wang et al. Nat Chem Biol, 2019), which showed that incubating cells with 10 mM PA for 5 min provided sufficient labeling efficiency in multiple subcellular compartments, including endoplasmic reticulum membrane, outer mitochondrial membrane, and mitochondrial matrix. Notably, CAP- seq labeling in the mitochondrial matrix is particularly challenging, as the small molecule probe PA has to rapidly penetrate the barriers of both outer and inner mitochondrial membranes. Thus, we reasoned that incubating cells with 10 mM PA for 5 min should allow the probe to permeate through other subcellular structures, including LLPS condensates. In the
+
+<--- Page Split --->
+
+revised manuscript, we have added the following comments to clarify our choice of 10 mM PA and 5 min incubation time:  
+
+Page 4 lines 11 - 15: "For RNA labeling, our previous CAP- seq experiments showed that \(10~\mathrm{mM}\) PA was capable of permeating throughout the cytoplasm within 5 min (Wang, et al., Nat. Chem. Biol., 2019). Since the core- shell structure of SG might hinder probe penetration, we pre- incubated cells with \(10~\mathrm{mM}\) PA for 5 min and subsequently illuminated the sample with \(24~\mathrm{mW / cm^2}\) blue LED for 15 min (Figure S4)."  
+
+The reviewer has raised an important issue regarding the potential cytotoxicity of our method. In the previous submission, we have provided confocal fluorescence images of unstressed G3BP1- miniSOG or untargeted- miniSOG cells. We did not observe any visible cytoplasmic puncta following CAP- seq labeling, suggesting that the toxicity of CAP- seq treatment was quite minimal as compared to sodium arsenite treatment. However, we agree with the reviewer that we should evaluate the cytotoxicity effect more carefully. In the revised manuscript, we used elF2α phosphorylation as a marker for cellular stress and measured its levels by Western blotting of whole cell lysate following CAP- seq labeling (i.e. pre- incubating cells with \(10~\mathrm{mM}\) PA for 5 min at \(37^{\circ}C\) , before blue LED illumination at room temperature for 15 min at \(24~\mathrm{mM / cm^2}\) , same as our cellular labeling conditions).  
+
+![](images/Figure_unknown_3.jpg)
+
+<center>Figure S5. Western blot analysis of elF2α phosphorylation (p-elF2α) in G3BP1-miniSOG HEK293T cells after CAP-seq labeling. The intensity of p-elF2α is normalized with respect to \(\alpha\) -Tubulin. </center>  
+
+As shown in revised Figure S5, Western blot analysis reveals that CAP- seq labeling causes the level of elF2α phosphorylation to increase by \(\sim 70\%\) from the basal
+
+<--- Page Split --->
+
+level. Treating the cells with blue LED illumination alone or incubating cells with \(10 \text{mM}\) PA alone causes the elF2α phosphorylation levels to increase by \(\sim 25\%\) and \(\sim 60\%\) , respectively. We reasoned that, as these increases are modest as compared to arsenite- induced stress (elevating elF2α phosphorylation by \(200\%\) ), the threshold of SG assembly has not been reached. Our data analysis workflow requires comparing the enrichment levels of RNAs labeled in G3BP1- miniSOG cells versus untargeted- miniSOG cells, which offers a means of normalizing the slight cellular toxicity effect, as both cell samples were treated by CAP- seq in the identical manner. The observation of PA- induced toxicity also prompts us to reduce the incubation time in the future via improvement in CAP- seq labeling efficiency, which is underway in our lab through coordinated efforts of probe design/synthesis and directed evolution of the photocatalyst.  
+
+In the revised manuscript, we have acknowledged the cytotoxicity issue in the current study, discussed the necessity of using appropriate control samples to normalize the cytotoxicity effect, and future directions that aim to mitigate the toxicity via more efficient photocatalysis.  
+
+Page 4, lines 15 - 20: "We measured the level of elF2α phosphorylation as a marker for translation inhibition. Western blot analysis reveals that CAP- seq labeling elevates elF2α phosphorylation by \(\sim 70\%\) from the basal level. Treating cells with blue LED illumination alone or incubation with \(10 \text{mM}\) PA alone causes the elF2α phosphorylation levels to increase by \(\sim 25\%\) and \(\sim 60\%\) , respectively (Figure S5). These changes are modest as compared to arsenite- induced stress (Figure S2).  
+
+2. Cu-click protocol used in the immunofluorescence and RNA isolation experiments appear to be slightly different (e.g., copper conc. THPTA vs BTTAA), and the authors should provide a brief comment explaining this experimental difference.  
+
+Response: Thanks for pointing out this issue. In the revised manuscript, we have provided additional experiment data comparing THPTA and BTTAA ligands for their effect on the CuAAC efficiency in fixed cells. Fluorescence imaging reveals similar levels of biotinylation by the two ligands (see figure below). Our choice of THPTA ligand is based on a previous report that CuAAC with THPTA achieves higher reaction yield than BTTAA in the aqueous solution (PMID: 29636419). The CuAAC protocol used in this study is also the same as our previous work of CAP- seq (PMID: 31591565). In the revised manuscript, we have explained the reason of choosing THPTA as the ligand:  
+
+Page 4, lines 26 - 27: "We chose THPTA as the ligand for CuAAC because it offers higher reaction yield with RNA in the aqueous solution (Huang, et al., Proc. Natl. Acad. Sci. U.S.A., 2018)."
+
+<--- Page Split --->
+![](images/Figure_unknown_4.jpg)
+
+<center>Figure for Reviewers: Immunofluorescence microscopy of U-2 OS cells expressing G3BP1-miniSOG (green). Cells were treated with \(0.5 \mathrm{mM}\) sodium arsenite for \(60 \mathrm{min}\) and labeled by \(10 \mathrm{mM}\) PA under \(15 \mathrm{min}\) blue light illumination. THTPA and BTTAa represents CuAAC reaction with THPTA ( \(100 \mu \mathrm{M}\) \(\mathrm{N}_3\) -PEG3-biotin, \(2 \mathrm{mM}\) THPTA, \(0.5 \mathrm{mM}\) CuSO4 and \(5 \mathrm{mM}\) sodium ascorbate) or BTTAa ( \(50 \mu \mathrm{M}\) \(\mathrm{N}_3\) -PEG3-biotin, \(2 \mathrm{mM}\) CuSO4, \(1 \mathrm{mM}\) BTTAa and \(0.5 \mathrm{mg / mL}\) sodium ascorbate), respectively. Biotinylated signals are shown in magenta. Scale bars, \(10 \mu \mathrm{m}\) . </center>  
+
+3. The authors need to address the limitation of their analysis method, specifically the possibility of indirect identification of RNA. If this indirect issue is present in this study, it needs to be clearly mentioned in the Discussion section.  
+
+Response: We agree with the reviewer that, for the benefit of the readers and future user of CAP-seq, the limitations of our work should be discussed in more detail. These include the slight cytotoxicity effect, limited temporal resolution, and potential bias towards exposed RNAs. First, as we have discussed in our response to Minor Point #1, prolonged treatment of PA causes slight cellular stress. In HEK293T and U- 2 OS cells, this level of stress has not reached the threshold of inducing SG assembly and could be normalized by using appropriate controls. However, the mild toxicity of PA treatment may impede the application of CAP-seq to more sensitive and delicate cells such as primary neuron cultures.  
+
+Second, the current version of CAP- seq requires blue LED illumination for \(15 \mathrm{min}\) to achieve efficient RNA labeling. This temporal resolution is sufficient for mapping the SG transcriptome during its disassembly stage, which occurs on the time scale of hours. Yet, the low temporal resolution is incompatible for resolving SG components during its rapid assembly stage, which typically matures within \(15 - 30 \mathrm{min}\) (Wheeler, et al., Elife, 2016).  
+
+Third, proximity labeling method is potentially biased towards prey proteins/RNAs that are more solvent accessible, such that the reactive singlet oxygen could reach the exposed targets (tyrosine, lysine, guanosine nucleobase, histidine, etc.) for efficient labeling. Notably, this bias is shared by APEX2, BioID/TurboID, CAP- seq, RinID, etc.
+
+<--- Page Split --->
+
+In the revised manuscript, we have added the following discussions:  
+
+Page 23, lines 18 - 31: "Finally, this study is limited by mild cytotoxicity, insufficient temporal resolution, and potential bias towards exposed RNAs. First, prolonged treatment of PA causes cellular stress, as revealed by elevated eIF2α phosphorylation. Although the level of stress has not reached the threshold of inducing SG assembly and could be normalized by using appropriate controls, it might still impede the application of CAP- seq to more sensitive and delicate cells such as primary neuron cultures. Second, CAP- seq labeling is potentially biased towards prey RNAs that are more solvent exposed, such that the reactive singlet oxygen could access guanosine nucleobase for successful labeling. Notably, this bias is shared by other proximity labeling methods, such as APEX2 (Fazal, et al., Cell, 2019), BioID (Roux, et al., J. Cell Biol, 2012)/TurboID (Branon, et al., Nat. Biotechnol., 2018), and RinID (Zheng, et al., Nat. Commun., 2023). Finally, CAP- seq requires blue LED illumination for 15 min to achieve efficient RNA labeling. This temporal resolution is still insufficient for resolving SG components during its rapid assembly stage, which typically matures within 15 - 30 min. In the future, more effective tools for photosensitive catalysis could be developed by directional evolution."  
+
+Regarding the issue of indirect identification of RNA, as pointed out by the reviewer, we reason that the labeling radius of CAP- seq is the key. While the exact radius has not been measured in the context of live cells, the half- life and diffusion radius of singlet oxygen were estimated to be \(0.6 \mu \mathrm{s}\) and \(70 \mathrm{nm}\) , respectively (Moan, J. Photoch. Photobio., 1990). Since the half- life of singlet oxygen depends critically on the local environment and considering the quenching reaction from cellular metabolites such as thiols and amines, the CAP- seq labeling is likely quite restricted. In our previous CAP- seq profiling at the surface of endoplasmic reticulum membrane, \(96.2\%\) of the captured mRNAs encode for the secretory pathway proteins, thus suggesting a spatial resolution on the scale of the size of a ribosome, i.e. \(20 \mathrm{nm}\) . As the size of SGs is in the range of \(100 \mathrm{nm} - 1 \mu \mathrm{m}\) , a resolution on the scale of tens of nanometers is sufficient for mapping SG components.  
+
+Nevertheless, the labeling radius of CAP- seq is still much larger than the size of a typical protein (e.g. 3 - 5 nm). Thus, CAP- seq is indeed capable of labeling RNAs that are indirectly interacting with the bait. This scenario is shared with other proximity labeling approaches and is quite different from direct capture methods such as chemical- or photo- crosslinking (e.g. CLIP, PAR- CLIP). In the revised manuscript, we have discussed the differences in methodology.  
+
+Page 3, lines 10 - 17: "Due to the limited half- life and diffusion distance of singlet oxygen (0.6 \(\mu \mathrm{s}\) and \(70 \mathrm{nm}\) in cells) (Moan, J. Photoch. Photobio., 1990), CAP- seq have been applied in profiling the subcellular transcriptome in the mitochondria and
+
+<--- Page Split --->
+
+endoplasmic reticulum with high spatial resolution in live cells. As the size of SGs is in the range of \(100 \text{nm} - 1 \mu \text{m}\) (Wolozin & Ivanov, Nat. Rev. Neurosci., 2019), a resolution on the scale of tens of nanometers is sufficient for mapping SG components. Notably, CAP- seq is capable of labeling RNAs that are indirectly interacting with the bait, which could complement direct capture methods such as chemical- or photo- crosslinking (e.g. CLIP, PAR- CLIP).  
+
+Reviewer #2 (Remarks to the Author):  
+
+The manuscript, "Profiling stress- triggered RNA condensation with photocatalytic proximity labeling" by Ren et al., contributes a novel approach to profile the transcripts that are enriched in stress granules called CAP- seq. The CAP- seq approach employs a miniSOG fusion protein to label and subsequently purify and identify the RNAs that are in proximity to G3BP1- miniSOG in stressed and unstressed conditions. The study demonstrates that there are similarities and differences in the RNAs that are differentially enriched in stress granules depending on cell type, stress type, and time over a stress time- course experiment. The primary novelty of this study is in defining the G3BP1- associated transcriptome during the recovery from stress, which is an important contribution to the field. However, major weaknesses in the data presented and the interpretation of the data must be addressed for this manuscript to be considered for publication. For this reason, I do not recommend publication of this manuscript in Nature Communications. Specific areas to be addressed are listed below.  
+
+## Major comments:  
+
+1. The observations that RNA length, AU rich elements, and m6A elements are associated with RNA enrichment in stress granules are not novel and have already been shown by several groups and published in articles including Khong et al., Mol Cell 2017; Namkoong et al., Mol Cell 2018; Van Treeck et al., PNAS 2018; Ries et al., 2019 Nature; and Matheny et al., Mol Cell Biol 2019. Therefore, these observations should be used to bolster the validity of the novel CAP-seq approach rather than portrayed as novel findings throughout the manuscript.  
+
+Response: We thank the reviewer for pointing out this issue. Indeed, our observed strong correlations between transcript length, AU content, and m6A levels with SG RNA enrichment are overall consistent with previous reports. In the previous submission, we have briefly mentioned this point in the Introduction and provided three references (refs 7- 9), but we agree with the reviewer that we should add more references to better
+
+<--- Page Split --->
+
+place our findings in the context of existing literature. In the revised manuscript, we have referenced the papers recommended by the reviewer and compared our data with these previous reports, as follows:  
+
+In the work by Namkoong et al. (PMID: 29576526), RNA granules were purified from NIH- 3T3 cells stressed with thapsigargin, heat shock, or sodium arsenite. The authors found that these stress- induced RNA granules tend to enrich transcripts with longer length and more AU- elements. Notably, 181 out of 457 transcripts enriched in our CAP- seq dataset are also enriched in their arsenite treated NIH- 3T3 dataset (Clusters 1 and 2 in Figure 7C in their paper, 2920 transcripts in total), and both our studies identified the same trend of length distributions in SG- enriched RNAs. However, the purification- based methods may have lower SG- specificity, as the purified RNA granules might contain components from other cytosolic membrane- less organelle such as processing bodies.  
+
+The correlation between RNA length and SG- enrichment have also been investigated by Van Treeck et al. (PMID: 29483269) and Matheny et al. (PMID: 31591142). Van Treeck et al. analyzed the components of liquid- liquid phase separation (LLPS) droplets formed by yeast RNAs in vitro. They found similar RNA components in these droplets as those in the SGs in yeast, and droplet- enriched RNAs contained longer transcripts than droplet- depleted RNAs. Matheny et al. analyzed the RNA components of purified RNA granules from arsenite- treated U- 2 OS cells, where they found impaired translation efficiency and longer transcript length.  
+
+Regarding RNA methylations, Ries et al. (PMID: 31292544) found that RNA methylation \(\mathsf{m}^6\mathsf{A}\) could induce LLPS condensate formation between \(\mathsf{m}^6\mathsf{A}\) - mRNAs and their binding proteins. By comparing their RNA \(\mathsf{m}^6\mathsf{A}\) dataset with a previously reported list of RNA components derived from purified granules (Namkoong et al. PMID: 29576526), they discovered that SG RNAs contain more \(\mathsf{m}^6\mathsf{A}\) modification.  
+
+More recently, Kleiner and co- workers applied proximity- dependent RNA editing technique (TRIBE- ID) to profile G3BP1- associated RNAs in HEK293T cells challenged with sodium arsenite stress. Their data revealed that G3BP1- associated RNAs are positively correlated with their transcript length and negatively correlated with translation efficiency, which is consistent with our findings. Among the 745 G3BP1- associated RNAs identified in their dataset, 106 were also enriched in our CAP- seq SG RNA datasets derived from arsenite stressed HEK293T.  
+
+All of the above references have been added to the revised manuscripts.  
+
+Page 9, lines 12 - 13: "Previous studies have revealed a negative correlation between mRNA association with SG and translational efficiency (TE) (NamKoong, et al., Mol. Cell, 2017; Khong, et al., Mol. Cell, 2017; Padron, et al., Mol. Cell, 2019; Matheny, et al., Mol. Cell Biol., 2019)"
+
+<--- Page Split --->
+
+Page 9, lines 15 - 20: "We also observed longer transcript length and higher AU content in SG-proximal versus SG- excluded mRNAs (Figure S9, Table S2B), which follows a similar trend as previous reports that \(\mathsf{S G}_{\mathsf{C o r e R N A}}\) contained less GC content. (Khong, et al., Mol. Cell, 2017). The above analysis shows that data acquired with CAP- seq are generally in agreement with prior knowledge of SG transcriptome."  
+
+Page 19, line 24 - page 20, line 8: "For example, mRNAs SRRM2, PLXNB2, PKD1 and MT- ND4, were previously enriched from purified SG (Khong, et al., Mol. Cell, 2017) but were found to be SG- excluded by CAP- seq. Our quantitative smFISH imaging analysis shows less than \(10\%\) co- localization of these RNAs with SG, thus supporting the CAP- seq dataset. We also compared our SG- proximal datasets with RNAs enriched in granules isolated from arsenite- stressed NIH- 3T3 cells (NamKoong, et al., Mol. Cell, 2017), revealing an overlap of 181 transcripts. Notably, both studies identified a positive correlation between transcript length and SG- enrichment. Such correlation has also been reported by previous research on liquid- liquid phase separation of yeast RNAs in vitro (Van Treeck, et al., Proc. Natl. Acad. Sci. U.S.A., 2018) and RNA granules purified from arsenite- stressed U- 2 OS cells (Matheny, et al., Mol. Cell Biol., 2019)."  
+
+2. Several key manuscripts are not discussed or included in citations in the introduction of the manuscript. These include sources listed in comment 1, in addition to the study by Khong et al., Nat Commun 2022 which describes a limited effect of m6A on mRNA recruitment to stress granules. Additionally, Moon et al., Nat Cell Biol 2019 demonstrated that translating mRNAs can transiently colocalize with stress granules, which should be included with the Mateju et al., 2020 Cell citation.  
+
+Response: We agree with the reviewer that these key references should be included in the manuscript. In the work by Khong et al., an RNA \(\mathsf{m}^6\mathsf{A}\) methylation- deficient mouse embryonic stem cell line was created to compare the level of SG partitioning between \(\mathsf{m}^6\mathsf{A}\) - modified RNAs and \(\mathsf{m}^6\mathsf{A}\) - deficient RNAs. Fluorescence microscopy analysis reveals a similar level of SG enrichment in RNA methylation- deficient cells and wild- type cells, leading the authors to conclude that transcript lengths, rather than \(\mathsf{m}^6\mathsf{A}\) levels, are correlated with SG enrichment. In the revised manuscript, we have cited this reference in the Introduction and added the following discussions:  
+
+Page 2, lines 12 - 17: "For example, \(\mathsf{m}^6\mathsf{A}\) modification is found to be enriched in granules isolated from arsenite- stressed NIH- 3T3 cells (NamKoong, et al., Mol. Cell, 2017; Ries, Nature, 2019). However, fluorescence microscopy analysis reveals a similar level of SG enrichment in RNA methylation- deficient mouse embryonic stem cells versus wild- type cells, leading to the conclusion that transcript lengths, rather than \(\mathsf{m}^6\mathsf{A}\) levels, are correlated with SG enrichment (Khong, et al., Nat. Commun., 2022)."
+
+<--- Page Split --->
+
+Moon et al. applied single- molecule tracking on reporter mRNAs to uncover colocalization with and bi- directional trafficking between stress granules and processing bodies. They demonstrated that translating mRNAs can transiently co- localize with stress granules, whereas non- translating mRNAs form stable mRNA granule structures. Notably, the KDM5B mRNA used in this study was also enriched in our SG CAP- seq datasets in both HEK293T and U- 2 OS cells. As suggested by the reviewer, we have cited this work along with the Mateju et al., 2020 Cell citation in the Introduction. The following comment on KDM5B is added to the revised manuscript:  
+
+Page 5, lines 13 - 16: "As a demonstration of the high spatial specificity of CAP- seq, known SG- localized transcripts, such as lncRNA NORAD, mRNA DYNCH1H1, and mRNA KDM5B (Khong, et al., Mol. Cell, 2017; Moon, et al., Nat. Cell Biol., 2019), were enriched in our SG dataset, whereas mitochondrial RNAs were significantly depleted."  
+
+Page 9, lines 1 - 3: "Among 1135 CAP- seq enriched RNAs, 296 (26%) have also been identified as SG transcripts in the previous study, including AHNAK, NORAD, PEG3, CDK6, and KDM5B (Khong, et al., Mol. Cell, 2017; Moon, et al., Nat. Cell Biol., 2019) were also found enrichment in the datasets."  
+
+3. G3BP1 is an RNA binding protein, therefore it is expected that mRNAs that bind to G3BP1 will be present in the unstressed and stressed G3BP1-miniSOG datasets. The statement, "It appears that a sub-population of mRNAs remain proximal to G3BP1 in the cytoplasm despite apparent SG disassembly" implies that these detected RNAs are within stress granule cores, but no experimental evidence is presented to support that statement.  
+
+Response: We agree with the reviewer that since G3BP1 is an RNA binding protein, it is expected to interact with mRNAs in both stressed and unstressed states. Indeed, previous live cell proximity- dependent protein labeling experiments have portrayed a pre- existing protein- protein interaction network formed by SG core components in the absence of stress (PMID: 29373831, PMID: 29395067). The concept of SG cores has been further implicated in live cell imaging of the SG disassembly process, which reveals the non- uniform SG disintegration into smaller foci before undergoing further clearance (PMID: 27602576). Given the above literature findings about SG cores, our observation that a sub- population of mRNAs appears to remain proximal to G3BP1 may likely arise from interaction with SG core components, although we do not have experimental data to directly support this speculation. Future super- resolution imaging experiments may provide further evidence for resolving this issue.  
+
+In the revised manuscript, we have added the following discussion to clarify the meaning of "proximity to G3BP1":
+
+<--- Page Split --->
+
+Page 22, lines 1 - 5: "Proximity- labeling experiments have also portrayed an extensive protein- protein interaction network formed by SG core components even in the absence of stress (Youn, et al., Mol. Cell, 2018; Markmiller, et al., Cell, 2018). Since G3BP1 is an RNA binding protein, it is expected to interact with mRNAs in both stressed and unstressed states."  
+
+Page 22, lines 10 - 11: "Live cell imaging of the SG disassembly has revealed the non- uniform SG disintegration into smaller foci before undergoing further clearance (Wheeler, et al., Elife, 2016)."  
+
+Page 22, lines 22 - 27: "While SGs are typically invisible in most cells after 3 hr of recovery, it remains unknown whether the RNA interaction network has been reset to the basal state. Our observation that a sub- population of mRNAs appears to remain proximal to G3BP1 may likely arise from interactions with SG core components, although we do not have experimental data to directly support this speculation. Future super- resolution imaging experiments may provide further evidence for resolving this issue."  
+
+4. The m6A enrichment data shown in Figure 4H and 5G must be normalized to total transcript length to be meaningful.  
+
+Response: We thank the reviewer for the advice. In the revised manuscript, we have re- analyzed our CAP- seq datasets by normalizing the number of \(\mathsf{m}^6\mathsf{A}\) sites with total transcript length (kb) (i.e. number of sites per kilobase). A similar trend was observed as our previous analysis using \(\mathsf{m}^6\mathsf{A}\) sites alone, revealing that the density of \(\mathsf{m}^6\mathsf{A}\) is higher in pre- existing G3BP1- proximal RNAs (Figure S23). During the disassembly stage, the density of \(\mathsf{m}^6\mathsf{A}\) is also the highest in Cluster 2 (RNAs that dissociate more slowly during the recovery phase), which is consistent with our previous observation using \(\mathsf{m}^6\mathsf{A}\) sites alone (Figure S26). In the revised manuscript, we have provided these figures in the supplementary file and added the following descriptions in the main text:  
+
+Page 14, line 30: "We thus analyzed both the number of \(\mathsf{m}^6\mathsf{A}\) sites and \(\mathsf{m}^6\mathsf{A}\) site density (number per kilobase) in our datasets, using published \(\mathsf{m}^6\mathsf{A}\) database in HEK293T cells."  
+
+Page 17, line 34: "We therefore compared the extent of \(\mathsf{m}^6\mathsf{A}\) sites and \(\mathsf{m}^6\mathsf{A}\) site density (number per kilobase) in our dataset."
+
+<--- Page Split --->
+![](images/Figure_unknown_5.jpg)
+
+<center>Figure S23. Comparing the proportion of mRNA with different \(\mathsf{m}^6\mathsf{A}\) density ( \(\mathsf{m}^6\mathsf{A}\) sites per kilobase) in basal-specific, pre-existing and de novo datasets. </center>  
+
+![](images/Figure_1B.jpg)
+
+<center>Figure S26. Comparing the proportion of mRNA with different \(\mathsf{m}^6\mathsf{A}\) density ( \(\mathsf{m}^6\mathsf{A}\) sites per kilobase) between 4 clusters. </center>  
+
+5. The details of the CAP-seq approach must be clarified to indicate the limitations of the assay. The following should be addressed to give confidence that the method is resulting in specific labeling of RNAs in stress granules: a. The distance between the miniSOG protein and the RNA target that allows labeling should be stated.  
+
+Response: We thank the reviewer for pointing out the issue of labeling radius. While the exact radius has not been measured in the context of live cells, the half- life and diffusion radius of singlet oxygen were estimated to be \(0.6 \mu \mathrm{s}\) and \(70 \mathrm{nm}\) , respectively (Moan, J. Photoch. Photobio., 1990). Since the half- life of singlet oxygen depends critically on the local environment and considering the quenching reaction from cellular metabolites such as thiols and amines, the CAP- seq labeling is likely quite restricted. In our previous CAP- seq profiling at the surface of endoplasmic reticulum membrane, \(96.2\%\) of the captured mRNAs encode for the secretory pathway proteins, thus
+
+<--- Page Split --->
+
+suggesting a spatial resolution on the scale of the size of a ribosome, i.e. \(20 \text{nm}\) . As the size of SGs is in the range of \(100 \text{nm} - 1 \mu \text{m}\) , a resolution on the scale of tens of nanometers is sufficient for mapping SG components. In the revised manuscript, we have added the above information of labeling radius in the Introduction:  
+
+Page 3, lines 10 - 17: "Due to the limited half- life and diffusion distance of singlet oxygen (0.6 \(\mu \text{s}\) and \(70 \text{nm}\) in cells) (Moan, J. Photoch. Photobio., 1990), CAP- seq have been applied in profiling the subcellular transcriptome in the mitochondria and endoplasmic reticulum with high spatial resolution in live cells."  
+
+b. The degree of G3BP1-miniSOG localization to stress granules and the degree of the control miniSOG in stress granules must be quantified and reported, especially considering that miniSOG is present in stress granules.  
+
+Response: In the revised manuscript, we performed additional immunofluorescence microscopy experiments to quantify the percentage of G3BP1- miniSOG that co- localizes with the SG marker, G3BP2, in both HEK293T and U- 2 OS cells. In cells challenged with arsenic stress, \(57 \pm 7.5\%\) and \(46\% \pm 2.4\%\) of G3BP1- miniSOG overlaps with SGs in HEK293T and U- 2 OS cells, respectively. In contrast, the degree of co- localization with SG is significantly lower for untargeted- miniSOG in these cells (HEK293T: \(21 \pm 3.2\%\) ; U- 2 OS: \(9.8\% \pm 0.4\%\) ). Notably, these values are quite similar to the ratios of RNA enrichment (SG vs. whole cell) in our smFISH experiments (Figure 1F). In the revised manuscript, we have added the above data in Supplementary Figures S2 and S9. We have also added the following information in the main text:  
+
+Page 4, lines 7 - 10: "Immunofluorescence imaging confirmed the co- localization of G3BP1- miniSOG ( \(57 \pm 7.5\%\) , mean \(\pm\) s.d., \(n = 30\) cells) with the SG marker, G3BP2, whereas untargeted miniSOG remains dispersed throughout the cell ( \(21 \pm 3.2\%\) , \(n = 30\) cells) (Figure S3)."  
+
+Page 7, lines 10 - 15: "Immunofluorescence imaging showed good co- localization between miniSOG and SG marker G3BP2, and \(46\% \pm 2.4\%\) ( \(n = 30\) ) of the G3BP1- miniSOG fusion protein were localized to SG (Figure S9). In contrast, only \(9.8\% \pm 0.4\%\) of untargeted miniSOG protein ( \(n = 30\) ) overlaps with SGs (Figure S9). Following RNA labeling, biotinylation signal is highly co- localized with both miniSOG and the SG marker, G3BP2 (Figure 2A)."  
+
+c. The definition of 'post-versus pre-enrichment of RNA' should be given. Is this equivalent to an 'input' sample in an RNA-protein co-immunoprecipitation experiment?  
+
+Response: Yes, the "pre- enrichment" sample is the same as "input" sample used in an RNA- protein co- IP experiment. The pre- enrichment RNAs refers to the total RNAs
+
+<--- Page Split --->
+
+before loading onto streptavidin- coated beads for affinity purification. The post- enrichment RNAs refer to those after streptavidin beads purification. We have clarified this statement in the manuscript:  
+
+Page 5, lines 2 - 4: "1) post- versus pre- enrichment of RNA labeled with G3BP1- miniSOG (post- enrichment: RNAs eluted from streptavidin- coated beads; pre- enrichment: total RNAs before loading onto streptavidin- coated beads);"  
+
+d. The rationale for pooling the three negative controls (post-versus pre-enrichment, PA-omitted, and untargeted miniSOG) should be clarified and results should be shown for each individual treatment as they are controlling for different aspects of the approach.  
+
+Response: The rationale of using three negative controls is to provide a stringent list of SG-proximal RNAs that we hope to exclude false positives as much as possible. These controls are necessary to remove non-specific background adsorption to streptavidin-coated beads (by omitting PA as the control) and to remove background RNA labeling outside the SGs (by using untargeted miniSOG as the control). In the revised manuscript, we have provided the results from each control in Figures S4, S7, S11, S15, and S19 (Figure S6, Figure S10, Figure S16, Figure S20 in the latest manuscript), and emphasized the rationale of using these control samples in the main text:  
+
+Page 4, lines 30 - 33: "In addition, as G3BP1 was only partially localized to SG under sodium arsenite stress (Wheeler, et al., Methods, 2017) (Fiuqre S3), we used HEK293T cells expressing untargeted miniSOG as another control to subtract background RNA labeling in the cytoplasm."  
+
+e. It is unclear whether this same approach of pooling all three negative control datasets is used for the unstressed HEK293T cell and the recovery from stress datasets.  
+
+Response: Yes, we applied the same criteria to determine the G3BP1- proximal RNA dataset throughout this study, i.e. for stressed, unstressed, and recovery experiments. For example, Figure 4C shows the Venn diagram of DESeq analysis against all three negative controls. In the revised manuscript, we have clarified this issue by adding the following information to the main text.  
+
+Page 13, lines 31 - 36: "Following the same DESeq2 work flow and cutoff values (log2FC > 0.3 and padj < 0.05) as in the previous analysis of stressed cells, CAP- seq identifies 3302, 2136, and 1490 transcripts in post- vs. pre- enrichment, labeling vs. omitting PA, and G3BP1- miniSOG vs. untargeted miniSOG analyses, respectively.
+
+<--- Page Split --->
+
+The overlap of these three datasets yields 731 RNAs, which we define as G3BP1- proximal transcripts under the basal condition (Figure 4C, Figure S20, Table S1C)."  
+
+Page 17, lines 4 - 9: "Following the identical RNA labeling protocol (10 mM PA and 15 min blue LED illumination), DESeq2 work flow (post- vs. pre- enrichment, labeling vs. omitting PA, and G3BP1- miniSOG vs. untargeted miniSOG) and cutoff values \((\log_2\mathrm{FC} > 0.3\) and \(\mathsf{p}_{\mathsf{adj}}< 0.05)\) as in the previous analyses of stressed and unstressed cells, a total of 533 and 428 G3BP1- proximal transcripts were captured at T1 and T3, respectively (Figures S24, Table S1D - E)"  
+
+6. Poor data quality throughout the manuscript makes the study results difficult to interpret. In particular:  
+
+a. Unstressed cells must be shown in Figure 1B, Figure 2A, Figure 3A, Figure S2  
+
+Response: We thank the reviewer for this suggestion. We have added fluorescence images of unstressed cells in Figures 1B, 2A, 3A, and S2, as requested by the reviewer.  
+
+![](images/Figure_2A.jpg)
+
+<center>Figure 1B Immunofluorescence microscopy of HEK293T cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Cells were treated with \(0.5 \mathrm{mM}\) sodium arsenite for 60 min and labeled by \(10 \mathrm{mM}\) PA under 15 min blue light illumination. Endogenous SG marker G3BP2 and biotinylated signal are shown in cyan and magenta, respectively. Scale bars, \(10 \mu \mathrm{m}\) . </center>
+
+<--- Page Split --->
+![](images/Figure_3A.jpg)
+
+<center>Figure 2A Immunofluorescence microscopy of U-2 OS cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Cells were treated with 0.5 mM sodium arsenite for 60 min and labeled by 10 mM PA under 15 min blue light illumination. Endogenous SG marker G3BP2 and biotinylated signal are shown in cyan and magenta, respectively. Scale bars, 10 μm. </center>  
+
+![](images/Figure_unknown_6.jpg)
+
+<center>Figure 3A Immunofluorescence microscopy of HEK293T cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Cells were treated with 0.4 M sorbitol for 150 min and labeled by 10 mM PA under 15 min blue light illumination. Endogenous SG marker G3BP2 and biotinylated signal are shown in cyan and magenta, respectively. Scale bars, 10 μm. </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_7.jpg)
+
+<center>Figure S3. Co-localization of G3BP1-miniSOG and untargeted-miniSOG with SG. (A) Immunofluorescence microscopy of HEK293T cells expressing G3BP1-miniSOG and untargeted miniSOG (green). Stressed cells were treated with \(0.5 \text{mM}\) sodium arsenite for 60 min. Endogenous SG marker G3BP2 or TIA1 signal are shown in cyan. Scale bars, \(10 \mu \text{m}\) . (B) Co-localization of G3BP1-miniSOG and untargeted-miniSOG with the SG marker G3BP2. Quantification was performed with a provided script written in MATLAB. ( \(n = 30\) from three biological replicates). </center>  
+
+b. Statistical analyses must be done for all Venn diagrams to enable a meaningful interpretation to be made (Figure 1D, Figure 2C, Figure 2D, Figure 3C, Figure 3D, Figure S5, Figure S12, Figure S16; Figure S19, Figure S20)  
+
+Response: We thank the reviewer for this suggestion. In the revised manuscript, we applied hypergeometric test to calculate the relevant p-values of Venn diagrams. Figure 2D compares our CAP-seq SG-proximal and SG-excluded RNAs from arsenite-stressed U-2 OS with the previously published SG RNA dataset, the p-values of these comparisons are \(2 \times 10^{- 183}\) and 0.0001, respectively. Figure 3D compares CAP-seq datasets for arsenite stress and sorbitol stress. The p-value of this comparison is \(2 \times 10^{- 150}\) . Figure S20 compares CAP-seq defined SG-proximal dataset under T0 and G3BP1-proximal datasets under T1 and T3. The p values of comparisons between T0 vs. T1, T0 vs. T3 and T1 vs. T3 are \(9 \times 10^{- 321}\) , \(1 \times 10^{- 150}\) and \(5 \times 10^{- 266}\) , respectively. We also added p value of other Venn diagrams that are not mentioned here. Figure 4D compares the CAP-seq enriched RNAs under unstressed, arsenite stress and sorbitol stress, the p value of the pre-existing datasets is \(4 \times 10^{- 246}\) . Figure 6A compares RNAs enriched at basal and T3 time point, the p-value of this comparison is \(5 \times 10^{- 135}\) . The p values have been attached to the mentioned Venn diagrams.  
+
+We note that the Venn diagrams shown in Figures 1D, 2C, S5, S12, S16, and S19 are used for defining our CAP- seq datasets of G3BP1- proximal or G3BP1- excluded transcripts. In these Venn diagrams, each circle represents a unique control
+
+<--- Page Split --->
+
+experiment. The overlap provides a stringent list of RNA with high spatial specificity, rather than representing a comparison of datasets. Thus, we did not calculate the p-values of these Venn diagrams.  
+
+c. Loading controls and results from experimental replicates must be included for all western blots (Figure S1, Figure S6, Figure S18)  
+
+Response: In the revised manuscript, we have added the loading controls to the Western blots. We performed biological triplicated experiments ( \(n = 3\) ) for each blot and reported the mean \(\pm\) standard error of image quantitation in the main text. We also added error bars to revised Figure S18 (Figure S2 in the latest supplementary information).  
+
+Page 3, lines 33 - 34: "The expression level of G3BP1- miniSOG was \(65\% \pm 4.9\%\) (mean \(\pm\) s.d.) that of the endogenous G3BP1 level, as measured by western blot (Figure S1)"  
+
+Page 7, lines 8 - 9: "Western blot analysis revealed that the expression level of G3BP1- miniSOG was approximately \(63\% \pm 2\%\) of the endogenous G3BP1 (Figure S8)."  
+
+![](images/Figure_unknown_8.jpg)
+
+<center>Figure S1. Western blot detection of G3BP1 overexpression in HEK293T cells stably expressing G3BP1-miniSOG. elF2α was set as loading control. </center>  
+
+![](images/Figure_unknown_9.jpg)
+
+<center>Figure S8. Western blot detection of G3BP1 overexpression in U-2 OS cells stably expressing G3BP1-miniSOG. elF2α was set as loading control. </center>
+
+<--- Page Split --->
+![](images/Figure_1E.jpg)
+
+<center>Figure S2. Western blot analysis of elF2α phosphorylation (p- elF2α) in G3BP1-miniSOG HEK293T cells in unstressed (basal), stressed (T0), and recovery (T1 and T3) stages. The intensity of elF2α phosphorylation are normalized with elF2α expression. Error bars represent standard deviation. </center>  
+
+d. Some smFISH images do not appear to have detectable RNA and/or stress granules in the cytoplasm, making them uninterpretable (Figure 1E CCNL2, Figure 2E SRRM2, Figure S8 PKD1)  
+
+Response: Since CCNL2, SRRM2 and PKD1 are all candidates from our SG- excluded CAP- seq datasets, it is expected that little or no co- localization exists between their smFISH signal and SGs. Figures 1E, 2E, and S8 (S11 in the latest manuscript) are consistent with this expectation. In the revised manuscript, we have emphasized in the figure captions that these transcripts are from SG- excluded datasets.
+
+<--- Page Split --->
+![](images/Figure_2E.jpg)
+
+<center>Figure 1E smFISH of SG-proximal (GAS1, BMS1, APLP2, USP7) and SG-excluded mRNAs (CCNL2, PCBP2). Scale bars, 5 μm. (F) Quantitation of the ratio of smFISH within SG vs. whole cell ( \(n = 20\) cells from three biological replicates). </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_10.jpg)
+
+<center>Figure 2E smFISH validation of CAP-seq uniquely captured SG-proximal mRNA (SMC1A) and SG-excluded mRNA (SRRM2). Scale bars, \(5 \mu \mathrm{m}\) . (F) Quantitation of the ratio of smFISH within SG vs. whole cell \((n = 20\) cells from three biological replicates). </center>  
+
+![](images/Figure_unknown_11.jpg)
+
+<center>Figure S11. smFISH validation of CAP-seq defined SG-excluded mRNAs (PLXNB2, PKD1, MT-ND4) in arsenite-stressed U-2 OS cells. Scale bars, \(5 \mu \mathrm{m}\) . Quantitation of the ratio of smFISH within SG vs. whole cell \((n = 20\) cells from three biological replicates) is shown on the right. </center>  
+
+e. The results of independent experimental replicates for the smFISH data (e.g., Figure 1F) should be reported rather than the number of individual cells quantified.
+
+<--- Page Split --->
+
+Response: We thank the reviewer for the advice. The smFISH data in Figure 1F are from three biological replicates with 20 cells in total. In the revised figure caption, we have provided information on the biological replicates:  
+
+Figure 1F: "Quantitation of the ratio of smFISH within SG vs. whole cell \((n = 20\) cells from three biological replicates)."  
+
+Figure 2F: "Quantitation of the ratio of smFISH within SG vs. whole cell \((n = 20\) cells from three biological replicates)."  
+
+Figure 3F: "Quantitation of the ratio of smFISH within SG vs. whole cell \((n = 20\) cells from three biological replicates)."  
+
+Figure S11: "smFISH validation of CAP- seq defined SG- excluded mRNAs (PLXNB2, PKD1, MT- ND4) in arsenite- stressed U- 2 OS cells. Scale bars, \(5 \mu \mathrm{m}\) . Quantitation of the ratio of smFISH within SG vs. whole cell \((n = 20\) cells from three biological replicates) is shown on the right."  
+
+Figure S18: "Quantitation of the level of smFISH targets (MAD2L1, PRDX3, MORF4L2) enrichment in SGs \((n = 20\) cells from three biological replicates)."  
+
+7. The stress granule transcriptome generated by Khong et al. 2017 Mol Cell is treated throughout the manuscript as if all the RNAs within it are enriched in stress granules, however, this dataset contains all detectable RNAs that are in stress granules at varying degrees. This is an important distinction because CAP-seq results throughout the paper are not generally showing different results from previously published data in Khong et al., 2017.  
+
+Response: We agree with the reviewer that, in terms of RNA features that correlate with SG localization, our datasets are overall consistent with previously published data in Khong et al., 2017. The advantage of CAP- seq lies in its high spatiotemporal resolution which enables the detection of G3BP1- proximal transcripts in unstressed state or in the disassembly stage. In the revised manuscript, we have added the following discussion:  
+
+Page 19, line 24 - page 20, line 3: "For example, mRNAs SRRM2, PLXNB2, PKD1 and MT- ND4, were previously enriched from purified SG (Khong, et al., Mol. Cell, 2017) but were found to be SG- excluded by CAP- seq. Our quantitative smFISH imaging analysis shows less than \(10\%\) co- localization of these RNAs with SG, thus supporting the CAP- seq dataset."  
+
+In the Khong et al., 2017 paper, the SGcoreRNA dataset is defined by comparing FPKM values of RNAs from purified SG core against those of the total RNAs. Among the list of 1841 SGcoreRNA, however, 8 mitochondrial mRNAs (MT- mRNAs) appear enriched, with Fold Change (SG core vs. total RNA) \(>2.5\) . Notably, MT- ND5 is ranked
+
+<--- Page Split --->
+
+as the top 5 of SG enriched RNAs. These results are surprising as MT- mRNAs are transcribed from the mitochondrial genome and strictly localized within the mitochondrial matrix. Thus, they should be considered as contaminants in the \(\mathrm{SG}_{\mathrm{coreRNA}}\) dataset.  
+
+Our CAP- seq datasets were defined by the overlap of three differential analysis, which provides a more stringent list of SG- enriched RNAs. We compared our CAP- seq datasets with the published list, performed FISH imaging of target RNAs SRRM2, PLXNB2, PKD1 and MT- ND4, which were found to be SG- excluded by CAP- seq but defined by Khong et al as SG- enriched. Quantitative analysis of FISH imaging revealed less than \(10\%\) co- localization with SG of these RNAs, thus demonstrating the high spatial specificity of CAP- seq.  
+
+8. Several statements in the text are misrepresenting the presented data: The statement "No significant differences in TE and transcript length were found among mRNAs in stress-independent, sorbitol-specific and arsenite-specific datasets" is not supported by the data shown in Figure S14, which shows significant differences in TE and transcript length in the arsenite-specific dataset.  
+
+Response: We have double- checked the data in Figure S14 (Figure S19 in the latest manuscript) and confirm that the three datasets (stress- independent, sorbitol- specific, and arsenite- specific) do not show statistical differences in the TE and transcript length (p values \(>0.05\) , Mann- Whitney test). The data used for statistical analysis are listed in Table S3. The statistical differences shown in the Figure are between total mRNAs and some of the above datasets, rather than among the three datasets.  
+
+![](images/Figure_unknown_12.jpg)
+
+<center>Figure S19. Box plot comparing the translation efficiencies (left) and transcript length (right) in </center>
+
+<--- Page Split --->
+
+arsenite- specific, stress- independent, sorbitol- specific SG- proximal mRNAs in HEK293T cells. Translation efficiencies are counted from a previous study (Sidrauski, et al., Elife, 2015). AU content and transcript length features are referenced from Ensembl website. ns, not significant \((p > 0.05)\) ; \(^{*}p < 0.05\) ; \(^{**}p < 0.01\) ; \(^{***}p < 0.001\) ; \(^{****}p < 0.0001\) (Mann- Whitney test).  
+
+9. The statement "At 3 hr post-stress, the elF2a phosphorylation appears fully restored to its basal level" is not supported by the data in Figure S18 which shows P-elF2 is 2-3x higher at 3 hr compared to the basal unstressed condition.  
+
+Response: We quantified the intensity of elF2a phosphorylation normalized to the intensity of Tubulin. The level of elF2a phosphorylation in T3 was \(29 \pm 15\%\) higher than basal level. The statement and Figure S2 have been updated in the latest manuscript.  
+
+![](images/Figure_unknown_13.jpg)
+
+<center>Figure S2. Western blot analysis of elF2α phosphorylation (p- elF2α) in G3BP1-miniSOG HEK293T cells in unstressed (basal), stressed (T0), and recovery (T1 and T3) stages. The intensity of elF2α phosphorylation are normalized with elF2α expression. Error bars represent standard deviation. </center>  
+
+## Minor comments:  
+
+1. References to Parker et al., and Jeffrey (Jaffrey) et al., should be replaced with the citation in proper format.  
+
+Response: Thanks for pointing out this issue. We have corrected this in the manuscript.  
+
+2. The observation that mitochondrial gene encoded RNAs are depleted from stress granules using CAP-seq is not novel as it is stated in Khong et al., Mol Cell 2017 that mitochondrial RNAs present in their transcriptome dataset may be contaminants introduced during purification.
+
+<--- Page Split --->
+
+Response: In Khong et al, mitochondrial gene encoded RNAs were found to be enriched in the SG, with Fold change (SG core vs. Total RNAs) no less than 2.5.  
+
+3. The source of the data used to create translation efficiency and transcript length plots throughout should be defined in the figure legends.  
+
+Response: Thanks for pointing out this issue, we have already added this source in the figure legends.  
+
+We used same translation efficiency data source as Khong et al. and TRIBE- ID, which counted from Sidrauski, et al., Elife, 2015. Transcript length were downloaded from Ensembl website.  
+
+Figure 3G: Box plot comparing the length features between sorbitol- specific, stress- independent and arsenite- specific SG- proximal mRNAs. ns, not significant \((p > 0.05)\) ; \*\*\* \(p < 0.0001\) (Mann- Whitney test). Transcript length features are referenced from Ensembl website.  
+
+Figure 4E - G: Bar plot showing the comparison of AU content in different regions (E), length features (F) and translation efficiencies. (G) between mRNAs in basal- specific, pre- existing and de novo datasets. ns, not significant \((p > 0.05)\) ; \* \(p < 0.05\) ; \*\* \(p < 0.01\) ; \*\* \(p < 0.001\) ; \*\*\* \(p < 0.0001\) (Mann- Whitney test). Translation efficiencies are counted from a previous report (Sidrauski, et al., Elife, 2015). Transcript length features and AU content are referenced from Ensembl website.  
+
+Same statement is also attached to the legends of Figure 5E - F, Figure 6B - D, Figure S12, Figure S14, Figure S19 and Figure S22.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In this manuscript, Ren and colleagues use the RNA proximity labeling approach CAP- seq to identify and quantify transcripts localized to stress granules (SG). They use this technique to profile how the SG transcriptome changes in response to different stresses (arsenite vs. sorbitol) and how its RNA content dynamically changes upon SG disassembly. Generally, the conclusions in the manuscript are supported by the data. I have only a few comments that may improve the manuscript.  
+
+## MAJOR COMMENTS  
+
+1. Again and again, the authors find that SG-proximal RNAs are (1) long, (2) AU-rich, and (3) poorly translated. They ascribe function to all of these characteristics and note that they may be involved in localizing RNAs to SG. However, across the transcriptome,
+
+<--- Page Split --->
+
+all 3 of these characteristics are correlated with each other. That is, RNAs that are long also tend to be AU- rich, and vice versa (Marin et al Yeast 2003; Lopez et al Frontiers in Genetics 2021, and others). It could be then that just one of these characteristics is important and the others come along for the ride through these correlations but are not actually functional themselves in terms of getting RNAs to SGs. This point should at least be noted.  
+
+Response: We thank the reviewer for the valuable advice. Indeed, RNA features such as AU content have been reported to be correlated with transcript length. When analyzing our CAP- seq dataset, in addition to examining the correlation between SG enrichment with total RNA length and AU% at the transcript level, we also break down the transcripts into 5'UTR, CDS, and 3'UTR segments. We observed quite different trends at each segment. For example, stress- de novo RNAs feature both longer length and higher AU% in their 3'UTR, whereas their CDSs are shorter with higher AU% (Figure 4E- F). Another example is the G3BP1- proximal RNA Cluster #4 during SG disassembly, which are characterized with shorter 3'UTR but higher AU%. These observations indicate that AU% and length may contribute differently to SG enrichment, as least in a few cases presented in our study. In the revised manuscript, we have noted this issue to the readers and cited the references of Marin et al Yeast 2003 and Lopez et al Frontiers in Genetics 2021:  
+
+Page 22, line 33 - page 23 line 7: "It should be noted that RNA features such as AU content have been reported to be correlated with transcript length (Marin, et al., Yeast, 2003). To better examine the correlation between SG enrichment with total RNA length and AU content at the transcript level, we broke down the transcripts into 5'UTR, CDS, and 3'UTR segments. We observed quite different trends at each segment. For example, stress- de novo RNAs feature both longer length and higher AU content in their 3'UTR, whereas their CDSs are shorter with higher AU content (Figure 4E- F). RNAs remained proximal to G3BP1 during final stages of SG disassembly (cluster 4) were also characterized with shorter 3'UTR but more AU. These observations indicate that AU content and length may contribute differently to SG enrichment."  
+
+Prompted by the reviewer's advice, we have also sought to disentangle the effects of \(\mathsf{m}^6\mathsf{A}\) site number and transcript length, which are strongly and positively correlated to each other. In the revised manuscript, we have re- analyzed our CAP- seq datasets by normalizing the number of \(\mathsf{m}^6\mathsf{A}\) sites with total transcript length (kb) (i.e. number of sites per kilobase). A similar trend was observed as our previous analysis using \(\mathsf{m}^6\mathsf{A}\) sites alone, revealing that the density of \(\mathsf{m}^6\mathsf{A}\) is higher in pre- existing G3BP1- proximal RNAs (Figure S23). During the disassembly stage, the density of \(\mathsf{m}^6\mathsf{A}\) is also the highest in Cluster 2 (RNAs that dissociate more slowly during the recovery phase),
+
+<--- Page Split --->
+
+which is consistent with our previous observation using \(\mathsf{m}^6\mathsf{A}\) sites alone (Figure S26). In the revised manuscript, we have provided these figures in the supplementary file and added the following descriptions in the main text:  
+
+Page 14, line 30: "We thus analyzed both the number of \(\mathsf{m}^6\mathsf{A}\) sites and \(\mathsf{m}^6\mathsf{A}\) site density (number per kilobase) in our datasets, using published \(\mathsf{m}^6\mathsf{A}\) database in HEK293T cells."  
+
+Page 17, line 34: "We therefore compared the extent of \(\mathsf{m}^6\mathsf{A}\) sites and \(\mathsf{m}^6\mathsf{A}\) site density (number per kilobase) in our dataset."  
+
+![](images/Figure_unknown_14.jpg)
+
+<center>Figure S23. Comparing the proportion of mRNA with different \(\mathsf{m}^6\mathsf{A}\) density ( \(\mathsf{m}^6\mathsf{A}\) sites per kilobase) in basal-specific, pre-existing and de novo datasets. </center>  
+
+![PLACEHOLDER_37_1]
+
+<center>Figure S26. Comparing the proportion of mRNA with different \(\mathsf{m}^6\mathsf{A}\) density ( \(\mathsf{m}^6\mathsf{A}\) sites per kilobase) between 4 clusters. </center>  
+
+2. Related to point 1, I particularly disagree with the statement that "RNA binding proteins generally prefer AU-rich motifs". Yes, many RBPs do bind AU-rich sequences, but many also do not. Characterizing the entirety of the RNA binding proteome in this way is not accurate.
+
+<--- Page Split --->
+
+Response: We thank the reviewer for pointing out this issue. We realize that we should not equate AU% with AU- rich elements, and we agree that it is not accurate to claim that RBPs prefer AU- rich sequences. In the revised manuscript, we have removed the above statement:  
+
+Page 14, lines 18 - 22: "In contrast, no significant differences in AU content was observed in the 5' UTR. In terms of transcript length, RNAs in all three G3BP1- /SG- proximal datasets are overall longer than total cellular mRNAs (Figure S22, Table S3B- C), which is in accordance with the previous observation that SG- localized mRNAs are longer than average."  
+
+3. Any time that membership in a group is tested across experiments (e.g. Venn diagrams comparing SG-enriched RNAs across stresses or cell types, etc.), p-values should be calculated for the overlap. Without some kind of test, it's difficult for me to know whether or not the overlap observed is more or less than expected. This is important as observing a greater overlap than expected would lend additional confidence to the results.  
+
+Response: We thank the reviewer for this suggestion. In the revised manuscript, we have applied hypergeometric test to calculate the relevant p-values of Venn diagrams.  
+
+Figure 2D compares our CAP- seq SG- proximal and SG- excluded RNAs from arsenite- stressed U- 2 OS with the previously published SG RNA dataset, the p- value of this comparison is \(2 \times 10^{- 183}\) and 0.0001, respectively.  
+
+Figure 3D compares CAP- seq datasets for arsenite stress and sorbitol stress. The p- value of this comparison is \(2 \times 10^{- 150}\) . Figure S20 compares CAP- seq defined SG- proximal dataset under T0 and G3BP1- proximal datasets under T1 and T3. The p values of comparisons between T0 vs. T1, T0 vs. T3 and T1 vs. T3 are \(9 \times 10^{- 321}\) , \(1 \times 10^{- 150}\) and \(5 \times 10^{- 266}\) , respectively.  
+
+We also added p value of other Venn diagrams that are not mentioned here. Figure 4D compares the CAP- seq enriched RNAs under unstressed, arsenite stress and sorbitol stress, the p value of the pre- existing datasets is \(4 \times 10^{- 246}\) . Figure 6A compares RNAs enriched at basal and T3 time point, the p- value of this comparison is \(4 \times 10^{- 135}\) . The p values have been attached to the mentioned Venn diagrams.  
+
+## MINOR COMMENTS  
+
+1. Where did the translational efficiency data come from? I may have missed it, but it was not obvious to me. This should be made clearer.
+
+<--- Page Split --->
+
+Response: Thanks for pointing out this issue. We have already added this source in the figure legends.  
+
+We used same translation efficiency data source as Khong et al. and TRIBE- ID, which counted from Sidrauski, et al., Elife, 2015. Transcript length were downloaded from Ensembl website.  
+
+Figure 3G: Box plot comparing the length features between sorbitol- specific, stress- independent and arsenite- specific SG- proximal mRNAs. ns, not significant \((p > 0.05)\) ; \(*** p < 0.0001\) (Mann- Whitney test). Transcript length features are referenced from Ensembl website.  
+
+Figure 4E - G: Bar plot showing the comparison of AU content in different regions (E), length features (F) and translation efficiencies. (G) between mRNAs in basal- specific, pre- existing and de novo datasets. ns, not significant \((p > 0.05)\) ; \(* p < 0.05\) ; \(** p < 0.01\) ; \(*** p < 0.001\) ; \(**** p < 0.0001\) (Mann- Whitney test). Translation efficiencies are counted from a previous report (Sidrauski, et al., Elife, 2015). Transcript length features and AU content are referenced from Ensembl website.  
+
+Same statement is also attached to the legends of Figure 5E - F, Figure 6B - D, Figure S12, Figure S14, Figure S19 and Figure S22.  
+
+2. "Transcription factors binding to AU-rich elements [have] been found in SG." Are you sure you mean transcription factors? Those don't normally bind RNA.  
+
+Response: Thanks for pointing out this mistake. We have removed this statement in the revised manuscript (page 17, line 21).  
+
+3. "However, the purification procedure is prone to contamination and loss of weakly associated material, thus causing high false positive rate." The loss of material that in truth really is associated with SG would result in false negatives, not false positives.  
+
+Response: Thanks for pointing out this mistake. We have revised the sentence (page 2, lines 32 - 34): "However, the purification procedure is prone to contamination and loss of weakly associated material, thus causing high false positive and high false negative rates, respectively."
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Throughout the revision period, the authors have carried out additional experiments and made commendable enhancements to the manuscript. Given that the current paper introduces a conventional yet innovative approach for charting spatial RNA molecules, while also pushing the boundaries of our understanding regarding stress granules and the G3BP1 interactome, this reviewer strongly recommends the acceptance of this manuscript in Nature Communications.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Overall the authors addressed the stated concerns. The remaining issues that must be addressed prior to publication are as follows.  
+
+1. While replicates were shown for one of the western blots (Figure S2), the number of independent experimental replicates for the western blots shown in Figure S1, S5, and S8 must be stated.  
+
+2. P-elf2 levels relative to total elf2 levels from three independent experimental replicates must be presented for Figure S5 to be meaningful.  
+
+3. If the unstressed controls now included in Figure 1B, Figure 2A, and Figure 3A were done at a separate time than the other conditions, this must be indicated in the figure legend or it assumed they were done at the same time.  
+
+4. Define the term "biological replicates" used throughout the manuscript.  
+
+5. The Figure names throughout the text should be checked (for example Figure S9 is mistakenly referenced in Page 9 line 17)  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have addressed my concerns.
+
+<--- Page Split --->
+
+## REVIEWERS COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Throughout the revision period, the authors have carried out additional experiments and made commendable enhancements to the manuscript. Given that the current paper introduces a conventional yet innovative approach for charting spatial RNA molecules, while also pushing the boundaries of our understanding regarding stress granules and the G3BP1 interactome, this reviewer strongly recommends the acceptance of this manuscript in Nature Communications.  
+
+Response: We thank the reviewer for the recommendation.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Overall the authors addressed the stated concerns. The remaining issues that must be addressed prior to publication are as follows.  
+
+1. While replicates were shown for one of the western blots (Figure S2), the number of independent experimental replicates for the western blots shown in Figure S1, S5, and S8 must be stated.  
+
+Response: We have added the statement in the figure legend. Three biological replicates were performed for the western blots mentioned above with the raw data provided in the Source Data file.  
+
+2. P-elF2 levels relative to total elF2 levels from three independent experimental replicates must be presented for Figure S5 to be meaningful.  
+
+Response: We have presented the results as request.  
+
+![PLACEHOLDER_41_0]
+  
+
+Supplementary Figure 5. Western blot analysis of elF2α phosphorylation (p- elF2α) in G3BP1- miniSOG HEK293T cells after CAP- seq labeling. The intensities of p- elF2α are normalized with respect to elF2α expression. The bars indicate mean values, lines indicate SD. The gel image presented here is a representative example from three independent experiments.
+
+<--- Page Split --->
+
+3. If the unstressed controls now included in Figure 1B, Figure 2A, and Figure 3A were done at a separate time than the other conditions, this must be indicated in the figure legend or it assumed they were done at the same time.  
+
+Response: We thank the reviewer for pointing out this issue. We have mentioned that unstressed controls were performed at a sperate time than the other conditions in the figure legend.  
+
+4. Define the term "biological replicates" used throughout the manuscript. Response: We thank the reviewer for pointing out this issue. We have changed the statement to independent experiments.  
+
+5. The Figure names throughout the text should be checked (for example Figure S9 is mistakenly referenced in Page 9 line 17)  
+
+Response: We thank the reviewer for pointing out the mistakes. We have checked the figure names and corrected the mistakes.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have addressed my concerns. Response: We thank the reviewer for the comment.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a2a6039f1ff1b8c1ad9861734d2f774acfb0dca9f56ff7064677e01d82329ecd/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__a2a6039f1ff1b8c1ad9861734d2f774acfb0dca9f56ff7064677e01d82329ecd/supplementary_0_Peer Review File__a2a6039f1ff1b8c1ad9861734d2f774acfb0dca9f56ff7064677e01d82329ecd.mmd

@@ -0,0 +1,395 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Hydrodynamic spin- orbit coupling in asynchronous optically driven micro- rotors  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The manuscript presents an experimental realization of spinning spherical particles with axis of rotation normal to a bottom wall. The rotation is driven by a focused beam of circularly polarized light; I could not follow the details but this seems like a clever idea. The particle rotation rate is set by the balance of optical and viscous torques. Particle translations and rotations are measured and the authors conclude that hydrodynamic interactions give rise to rotor pairing and orbiting around each other. The authors claim "universal hydrodynamic spin- orbit coupling" which is "geometrical in nature" but, unless I am missing something, it is expected that HD interactions between solid particles in Stokes flow depend only on geometry (Kim and Karrila, Microhydrodynamics, 1991)  
+
+It is not clear what the significance of the fact that particles optical axes are asynchronized - wouldn't that just be set by the initial (random) conditions?  
+
+More details need to be provided about the model, at least in the supplemental material. Distinction should be made between the 3D r and the 2D r in Eq. 5. Eq. 8 takes the derivative of \(1 / r^{\wedge}\) alpha to claim that the rotation rate decreases by a factor of \(1 - \lambda\) alpha. However, in Eq. 5 there is prefactor \(3a^{\wedge}3\backslash \mathrm{delta}^{\wedge}2 / r^{\wedge}5\) . Is this really \(O(1)\) constant? It does depend on the separation between the rotors r. Is this the universality proposed by the authors - that the dependence on size, separation, and \delta delta somehow compensate each other to give order 1 constant? If this is the case, this point should be clarified. However, the data in fig. 6d is pretty scattered and not very convincing  
+
+In a study that emphasizes the importance of HD interactions, it seems the authors ignore a significant body of work dedicated on the emergence of hydrodynamically bound states in systems of rollers, e.g., Martinez- Pedrero et al (Sci. Advances, 2018) Delmotte (Phys. Rev. Fluids, 2019), and spinners (confined to a 2D plane)- in addition to Ref. 24, Goto et al. (Nat. Comm, 2015), Kokot et al (PNAS, 2017). In addition to Ref. 7- 9, the ordering in monolayer of rotors has been analyzed by Lushi and Vlahovska (J Non- linear Science, 2015), where orbital motion in such 2D systems of rotors is predicted (albeit in free space, no wall).  
+
+In conclusion, while the experiment is neat, the novelty beyond the new experiment is limited. While orbital motion is indeed observed for a first time experimentally, it has been theoretically predicted in systems of plane- confined rotors.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+See attached.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+The manuscript "Hydrodynamic spin- orbit coupling in asynchronous optically driven microrotors" presents a synthetic system of active self- rotating particles which have both rotational and translational degrees of freedom.  
+
+The manuscript presents a novel biomimetic system of active rotors. Other examples of synthetic active rotors have translational, but not rotational, degrees of freedom because their active rotations closely follow an external field. By contrast, in this manuscript, the particles are able to rotate with different speeds and different phases, much like self- spinning
+
+<--- Page Split --->
+
+living cells, which interact only hydrodynamically. The manuscript is well written and includes a concrete set of experimental results that convincingly test quantitative predictions for these kinds of active rotors. I recommend publication.  
+
+Before publication, I suggest that the authors consider the following two points:  
+
+- The inset in Fig. 6c was not clear to me. Is it the spinning rate \Omegaega or its change \DeltaOmega which is plotted? What are the theoretical predictions plotted along with the experimental data?  
+
+- The following recent work may be of interest as an analogous biological system: Odd dynamics of living chiral crystals Tan et al Nature 607, 287 (2022).
+
+<--- Page Split --->
+
+## Review of Hydrodynamic spin-orbit coupling in asynchronous optically driven micro-rotors by A. Modin et al.  
+
+The manuscript reports a study of the dynamics of a sample of micro- particles immersed in water. The particles are birefringent, and some effort was made with particle synthesis to ensure they were stable in water. The particles can be optically rotated by illuminating them with circularly polarised light - and an unfocussed laser beam is chosen such that the particles are not translationally optically trapped, and so are still able to freely diffuse while rotating. 2D particle tracking is performed allowing the trajectories of the particles, and their hydrodynamic interactions to be analysed. This analysis compares well with analytical theory of low Reynolds number hydrodynamics. In particular, the interaction of pairs of particles is observed, in which the particles become transiently coupled in their translational motion.  
+
+I find the work interesting, and as far as I am aware, this is the first such analysis of a complex sample of this nature. The experiments and theoretical analysis appear to be rigorously carried out, however I do think the explanation of the theory could be clearer in places, see more details below. Perhaps I am missing something deeper, however it does not seem unexpected that a spinning particle should cause nearby particles to orbit. Therefore overall I am unsure of the wider significance of these findings. A few examples are given in the conclusion, however it is unclear where or how the theory developed in this work could be applied to give a deeper understanding of any other situation/type of sample. Therefore, I suggest this wider significance is articulated more clearly for consideration in a broad interest journal such as nature communications.  
+
+Small points:  
+
+line 106: MSD - Mean squared displacement? is not defined. I think should be for a broad audience journal like nat. comms.  
+
+line 113: The manuscript states '..monitoring the particles' vertical rising speed..' How is this out of plane motion monitored?  
+
+line 161: 'The magnitude of the Fourier of the individual particles,..' should instead be something like: 'The magnitude of the Fourier Transform of the intensity transmitted by individual particles as a function of time,..'?  
+
+line 162: 'corresponding to four times the typical rotation frequency (- 0.125 Hz)' It is not clear to me why this is four times rather than twice the typical rotation frequency?  
+
+line 163: 'The asynchronous phases of the light intensities add- up destructively,' I know what is meant here, however I find this use of the term 'destructive' when speaking of purely real positive functions which can't 'cancel each other out to zero' a bit misleading. Consider re- phasing this sentence?  
+
+Equation 4 - might be helpful to give more detail of how this equation was constructed, maybe in the supplementary.  
+
+Figure 3b inset: I don't follow why the data points appear to form into two parallel lines. Is this showing particles of different sizes? I think this inset needs more explanation in the caption/main text. Also, what sized particle is this data plotted for?  
+
+Line 173: The Green's function equation - I suggest all variables (epsilon, r - is at distance in any direction, or distance perpendicular to the rotation axis?) and indices (ijk) are specifically defined here, along with some additional explanation, to help those readers unfamiliar with how the Stokelet and Rotlet descriptions of hydrodynamic interactions operate.  
+
+Line 178- 187: I suggest clarifying this section. I don't follow how the statement on line 179- 180 about the flow field decaying as \(\sim 1 / r^{\wedge}3\) then ties in with the following sentences and eq5 which has a \(1 / r^{\wedge}4\) scaling. These equations are introduced quickly, without proper definition of the many terms (e.g. r, \(x^{\wedge}\) , \(y^{\wedge}\) , theta^). Also is R superscript(+) the same as R subscript(+)? I assume so but it is a little confusing seeing both. Figure 5 is helpful, and I understand these equations are well
+
+<--- Page Split --->
+
+used in many body low Reynolds number hydrodynamics (as is the method of images near a boundary), however for readers unfamiliar with this field, I think more explanation is needed.  
+
+Line 197: 'microscopic organism' > 'microscopic organisms'.
+
+<--- Page Split --->
+
+# Response to Referees for Manuscript NCOMMS-22-37460  
+
+March 17, 2023  
+
+## Response summary  
+
+Dear Referees,  
+
+We are writing to resubmit a revised version of our manuscript entitled "Hydrodynamic spin- orbit coupling in asynchronous optically driven micro- rotors" (NCOMMS- 22- 37460).  
+
+We thank the Referees for their kind words, finding our work "clever" (Referee 1) and "rigorously carried out" (Referee 2). We are also grateful to Referee 3, who found our synthetic rotors to be "much like self- spinning living cells."  
+
+Please find below a point- by- point response for each Referee's professional critique, separated by Referee. Referees' questions are highlighted in bold. Changes in the research article are highlighted in green in the new version of the manuscript.  
+
+We appreciate all of the Referees' comments and questions, helping to raise the manuscript's scientific standards to match Nature Communications. Thank you for considering our re- submission.  
+
+Sincerely,  
+
+Matan Yah Ben Zion  
+
+Alvin Modin  
+
+Paul Chaikin
+
+<--- Page Split --->
+
+## Response to Referee 1  
+
+1. The manuscript presents an experimental realization of spinning spherical particles with axis of rotation normal to a bottom wall. The rotation is driven by a focused beam of circularly polarized light; I could not follow the details but this seems like a clever idea. The particle rotation rate is set by the balance of optical and viscous torques. Particle translations and rotations are measured and the authors conclude that hydrodynamic interactions give rise to rotor pairing and orbiting around each other.  
+
+The authors claim "universal hydrodynamic spin-orbit coupling" which is "geometrical in nature" but, unless I am missing something, it is expected that HD interactions between solid particles in Stokes flow depend only on geometry (Kim and Karrila, Microhydrodynamics, 1991)  
+
+We agree that, in theory, Stokes flow is well known to be geometric in nature. However, experimental work studying hydrodynamic coupling between rotating particles thus far showed mixed contributions, including steric interactions \(^{1,2}\) , magnetic interactions \(^{3}\) , or phoretic interactions \(^{4}\) . We designed a new experimental approach for rotating micro- particles to decouple the different contributions. We did not use a focused beam of light (as was done before \(^{5 - 10}\) ). Instead, we used a collimated beam (as illustrated in Fig. 1 a,b). This distinction allows us to experimentally investigate previously inaccessible conditions for the following reasons:  
+
+(a) Focused light beams create sharp light intensity gradients that tweeze particles to the tight focal point of the beam. This typically restricts the translational motion of the particle to roughly its size, obscuring the coupling between translation and rotation. 
+(b) To date, the focused light beams used to rotate particles were too tight to host more than one particle at a time, making hydrodynamic particle-particle interactions inaccessible.  
+
+Using a broad, collimated beam of light, we were able to observe the geometric nature of Stokes flow directly, derive the universal hydrodynamic spin-orbit coupling, and experimentally support theoretical predictions found in Kim and Karrila (which we now cite).  
+
+The distinction between a focused and collimated photonic torque field is particularly important when testing the significance of pair interactions in search for theoretically predicted emergent behavior \(^{2,11 - 14}\) . We added the following in the text to illustrate this:
+
+<--- Page Split --->
+
+When gradients in the beam's intensity are present, particles are constrained to the narrow waist of the focused light. A narrow- waisted beam generates tweezing forces that typically restrict the translational motion of a particle to roughly its diameter, obscuring the coupling between translation and rotation. To date, focused light beams used to rotate particles were too tight to host an ensemble of particles, making hydrodynamic particle- particle interactions inaccessible \(^{9,15}\) .  
+
+## 2. It is not clear what the significance of the fact that particles optical axes are asynchronized – wouldn't that just be set by the initial (random) conditions?  
+
+The Referee correctly points out that having rotors with different initial orientations would be sufficient to show the decay of the magnitude of the Fourier transform of the summed intensities (Fig 1d). This alone is an experimental observation previously inaccessible for synthetic microrotors – magnetic particles, for example, will always align their dipole moment with the direction of the applied external magnetic field. The applied field artificially “freezes” the orientational degrees of freedom of magnetic rotors.  
+
+Following the Referee's comment, we identify the need to point out further sources of asynchronicity:  
+
+(a) Different initial conditions  
+(b) Difference in rotation frequency  
+(c) Difference in the stochastic torque  
+
+The inset of Fig. 1d shows that the two oscillating signals change their relative phase. This can be interpreted as a difference in the stochastic torque experienced by either particle, indicating that the relative orientation of the particles changes with time. By contrast, magnetically rotated particles (whose orientational degrees of freedom are frozen) have the same orientation (and relative orientation) throughout an experiment. Asynchronicity is the crucial ingredient that allows observation of hydrodynamic spin- orbit coupling. If a particle's orientation is constrained to the orientation of the external drive, it will not respond to the hydrodynamic torque generated by neighboring particles. For clarification, we added the following sentence in the main text:  
+
+The relative orientation of an ensemble of vaterite particles is free to vary.  
+
+We highlight this point once again in the conclusion of the manuscript:  
+
+We systematically quantified the micro- rotors' optical and hydrodynamic properties and found that particles rotate asynchronously, unlike any previous synthetic micro- rotor system. The particles' asyn
+
+<--- Page Split --->
+
+chronous rotation indicates that their orientational degrees of freedom are dynamic variables; this is in contrast to magnetic rotors, whose orientational degrees of freedom are "frozen" by an applied magnetic field.  
+
+We have also added a section to the Supplementary Information:  
+
+- Measuring the frequency and global phase of a rotating particle  
+
+The section contains the following supporting information:  
+
+(a) A detailed calculation of the Fourier transform of an individual rotor.  
+(b) A detailed calculation of the Fourier transform of the sum of the transmitted light intensity of individual rotors.  
+
+3. More details need to be provided about the model, at least in the supplemental material. Distinction should be made between the 3D r and the 2D r in Eq. 5. Eq. 8 takes the derivative of \(1 / r^{\alpha}\) to claim that the rotation rate decreases by a factor of \(1 - \alpha\) . However, in Eq. 5 there is prefactor \(3a^{3}\delta^{2} / r^{5}\) . Is this really O(1) constant? It does depend on the separation between the rotors r. Is this the universality proposed by the authors – that the dependence on size, separation, and \(\delta\) somehow compensate each other to give order 1 constant? If this is the case, this point should be clarified.  
+
+We thank the Referee for pointing out the need to clarify our model further. The relation (Eq. 8) is universal and is obtained by combining Eqs. 6 and 7 using the flow field given by Eq. 5.  
+
+The Referee is correct in noting that the pre- factor \(3a^{3}\delta^{2} / r^{5}\) in Eq. 5 depends on the separation between the rotors, \(r\) . However, because of geometry, every translation is accompanied by a proportional amount of rotation. When we re- scale the data in Figure 6c according to the particles' radii \(a\) , separations \(r\) , and distances from the wall of the capillary \(\delta\) , we find that these parameters compensate for each other in a way that results in an \(\mathcal{O}(1)\) constant. For example, the maximum value of advective flow \(\boldsymbol {u}(r)\) experienced by a neighboring particle occurs when particles touch. Assuming two identically sized spheres (with radius \(a\) ), the minimum distance \(r\) between their two centers is \(r = 2a\) , resulting in a pre- factor \(3a^{3}\delta^{2} / r^{5} \approx 3 / 32\) .  
+
+This allows us to obtain a scaling law for spin- orbit coupling that depends only on the type of confining geometry (Eq. 8) being considered. For example, a rotor near a plane has an \(\alpha = 4\) , but a rotor confined between two planes or next to a fluid- fluid interface will have a different value of \(\alpha\) and thus
+
+<--- Page Split --->
+
+a different "strength" of spin- orbit coupling. The spin- orbit scaling law depends only on the flow- field generated and not on the material parameters of the particles. When accounting for these parameters, the experimental results presented in Fig. 6d fall onto a line with a slope of \(1 - \alpha = - 3\) , as predicted. To provide more clarity regarding the model, we have added a new section in the Supplementary Information, titled:  
+
+- Flow generated by a rotating sphere near a wall in the Stokes-flow regime  
+
+This section includes a step- by- step derivation of Eqs. 5, 8, and 9. In the text, we added key derivation points that emphasize the distinction between 2D and 3D \(r\) . For completeness, on lines 203- 205: we define all variables as:  
+
+Here \(\hat{x},\hat{y}\) are Cartesian unit vectors, and \(|R_{\pm}|\equiv \left(x^{2} + y^{2} + (z\mp \delta)^{2}\right)^{\frac{1}{2}}\) , representing the distance to a point \((x,y,z)\) in space from the source and image charges, respectively. In the far- field limit, \(|R_{\pm}|^{- 3}\approx \frac{1}{r^{3}}\left(1\pm \frac{3\delta^{2}}{r^{2}}\right)\) .  
+
+And after Eq. 8, we have added the following:  
+
+...where \(r\) is now a two- dimensional distance.  
+
+We have also stated that Eq. 9 is explicitly independent of particle size, separation, and height above the rotating plane:  
+
+This relation is general - independent of \(a\) , \(r\) , and \(\delta\) . Every translation is accompanied by a proportional amount of rotation. Eq. 8 holds regardless of whether the flow is three- dimensional (in bulk), quasi- two- dimensional (near a wall), or strictly two- dimensional (in a liquid film).  
+
+## 4. However, the data in fig. 6d is pretty scattered and not very convincing.  
+
+The scatter in the spin- orbit coupling measurement (Fig. 6d) can be estimated from the thermal fluctuations of a Brownian rotor. The time evolution of the variance of the orientation, \(\langle \Delta \theta^{2}(t)\rangle\) , of a rotor with a nominal spinning rate of \(\Omega_{0}\) , which is subjected to rotational diffusion with rotational diffusion constant \(D_{r}\) , follows an equation similar to a 1D Brownian particle subjected to an external drift: \(\langle \Delta \theta^{2}(t)\rangle = 2D_{r}t + (\Omega_{0}t)^{2}\) (see for example Doi, Oxford University Press, 2013). At short times \((t \ll \frac{2D_{r}}{\Omega_{0}^{2}})\) , the motion is diffusion dominated, and at longer times \((t \gg \frac{2D_{r}}{\Omega_{0}^{2}})\) , the motion is drift (or activity) dominated. This means that if the orientation is monitored over a short duration, we should expect inherent fluctuations stemming from the diffusive term, with their relative significance depending on the ratio of the drift to the diffusive contributions. Note that this is analogous to the Péclet
+
+<--- Page Split --->
+
+number, which is typically related to translational motion in the literature. In our work, we measured the different parameters and can estimate quantitatively that for a typical \(4\mu \mathrm{m}\) particle, rotational diffusion is \(D_{r}\approx 0.02\mathrm{rad}^{2} / \mathrm{s}\) (see Fig. 3c in the main text), and a nominal spinning rate is \(\Omega_{0}\approx 1\mathrm{rad / s}\) . Since the particles also undergo translational diffusion, their orbits are transient, limiting the duration over which the instantaneous rotation rate can be extracted. This requires striking a balance between the following extremes: on one end, if a particle's orientation is monitored over a long duration, it will average over different orbital separations. On the other hand, if the particle's orientation is measured over too short of a period, its dynamics will be dominated by thermal diffusion. To balance these, we extract the instantaneous spinning rate in the period between two blinks, \(\tau_{\mathrm{blink}}\approx 2\mathrm{s}\) (see Fig. 1c). This gives a relative error in the measured "instantaneous" spinning rate of \(\sqrt{2D_{r}\tau_{\mathrm{blink}} / \left(\Omega_{0}\tau_{\mathrm{blink}}\right)^{2}}\approx 0.14\) . By comparison, the relative contribution of the spin- orbit coupling to the instantaneous spinning rate is about \(\Delta \Omega /\Omega_{0}\lesssim 0.3\) , as measured in our work (see Fig. 6c inset) and also theoretically predicted in the past (Davis1969). Thus, the contribution of the spin- orbit coupling to the relative change in spinning rate is larger but comparable to the relative thermal fluctuations, consistent with the scatter observed in Fig. 6d.  
+
+We thank the reviewer for raising this important point, as it emphasizes the unique dynamics of free rotors that undergo rotational and translation diffusion. In the revised manuscript, we added two subsections in the Supporting Information detailing the instantaneous spinning rate measurement process and evaluating their fluctuations. These subsections are titled:  
+
+- Expected fluctuations in the extracted spin rate for measuring the spin-orbit coupling- Measuring the change in the instantaneous spinning rate \(\Delta \Omega\) of a rotor  
+
+We also added the following in the caption of Fig. 6:  
+
+The scatter relative to the trend line originates from thermal fluctuations in transient orbits of freely diffusing Brownian rotors (see Supplementary Information).  
+
+5. In a study that emphasizes the importance of HD interactions, it seems the authors ignore a significant body of work dedicated on the emergence of hydrodynamically bound states in systems of rollers, e.g., Martinez-Pedrero et al (Sci. Advances, 2018) Delmotte (Phys. Rev. Fluids, 2019), and spinners (confined to a 2D plane)-in addition to Ref. 24, Goto et al. (Nat. Comm, 2015), Kokot et al (PNAS, 2017). In addition to Ref. 7-9, the ordering in monolayer of rotors has been analyzed by Lushi and Vlahovska (J Non-linear Science,
+
+<--- Page Split --->
+
+2015), where orbital motion in such 2D systems of rotors is predicted (albeit in free space, no wall).  
+
+We thank the Referee for drawing our attention to past numerical simulations and experiments investigating hydrodynamically bound states in rollers and spinners. We have added the references to the manuscript, specifically those treating systems of spinners. The work by Lushi and Vlahovska is especially interesting, offering predictions on counter- rotating particles, which the system presented in our work makes experimentally accessible. We specifically emphasize this work (as well as the related work by Kokot et al. 2017) by including the following sentence in the conclusion of the manuscript:  
+
+Moreover, combining our system of optical rotors with rotors driven by an external magnetic field could enable the experimental study of ensembles of counter- rotating particles, where optical rotors rotate independently from the magnetic rotors. Experimental investigation of an ensemble of counter- rotors would elucidate recent predictions on self- assembly, phase separation, and edge modes, expanding our understanding of far- from- equilibrium states of matter \(^{12,13,16,17}\) .  
+
+6. In conclusion, while the experiment is neat, the novelty beyond the new experiment is limited. While orbital motion is indeed observed for a first time experimentally, it has been theoretically predicted in systems of plane-confined rotors.  
+
+We thank the Referee for finding the experiment to be neat. Our findings indeed show hydrodynamic spin- orbit coupling in a synthetic system for the first time and also offer researchers who pursue emergence in ensembles of coupled rotors a new experimental test bed to revise previous findings where hydrodynamic spin- orbit coupling was inaccessible by construction.  
+
+We sincerely thank the Referee for their in- depth comments and suggestions.
+
+<--- Page Split --->
+
+## Response to Referee 2  
+
+1. The manuscript reports a study of the dynamics of a sample of micro-particles immersed in water. The particles are birefringent, and some effort was made with particle synthesis to ensure they were stable in water. The particles can be optically rotated by illuminating them with circularly polarised light - and an unfocussed laser beam is chosen such that the particles are not translationally optically trapped, and so are still able to freely diffuse while rotating. 2D particle tracking is performed allowing the trajectories of the particles, and their hydrodynamic interactions to be analysed. This analysis compares well with analytical theory of low Reynolds number hydrodynamics. In particular, the interaction of pairs of particles is observed, in which the particles become transiently coupled in their translational motion.  
+
+I find the work interesting, and as far as I am aware, this is the first such analysis of a complex sample of this nature. The experiments and theoretical analysis appear to be rigorously carried out, however I do think the explanation of the theory could be clearer in places, see more details below. Perhaps I am missing something deeper, however, it does not seem unexpected that a spinning particle should cause nearby particles to orbit. Therefore overall I am unsure of the wider significance of these findings. A few examples are given in the conclusion, however it is unclear where or how the theory developed in this work could be applied to give a deeper understanding of any other situation/type of sample. Therefore, I suggest this wider significance is articulated more clearly for consideration in a broad interest journal such as nature communications.  
+
+We thank the Referee for finding our work interesting and rigorous. While we agree that pairs of rotating particles are expected to couple hydrodynamically, previous experimental work showed incompatible results: biological micro- rotors \(^{18,19}\) and synthetic micro- rotors \(^{20,21}\) show different dynamics. This discrepancy manifests itself in large ensembles of rotors, so- called chiral fluids \(^{1,22}\) or crystals \(^{3,20,21}\) . Moreover, isotropic materials made of rotors are theorized to have unique material properties impossible at equilibrium \(^{11}\) , and it is unclear if such materials are experimentally accessible in a non- isotropic system where all particles point in the same direction (such as magnetically stirred particles \(^{1,22}\) ). Our work focuses on pairs of micro- rotors as a first step in paving the way to study large ensembles of asynchronous rotors in search of novel emergent behavior.
+
+<--- Page Split --->
+
+We added the following in the main text to emphasize the broader significance of our findings: The motion of synthetic micro- rotors studied so far is incompatible with the \(\mathbf{R} - \theta\) hydrodynamic coupling observed in pairs of biological micro- rotors \(^{3,18}\) . Moreover, previous studies with ensembles of synthetic micro- rotors \(^{20,22}\) spin by an externally imposed field and can not show spontaneous symmetry breaking as seen in ensembles of biological rotors \(^{19}\) .  
+
+2. line 106: MSD - Mean squared displacement? is not defined. I think should be for a broad audience journal like nat. comms.  
+
+Thank you for the suggestion. We added the following in the main text to define MSD:  
+
+Measuring the particles' mean squared displacements (MSDs), \(\langle \Delta r^2 \rangle = 4D_t \tau\) , we observe a reduction in their diffusion constants \(D_t\) compared to the bulk value, \(D_t^{\text{bulk}} = k_B T / 3 \pi \eta d\) (Fig. 3b).  
+
+3. line 113: The manuscript states '..monitoring the particles' vertical rising speed..' How is this out of plane motion monitored?  
+
+In the original version of the manuscript, the measurement of the particle's vertical rising speed was explained in the Supplementary Information (see Fig. S3). Following the Referee's question, we added a detailed sub- section in the Supplementary Information:  
+
+- Measuring the average rise velocity \(v\) of a particle in the presence of an optical flux  
+
+We also added the following sentences in the main text describing the measurement process:  
+
+At higher fluxes where \(F_{rad}\) exceeds \(F_g\) , vaterite particles begin to steadily rise from the capillary's bottom surface at a constant speed. Vertically shifting the imaging focal plane from the bottom of the capillary to its top surface, we monitor the time it takes for particles to travel \(100 \mu \mathrm{m}\) , corresponding to when a focused image of a particle re- appears (see Supplementary Information for additional experimental details.  
+
+4. line 161: 'The magnitude of the Fourier of the individual particles,..' should instead be something like: 'The magnitude of the Fourier Transform of the intensity transmitted by individual particles as a function of time,..'?  
+
+Corrected – thank you for pointing out this oversight.  
+
+5. line 162: 'corresponding to four times the typical rotation frequency (0.125 Hz)' It is not clear to me why this is four times rather than twice the typical rotation frequency?
+
+<--- Page Split --->
+
+For light to reach the detector (camera) through crossed-polarizers, the particle must de- polarize the linearly polarized illumination beam. In a birefringent particle, peak de- polarization happens when the light is polarized at 45 degrees relative to the optical axis of the particle. This happens four times per full revolution \(^{9,23}\) . We illustrate this with a cartoon in Fig. 1c. There, the measured light intensity peaks once while the schematic particle has rotated by only a quarter period.  
+
+To clarify this in the text, we revised the schematic in Figure 1c and added the direction of the illumination's electric field (set by the polarizer). The new caption in Figure 1 now reads:  
+
+Experimental set- up to drive micro- rotors asynchronously. a Optical setup for introducing a broad ( \(D\approx 440\mu \mathrm{m}\) ) circularly polarized beam into a microscope sample. b Schematic and polarized microscopy image of birefringent vaterite particles rotating while moving freely in the illuminated region. The transmitted light intensities of two particles (blue and green) are tracked throughout the experiment and are shown in c and the inset of d. c One- half of the particles' (blue and green) blinking cycle, demonstrating that their optical axes are asynchronous. The incident electric field - whose direction is set by the orientation of the polarizer (P) - is de- polarized whenever the optical axis of the rotating particles is aligned with neither the polarizer nor analyzer (A). d Computing the magnitude of the Fourier transform \((\sqrt{\mathcal{F}\mathcal{F}^{*}})\) of the blinking patterns (inset) of the two particles in b shows that the frequencies at which the particles de- polarize the incident L.E.D. light are centered around \(0.5\mathrm{Hz}\) , corresponding to a rotation frequency of \(0.125\mathrm{Hz}\) . However, the magnitude of the sum of transforms, \(|\sum_{i}\mathcal{F}_{i}|^{2}\) (solid line), decays, confirming that the particles are out of phase. Scale bar: \(5\mu \mathrm{m}\) ;  
+
+We also added a new subsection in the Supplementary Information entitled:  
+
+- Measuring the frequency and global phase of a rotating particle  
+
+The new subsection provides additional details relating a particle's rotation frequency to the intensity of light that it transmits over one period. As the 4- fold blinking per revolution is more evident for a birefringent particle that is not perfectly spherical, we attach a video showing this for the Referee's reference.  
+
+6. line 163: 'The asynchronous phases of the light intensities add-up destructively,' I know what is meant here, however I find this use of the term 'destructive' when speaking of purely real positive functions which can't 'cancel each other out to zero' a bit misleading. Consider re-phasing this sentence?
+
+<--- Page Split --->
+
+We thank the Referee for pointing out this potential ambiguity. We rephrased the sentence, which now says:  
+
+The different phases of the light intensities do not necessarily add up constructively.  
+
+7. Equation 4 - might be helpful to give more detail of how this equation was constructed, maybe in the supplementary.  
+
+In response to the Referee's suggestion, we have included a new section in the Supplementary Information that presents the derivation of Equation 4, entitled:  
+
+- Spinning angular frequency \(\Omega\) of a birefringent particle  
+
+The section details the relationship between elliptically polarized light's optical torque and a birefringent particle's spinning frequency.  
+
+8. Figure 3b inset: I don't follow why the data points appear to form into two parallel lines. Is this showing particles of different sizes? I think this inset needs more explanation in the caption/main text. Also, what sized particle is this data plotted for?  
+
+We appreciate the Referee for noticing that the caption describing Figure 3 needed additional details. The inset of Fig. 3b serves to show that the translational mean square displacement (MSD) is linear with time (Brownian motion). The diffusion constants presented in the main panel were extracted from the y-intercept of these slopes. Each curve (line) in the inset represents an individual particle of a different size. We added a dashed line in the inset to indicate the expected linear trend. In the main panel, we included grey arrows to indicate the corresponding particle sizes. To describe the process of computing the translational diffusion constant, we have included a new subsection in the Supplementary Information:  
+
+- Measuring the translational diffusion constant of rotors  
+
+Our procedure follows the methods described in Refs. 28 and 29. For completeness, we have added a blue arrow in the main panel of Fig. 3c to indicate the particle size for which the correlation function presented in the inset is measured. The updated caption in the new version of the manuscript now reads:  
+
+Inset shows the translational mean- squared displacement (MSDs) for particles with diameters \(d = 1.92 \pm 0.11\mu \mathrm{m}\) and \(5.92 \pm 0.35\mu \mathrm{m}\) . The MSD \(\propto \tau^{1}\) , consistent with particles undergoing Brownian
+
+<--- Page Split --->
+
+motion. The translational diffusion constants \(D_{\mathrm{t}}\) obtained from these two MSDs are indicated by grey arrows in the main panel. c Rotational diffusion perpendicular to the wall, (spinning) \(D_{\mathrm{r},\perp}\) , measured using depolarization intensity decorrelation, \(g_{\mathrm{PA}}\) . The blue arrow in the main panel indicates \(D_{\mathrm{r},\perp}\) obtained from fitting \(g_{\mathrm{PA}}\) of a \(d = 2.31\pm 0.14\mu \mathrm{m}\) particle (inset).  
+
+9. Line 173: The Green's function equation - I suggest all variables (epsilon, r - is at distance in any direction, or distance perpendicular to the rotation axis?) and indices (ijk) are specifically defined here, along with some additional explanation, to help those readers unfamiliar with how the Stokelet and Rotlet descriptions of hydrodynamic interactions operate.  
+
+In the new version of the manuscript, we added a complete description of all variables, and we hope this makes the notation clearer for researchers from different disciplines. On lines 190- 193:  
+
+We introduce a singularity at position \((x_{0},y_{0},z_{0})\) , acting as a point- torque distrubance (rotlet). The corresponding Green's function, \(G_{ij}\) , satisfying the Stoke's equations is \(G_{ij} = \frac{\epsilon_{ijk}r_{k}}{r^{3}}\) , where \(\epsilon_{ijk}\) is the Levi- Cevita symbol whose indices represent components of the rotlet's position in the Cartesian coordinate system, and \(r\) is a 3D vector pointing from the rotlet to a point \((x,y,z)\) in space  
+
+And on lines 203- 205:  
+
+Here \(\hat{x},\hat{y}\) are Cartesian unit vectors, and \(|R_{\pm}|\equiv \left(x^{2} + y^{2} + (z\mp \delta)^{2}\right)^{\frac{1}{2}}\) , representing the distance to a point \((x,y,z)\) in space from the source and image charges, respectively. In the far- field limit, \(|R_{\pm}|^{- 3}\approx \frac{1}{r^{3}}\left(1\pm \frac{3\delta^{2}}{r^{2}}\right)\) .  
+
+In the revised version, Fig. 5 also graphically depicts these quantities.  
+
+10. Line 178-187: I suggest clarifying this section. I don't follow how the statement on line 179-180 about the flow field decaying as \(1 / r^{3}\) then ties in with the following sentences and eq5 which has a \(1 / r^{4}\) scaling. These equations are introduced quickly, without proper definition of the many terms (e.g. \(r\) , \(\hat{x}\) , \(\hat{y}\) , \(\hat{\theta}\) ). Also is R superscript(+) the same as R subscript(+)? I assume so but it is a little confusing seeing both. Figure 5 is helpful, and I understand these equations are well used in many body low Reynolds number hydrodynamics (as is the method of images near a boundary), however for readers unfamiliar with this field, I think more explanation is needed.  
+
+We thank the Referee for this comment. We also added a detailed step- by- step derivation in the Supplementary Information, entitled:
+
+<--- Page Split --->
+
+- Flow generated by a rotating sphere near a wall in the Stokes-flow regime.  
+
+In the subsection, we detail the calculation that leads to Eqs. 5 and 8. We also added the following sentence (line 207) in the main text to fully show how the \(1 / r^{4}\) scaling arises:  
+
+This scaling arises from noting that \(- y\hat{x} +x\hat{y} = r\hat{\theta}\) .  
+
+11. Line 197: 'microscopic organism' \(\mathcal{L}\) 'microscopic organisms'.  
+
+We have made this change and thank the Referee for noticing this oversight.  
+
+We sincerely thank Referee 2 for their detailed review of our manuscript, helping to clarify our experiments and theory, and making the manuscript more accessible to a broader audience.
+
+<--- Page Split --->
+
+## Response to Referee 3  
+
+1. The manuscript "Hydrodynamic spin-orbit coupling in asynchronous optically driven micro-rotors" presents a synthetic system of active self-rotating particles which have both rotational and translational degrees of freedom.  
+
+The manuscript presents a novel biomimetic system of active rotors. Other examples of synthetic active rotors have translational, but not rotational, degrees of freedom because their active rotations closely follow an external field. By contrast, in this manuscript, the particles are able to rotate with different speeds and different phases, much like self- spinning living cells, which interact only hydrodynamically. The manuscript is well written and includes a concrete set of experimental results that convincingly test quantitative predictions for these kinds of active rotors. I recommend publication.  
+
+We thank the Referee for finding our manuscript well- written and identifying that the synthetic swimmers presented here resemble living matter.  
+
+2. Before publication, I suggest that the authors consider the following two points: The inset in Fig. 6c was not clear to me. Is it the spinning rate \(\Omega\) or its change \(\Delta \Omega\) which is plotted? We thank the Referee for pointing out a potential ambiguity in the caption of Fig 6c. In the new version of the manuscript, the caption reads: Angular speed \(\omega\) of two orbiting spheres (diameters \(6.7 \mu \mathrm{m}\) and \(5.1 \mu \mathrm{m}\) ) at different separations, along with the dependence of the spinning rate \(\Omega\) on the normalized separation (inset).  
+
+3. What are the theoretical predictions plotted along with the experimental data? The theoretical predictions in the inset of Fig. 6c are derived from Faxen's 1st and 2nd laws (Eq. 6 and 7). Using Eq. 5, the flow-field generated by a single rotor \(u(\boldsymbol {r})\) , we compute the dependence of the spinning rate of each rotor, \(\Omega_{i,j}\) , on the particles' separation \(r\) , given their asymptotic spinning rates at large separation \(\Omega_{i,j}^{0}\) . We added the following comment in the caption to clarify this: Curves show the predicted spinning rates for each particle at different separations – derived from Faxen's laws (Eqs. 6 and 7 in the main text) – given the particles' asymptotic spinning rates, \(\Omega_{i,j}^{0}\) , measured at large separations  
+
+4. The following recent work may be of interest as an analogous biological system: Odd dynamics of living chiral crystals Tan et al Nature 607, 287 (2022).
+
+<--- Page Split --->
+
+The recent findings by the Fakhri group (now cited in our work) demonstrate a beautiful example of emergence found in asynchronously rotating particles. We aspire that the dialogue between biological and synthetic model systems will enhance our understanding of new states of matter far from equilibrium.  
+
+We thank Referee 3 for their kind words and critical comments.  
+
+## References  
+
+[1] Bililign, E. S. et al. Motile dislocations knead odd crystals into whorls. Nature Physics 18, 212- 218 (2022). URL https://www.nature.com/articles/s41567- 021- 01429- 3.  
+
+[2] Oppenheimer, N., Stein, D. B. & Shelley, M. J. Rotating membrane inclusions crystallize through hydrodynamic and steric interactions. Physical Review Letters 123, 148101 (2019).  
+
+[3] Massana- Cid, H., Meng, F., Matsunaga, D., Golestanian, R. & Tierno, P. Tunable self- healing of magnetically propelling colloidal carpets. Nature Communications 10 (2019). URL https://doi.org/10.1038/s41467- 019- 10255- 4.  
+
+[4] Aubret, A., Youssef, M., Sacanna, S. & Palacci, J. Targeted assembly and synchronization of self- spinning microgears. Nature Physics 14, 1114- 1118 (2018). URL http://www.nature.com/articles/s41567- 018- 0227- 4.  
+
+[5] Vogel, R. et al. Synthesis and surface modification of birefringent vaterite microspheres. Langmuir 25, 11672- 11679 (2009).  
+
+[6] Friese, M. E. J., Nieminen, T. A., Heckenberg, N. R. & Rubinsztein- Dunlop, H. Optical alignment and spinning of laser- trapped microscopic particles. Nature 394, 348- 350 (1998).  
+
+[7] Donato, M. G. et al. Light- induced rotations of chiral birefringent microparticles in optical tweezers. Scientific Reports 6 (2016).  
+
+[8] Leach, J., Mushfique, H., di Leonardo, R., Padgett, M. & Cooper, J. An optically driven pump for microfluidics. Lab on a Chip 6, 735 (2006). URL http://xlink.rsc.org/?DOI=b601886f.  
+
+[9] Parkin, S. J. et al. Highly birefringent vaterite microspheres: production, characterization and applications for optical micromanipulation. Optics Express 17, 21944 (2009).
+
+<--- Page Split --->
+
+[10] Zhang, S. et al. Ultrasensitive rotating photonic probes for complex biological systems. Optica 4, 1103 (2017). URL https://opg.optica.org/abstract.cfm?URI=optica- 4- 9- 1103. [11] Avron, J. E. Odd viscosity. Journal of Statistical Physics 92, 543- 557 (1998).[12] Nguyen, N. H., Klotsa, D., Engel, M. & Glotzer, S. C. Emergent collective phenomena in a mixture of hard shapes through active rotation. Physical Review Letters 112 (2014).[13] Yeo, K., Lushi, E. & Vlahovska, P. M. Collective dynamics in a binary mixture of hydrodynamically coupled microrotors. Physical Review Letters 114, 188301 (2015).[14] Oppenheimer, N., Stein, D. B., Ben Zion, M. Y. & Shelley, M. J. Hyperuniformity and phase enrichment in vortex and rotor assemblies. Nature Communications 13, 804 (2022).[15] Molloy, J. E. & Padgett, M. J. Lights, action: Optical tweezers. Contemporary Physics 43, 241- 258 (2002). URL https://doi.org/10.1080/00107510110116051. [16] Lushi, E. & Vlahovska, P. M. Periodic and Chaotic Orbits of Plane- Confined Microrotors in Creeping Flows. Journal of Nonlinear Science 25, 1111- 1123 (2015). URL https://doi.org/10.1007/s00332- 015- 9254- 9. [17] Kokot, G. et al. Active turbulence in a gas of self- assembled spinners. Proceedings of the National Academy of Sciences 114, 12870- 12875 (2017). URL https://pnas.org/doi/full/10.1073/pnas.1710188114. [18] Drescher, K. et al. Dancing volvox: Hydrodynamic bound states of swimming algae. Physical Review Letters 102, 1- 4 (2009). 0901.2087. [19] Tan, T. H. et al. Odd dynamics of living chiral crystals. Nature 607, 287- 293 (2022).[20] Yan, J., Bae, S. C. & Granick, S. Rotating crystals of magnetic janus colloids. Soft Matter 11, 147- 153 (2015).[21] Massana- Cid, H., Levis, D., Hernández, R. J. H., Pagonabarraga, I. & Tierno, P. Arrested phase separation in chiral fluids of colloidal spinners. Physical Review Research 3 (2021).[22] Soni, V. et al. The odd free surface flows of a colloidal chiral fluid. Nature Physics 15, 1188- 1194 (2019).[23] Hecht, E. Optics (Pearson Education, Inc, Boston, 2017), 5 ed edn.
+
+<--- Page Split --->
+
+[24] Blake, J. R. & Chwang, A. T. Fundamental singularities of viscous flow. Journal of Engineering Mathematics 8, 23- 29 (1974).
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The revised manuscript now clearly shows to novelty of the contribution. I no longer have concerns about the model. I recommend publication.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The authors have made a good effort to answer all of my questions, and also those of the other reviewers. Furthermore, for me they have articulated more clearly the importance of the work, and so I suggest publication as is.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+The authors have addressed my comments well, and I continue to support publication of this manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a2a6039f1ff1b8c1ad9861734d2f774acfb0dca9f56ff7064677e01d82329ecd/supplementary_0_Peer Review File__a2a6039f1ff1b8c1ad9861734d2f774acfb0dca9f56ff7064677e01d82329ecd_det.mmd

@@ -0,0 +1,542 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[102, 110, 374, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[123, 155, 874, 209]]<|/det|>
+Hydrodynamic spin- orbit coupling in asynchronous optically driven micro- rotors  
+
+<|ref|>image<|/ref|><|det|>[[94, 732, 262, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[270, 732, 880, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 83, 332, 99]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 113, 449, 129]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 143, 876, 280]]<|/det|>
+The manuscript presents an experimental realization of spinning spherical particles with axis of rotation normal to a bottom wall. The rotation is driven by a focused beam of circularly polarized light; I could not follow the details but this seems like a clever idea. The particle rotation rate is set by the balance of optical and viscous torques. Particle translations and rotations are measured and the authors conclude that hydrodynamic interactions give rise to rotor pairing and orbiting around each other. The authors claim "universal hydrodynamic spin- orbit coupling" which is "geometrical in nature" but, unless I am missing something, it is expected that HD interactions between solid particles in Stokes flow depend only on geometry (Kim and Karrila, Microhydrodynamics, 1991)  
+
+<|ref|>text<|/ref|><|det|>[[118, 294, 872, 324]]<|/det|>
+It is not clear what the significance of the fact that particles optical axes are asynchronized - wouldn't that just be set by the initial (random) conditions?  
+
+<|ref|>text<|/ref|><|det|>[[118, 339, 872, 460]]<|/det|>
+More details need to be provided about the model, at least in the supplemental material. Distinction should be made between the 3D r and the 2D r in Eq. 5. Eq. 8 takes the derivative of \(1 / r^{\wedge}\) alpha to claim that the rotation rate decreases by a factor of \(1 - \lambda\) alpha. However, in Eq. 5 there is prefactor \(3a^{\wedge}3\backslash \mathrm{delta}^{\wedge}2 / r^{\wedge}5\) . Is this really \(O(1)\) constant? It does depend on the separation between the rotors r. Is this the universality proposed by the authors - that the dependence on size, separation, and \delta delta somehow compensate each other to give order 1 constant? If this is the case, this point should be clarified. However, the data in fig. 6d is pretty scattered and not very convincing  
+
+<|ref|>text<|/ref|><|det|>[[118, 474, 878, 580]]<|/det|>
+In a study that emphasizes the importance of HD interactions, it seems the authors ignore a significant body of work dedicated on the emergence of hydrodynamically bound states in systems of rollers, e.g., Martinez- Pedrero et al (Sci. Advances, 2018) Delmotte (Phys. Rev. Fluids, 2019), and spinners (confined to a 2D plane)- in addition to Ref. 24, Goto et al. (Nat. Comm, 2015), Kokot et al (PNAS, 2017). In addition to Ref. 7- 9, the ordering in monolayer of rotors has been analyzed by Lushi and Vlahovska (J Non- linear Science, 2015), where orbital motion in such 2D systems of rotors is predicted (albeit in free space, no wall).  
+
+<|ref|>text<|/ref|><|det|>[[119, 594, 851, 640]]<|/det|>
+In conclusion, while the experiment is neat, the novelty beyond the new experiment is limited. While orbital motion is indeed observed for a first time experimentally, it has been theoretically predicted in systems of plane- confined rotors.  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 684, 449, 700]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 714, 233, 729]]<|/det|>
+See attached.  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 759, 449, 775]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 789, 878, 835]]<|/det|>
+The manuscript "Hydrodynamic spin- orbit coupling in asynchronous optically driven microrotors" presents a synthetic system of active self- rotating particles which have both rotational and translational degrees of freedom.  
+
+<|ref|>text<|/ref|><|det|>[[119, 850, 878, 910]]<|/det|>
+The manuscript presents a novel biomimetic system of active rotors. Other examples of synthetic active rotors have translational, but not rotational, degrees of freedom because their active rotations closely follow an external field. By contrast, in this manuscript, the particles are able to rotate with different speeds and different phases, much like self- spinning
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 876, 130]]<|/det|>
+living cells, which interact only hydrodynamically. The manuscript is well written and includes a concrete set of experimental results that convincingly test quantitative predictions for these kinds of active rotors. I recommend publication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 144, 763, 160]]<|/det|>
+Before publication, I suggest that the authors consider the following two points:  
+
+<|ref|>text<|/ref|><|det|>[[118, 174, 855, 220]]<|/det|>
+- The inset in Fig. 6c was not clear to me. Is it the spinning rate \Omegaega or its change \DeltaOmega which is plotted? What are the theoretical predictions plotted along with the experimental data?  
+
+<|ref|>text<|/ref|><|det|>[[118, 235, 785, 280]]<|/det|>
+- The following recent work may be of interest as an analogous biological system: Odd dynamics of living chiral crystals Tan et al Nature 607, 287 (2022).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[92, 66, 890, 97]]<|/det|>
+## Review of Hydrodynamic spin-orbit coupling in asynchronous optically driven micro-rotors by A. Modin et al.  
+
+<|ref|>text<|/ref|><|det|>[[92, 110, 900, 239]]<|/det|>
+The manuscript reports a study of the dynamics of a sample of micro- particles immersed in water. The particles are birefringent, and some effort was made with particle synthesis to ensure they were stable in water. The particles can be optically rotated by illuminating them with circularly polarised light - and an unfocussed laser beam is chosen such that the particles are not translationally optically trapped, and so are still able to freely diffuse while rotating. 2D particle tracking is performed allowing the trajectories of the particles, and their hydrodynamic interactions to be analysed. This analysis compares well with analytical theory of low Reynolds number hydrodynamics. In particular, the interaction of pairs of particles is observed, in which the particles become transiently coupled in their translational motion.  
+
+<|ref|>text<|/ref|><|det|>[[92, 252, 899, 380]]<|/det|>
+I find the work interesting, and as far as I am aware, this is the first such analysis of a complex sample of this nature. The experiments and theoretical analysis appear to be rigorously carried out, however I do think the explanation of the theory could be clearer in places, see more details below. Perhaps I am missing something deeper, however it does not seem unexpected that a spinning particle should cause nearby particles to orbit. Therefore overall I am unsure of the wider significance of these findings. A few examples are given in the conclusion, however it is unclear where or how the theory developed in this work could be applied to give a deeper understanding of any other situation/type of sample. Therefore, I suggest this wider significance is articulated more clearly for consideration in a broad interest journal such as nature communications.  
+
+<|ref|>text<|/ref|><|det|>[[93, 394, 201, 409]]<|/det|>
+Small points:  
+
+<|ref|>text<|/ref|><|det|>[[92, 422, 833, 452]]<|/det|>
+line 106: MSD - Mean squared displacement? is not defined. I think should be for a broad audience journal like nat. comms.  
+
+<|ref|>text<|/ref|><|det|>[[92, 465, 890, 495]]<|/det|>
+line 113: The manuscript states '..monitoring the particles' vertical rising speed..' How is this out of plane motion monitored?  
+
+<|ref|>text<|/ref|><|det|>[[92, 508, 899, 552]]<|/det|>
+line 161: 'The magnitude of the Fourier of the individual particles,..' should instead be something like: 'The magnitude of the Fourier Transform of the intensity transmitted by individual particles as a function of time,..'?  
+
+<|ref|>text<|/ref|><|det|>[[92, 565, 890, 595]]<|/det|>
+line 162: 'corresponding to four times the typical rotation frequency (- 0.125 Hz)' It is not clear to me why this is four times rather than twice the typical rotation frequency?  
+
+<|ref|>text<|/ref|><|det|>[[92, 609, 896, 667]]<|/det|>
+line 163: 'The asynchronous phases of the light intensities add- up destructively,' I know what is meant here, however I find this use of the term 'destructive' when speaking of purely real positive functions which can't 'cancel each other out to zero' a bit misleading. Consider re- phasing this sentence?  
+
+<|ref|>text<|/ref|><|det|>[[92, 681, 894, 711]]<|/det|>
+Equation 4 - might be helpful to give more detail of how this equation was constructed, maybe in the supplementary.  
+
+<|ref|>text<|/ref|><|det|>[[92, 724, 896, 768]]<|/det|>
+Figure 3b inset: I don't follow why the data points appear to form into two parallel lines. Is this showing particles of different sizes? I think this inset needs more explanation in the caption/main text. Also, what sized particle is this data plotted for?  
+
+<|ref|>text<|/ref|><|det|>[[92, 781, 889, 839]]<|/det|>
+Line 173: The Green's function equation - I suggest all variables (epsilon, r - is at distance in any direction, or distance perpendicular to the rotation axis?) and indices (ijk) are specifically defined here, along with some additional explanation, to help those readers unfamiliar with how the Stokelet and Rotlet descriptions of hydrodynamic interactions operate.  
+
+<|ref|>text<|/ref|><|det|>[[92, 852, 899, 924]]<|/det|>
+Line 178- 187: I suggest clarifying this section. I don't follow how the statement on line 179- 180 about the flow field decaying as \(\sim 1 / r^{\wedge}3\) then ties in with the following sentences and eq5 which has a \(1 / r^{\wedge}4\) scaling. These equations are introduced quickly, without proper definition of the many terms (e.g. r, \(x^{\wedge}\) , \(y^{\wedge}\) , theta^). Also is R superscript(+) the same as R subscript(+)? I assume so but it is a little confusing seeing both. Figure 5 is helpful, and I understand these equations are well
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 65, 860, 95]]<|/det|>
+used in many body low Reynolds number hydrodynamics (as is the method of images near a boundary), however for readers unfamiliar with this field, I think more explanation is needed.  
+
+<|ref|>text<|/ref|><|det|>[[92, 108, 595, 124]]<|/det|>
+Line 197: 'microscopic organism' > 'microscopic organisms'.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[171, 111, 825, 137]]<|/det|>
+# Response to Referees for Manuscript NCOMMS-22-37460  
+
+<|ref|>text<|/ref|><|det|>[[432, 213, 564, 230]]<|/det|>
+March 17, 2023  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 274, 301, 293]]<|/det|>
+## Response summary  
+
+<|ref|>text<|/ref|><|det|>[[113, 315, 226, 330]]<|/det|>
+Dear Referees,  
+
+<|ref|>text<|/ref|><|det|>[[112, 346, 886, 390]]<|/det|>
+We are writing to resubmit a revised version of our manuscript entitled "Hydrodynamic spin- orbit coupling in asynchronous optically driven micro- rotors" (NCOMMS- 22- 37460).  
+
+<|ref|>text<|/ref|><|det|>[[112, 396, 886, 465]]<|/det|>
+We thank the Referees for their kind words, finding our work "clever" (Referee 1) and "rigorously carried out" (Referee 2). We are also grateful to Referee 3, who found our synthetic rotors to be "much like self- spinning living cells."  
+
+<|ref|>text<|/ref|><|det|>[[112, 472, 886, 540]]<|/det|>
+Please find below a point- by- point response for each Referee's professional critique, separated by Referee. Referees' questions are highlighted in bold. Changes in the research article are highlighted in green in the new version of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[112, 548, 886, 591]]<|/det|>
+We appreciate all of the Referees' comments and questions, helping to raise the manuscript's scientific standards to match Nature Communications. Thank you for considering our re- submission.  
+
+<|ref|>text<|/ref|><|det|>[[700, 635, 769, 650]]<|/det|>
+Sincerely,  
+
+<|ref|>text<|/ref|><|det|>[[616, 677, 770, 693]]<|/det|>
+Matan Yah Ben Zion  
+
+<|ref|>text<|/ref|><|det|>[[675, 703, 770, 718]]<|/det|>
+Alvin Modin  
+
+<|ref|>text<|/ref|><|det|>[[674, 728, 770, 743]]<|/det|>
+Paul Chaikin
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[113, 112, 329, 131]]<|/det|>
+## Response to Referee 1  
+
+<|ref|>text<|/ref|><|det|>[[133, 150, 888, 295]]<|/det|>
+1. The manuscript presents an experimental realization of spinning spherical particles with axis of rotation normal to a bottom wall. The rotation is driven by a focused beam of circularly polarized light; I could not follow the details but this seems like a clever idea. The particle rotation rate is set by the balance of optical and viscous torques. Particle translations and rotations are measured and the authors conclude that hydrodynamic interactions give rise to rotor pairing and orbiting around each other.  
+
+<|ref|>text<|/ref|><|det|>[[152, 306, 886, 401]]<|/det|>
+The authors claim "universal hydrodynamic spin-orbit coupling" which is "geometrical in nature" but, unless I am missing something, it is expected that HD interactions between solid particles in Stokes flow depend only on geometry (Kim and Karrila, Microhydrodynamics, 1991)  
+
+<|ref|>text<|/ref|><|det|>[[152, 411, 887, 581]]<|/det|>
+We agree that, in theory, Stokes flow is well known to be geometric in nature. However, experimental work studying hydrodynamic coupling between rotating particles thus far showed mixed contributions, including steric interactions \(^{1,2}\) , magnetic interactions \(^{3}\) , or phoretic interactions \(^{4}\) . We designed a new experimental approach for rotating micro- particles to decouple the different contributions. We did not use a focused beam of light (as was done before \(^{5 - 10}\) ). Instead, we used a collimated beam (as illustrated in Fig. 1 a,b). This distinction allows us to experimentally investigate previously inaccessible conditions for the following reasons:  
+
+<|ref|>text<|/ref|><|det|>[[159, 600, 886, 720]]<|/det|>
+(a) Focused light beams create sharp light intensity gradients that tweeze particles to the tight focal point of the beam. This typically restricts the translational motion of the particle to roughly its size, obscuring the coupling between translation and rotation. 
+(b) To date, the focused light beams used to rotate particles were too tight to host more than one particle at a time, making hydrodynamic particle-particle interactions inaccessible.  
+
+<|ref|>text<|/ref|><|det|>[[153, 740, 886, 810]]<|/det|>
+Using a broad, collimated beam of light, we were able to observe the geometric nature of Stokes flow directly, derive the universal hydrodynamic spin-orbit coupling, and experimentally support theoretical predictions found in Kim and Karrila (which we now cite).  
+
+<|ref|>text<|/ref|><|det|>[[152, 822, 886, 890]]<|/det|>
+The distinction between a focused and collimated photonic torque field is particularly important when testing the significance of pair interactions in search for theoretically predicted emergent behavior \(^{2,11 - 14}\) . We added the following in the text to illustrate this:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 105, 886, 223]]<|/det|>
+When gradients in the beam's intensity are present, particles are constrained to the narrow waist of the focused light. A narrow- waisted beam generates tweezing forces that typically restrict the translational motion of a particle to roughly its diameter, obscuring the coupling between translation and rotation. To date, focused light beams used to rotate particles were too tight to host an ensemble of particles, making hydrodynamic particle- particle interactions inaccessible \(^{9,15}\) .  
+
+<|ref|>sub_title<|/ref|><|det|>[[131, 241, 884, 284]]<|/det|>
+## 2. It is not clear what the significance of the fact that particles optical axes are asynchronized – wouldn't that just be set by the initial (random) conditions?  
+
+<|ref|>text<|/ref|><|det|>[[152, 296, 886, 440]]<|/det|>
+The Referee correctly points out that having rotors with different initial orientations would be sufficient to show the decay of the magnitude of the Fourier transform of the summed intensities (Fig 1d). This alone is an experimental observation previously inaccessible for synthetic microrotors – magnetic particles, for example, will always align their dipole moment with the direction of the applied external magnetic field. The applied field artificially “freezes” the orientational degrees of freedom of magnetic rotors.  
+
+<|ref|>text<|/ref|><|det|>[[152, 453, 877, 470]]<|/det|>
+Following the Referee's comment, we identify the need to point out further sources of asynchronicity:  
+
+<|ref|>text<|/ref|><|det|>[[160, 490, 437, 566]]<|/det|>
+(a) Different initial conditions  
+(b) Difference in rotation frequency  
+(c) Difference in the stochastic torque  
+
+<|ref|>text<|/ref|><|det|>[[152, 586, 887, 780]]<|/det|>
+The inset of Fig. 1d shows that the two oscillating signals change their relative phase. This can be interpreted as a difference in the stochastic torque experienced by either particle, indicating that the relative orientation of the particles changes with time. By contrast, magnetically rotated particles (whose orientational degrees of freedom are frozen) have the same orientation (and relative orientation) throughout an experiment. Asynchronicity is the crucial ingredient that allows observation of hydrodynamic spin- orbit coupling. If a particle's orientation is constrained to the orientation of the external drive, it will not respond to the hydrodynamic torque generated by neighboring particles. For clarification, we added the following sentence in the main text:  
+
+<|ref|>text<|/ref|><|det|>[[153, 794, 685, 810]]<|/det|>
+The relative orientation of an ensemble of vaterite particles is free to vary.  
+
+<|ref|>text<|/ref|><|det|>[[153, 823, 664, 840]]<|/det|>
+We highlight this point once again in the conclusion of the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[153, 853, 886, 895]]<|/det|>
+We systematically quantified the micro- rotors' optical and hydrodynamic properties and found that particles rotate asynchronously, unlike any previous synthetic micro- rotor system. The particles' asyn
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[154, 105, 886, 173]]<|/det|>
+chronous rotation indicates that their orientational degrees of freedom are dynamic variables; this is in contrast to magnetic rotors, whose orientational degrees of freedom are "frozen" by an applied magnetic field.  
+
+<|ref|>text<|/ref|><|det|>[[154, 186, 620, 202]]<|/det|>
+We have also added a section to the Supplementary Information:  
+
+<|ref|>text<|/ref|><|det|>[[173, 221, 640, 238]]<|/det|>
+- Measuring the frequency and global phase of a rotating particle  
+
+<|ref|>text<|/ref|><|det|>[[155, 256, 575, 273]]<|/det|>
+The section contains the following supporting information:  
+
+<|ref|>text<|/ref|><|det|>[[160, 293, 886, 365]]<|/det|>
+(a) A detailed calculation of the Fourier transform of an individual rotor.  
+(b) A detailed calculation of the Fourier transform of the sum of the transmitted light intensity of individual rotors.  
+
+<|ref|>text<|/ref|><|det|>[[133, 384, 886, 555]]<|/det|>
+3. More details need to be provided about the model, at least in the supplemental material. Distinction should be made between the 3D r and the 2D r in Eq. 5. Eq. 8 takes the derivative of \(1 / r^{\alpha}\) to claim that the rotation rate decreases by a factor of \(1 - \alpha\) . However, in Eq. 5 there is prefactor \(3a^{3}\delta^{2} / r^{5}\) . Is this really O(1) constant? It does depend on the separation between the rotors r. Is this the universality proposed by the authors – that the dependence on size, separation, and \(\delta\) somehow compensate each other to give order 1 constant? If this is the case, this point should be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[154, 566, 886, 610]]<|/det|>
+We thank the Referee for pointing out the need to clarify our model further. The relation (Eq. 8) is universal and is obtained by combining Eqs. 6 and 7 using the flow field given by Eq. 5.  
+
+<|ref|>text<|/ref|><|det|>[[153, 620, 886, 816]]<|/det|>
+The Referee is correct in noting that the pre- factor \(3a^{3}\delta^{2} / r^{5}\) in Eq. 5 depends on the separation between the rotors, \(r\) . However, because of geometry, every translation is accompanied by a proportional amount of rotation. When we re- scale the data in Figure 6c according to the particles' radii \(a\) , separations \(r\) , and distances from the wall of the capillary \(\delta\) , we find that these parameters compensate for each other in a way that results in an \(\mathcal{O}(1)\) constant. For example, the maximum value of advective flow \(\boldsymbol {u}(r)\) experienced by a neighboring particle occurs when particles touch. Assuming two identically sized spheres (with radius \(a\) ), the minimum distance \(r\) between their two centers is \(r = 2a\) , resulting in a pre- factor \(3a^{3}\delta^{2} / r^{5} \approx 3 / 32\) .  
+
+<|ref|>text<|/ref|><|det|>[[154, 828, 886, 895]]<|/det|>
+This allows us to obtain a scaling law for spin- orbit coupling that depends only on the type of confining geometry (Eq. 8) being considered. For example, a rotor near a plane has an \(\alpha = 4\) , but a rotor confined between two planes or next to a fluid- fluid interface will have a different value of \(\alpha\) and thus
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 103, 886, 230]]<|/det|>
+a different "strength" of spin- orbit coupling. The spin- orbit scaling law depends only on the flow- field generated and not on the material parameters of the particles. When accounting for these parameters, the experimental results presented in Fig. 6d fall onto a line with a slope of \(1 - \alpha = - 3\) , as predicted. To provide more clarity regarding the model, we have added a new section in the Supplementary Information, titled:  
+
+<|ref|>text<|/ref|><|det|>[[172, 248, 715, 266]]<|/det|>
+- Flow generated by a rotating sphere near a wall in the Stokes-flow regime  
+
+<|ref|>text<|/ref|><|det|>[[153, 285, 886, 353]]<|/det|>
+This section includes a step- by- step derivation of Eqs. 5, 8, and 9. In the text, we added key derivation points that emphasize the distinction between 2D and 3D \(r\) . For completeness, on lines 203- 205: we define all variables as:  
+
+<|ref|>text<|/ref|><|det|>[[153, 357, 886, 440]]<|/det|>
+Here \(\hat{x},\hat{y}\) are Cartesian unit vectors, and \(|R_{\pm}|\equiv \left(x^{2} + y^{2} + (z\mp \delta)^{2}\right)^{\frac{1}{2}}\) , representing the distance to a point \((x,y,z)\) in space from the source and image charges, respectively. In the far- field limit, \(|R_{\pm}|^{- 3}\approx \frac{1}{r^{3}}\left(1\pm \frac{3\delta^{2}}{r^{2}}\right)\) .  
+
+<|ref|>text<|/ref|><|det|>[[153, 446, 488, 464]]<|/det|>
+And after Eq. 8, we have added the following:  
+
+<|ref|>text<|/ref|><|det|>[[154, 476, 477, 493]]<|/det|>
+...where \(r\) is now a two- dimensional distance.  
+
+<|ref|>text<|/ref|><|det|>[[153, 506, 886, 550]]<|/det|>
+We have also stated that Eq. 9 is explicitly independent of particle size, separation, and height above the rotating plane:  
+
+<|ref|>text<|/ref|><|det|>[[153, 560, 886, 630]]<|/det|>
+This relation is general - independent of \(a\) , \(r\) , and \(\delta\) . Every translation is accompanied by a proportional amount of rotation. Eq. 8 holds regardless of whether the flow is three- dimensional (in bulk), quasi- two- dimensional (near a wall), or strictly two- dimensional (in a liquid film).  
+
+<|ref|>sub_title<|/ref|><|det|>[[133, 645, 753, 664]]<|/det|>
+## 4. However, the data in fig. 6d is pretty scattered and not very convincing.  
+
+<|ref|>text<|/ref|><|det|>[[152, 675, 887, 896]]<|/det|>
+The scatter in the spin- orbit coupling measurement (Fig. 6d) can be estimated from the thermal fluctuations of a Brownian rotor. The time evolution of the variance of the orientation, \(\langle \Delta \theta^{2}(t)\rangle\) , of a rotor with a nominal spinning rate of \(\Omega_{0}\) , which is subjected to rotational diffusion with rotational diffusion constant \(D_{r}\) , follows an equation similar to a 1D Brownian particle subjected to an external drift: \(\langle \Delta \theta^{2}(t)\rangle = 2D_{r}t + (\Omega_{0}t)^{2}\) (see for example Doi, Oxford University Press, 2013). At short times \((t \ll \frac{2D_{r}}{\Omega_{0}^{2}})\) , the motion is diffusion dominated, and at longer times \((t \gg \frac{2D_{r}}{\Omega_{0}^{2}})\) , the motion is drift (or activity) dominated. This means that if the orientation is monitored over a short duration, we should expect inherent fluctuations stemming from the diffusive term, with their relative significance depending on the ratio of the drift to the diffusive contributions. Note that this is analogous to the Péclet
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[151, 103, 888, 476]]<|/det|>
+number, which is typically related to translational motion in the literature. In our work, we measured the different parameters and can estimate quantitatively that for a typical \(4\mu \mathrm{m}\) particle, rotational diffusion is \(D_{r}\approx 0.02\mathrm{rad}^{2} / \mathrm{s}\) (see Fig. 3c in the main text), and a nominal spinning rate is \(\Omega_{0}\approx 1\mathrm{rad / s}\) . Since the particles also undergo translational diffusion, their orbits are transient, limiting the duration over which the instantaneous rotation rate can be extracted. This requires striking a balance between the following extremes: on one end, if a particle's orientation is monitored over a long duration, it will average over different orbital separations. On the other hand, if the particle's orientation is measured over too short of a period, its dynamics will be dominated by thermal diffusion. To balance these, we extract the instantaneous spinning rate in the period between two blinks, \(\tau_{\mathrm{blink}}\approx 2\mathrm{s}\) (see Fig. 1c). This gives a relative error in the measured "instantaneous" spinning rate of \(\sqrt{2D_{r}\tau_{\mathrm{blink}} / \left(\Omega_{0}\tau_{\mathrm{blink}}\right)^{2}}\approx 0.14\) . By comparison, the relative contribution of the spin- orbit coupling to the instantaneous spinning rate is about \(\Delta \Omega /\Omega_{0}\lesssim 0.3\) , as measured in our work (see Fig. 6c inset) and also theoretically predicted in the past (Davis1969). Thus, the contribution of the spin- orbit coupling to the relative change in spinning rate is larger but comparable to the relative thermal fluctuations, consistent with the scatter observed in Fig. 6d.  
+
+<|ref|>text<|/ref|><|det|>[[152, 487, 887, 581]]<|/det|>
+We thank the reviewer for raising this important point, as it emphasizes the unique dynamics of free rotors that undergo rotational and translation diffusion. In the revised manuscript, we added two subsections in the Supporting Information detailing the instantaneous spinning rate measurement process and evaluating their fluctuations. These subsections are titled:  
+
+<|ref|>text<|/ref|><|det|>[[172, 597, 796, 644]]<|/det|>
+- Expected fluctuations in the extracted spin rate for measuring the spin-orbit coupling- Measuring the change in the instantaneous spinning rate \(\Delta \Omega\) of a rotor  
+
+<|ref|>text<|/ref|><|det|>[[153, 662, 533, 680]]<|/det|>
+We also added the following in the caption of Fig. 6:  
+
+<|ref|>text<|/ref|><|det|>[[152, 692, 886, 736]]<|/det|>
+The scatter relative to the trend line originates from thermal fluctuations in transient orbits of freely diffusing Brownian rotors (see Supplementary Information).  
+
+<|ref|>text<|/ref|><|det|>[[131, 752, 886, 897]]<|/det|>
+5. In a study that emphasizes the importance of HD interactions, it seems the authors ignore a significant body of work dedicated on the emergence of hydrodynamically bound states in systems of rollers, e.g., Martinez-Pedrero et al (Sci. Advances, 2018) Delmotte (Phys. Rev. Fluids, 2019), and spinners (confined to a 2D plane)-in addition to Ref. 24, Goto et al. (Nat. Comm, 2015), Kokot et al (PNAS, 2017). In addition to Ref. 7-9, the ordering in monolayer of rotors has been analyzed by Lushi and Vlahovska (J Non-linear Science,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[153, 105, 885, 149]]<|/det|>
+2015), where orbital motion in such 2D systems of rotors is predicted (albeit in free space, no wall).  
+
+<|ref|>text<|/ref|><|det|>[[152, 161, 888, 305]]<|/det|>
+We thank the Referee for drawing our attention to past numerical simulations and experiments investigating hydrodynamically bound states in rollers and spinners. We have added the references to the manuscript, specifically those treating systems of spinners. The work by Lushi and Vlahovska is especially interesting, offering predictions on counter- rotating particles, which the system presented in our work makes experimentally accessible. We specifically emphasize this work (as well as the related work by Kokot et al. 2017) by including the following sentence in the conclusion of the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[152, 317, 886, 435]]<|/det|>
+Moreover, combining our system of optical rotors with rotors driven by an external magnetic field could enable the experimental study of ensembles of counter- rotating particles, where optical rotors rotate independently from the magnetic rotors. Experimental investigation of an ensemble of counter- rotors would elucidate recent predictions on self- assembly, phase separation, and edge modes, expanding our understanding of far- from- equilibrium states of matter \(^{12,13,16,17}\) .  
+
+<|ref|>text<|/ref|><|det|>[[132, 452, 886, 522]]<|/det|>
+6. In conclusion, while the experiment is neat, the novelty beyond the new experiment is limited. While orbital motion is indeed observed for a first time experimentally, it has been theoretically predicted in systems of plane-confined rotors.  
+
+<|ref|>text<|/ref|><|det|>[[152, 533, 886, 626]]<|/det|>
+We thank the Referee for finding the experiment to be neat. Our findings indeed show hydrodynamic spin- orbit coupling in a synthetic system for the first time and also offer researchers who pursue emergence in ensembles of coupled rotors a new experimental test bed to revise previous findings where hydrodynamic spin- orbit coupling was inaccessible by construction.  
+
+<|ref|>text<|/ref|><|det|>[[112, 647, 664, 664]]<|/det|>
+We sincerely thank the Referee for their in- depth comments and suggestions.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[114, 113, 330, 131]]<|/det|>
+## Response to Referee 2  
+
+<|ref|>text<|/ref|><|det|>[[135, 150, 888, 395]]<|/det|>
+1. The manuscript reports a study of the dynamics of a sample of micro-particles immersed in water. The particles are birefringent, and some effort was made with particle synthesis to ensure they were stable in water. The particles can be optically rotated by illuminating them with circularly polarised light - and an unfocussed laser beam is chosen such that the particles are not translationally optically trapped, and so are still able to freely diffuse while rotating. 2D particle tracking is performed allowing the trajectories of the particles, and their hydrodynamic interactions to be analysed. This analysis compares well with analytical theory of low Reynolds number hydrodynamics. In particular, the interaction of pairs of particles is observed, in which the particles become transiently coupled in their translational motion.  
+
+<|ref|>text<|/ref|><|det|>[[151, 406, 888, 653]]<|/det|>
+I find the work interesting, and as far as I am aware, this is the first such analysis of a complex sample of this nature. The experiments and theoretical analysis appear to be rigorously carried out, however I do think the explanation of the theory could be clearer in places, see more details below. Perhaps I am missing something deeper, however, it does not seem unexpected that a spinning particle should cause nearby particles to orbit. Therefore overall I am unsure of the wider significance of these findings. A few examples are given in the conclusion, however it is unclear where or how the theory developed in this work could be applied to give a deeper understanding of any other situation/type of sample. Therefore, I suggest this wider significance is articulated more clearly for consideration in a broad interest journal such as nature communications.  
+
+<|ref|>text<|/ref|><|det|>[[152, 663, 888, 882]]<|/det|>
+We thank the Referee for finding our work interesting and rigorous. While we agree that pairs of rotating particles are expected to couple hydrodynamically, previous experimental work showed incompatible results: biological micro- rotors \(^{18,19}\) and synthetic micro- rotors \(^{20,21}\) show different dynamics. This discrepancy manifests itself in large ensembles of rotors, so- called chiral fluids \(^{1,22}\) or crystals \(^{3,20,21}\) . Moreover, isotropic materials made of rotors are theorized to have unique material properties impossible at equilibrium \(^{11}\) , and it is unclear if such materials are experimentally accessible in a non- isotropic system where all particles point in the same direction (such as magnetically stirred particles \(^{1,22}\) ). Our work focuses on pairs of micro- rotors as a first step in paving the way to study large ensembles of asynchronous rotors in search of novel emergent behavior.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 105, 886, 224]]<|/det|>
+We added the following in the main text to emphasize the broader significance of our findings: The motion of synthetic micro- rotors studied so far is incompatible with the \(\mathbf{R} - \theta\) hydrodynamic coupling observed in pairs of biological micro- rotors \(^{3,18}\) . Moreover, previous studies with ensembles of synthetic micro- rotors \(^{20,22}\) spin by an externally imposed field and can not show spontaneous symmetry breaking as seen in ensembles of biological rotors \(^{19}\) .  
+
+<|ref|>text<|/ref|><|det|>[[132, 241, 886, 285]]<|/det|>
+2. line 106: MSD - Mean squared displacement? is not defined. I think should be for a broad audience journal like nat. comms.  
+
+<|ref|>text<|/ref|><|det|>[[152, 297, 780, 315]]<|/det|>
+Thank you for the suggestion. We added the following in the main text to define MSD:  
+
+<|ref|>text<|/ref|><|det|>[[152, 327, 886, 371]]<|/det|>
+Measuring the particles' mean squared displacements (MSDs), \(\langle \Delta r^2 \rangle = 4D_t \tau\) , we observe a reduction in their diffusion constants \(D_t\) compared to the bulk value, \(D_t^{\text{bulk}} = k_B T / 3 \pi \eta d\) (Fig. 3b).  
+
+<|ref|>text<|/ref|><|det|>[[133, 387, 884, 431]]<|/det|>
+3. line 113: The manuscript states '..monitoring the particles' vertical rising speed..' How is this out of plane motion monitored?  
+
+<|ref|>text<|/ref|><|det|>[[152, 443, 886, 512]]<|/det|>
+In the original version of the manuscript, the measurement of the particle's vertical rising speed was explained in the Supplementary Information (see Fig. S3). Following the Referee's question, we added a detailed sub- section in the Supplementary Information:  
+
+<|ref|>text<|/ref|><|det|>[[171, 529, 787, 547]]<|/det|>
+- Measuring the average rise velocity \(v\) of a particle in the presence of an optical flux  
+
+<|ref|>text<|/ref|><|det|>[[152, 564, 815, 581]]<|/det|>
+We also added the following sentences in the main text describing the measurement process:  
+
+<|ref|>text<|/ref|><|det|>[[152, 593, 886, 712]]<|/det|>
+At higher fluxes where \(F_{rad}\) exceeds \(F_g\) , vaterite particles begin to steadily rise from the capillary's bottom surface at a constant speed. Vertically shifting the imaging focal plane from the bottom of the capillary to its top surface, we monitor the time it takes for particles to travel \(100 \mu \mathrm{m}\) , corresponding to when a focused image of a particle re- appears (see Supplementary Information for additional experimental details.  
+
+<|ref|>text<|/ref|><|det|>[[133, 729, 886, 799]]<|/det|>
+4. line 161: 'The magnitude of the Fourier of the individual particles,..' should instead be something like: 'The magnitude of the Fourier Transform of the intensity transmitted by individual particles as a function of time,..'?  
+
+<|ref|>text<|/ref|><|det|>[[152, 811, 540, 828]]<|/det|>
+Corrected – thank you for pointing out this oversight.  
+
+<|ref|>text<|/ref|><|det|>[[133, 846, 884, 890]]<|/det|>
+5. line 162: 'corresponding to four times the typical rotation frequency (0.125 Hz)' It is not clear to me why this is four times rather than twice the typical rotation frequency?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 105, 886, 224]]<|/det|>
+For light to reach the detector (camera) through crossed-polarizers, the particle must de- polarize the linearly polarized illumination beam. In a birefringent particle, peak de- polarization happens when the light is polarized at 45 degrees relative to the optical axis of the particle. This happens four times per full revolution \(^{9,23}\) . We illustrate this with a cartoon in Fig. 1c. There, the measured light intensity peaks once while the schematic particle has rotated by only a quarter period.  
+
+<|ref|>text<|/ref|><|det|>[[153, 236, 886, 279]]<|/det|>
+To clarify this in the text, we revised the schematic in Figure 1c and added the direction of the illumination's electric field (set by the polarizer). The new caption in Figure 1 now reads:  
+
+<|ref|>text<|/ref|><|det|>[[152, 291, 887, 590]]<|/det|>
+Experimental set- up to drive micro- rotors asynchronously. a Optical setup for introducing a broad ( \(D\approx 440\mu \mathrm{m}\) ) circularly polarized beam into a microscope sample. b Schematic and polarized microscopy image of birefringent vaterite particles rotating while moving freely in the illuminated region. The transmitted light intensities of two particles (blue and green) are tracked throughout the experiment and are shown in c and the inset of d. c One- half of the particles' (blue and green) blinking cycle, demonstrating that their optical axes are asynchronous. The incident electric field - whose direction is set by the orientation of the polarizer (P) - is de- polarized whenever the optical axis of the rotating particles is aligned with neither the polarizer nor analyzer (A). d Computing the magnitude of the Fourier transform \((\sqrt{\mathcal{F}\mathcal{F}^{*}})\) of the blinking patterns (inset) of the two particles in b shows that the frequencies at which the particles de- polarize the incident L.E.D. light are centered around \(0.5\mathrm{Hz}\) , corresponding to a rotation frequency of \(0.125\mathrm{Hz}\) . However, the magnitude of the sum of transforms, \(|\sum_{i}\mathcal{F}_{i}|^{2}\) (solid line), decays, confirming that the particles are out of phase. Scale bar: \(5\mu \mathrm{m}\) ;  
+
+<|ref|>text<|/ref|><|det|>[[153, 600, 698, 617]]<|/det|>
+We also added a new subsection in the Supplementary Information entitled:  
+
+<|ref|>text<|/ref|><|det|>[[172, 635, 640, 652]]<|/det|>
+- Measuring the frequency and global phase of a rotating particle  
+
+<|ref|>text<|/ref|><|det|>[[152, 670, 886, 761]]<|/det|>
+The new subsection provides additional details relating a particle's rotation frequency to the intensity of light that it transmits over one period. As the 4- fold blinking per revolution is more evident for a birefringent particle that is not perfectly spherical, we attach a video showing this for the Referee's reference.  
+
+<|ref|>text<|/ref|><|det|>[[131, 780, 886, 874]]<|/det|>
+6. line 163: 'The asynchronous phases of the light intensities add-up destructively,' I know what is meant here, however I find this use of the term 'destructive' when speaking of purely real positive functions which can't 'cancel each other out to zero' a bit misleading. Consider re-phasing this sentence?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 105, 884, 148]]<|/det|>
+We thank the Referee for pointing out this potential ambiguity. We rephrased the sentence, which now says:  
+
+<|ref|>text<|/ref|><|det|>[[152, 161, 757, 178]]<|/det|>
+The different phases of the light intensities do not necessarily add up constructively.  
+
+<|ref|>text<|/ref|><|det|>[[133, 196, 884, 240]]<|/det|>
+7. Equation 4 - might be helpful to give more detail of how this equation was constructed, maybe in the supplementary.  
+
+<|ref|>text<|/ref|><|det|>[[152, 252, 884, 294]]<|/det|>
+In response to the Referee's suggestion, we have included a new section in the Supplementary Information that presents the derivation of Equation 4, entitled:  
+
+<|ref|>text<|/ref|><|det|>[[170, 313, 583, 329]]<|/det|>
+- Spinning angular frequency \(\Omega\) of a birefringent particle  
+
+<|ref|>text<|/ref|><|det|>[[152, 348, 884, 391]]<|/det|>
+The section details the relationship between elliptically polarized light's optical torque and a birefringent particle's spinning frequency.  
+
+<|ref|>text<|/ref|><|det|>[[133, 408, 884, 477]]<|/det|>
+8. Figure 3b inset: I don't follow why the data points appear to form into two parallel lines. Is this showing particles of different sizes? I think this inset needs more explanation in the caption/main text. Also, what sized particle is this data plotted for?  
+
+<|ref|>text<|/ref|><|det|>[[152, 488, 886, 682]]<|/det|>
+We appreciate the Referee for noticing that the caption describing Figure 3 needed additional details. The inset of Fig. 3b serves to show that the translational mean square displacement (MSD) is linear with time (Brownian motion). The diffusion constants presented in the main panel were extracted from the y-intercept of these slopes. Each curve (line) in the inset represents an individual particle of a different size. We added a dashed line in the inset to indicate the expected linear trend. In the main panel, we included grey arrows to indicate the corresponding particle sizes. To describe the process of computing the translational diffusion constant, we have included a new subsection in the Supplementary Information:  
+
+<|ref|>text<|/ref|><|det|>[[172, 704, 586, 720]]<|/det|>
+- Measuring the translational diffusion constant of rotors  
+
+<|ref|>text<|/ref|><|det|>[[152, 740, 886, 832]]<|/det|>
+Our procedure follows the methods described in Refs. 28 and 29. For completeness, we have added a blue arrow in the main panel of Fig. 3c to indicate the particle size for which the correlation function presented in the inset is measured. The updated caption in the new version of the manuscript now reads:  
+
+<|ref|>text<|/ref|><|det|>[[152, 845, 884, 888]]<|/det|>
+Inset shows the translational mean- squared displacement (MSDs) for particles with diameters \(d = 1.92 \pm 0.11\mu \mathrm{m}\) and \(5.92 \pm 0.35\mu \mathrm{m}\) . The MSD \(\propto \tau^{1}\) , consistent with particles undergoing Brownian
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 105, 886, 199]]<|/det|>
+motion. The translational diffusion constants \(D_{\mathrm{t}}\) obtained from these two MSDs are indicated by grey arrows in the main panel. c Rotational diffusion perpendicular to the wall, (spinning) \(D_{\mathrm{r},\perp}\) , measured using depolarization intensity decorrelation, \(g_{\mathrm{PA}}\) . The blue arrow in the main panel indicates \(D_{\mathrm{r},\perp}\) obtained from fitting \(g_{\mathrm{PA}}\) of a \(d = 2.31\pm 0.14\mu \mathrm{m}\) particle (inset).  
+
+<|ref|>text<|/ref|><|det|>[[131, 214, 886, 333]]<|/det|>
+9. Line 173: The Green's function equation - I suggest all variables (epsilon, r - is at distance in any direction, or distance perpendicular to the rotation axis?) and indices (ijk) are specifically defined here, along with some additional explanation, to help those readers unfamiliar with how the Stokelet and Rotlet descriptions of hydrodynamic interactions operate.  
+
+<|ref|>text<|/ref|><|det|>[[152, 345, 886, 388]]<|/det|>
+In the new version of the manuscript, we added a complete description of all variables, and we hope this makes the notation clearer for researchers from different disciplines. On lines 190- 193:  
+
+<|ref|>text<|/ref|><|det|>[[152, 399, 886, 494]]<|/det|>
+We introduce a singularity at position \((x_{0},y_{0},z_{0})\) , acting as a point- torque distrubance (rotlet). The corresponding Green's function, \(G_{ij}\) , satisfying the Stoke's equations is \(G_{ij} = \frac{\epsilon_{ijk}r_{k}}{r^{3}}\) , where \(\epsilon_{ijk}\) is the Levi- Cevita symbol whose indices represent components of the rotlet's position in the Cartesian coordinate system, and \(r\) is a 3D vector pointing from the rotlet to a point \((x,y,z)\) in space  
+
+<|ref|>text<|/ref|><|det|>[[153, 506, 311, 522]]<|/det|>
+And on lines 203- 205:  
+
+<|ref|>text<|/ref|><|det|>[[152, 528, 886, 604]]<|/det|>
+Here \(\hat{x},\hat{y}\) are Cartesian unit vectors, and \(|R_{\pm}|\equiv \left(x^{2} + y^{2} + (z\mp \delta)^{2}\right)^{\frac{1}{2}}\) , representing the distance to a point \((x,y,z)\) in space from the source and image charges, respectively. In the far- field limit, \(|R_{\pm}|^{- 3}\approx \frac{1}{r^{3}}\left(1\pm \frac{3\delta^{2}}{r^{2}}\right)\) .  
+
+<|ref|>text<|/ref|><|det|>[[152, 613, 655, 631]]<|/det|>
+In the revised version, Fig. 5 also graphically depicts these quantities.  
+
+<|ref|>text<|/ref|><|det|>[[125, 645, 886, 841]]<|/det|>
+10. Line 178-187: I suggest clarifying this section. I don't follow how the statement on line 179-180 about the flow field decaying as \(1 / r^{3}\) then ties in with the following sentences and eq5 which has a \(1 / r^{4}\) scaling. These equations are introduced quickly, without proper definition of the many terms (e.g. \(r\) , \(\hat{x}\) , \(\hat{y}\) , \(\hat{\theta}\) ). Also is R superscript(+) the same as R subscript(+)? I assume so but it is a little confusing seeing both. Figure 5 is helpful, and I understand these equations are well used in many body low Reynolds number hydrodynamics (as is the method of images near a boundary), however for readers unfamiliar with this field, I think more explanation is needed.  
+
+<|ref|>text<|/ref|><|det|>[[152, 853, 885, 896]]<|/det|>
+We thank the Referee for this comment. We also added a detailed step- by- step derivation in the Supplementary Information, entitled:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[173, 106, 718, 123]]<|/det|>
+- Flow generated by a rotating sphere near a wall in the Stokes-flow regime.  
+
+<|ref|>text<|/ref|><|det|>[[151, 141, 886, 185]]<|/det|>
+In the subsection, we detail the calculation that leads to Eqs. 5 and 8. We also added the following sentence (line 207) in the main text to fully show how the \(1 / r^{4}\) scaling arises:  
+
+<|ref|>text<|/ref|><|det|>[[152, 197, 530, 214]]<|/det|>
+This scaling arises from noting that \(- y\hat{x} +x\hat{y} = r\hat{\theta}\) .  
+
+<|ref|>text<|/ref|><|det|>[[125, 232, 644, 249]]<|/det|>
+11. Line 197: 'microscopic organism' \(\mathcal{L}\) 'microscopic organisms'.  
+
+<|ref|>text<|/ref|><|det|>[[152, 261, 701, 278]]<|/det|>
+We have made this change and thank the Referee for noticing this oversight.  
+
+<|ref|>text<|/ref|><|det|>[[113, 299, 884, 343]]<|/det|>
+We sincerely thank Referee 2 for their detailed review of our manuscript, helping to clarify our experiments and theory, and making the manuscript more accessible to a broader audience.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[114, 113, 330, 131]]<|/det|>
+## Response to Referee 3  
+
+<|ref|>text<|/ref|><|det|>[[133, 150, 886, 219]]<|/det|>
+1. The manuscript "Hydrodynamic spin-orbit coupling in asynchronous optically driven micro-rotors" presents a synthetic system of active self-rotating particles which have both rotational and translational degrees of freedom.  
+
+<|ref|>text<|/ref|><|det|>[[151, 230, 886, 399]]<|/det|>
+The manuscript presents a novel biomimetic system of active rotors. Other examples of synthetic active rotors have translational, but not rotational, degrees of freedom because their active rotations closely follow an external field. By contrast, in this manuscript, the particles are able to rotate with different speeds and different phases, much like self- spinning living cells, which interact only hydrodynamically. The manuscript is well written and includes a concrete set of experimental results that convincingly test quantitative predictions for these kinds of active rotors. I recommend publication.  
+
+<|ref|>text<|/ref|><|det|>[[152, 410, 884, 453]]<|/det|>
+We thank the Referee for finding our manuscript well- written and identifying that the synthetic swimmers presented here resemble living matter.  
+
+<|ref|>text<|/ref|><|det|>[[133, 468, 886, 612]]<|/det|>
+2. Before publication, I suggest that the authors consider the following two points: The inset in Fig. 6c was not clear to me. Is it the spinning rate \(\Omega\) or its change \(\Delta \Omega\) which is plotted? We thank the Referee for pointing out a potential ambiguity in the caption of Fig 6c. In the new version of the manuscript, the caption reads: Angular speed \(\omega\) of two orbiting spheres (diameters \(6.7 \mu \mathrm{m}\) and \(5.1 \mu \mathrm{m}\) ) at different separations, along with the dependence of the spinning rate \(\Omega\) on the normalized separation (inset).  
+
+<|ref|>text<|/ref|><|det|>[[133, 639, 886, 840]]<|/det|>
+3. What are the theoretical predictions plotted along with the experimental data? The theoretical predictions in the inset of Fig. 6c are derived from Faxen's 1st and 2nd laws (Eq. 6 and 7). Using Eq. 5, the flow-field generated by a single rotor \(u(\boldsymbol {r})\) , we compute the dependence of the spinning rate of each rotor, \(\Omega_{i,j}\) , on the particles' separation \(r\) , given their asymptotic spinning rates at large separation \(\Omega_{i,j}^{0}\) . We added the following comment in the caption to clarify this: Curves show the predicted spinning rates for each particle at different separations – derived from Faxen's laws (Eqs. 6 and 7 in the main text) – given the particles' asymptotic spinning rates, \(\Omega_{i,j}^{0}\) , measured at large separations  
+
+<|ref|>text<|/ref|><|det|>[[133, 853, 884, 896]]<|/det|>
+4. The following recent work may be of interest as an analogous biological system: Odd dynamics of living chiral crystals Tan et al Nature 607, 287 (2022).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[152, 105, 886, 200]]<|/det|>
+The recent findings by the Fakhri group (now cited in our work) demonstrate a beautiful example of emergence found in asynchronously rotating particles. We aspire that the dialogue between biological and synthetic model systems will enhance our understanding of new states of matter far from equilibrium.  
+
+<|ref|>text<|/ref|><|det|>[[135, 219, 593, 235]]<|/det|>
+We thank Referee 3 for their kind words and critical comments.  
+
+<|ref|>sub_title<|/ref|><|det|>[[113, 266, 219, 284]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[119, 303, 886, 350]]<|/det|>
+[1] Bililign, E. S. et al. Motile dislocations knead odd crystals into whorls. Nature Physics 18, 212- 218 (2022). URL https://www.nature.com/articles/s41567- 021- 01429- 3.  
+
+<|ref|>text<|/ref|><|det|>[[120, 364, 886, 410]]<|/det|>
+[2] Oppenheimer, N., Stein, D. B. & Shelley, M. J. Rotating membrane inclusions crystallize through hydrodynamic and steric interactions. Physical Review Letters 123, 148101 (2019).  
+
+<|ref|>text<|/ref|><|det|>[[120, 424, 886, 494]]<|/det|>
+[3] Massana- Cid, H., Meng, F., Matsunaga, D., Golestanian, R. & Tierno, P. Tunable self- healing of magnetically propelling colloidal carpets. Nature Communications 10 (2019). URL https://doi.org/10.1038/s41467- 019- 10255- 4.  
+
+<|ref|>text<|/ref|><|det|>[[120, 509, 886, 580]]<|/det|>
+[4] Aubret, A., Youssef, M., Sacanna, S. & Palacci, J. Targeted assembly and synchronization of self- spinning microgears. Nature Physics 14, 1114- 1118 (2018). URL http://www.nature.com/articles/s41567- 018- 0227- 4.  
+
+<|ref|>text<|/ref|><|det|>[[120, 595, 886, 640]]<|/det|>
+[5] Vogel, R. et al. Synthesis and surface modification of birefringent vaterite microspheres. Langmuir 25, 11672- 11679 (2009).  
+
+<|ref|>text<|/ref|><|det|>[[120, 655, 886, 701]]<|/det|>
+[6] Friese, M. E. J., Nieminen, T. A., Heckenberg, N. R. & Rubinsztein- Dunlop, H. Optical alignment and spinning of laser- trapped microscopic particles. Nature 394, 348- 350 (1998).  
+
+<|ref|>text<|/ref|><|det|>[[120, 716, 886, 761]]<|/det|>
+[7] Donato, M. G. et al. Light- induced rotations of chiral birefringent microparticles in optical tweezers. Scientific Reports 6 (2016).  
+
+<|ref|>text<|/ref|><|det|>[[120, 777, 886, 822]]<|/det|>
+[8] Leach, J., Mushfique, H., di Leonardo, R., Padgett, M. & Cooper, J. An optically driven pump for microfluidics. Lab on a Chip 6, 735 (2006). URL http://xlink.rsc.org/?DOI=b601886f.  
+
+<|ref|>text<|/ref|><|det|>[[120, 837, 886, 882]]<|/det|>
+[9] Parkin, S. J. et al. Highly birefringent vaterite microspheres: production, characterization and applications for optical micromanipulation. Optics Express 17, 21944 (2009).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[110, 102, 888, 899]]<|/det|>
+[10] Zhang, S. et al. Ultrasensitive rotating photonic probes for complex biological systems. Optica 4, 1103 (2017). URL https://opg.optica.org/abstract.cfm?URI=optica- 4- 9- 1103. [11] Avron, J. E. Odd viscosity. Journal of Statistical Physics 92, 543- 557 (1998).[12] Nguyen, N. H., Klotsa, D., Engel, M. & Glotzer, S. C. Emergent collective phenomena in a mixture of hard shapes through active rotation. Physical Review Letters 112 (2014).[13] Yeo, K., Lushi, E. & Vlahovska, P. M. Collective dynamics in a binary mixture of hydrodynamically coupled microrotors. Physical Review Letters 114, 188301 (2015).[14] Oppenheimer, N., Stein, D. B., Ben Zion, M. Y. & Shelley, M. J. Hyperuniformity and phase enrichment in vortex and rotor assemblies. Nature Communications 13, 804 (2022).[15] Molloy, J. E. & Padgett, M. J. Lights, action: Optical tweezers. Contemporary Physics 43, 241- 258 (2002). URL https://doi.org/10.1080/00107510110116051. [16] Lushi, E. & Vlahovska, P. M. Periodic and Chaotic Orbits of Plane- Confined Microrotors in Creeping Flows. Journal of Nonlinear Science 25, 1111- 1123 (2015). URL https://doi.org/10.1007/s00332- 015- 9254- 9. [17] Kokot, G. et al. Active turbulence in a gas of self- assembled spinners. Proceedings of the National Academy of Sciences 114, 12870- 12875 (2017). URL https://pnas.org/doi/full/10.1073/pnas.1710188114. [18] Drescher, K. et al. Dancing volvox: Hydrodynamic bound states of swimming algae. Physical Review Letters 102, 1- 4 (2009). 0901.2087. [19] Tan, T. H. et al. Odd dynamics of living chiral crystals. Nature 607, 287- 293 (2022).[20] Yan, J., Bae, S. C. & Granick, S. Rotating crystals of magnetic janus colloids. Soft Matter 11, 147- 153 (2015).[21] Massana- Cid, H., Levis, D., Hernández, R. J. H., Pagonabarraga, I. & Tierno, P. Arrested phase separation in chiral fluids of colloidal spinners. Physical Review Research 3 (2021).[22] Soni, V. et al. The odd free surface flows of a colloidal chiral fluid. Nature Physics 15, 1188- 1194 (2019).[23] Hecht, E. Optics (Pearson Education, Inc, Boston, 2017), 5 ed edn.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 105, 886, 149]]<|/det|>
+[24] Blake, J. R. & Chwang, A. T. Fundamental singularities of viscous flow. Journal of Engineering Mathematics 8, 23- 29 (1974).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 83, 350, 100]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 113, 450, 130]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 144, 848, 175]]<|/det|>
+The revised manuscript now clearly shows to novelty of the contribution. I no longer have concerns about the model. I recommend publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 203, 450, 220]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 234, 852, 280]]<|/det|>
+The authors have made a good effort to answer all of my questions, and also those of the other reviewers. Furthermore, for me they have articulated more clearly the importance of the work, and so I suggest publication as is.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 308, 450, 325]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 339, 875, 370]]<|/det|>
+The authors have addressed my comments well, and I continue to support publication of this manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a2ad6e45b5f42993bdc2177d2c6ec0e5345d6529a2c6268261fc35ce6a1a175a/images_list.json

@@ -0,0 +1,138 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1r.jpg",
+    "caption": "Figure 1r. Intermediate traping experiments, kinetic profiles and reaction process.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2r.jpg",
+    "caption": "Figure 2r. The oxidation of intermediates 3 to 4.",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        90,
+        821,
+        245
+      ]
+    ],
+    "page_idx": 3
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3r.jpg",
+    "caption": "Figure 3r. \\(^1\\mathrm{H}\\) NMR and \\(^1\\mathrm{H}\\) - \\(^1\\mathrm{H}\\) COSY NMR spectra of 3 (400 MHz, CDCl3)",
+    "footnote": [],
+    "bbox": [
+      [
+        216,
+        282,
+        780,
+        870
+      ]
+    ],
+    "page_idx": 4
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4r.jpg",
+    "caption": "Figure 4r. The pathway to generate ortho-product.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        395,
+        850,
+        550
+      ]
+    ],
+    "page_idx": 4
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5r.jpg",
+    "caption": "Figure 5r. The intermediates detected in reaction system of 1a.",
+    "footnote": [],
+    "bbox": [
+      [
+        196,
+        90,
+        800,
+        243
+      ]
+    ],
+    "page_idx": 5
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6r.jpg",
+    "caption": "Figure 6r. HRMS spectrum of the key intermediate species F.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        290,
+        848,
+        377
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7r.jpg",
+    "caption": "Figure 7r. HRMS spectrum of the key intermediate species H.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        411,
+        848,
+        495
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_8r.jpg",
+    "caption": "Figure 8r. Proposed reaction mechanism for synthesis of meta-carbonyl phenols.",
+    "footnote": [],
+    "bbox": [
+      [
+        216,
+        324,
+        780,
+        727
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_9r.jpg",
+    "caption": "Figure 9r. \\(^1\\mathrm{H}\\) NMR spectrum of compound deuterium-labeling 2a (400 MHz, CDCl3)",
+    "footnote": [],
+    "bbox": [
+      [
+        179,
+        142,
+        820,
+        525
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_10r.jpg",
+    "caption": "Figure 10r. \\(^1\\mathrm{H}\\) NMR spectrum of the recovered 1a (400 MHz, CDCl3)",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 8
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a2ad6e45b5f42993bdc2177d2c6ec0e5345d6529a2c6268261fc35ce6a1a175a/supplementary_0_Peer Review File__a2ad6e45b5f42993bdc2177d2c6ec0e5345d6529a2c6268261fc35ce6a1a175a.mmd

@@ -0,0 +1,143 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Synthesis of meta- carbonyl phenols and anilines  
+
+![](images/Figure_1r.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+The manuscript presents a copper- catalyzed dehydrogenation strategy to exclusively synthesize meta- functionalized phenols and anilines from carbonyl- substituted cyclohexanes. The approach is simple, selective, uses inexpensive copper catalysts, and avoids the subsequent purification processes, which renders its practical advantages. However, the substrate scope is restricted to only carbonyl functionalized cyclohexanes, which decreases the generality and significance of the protocol. More importantly, only several basic control experiments are performed. Consequently, some key insightful explanations are missing. For instance, why are only pure meta- substitution products observed without any ortho- and para- products? What are the key factors behind the excellent selectivity? How does the carbonyl group affect the reaction? As a directing group? More experimental and computational studies should be carried out. This reviewer thinks the manuscript lacks deeper understanding of the mentioned new protocol and does not provide enough inspiring take- home messages. Therefore, the manuscript is not suitable for publishing in Nat. Comm. in its current form.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+The authors have provided satisfactory responses to the earlier comments. I recommend proceeding with the publication of the manuscript.  
+
+Reviewer #3:  
+
+Remarks to the Author:  
+
+This manuscript is an improved version of a manuscript that I have previously reviewed for another journal. The results are certainly impressive and the mechanistic implications for the field of copper catalysis are profound. My concerns have been addressed by the revisions. I have no further comments on the paper, which is well prepared.
+
+<--- Page Split --->
+
+## Response Letter  
+
+## Reviewer #1 (Remarks to the Author):  
+
+"The manuscript presents a copper- catalyzed dehydrogenation strategy to exclusively synthesize meta- functionalized phenols and anilines from carbonyl- substituted cyclohexanes. The approach is simple, selective, uses inexpensive copper catalysts, and avoids the subsequent purification processes, which renders its practical advantages. However, the substrate scope is restricted to only carbonyl functionalized cyclohexanes, which decreases the generality and significance of the protocol. More importantly, only several basic control experiments are performed. Consequently, some key insightful explanations are missing. For instance, why are only pure meta- substitution products observed without any ortho- and para- products? What are the key factors behind the excellent selectivity? How does the carbonyl group affect the reaction? As a directing group? More experimental and computational studies should be carried out. This reviewer thinks the manuscript lacks deeper understanding of the mentioned new protocol and does not provide enough inspiring take- home messages. Therefore, the manuscript is not suitable for publishing in Nat. Comm. in its current form."  
+
+Response: We thank the reviewer very much for the valuable comments.  
+
+As suggested, more mechanistic experiments were carried out.  
+
+1. Based on the experimental results and the relevant references, below is our explanation why our reaction system favors the formation of meta-products.  
+
+1) We conducted intermediate species isolating experiments under the standard reaction conditions with 1a as the substrate. To our delight, intermediate products 3, 4, and 16 were isolated (Figure 1r). To gain deeper understanding into the reaction process, we conducted a kinetic time course analysis of the reaction using \(^1\mathrm{HNMR}\) . Monitoring the reaction process revealed that \(\alpha ,\beta\) -unsaturated ketone 3 is initially formed, allylation product 4 subsequently formed, and 1,4-enedione 16 ultimately
+
+<--- Page Split --->
+
+formed. Compounds **3, 4** and **16** are subsequently consumed as the product **2a** is formed. These results provided evidence to support that **3, 4** and **16** are reaction intermediates to produce the product **2a**. We reasoned that synthesis of meta- carbonyl phenols proceeds through the following cascade (1a→3→4→16→2a). To obtain kinetic rate constants of each step, the substrate **1a** and intermediates **3, 4** and **16** were subjected to the standard conditions, respectively. The rate constants for each step were determined as follows: \(k_{1} = 2.0959 \times 10^{-4} \mathrm{min}^{-1}\) , \(k_{2} = 4.9333 \times 10^{-4} \mathrm{min}^{-1}\) , \(k_{3} = 1.8284 \times 10^{-4} \mathrm{min}^{-1}\) , \(k_{4} = 2.6889 \times 10^{-4} \mathrm{min}^{-1}\) , giving a ratio of \(k_{1} / k_{2} / k_{3} / k_{4} = 1.15:2.70:1.00:1.47\) .  
+
+![](images/Figure_2r.jpg)
+
+<center>Figure 1r. Intermediate traping experiments, kinetic profiles and reaction process. </center>  
+
+2) The selectivity of ortho-, meta- or para-products is determined by the step of the oxidation of intermediates 3 to 4. As shown in Figure 2r, because the carbon of carbonyl group possesses the partial positive charge, so \(\alpha\) -position and \(\gamma\) -position of carbonyl group possesses the partial negative charge, while \(\beta\) -position and \(\beta\) -position possess the partial positive charge. Therefore, \(\gamma\) -position is easier oxidized than \(\beta\) -position and \(\beta\) -position. From \(^1\mathrm{H}\) NMR spectrum and \(^1\mathrm{H}\) - \(^1\mathrm{H}\) COSY NMR spectrum of compound 3 (figure 3r), the proton of \(\gamma\) position has lower chemical shift than the proton of \(\beta\) ' position, which also indicates that the \(\gamma\) position is more electron-rich and easier oxidized than \(\beta\) -position.
+
+<--- Page Split --->
+![](images/Figure_3r.jpg)
+
+<center>Figure 2r. The oxidation of intermediates 3 to 4. </center>  
+
+![](images/Figure_4r.jpg)
+
+<center>Figure 3r. \(^1\mathrm{H}\) NMR and \(^1\mathrm{H}\) - \(^1\mathrm{H}\) COSY NMR spectra of 3 (400 MHz, CDCl3) </center>
+
+<--- Page Split --->
+
+3) As shown in Figure 4r, there are main three pathways to generate ortho-product. Path i is allylic oxidation of \(\beta '\) -position. However, as mentioned above, \(\gamma\) -position is easier oxidized than \(\beta\) -position and \(\beta '\) -position. In this case, our reaction system favors the formation of meta-products. Path ii involves a Michael addition/oxidation/further dehydrogenation cascade process. However, Michael addition is typically under the basic condition, which does not match our acid reaction conditions. In this case, our reaction system disfavors the formation of ortho-products. The related example of base-assisted Michael addition reactions are reported (Curr. Org. Chem. 26, 1264-1293 (2022)). Path iii is the direct oxidation of \(\alpha ,\beta\) -unsaturated ketone 3 to 1,3-diketone. Generally, the oxidation hardly proceeded due to the electrical property.  
+
+![](images/Figure_5r.jpg)
+
+<center>Figure 4r. The pathway to generate ortho-product. </center>  
+
+4) To obtain para-product, the homoallylic oxidation of \(\delta\) -position must be proceeded. As we known, homoallylic oxidation is less reactive than allylic oxidation, thus homoallylic oxidation is hard to achieve under the reaction conditions.  
+
+2. To our delight, beside compounds 2a, 3, 4 and 16 (Figure 5r), the key intermediate species F and H were also detected by high-resolution mass spectrometry (HRMS) analysis. For intermediate F: HRMS (ESI) m/z: [M + Na]\(^+\) Calc'd for C\(_{15}\)H\(_{13}\)F\(_{3}\)O\(_{3}\)CuNa 384.0005; Found 383.9997 (Figure 6r); For intermediates H: HRMS (ESI) m/z: [M + Na]\(^+\) Calc'd for C\(_{15}\)H\(_{13}\)F\(_{3}\)O\(_{3}\)Na 321.0709; Found 321.0694 (Figure 7r).
+
+<--- Page Split --->
+![](images/Figure_6r.jpg)
+
+<center>Figure 5r. The intermediates detected in reaction system of 1a. </center>  
+
+![](images/Figure_7r.jpg)
+
+<center>Figure 6r. HRMS spectrum of the key intermediate species F. </center>  
+
+![](images/Figure_8r.jpg)
+
+<center>Figure 7r. HRMS spectrum of the key intermediate species H. </center>  
+
+Based on the experimental results and the related literatures (J. Am. Chem. Soc. 141, 14889- 14897 (2019); J. Am. Chem. Soc. 131, 5044- 5045 (2009); J. Am. Chem. Soc. 133, 15300- 15303 (2011).), a plausible reaction mechanism for synthesis of meta- carbonyl phenols was proposed as figure 8r (which could also be found in Supplementary Information as Supplementary Figure 23. ). First, \(\mathrm{Cu}^{1}\) species is oxidized in situ by \(\mathrm{AgOAc}\) in the presence of TFA under oxygen atmosphere, generating \(\mathrm{Cu}^{II}(\mathrm{O}_2\mathrm{CCF}_3)_2\) ; meanwhile, the ketone suffers from the enolization. The formation of copper(II) enolate followed by the oxidation or the disproportionation gives copper(III) enolate that undergoes \(\beta\) - hydride elimination to deliver the \(\alpha ,\beta\) - unsaturated ketone 3 along with a \(\mathrm{Cu}^{III}\) - hydride intermediate. The \(\mathrm{Cu}^{III}\) - hydride species eliminates a TFA, resulting in \(\mathrm{Cu}^{I}\mathrm{O}_2\mathrm{CCF}_3\) that is reoxidized to \(\mathrm{Cu}^{II}(\mathrm{O}_2\mathrm{CCF}_3)_2\) by \(\mathrm{AgOAc}\) and \(\mathrm{O}_2\) in the presence of TFA. Subsequently, \(\alpha ,\beta\) - unsaturated ketone 3 isomerizes into diene E. The terminal \(\mathrm{C} = \mathrm{C}\) double bond of diene is activated by \(\mathrm{Cu}^{II}(\mathrm{O}_2\mathrm{CCF}_3)_2\) , and then delivers into \(\mathrm{Cu}^{II}\) species F, meanwhile losing a TFA. \(\mathrm{Cu}^{II}\) species F can be detected by HRMS (HRMS (ESI) m/z: [M + Na]\(^+\) Calcd for \(\mathrm{C}_{15}\mathrm{H}_{13}\mathrm{F}_{3}\mathrm{O}_{3}\mathrm{CuNa}\) 384.0005; Found 383.9997). \(\mathrm{Cu}^{II}\) species F undergoes the oxidation
+
+<--- Page Split --->
+
+or the disproportionation to give \(\mathrm{Cu^{III}}\) intermediate \(\mathbf{G}\) which proceeds a reductive elimination to generate intermediate \(\mathbf{H}\) and \(\mathrm{Cu^{I}O_{2}C C F_{3}}\) that proceeds the same process as above to regenerate the active \(\mathrm{Cu^{II}(O_{2}C C F_{3})_{2}}\) . Intermediate \(\mathbf{H}\) can also be detected by HRMS (HRMS (ESI) m/z: \(\mathrm{[M + Na]^{+}}\) Calcul for \(\mathrm{C_{15}H_{13}F_{3}O_{3}Na}\) 321.0709; Found 321.0694). Then intermediate \(\mathbf{H}\) is hydrolyzed to furnish the product 4 (J. Org. Chem. 86, 7603- 7608 (2021).). 4 is oxidized producing 1,4- endione 16 which then undergoes the similar procedure like 1a to 3, affording the targeted product 2a. The related examples of \(\mathrm{Cu^{II}}\) disproportionation to give the \(\mathrm{Cu^{III}}\) intermediate are reported (Angew. Chem. Int. Ed. 50, 11062- 11087 (2011); J. Am. Chem. Soc. 131, 5044- 5045 (2009).).  
+
+![](images/Figure_9r.jpg)
+
+<center>Figure 8r. Proposed reaction mechanism for synthesis of meta-carbonyl phenols. </center>  
+
+3. In our reaction system, the carbonyl group might not act as a directing group role. Actually, the carbonyl group enables the activation of the \(\alpha\) -position and \(\gamma\) -position by the formation of enole \(\mathbf{A}\) and diene \(\mathbf{E}\) (Figure 8r). In mechanistic investigation section, we carried out deuterium labeling experiments of 1a with \(\mathrm{D_2O}\) under the standard reaction conditions, giving the deuterium-labeling product with 48%
+
+<--- Page Split --->
+
+D incorporation at the \(\gamma '\) position. Besides, the recovered 1a was found to contain \(50\%\) D at the \(\alpha\) position. Please see figures 9r and 10r.  
+
+![](images/Figure_10r.jpg)
+
+<center>Figure 9r. \(^1\mathrm{H}\) NMR spectrum of compound deuterium-labeling 2a (400 MHz, CDCl3) </center>  
+
+![PLACEHOLDER_8_1]
+
+
+<--- Page Split --->
+![PLACEHOLDER_9_0]
+
+<center>Figure 10r. \(^1\mathrm{H}\) NMR spectrum of the recovered 1a (400 MHz, CDCl3) </center>  
+
+## Reviewer #2 (Remarks to the Author):  
+
+"The authors have provided satisfactory responses to the earlier comments. I recommend proceeding with the publication of the manuscript."  
+
+Response: We thank the reviewer very much for supporting the publication of this work in Nature Communications!  
+
+## Reviewer #3 (Remarks to the Author):  
+
+"This manuscript is an improved version of a manuscript that I have previously reviewed for another journal. The results are certainly impressive and the mechanistic implications for the field of copper catalysis are profound. My concerns have been addressed by the revisions. I have no further comments on the paper, which is well prepared."  
+
+Response: We thank the reviewer for supporting the publication of this work in Nature Communications!
+
+<--- Page Split --->
+
++++  
+
+Reviewer #1:  
+
+My concerns have been well addressed in the response. I recommend further proceeding for publication.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a2ad6e45b5f42993bdc2177d2c6ec0e5345d6529a2c6268261fc35ce6a1a175a/supplementary_0_Peer Review File__a2ad6e45b5f42993bdc2177d2c6ec0e5345d6529a2c6268261fc35ce6a1a175a_det.mmd

@@ -0,0 +1,172 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[105, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[105, 155, 712, 181]]<|/det|>
+Synthesis of meta- carbonyl phenols and anilines  
+
+<|ref|>image<|/ref|><|det|>[[95, 732, 261, 782]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[270, 732, 880, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 90, 285, 103]]<|/det|>
+Reviewers' Comments:  
+
+<|ref|>text<|/ref|><|det|>[[116, 119, 216, 132]]<|/det|>
+Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[116, 135, 291, 148]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 149, 875, 328]]<|/det|>
+The manuscript presents a copper- catalyzed dehydrogenation strategy to exclusively synthesize meta- functionalized phenols and anilines from carbonyl- substituted cyclohexanes. The approach is simple, selective, uses inexpensive copper catalysts, and avoids the subsequent purification processes, which renders its practical advantages. However, the substrate scope is restricted to only carbonyl functionalized cyclohexanes, which decreases the generality and significance of the protocol. More importantly, only several basic control experiments are performed. Consequently, some key insightful explanations are missing. For instance, why are only pure meta- substitution products observed without any ortho- and para- products? What are the key factors behind the excellent selectivity? How does the carbonyl group affect the reaction? As a directing group? More experimental and computational studies should be carried out. This reviewer thinks the manuscript lacks deeper understanding of the mentioned new protocol and does not provide enough inspiring take- home messages. Therefore, the manuscript is not suitable for publishing in Nat. Comm. in its current form.  
+
+<|ref|>text<|/ref|><|det|>[[115, 373, 217, 386]]<|/det|>
+Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[116, 388, 291, 401]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 402, 865, 431]]<|/det|>
+The authors have provided satisfactory responses to the earlier comments. I recommend proceeding with the publication of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[116, 477, 216, 490]]<|/det|>
+Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[116, 493, 291, 505]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 506, 857, 565]]<|/det|>
+This manuscript is an improved version of a manuscript that I have previously reviewed for another journal. The results are certainly impressive and the mechanistic implications for the field of copper catalysis are profound. My concerns have been addressed by the revisions. I have no further comments on the paper, which is well prepared.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[148, 91, 290, 108]]<|/det|>
+## Response Letter  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 155, 479, 172]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[147, 191, 852, 599]]<|/det|>
+"The manuscript presents a copper- catalyzed dehydrogenation strategy to exclusively synthesize meta- functionalized phenols and anilines from carbonyl- substituted cyclohexanes. The approach is simple, selective, uses inexpensive copper catalysts, and avoids the subsequent purification processes, which renders its practical advantages. However, the substrate scope is restricted to only carbonyl functionalized cyclohexanes, which decreases the generality and significance of the protocol. More importantly, only several basic control experiments are performed. Consequently, some key insightful explanations are missing. For instance, why are only pure meta- substitution products observed without any ortho- and para- products? What are the key factors behind the excellent selectivity? How does the carbonyl group affect the reaction? As a directing group? More experimental and computational studies should be carried out. This reviewer thinks the manuscript lacks deeper understanding of the mentioned new protocol and does not provide enough inspiring take- home messages. Therefore, the manuscript is not suitable for publishing in Nat. Comm. in its current form."  
+
+<|ref|>text<|/ref|><|det|>[[148, 617, 738, 636]]<|/det|>
+Response: We thank the reviewer very much for the valuable comments.  
+
+<|ref|>text<|/ref|><|det|>[[148, 654, 655, 672]]<|/det|>
+As suggested, more mechanistic experiments were carried out.  
+
+<|ref|>text<|/ref|><|det|>[[148, 691, 852, 737]]<|/det|>
+1. Based on the experimental results and the relevant references, below is our explanation why our reaction system favors the formation of meta-products.  
+
+<|ref|>text<|/ref|><|det|>[[147, 746, 854, 905]]<|/det|>
+1) We conducted intermediate species isolating experiments under the standard reaction conditions with 1a as the substrate. To our delight, intermediate products 3, 4, and 16 were isolated (Figure 1r). To gain deeper understanding into the reaction process, we conducted a kinetic time course analysis of the reaction using \(^1\mathrm{HNMR}\) . Monitoring the reaction process revealed that \(\alpha ,\beta\) -unsaturated ketone 3 is initially formed, allylation product 4 subsequently formed, and 1,4-enedione 16 ultimately
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 853, 330]]<|/det|>
+formed. Compounds **3, 4** and **16** are subsequently consumed as the product **2a** is formed. These results provided evidence to support that **3, 4** and **16** are reaction intermediates to produce the product **2a**. We reasoned that synthesis of meta- carbonyl phenols proceeds through the following cascade (1a→3→4→16→2a). To obtain kinetic rate constants of each step, the substrate **1a** and intermediates **3, 4** and **16** were subjected to the standard conditions, respectively. The rate constants for each step were determined as follows: \(k_{1} = 2.0959 \times 10^{-4} \mathrm{min}^{-1}\) , \(k_{2} = 4.9333 \times 10^{-4} \mathrm{min}^{-1}\) , \(k_{3} = 1.8284 \times 10^{-4} \mathrm{min}^{-1}\) , \(k_{4} = 2.6889 \times 10^{-4} \mathrm{min}^{-1}\) , giving a ratio of \(k_{1} / k_{2} / k_{3} / k_{4} = 1.15:2.70:1.00:1.47\) .  
+
+<|ref|>image<|/ref|><|det|>[[155, 345, 848, 608]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[163, 626, 831, 645]]<|/det|>
+<center>Figure 1r. Intermediate traping experiments, kinetic profiles and reaction process. </center>  
+
+<|ref|>text<|/ref|><|det|>[[147, 659, 853, 901]]<|/det|>
+2) The selectivity of ortho-, meta- or para-products is determined by the step of the oxidation of intermediates 3 to 4. As shown in Figure 2r, because the carbon of carbonyl group possesses the partial positive charge, so \(\alpha\) -position and \(\gamma\) -position of carbonyl group possesses the partial negative charge, while \(\beta\) -position and \(\beta\) -position possess the partial positive charge. Therefore, \(\gamma\) -position is easier oxidized than \(\beta\) -position and \(\beta\) -position. From \(^1\mathrm{H}\) NMR spectrum and \(^1\mathrm{H}\) - \(^1\mathrm{H}\) COSY NMR spectrum of compound 3 (figure 3r), the proton of \(\gamma\) position has lower chemical shift than the proton of \(\beta\) ' position, which also indicates that the \(\gamma\) position is more electron-rich and easier oxidized than \(\beta\) -position.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[175, 90, 821, 245]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[298, 264, 696, 280]]<|/det|>
+<center>Figure 2r. The oxidation of intermediates 3 to 4. </center>  
+
+<|ref|>image<|/ref|><|det|>[[216, 282, 780, 870]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[188, 882, 808, 901]]<|/det|>
+<center>Figure 3r. \(^1\mathrm{H}\) NMR and \(^1\mathrm{H}\) - \(^1\mathrm{H}\) COSY NMR spectra of 3 (400 MHz, CDCl3) </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 852, 387]]<|/det|>
+3) As shown in Figure 4r, there are main three pathways to generate ortho-product. Path i is allylic oxidation of \(\beta '\) -position. However, as mentioned above, \(\gamma\) -position is easier oxidized than \(\beta\) -position and \(\beta '\) -position. In this case, our reaction system favors the formation of meta-products. Path ii involves a Michael addition/oxidation/further dehydrogenation cascade process. However, Michael addition is typically under the basic condition, which does not match our acid reaction conditions. In this case, our reaction system disfavors the formation of ortho-products. The related example of base-assisted Michael addition reactions are reported (Curr. Org. Chem. 26, 1264-1293 (2022)). Path iii is the direct oxidation of \(\alpha ,\beta\) -unsaturated ketone 3 to 1,3-diketone. Generally, the oxidation hardly proceeded due to the electrical property.  
+
+<|ref|>image<|/ref|><|det|>[[147, 395, 850, 550]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[290, 563, 706, 581]]<|/det|>
+<center>Figure 4r. The pathway to generate ortho-product. </center>  
+
+<|ref|>text<|/ref|><|det|>[[147, 595, 852, 670]]<|/det|>
+4) To obtain para-product, the homoallylic oxidation of \(\delta\) -position must be proceeded. As we known, homoallylic oxidation is less reactive than allylic oxidation, thus homoallylic oxidation is hard to achieve under the reaction conditions.  
+
+<|ref|>text<|/ref|><|det|>[[147, 679, 852, 838]]<|/det|>
+2. To our delight, beside compounds 2a, 3, 4 and 16 (Figure 5r), the key intermediate species F and H were also detected by high-resolution mass spectrometry (HRMS) analysis. For intermediate F: HRMS (ESI) m/z: [M + Na]\(^+\) Calc'd for C\(_{15}\)H\(_{13}\)F\(_{3}\)O\(_{3}\)CuNa 384.0005; Found 383.9997 (Figure 6r); For intermediates H: HRMS (ESI) m/z: [M + Na]\(^+\) Calc'd for C\(_{15}\)H\(_{13}\)F\(_{3}\)O\(_{3}\)Na 321.0709; Found 321.0694 (Figure 7r).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[196, 90, 800, 243]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[242, 256, 754, 274]]<|/det|>
+<center>Figure 5r. The intermediates detected in reaction system of 1a. </center>  
+
+<|ref|>image<|/ref|><|det|>[[147, 290, 848, 377]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 384, 653, 402]]<|/det|>
+<center>Figure 6r. HRMS spectrum of the key intermediate species F. </center>  
+
+<|ref|>image<|/ref|><|det|>[[147, 411, 848, 495]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 504, 657, 523]]<|/det|>
+<center>Figure 7r. HRMS spectrum of the key intermediate species H. </center>  
+
+<|ref|>text<|/ref|><|det|>[[147, 531, 852, 907]]<|/det|>
+Based on the experimental results and the related literatures (J. Am. Chem. Soc. 141, 14889- 14897 (2019); J. Am. Chem. Soc. 131, 5044- 5045 (2009); J. Am. Chem. Soc. 133, 15300- 15303 (2011).), a plausible reaction mechanism for synthesis of meta- carbonyl phenols was proposed as figure 8r (which could also be found in Supplementary Information as Supplementary Figure 23. ). First, \(\mathrm{Cu}^{1}\) species is oxidized in situ by \(\mathrm{AgOAc}\) in the presence of TFA under oxygen atmosphere, generating \(\mathrm{Cu}^{II}(\mathrm{O}_2\mathrm{CCF}_3)_2\) ; meanwhile, the ketone suffers from the enolization. The formation of copper(II) enolate followed by the oxidation or the disproportionation gives copper(III) enolate that undergoes \(\beta\) - hydride elimination to deliver the \(\alpha ,\beta\) - unsaturated ketone 3 along with a \(\mathrm{Cu}^{III}\) - hydride intermediate. The \(\mathrm{Cu}^{III}\) - hydride species eliminates a TFA, resulting in \(\mathrm{Cu}^{I}\mathrm{O}_2\mathrm{CCF}_3\) that is reoxidized to \(\mathrm{Cu}^{II}(\mathrm{O}_2\mathrm{CCF}_3)_2\) by \(\mathrm{AgOAc}\) and \(\mathrm{O}_2\) in the presence of TFA. Subsequently, \(\alpha ,\beta\) - unsaturated ketone 3 isomerizes into diene E. The terminal \(\mathrm{C} = \mathrm{C}\) double bond of diene is activated by \(\mathrm{Cu}^{II}(\mathrm{O}_2\mathrm{CCF}_3)_2\) , and then delivers into \(\mathrm{Cu}^{II}\) species F, meanwhile losing a TFA. \(\mathrm{Cu}^{II}\) species F can be detected by HRMS (HRMS (ESI) m/z: [M + Na]\(^+\) Calcd for \(\mathrm{C}_{15}\mathrm{H}_{13}\mathrm{F}_{3}\mathrm{O}_{3}\mathrm{CuNa}\) 384.0005; Found 383.9997). \(\mathrm{Cu}^{II}\) species F undergoes the oxidation
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 88, 852, 320]]<|/det|>
+or the disproportionation to give \(\mathrm{Cu^{III}}\) intermediate \(\mathbf{G}\) which proceeds a reductive elimination to generate intermediate \(\mathbf{H}\) and \(\mathrm{Cu^{I}O_{2}C C F_{3}}\) that proceeds the same process as above to regenerate the active \(\mathrm{Cu^{II}(O_{2}C C F_{3})_{2}}\) . Intermediate \(\mathbf{H}\) can also be detected by HRMS (HRMS (ESI) m/z: \(\mathrm{[M + Na]^{+}}\) Calcul for \(\mathrm{C_{15}H_{13}F_{3}O_{3}Na}\) 321.0709; Found 321.0694). Then intermediate \(\mathbf{H}\) is hydrolyzed to furnish the product 4 (J. Org. Chem. 86, 7603- 7608 (2021).). 4 is oxidized producing 1,4- endione 16 which then undergoes the similar procedure like 1a to 3, affording the targeted product 2a. The related examples of \(\mathrm{Cu^{II}}\) disproportionation to give the \(\mathrm{Cu^{III}}\) intermediate are reported (Angew. Chem. Int. Ed. 50, 11062- 11087 (2011); J. Am. Chem. Soc. 131, 5044- 5045 (2009).).  
+
+<|ref|>image<|/ref|><|det|>[[216, 324, 780, 727]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[167, 731, 826, 749]]<|/det|>
+<center>Figure 8r. Proposed reaction mechanism for synthesis of meta-carbonyl phenols. </center>  
+
+<|ref|>text<|/ref|><|det|>[[147, 762, 866, 892]]<|/det|>
+3. In our reaction system, the carbonyl group might not act as a directing group role. Actually, the carbonyl group enables the activation of the \(\alpha\) -position and \(\gamma\) -position by the formation of enole \(\mathbf{A}\) and diene \(\mathbf{E}\) (Figure 8r). In mechanistic investigation section, we carried out deuterium labeling experiments of 1a with \(\mathrm{D_2O}\) under the standard reaction conditions, giving the deuterium-labeling product with 48%
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 89, 850, 135]]<|/det|>
+D incorporation at the \(\gamma '\) position. Besides, the recovered 1a was found to contain \(50\%\) D at the \(\alpha\) position. Please see figures 9r and 10r.  
+
+<|ref|>image<|/ref|><|det|>[[179, 142, 820, 525]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[147, 533, 848, 552]]<|/det|>
+<center>Figure 9r. \(^1\mathrm{H}\) NMR spectrum of compound deuterium-labeling 2a (400 MHz, CDCl3) </center>  
+
+<|ref|>image<|/ref|><|det|>[[333, 561, 664, 625]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[195, 90, 802, 393]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[210, 404, 786, 423]]<|/det|>
+<center>Figure 10r. \(^1\mathrm{H}\) NMR spectrum of the recovered 1a (400 MHz, CDCl3) </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 442, 480, 460]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 479, 850, 525]]<|/det|>
+"The authors have provided satisfactory responses to the earlier comments. I recommend proceeding with the publication of the manuscript."  
+
+<|ref|>text<|/ref|><|det|>[[148, 544, 850, 589]]<|/det|>
+Response: We thank the reviewer very much for supporting the publication of this work in Nature Communications!  
+
+<|ref|>sub_title<|/ref|><|det|>[[148, 608, 480, 626]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[147, 644, 851, 775]]<|/det|>
+"This manuscript is an improved version of a manuscript that I have previously reviewed for another journal. The results are certainly impressive and the mechanistic implications for the field of copper catalysis are profound. My concerns have been addressed by the revisions. I have no further comments on the paper, which is well prepared."  
+
+<|ref|>text<|/ref|><|det|>[[148, 794, 850, 839]]<|/det|>
+Response: We thank the reviewer for supporting the publication of this work in Nature Communications!
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 92, 135, 103]]<|/det|>
++++  
+
+<|ref|>text<|/ref|><|det|>[[115, 109, 211, 123]]<|/det|>
+Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[112, 127, 881, 144]]<|/det|>
+My concerns have been well addressed in the response. I recommend further proceeding for publication.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a2e4cfd876998db6d4acbcedc4a2212a495eea77357accea0026d65cd7e6325f/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__a2e4cfd876998db6d4acbcedc4a2212a495eea77357accea0026d65cd7e6325f/supplementary_0_Peer Review File__a2e4cfd876998db6d4acbcedc4a2212a495eea77357accea0026d65cd7e6325f.mmd

@@ -0,0 +1,380 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+An opioid- gated thalamoaccumbal circuit for the suppression of reward seeking in mice  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS</B>  
+
+Reviewer #1 (Remarks to the Author):  
+
+The present study by Vollmer, Green et al. test the hypothesis that a thalamostriatal circuit inhibits reward seeking during exposure to fear- provoking stimuli and that opioid receptor activation promotes risky behavior in the face of these stimuli by inhibiting this circuit via a presynaptic site of action. First the authors demonstrate that thalamostriatal PVT cells display heterogenous calcium responses to lever press activity (type 1 and 3 responses) associated with sucrose administration, and some ensembles are inhibited in a tonic manner. The authors demonstrate that optogenetic activation of PVT- NAcc pathway inhibited sucrose self- administration, while optogenetic inhibition of thalamostriatal neurons reversed TMT, yohimbine, and extinction- mediated reductions in sucrose self- administration. The authors subsequently demonstrated that thalamostriatal neurons innervate MSNs and PV interneurons, with the latter containing CP- AMPARs. The authors then show that the effects of optogenetic activation of PVT to NAcc afferents on sucrose self- administration is blocked by antagonism of CP- AMPARs in the NAcc and chemogenetic inhibition of NAcc PV neurons. The authors claim that thalamostriatal neurons and their terminals in the NAcc express MOR and that acute heroin administration decreases ensemble decoding of sucrose self- administration and blocks the suppressive effects of PVT to NAcc pathway stimulation and suppression induced by fear provoking stimuli. Lastly, the authors demonstrate that intra- NAcc DAMGO injections inhibit optogenetic -, TMT- and yohimbine- induced decreases in sucrose self- administration. They show that synapses from PVT to NAcc PV neurons are inhibited by DAMGO and that MOR deletion from PVT blocks the ability of intra- NAcc DAMGO from reversing the inhibitory effects of thalamostriatal optogenetic stimulation on sucrose self- administration. Overall, the study is of importance and tackles an important question. The authors test their hypothesis using an impressive set of interdisciplinary approaches. However, the authors make strong conclusions about their data that are not entirely supported by the present findings. Specifically, there are significant gaps that need to be addressed for the conclusions in their present form to be drawn, which include additional experiments and/or dialing back some of the interpretations and discussing caveats and alternative hypotheses. Below are some comments that will aid the authors in strengthening their manuscript and improving the cohesiveness of the study.  
+
+## Comments  
+
+- There is a disconnect between the data across the different figures. The authors make strong claims that opioids acting through presynaptic receptors disrupt PVT inhibition, but this is largely decoupled from recordings of cell bodies and the effect of heroin presented earlier in the paper. Opioid receptors on terminals would inhibit terminal release and any potential influence on cell body activity in thalamostriatal neurons would likely be an indirect consequence. That is, is there any relationship between what MORs are doing at terminals and what is happening in the cell bodies of thalamostriatal neurons? Moreover, it is unclear whether the exciting findings in Fig 5 are related to the findings with heroin. Additional studies linking the role of MORs on PVT terminals, activity of thalamostriatal cells, and how this is involved in heroin's effects would be needed to reach the interpretations presented, such as
+
+<--- Page Split --->
+
+deleting MOR from thalamostriatal neurons and determining if that impacts the effect of heroin.  
+
+- The authors should discuss the caveat that IEM-1640 may be acting on other cell types not examined in the present paper or that CP-AMPARs may become engaged downstream of PVT inputs to NAcc, including the PV neurons. The authors' interpretation should be much more cautious without additional evidence.  
+
+- Caveats associated with electrophysiological characterization of synaptic connections onto NAcc neurons should be discussed/addressed. A between subjects design comparing MSNs and PV neurons is subject to variability from ChR2 expression, and as such additional measures would be useful for determining synaptic strength differences between cell types. Moreover, since virus was used to label D2 cells the caveat that unlabeled cells may constitute putative D1 and unlabeled D2 cells should be addressed. That PV neurons display larger currents relative to MSNs is not surprising, since PV interneurons are broadly characterized by the presence of CP-AMPARs and enhanced excitatory synaptic strength in striatal structures and cortex. Furthermore, NAcc PV neurons been implicated in mediating aversive behavior independent of the PVT (see work by Morales group (NIDA) for example).  
+
+- MOR activation may also inhibit glutamate release onto non-PV targets (e.g. MSNs), which may be contributing to the observed behavioral effects. The authors do not examine these synapses and make strong conclusions about PVT inputs to PV that warrant further investigation. Furthermore, their physiology does not directly demonstrate presynaptic effects with the data presented, though it's implied.  
+
+- In fig 1, The authors describe a tonic decrease in ensemble 2 and 3 activity over the course of the sucrose SA session that is blocked by fear provoking stimuli and extinction and reappears upon reinstatement. However, the emphasis is placed on ensemble 3 activity aligned to lever pressing based on decoding accuracy. Further discussion on this would be useful for the reader.  
+
+- The authors state that ensemble 1 and 3 decoding accuracy is decreased after heroin due to decreased changes in calcium activity, which is consistent with 4H. However, 4G suggest that decoding accuracy in ensemble 1 is increased relative to saline day. The same mismatch appears for ensemble 2, but not as pronounced as ensemble 1.  
+
+- The switch to real time place preference evoked by opto PVT to NAcc stim with heroin injection is very interesting and in some ways may be relevant to the potentiation of sucrose self-administration that is observed in response to opto stim, TMT, and yohimbine in fig 4 and 5. This suggest that MOR activity may be unmasking a population of pro-sucrose administration neurons upon inhibition of MOR sensitive neurons.  
+
+- In supplementary figure s2, cluster 1 seems to contain a subset of cells that are briefly inhibited prior to lever press and are excited during reward delivery late in acquisition and inhibited during
+
+<--- Page Split --->
+
+reinstatement, suggesting they are distinct from other cells in cluster 1. This cluster in some ways has both cluster 1 and 3 properties.  
+
+- The MOR statining of PVT neurons and their terminals could be due to chance alone as MOR-like immunoreactivity appears to be widespread across the image. In the zoom within the representative images it appears that background from adjacent regions was cropped or subtracted. Immunostaining for MOR with this in MOR loxP mice would be useful for validating the antibody and genetic approach.  
+
+- NAcc PV cells are sparse relative to other interneuron populations. Moreover, PV-Cre mice used in the present study have incomplete genetic penetrance in the NAcc and may select for subtypes of NAcc fast-spiking interneurons in addition to non-PV cells with CP-AMPARs (see Adam Carter's work and Yan Dong). The authors should mention the incomplete penetrance to let the reader know that this may pertain a selective sub-population of PV cells and overall broaden the discussion of how PV cells are limited in density in NAcc relative to striatum in the context of the present work.  
+
+- The authors should show the lever press – aligned population activity during the fear provoking stimuli in addition to the change across the session that is reported.  
+
+- The intrinsic properties of the MSNs in the representative traces do not show the well-characterized regular spiking properties of healthy MSNs ex-vivo, raising the possibility the cells / preparation was not optimal.  
+
+- The authors should cite relevant work on thalamostriatal circuitry where appropriate from the Penzo group at NIMH describing a role for thalamostriatal inputs in promoting avoidance and homeostatic feeding behavior.  
+
+- The ordering of the supplemental data makes it a little bit difficult to follow. I understand clumping all the inactive lever presses into one figure but other figures are out of order relative to how the data are presented in the main text and figures.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Vollmer et al. report the results of an interesting series of experiments providing much needed new knowledge on the organisation of the thalamostriatal pathway from paraventricular thalamus to nucleus accumbens in reward seeking behavior. The authors show, quite convincingly, that this circuit involves inputs onto parvalbumin neurons in the accumbens, is enriched in calcium permeable AMPA receptors, is opioid sensitive, and generally inhibits reward seeking across a variety of environmental conditions. These are all important knowledge gains.  
+
+There is a lot of work in this manuscript, some of it is very clever, all of it is very well done, and the
+
+<--- Page Split --->
+
+manuscript is very well presented. It significantly extends the field. I think the manuscript will be of interest to many readers and that it adds critical new knowledge. I enjoyed reading it and I think others will too. I had only the following comments on the manuscript:  
+
+1. It is not clear to me that the manuscript speaks to "risky reward motivated behaviors". For example, the predator odor is presented in the home cages, not when mice are seeking sucrose. So, there is little to no 'risk' here. There is stress etc, but risk implies an adverse consequence of seeking sucrose. Moreover, the authors show that the same circuit mechanism suppress sucrose seeing after extinction training, which clearly has no risk at all. I am well persuaded by these data that this circuit is a general one for suppressing reward seeking behaviour (as predicted by others), but I am less persuaded that the reader learns much about risk. I suggest that "risky" is removed from the title, abstract etc.  
+
+2. "goal-directed'. In a number of places, the selective effects of manipulations on the 'active' lever and no effect on the 'inactive' lever are used to support the claim that behavior and the effects of manipulations are 'goal directed'. This is hard to evaluate because there is no actual test of the goal directedness of the behavior here (e.g., contingency degradation; outcome devaluation). Moreover, it appears the sucrose delivery tube was physically located under the 'active' lever (Figure 1a), which reduces the utility and relevance of the 'inactive' lever to the task. I think what the authors mean is that manipulations were lever or behaviorally specific. I am certainly persuaded of this. It may be more helpful to use these terms rather than imputing untested mental states to the animals.  
+
+3. The decoder results are interesting, but these are based on only using the 1s before each lever press vs a random 1-s control. I wondered why only a 1s period was used and why this 1s was chosen over other possible durations? This could be better justified. Also, did the authors test a decoder using any other pre-lever press durations? If yes, how did they control for these exploratory analyses in their final statistical analyses?  
+
+4. The identification of these effects to PV interneurons is neat. PV neurons are sparse in the accumbens, so it is interesting that their manipulations here have such strong effects. However, the reader never learns how selective the hM4Di was to PV neurons in the accumbens in Figure 3i. Can the authors confirm that other cells in the accumbens did not express the DREADD?  
+
+## Minor  
+
+1. I may have missed it, but do the authors state the number of cells in each cluster in Figure 1i or I Figure 4? These are worth clearly stating. 
+2. State dose of DAMGO used in the manuscript proper. 
+3. Line 240: should refer to Figure 5H not 5F
+
+<--- Page Split --->
+
+The manuscript by Vollmer et al. describes the role of PVT- NAC projection in the suppression of sucrose seeking. Using the head- fixed sucrose seeking paradigm, authors performed two- photon imaging and optogenetic manipulation of PVT neurons projecting to NAC. Furthermore, the authors examine the role of the opioid receptors in this neuronal type. Overall, all experiments are well designed and performed. However, some interpretation is not fully convincing.  
+
+1. It is difficult to connect the contribution of PVT projection to NAC PV neurons with behavioral changes in sucrose seeking. Especially, since several papers showed that PVT inputs preferentially modify D2- MSN in opioid-induced behavioral changes, it is not clear how the contribution of PVT projection is specifically modulating NAC PV neurons in sucrose seeking. Unless authors can image three different neurons (D1-MSN, D2-MSN, PV neurons) while modulating PVT inputs, this interpretation is not fully convincing. Moreover, since it is very likely that the chemogenetic inhibition of PV neurons (Figure 3k-o) will have massive effects on NAC circuitry already, the optogenetic stimulation of PVT together with PV chemogenetic manipulation may not be PV neuron dependent.  
+
+2. Authors found that CP-AMPARs are selectively located in PV neurons in NAC. However, the authors didn't show any effort to describe the potential roles of these CP-AMPARs on sucrose seeking. What is the role of PVT inputs for CP-AMPARs dependent PV neuronal activity?  
+
+3. The effects of DAMGO application on synaptic transmission (Figure 5) can be due to the postsynaptic effects since NAC MSNs also express MORs. More extensive analysis of synaptic transmission is needed to fully validate the presynaptic roles of MORs in PVT axon fibers. Again, the behavioral effects of DAMGO (Figure 5h) can be also due to the postsynaptic effects due to the DAMGO-induced changes in postsynaptic MSNs.  
+
+4. The systemic injection of heroin can induce neural adaptation in other brain areas, not just PVT-NAC inputs. These other adaptations can change the efficacy of PVT-NAC stimulation with heroin injection. Thus, it is difficult to know how the heroin injection can reverse the effect of PVT-NAC stimulation clearly. Authors should show how PVT-NAC projections are specifically changed by the systemic heroin injection.  
+
+5. The effect of the optogenetic manipulation of PVT to NAC projection on TMT or Yohimbe mediated suppression of sucrose seeking was not examined.  
+
+6. Authors can fully take advantage of two-photon imaging. The longitudinal imaging of single neurons can provide the trends in the adaptation of individual neurons.
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+We would like to thank the Reviewers for their outstanding suggestions, which we have addressed point- by- point below (specific concerns are in BOLD text, followed by our responses and additions to the manuscript in ITALICIZED font). In response to these concerns and suggestions, we provide new data, analyses, and clarifications which we feel have significantly strengthened the manuscript.  
+
+## Reviewer #1 (Recommendations for the authors):  
+
+The present study by Vollmer, Green et al. test the hypothesis that a thalamostriatal circuit inhibits reward seeking during exposure to fear- provoking stimuli and that opioid receptor activation promotes risky behavior in the face of these stimuli by inhibiting this circuit via a presynaptic site of action. First the authors demonstrate that thalamostriatal PVT cells display heterogenous calcium responses to lever press activity (type 1 and 3 responses) associated with sucrose administration, and some ensembles are inhibited in a tonic manner. The authors demonstrate that optogenetic activation of PVT- NAcc pathway inhibited sucrose self- administration, while optogenetic inhibition of thalamostriatal neurons reversed TMT, yohimbine, and extinction- mediated reductions in sucrose self- administration. The authors subsequently demonstrated that thalamostriatal neurons innervate MSNs and PV interneurons, with the latter containing CP- AMPARs. The authors then show that the effects of optogenetic activation of PVT to NAcc afferents on sucrose self- administration is blocked by antagonism of CP- AMPARs in the NAcc and chemogenetic inhibition of NAcc PV neurons. The authors claim that thalamostriatal neurons and their terminals in the NAcc express MOR and that acute heroin administration decreases ensemble decoding of sucrose self- administration and blocks the suppressive effects of PVT to NAcc pathway stimulation and suppression induced by fear provoking stimuli. Lastly, the authors demonstrate that intra- NAcc DAMGO injections inhibit optogenetic -, TMT- and yohimbine- induced decreases in sucrose self- administration. They show that synapses from PVT to NAcc PV neurons are inhibited by DAMGO and that MOR deletion from PVT blocks the ability of intra- NAcc DAMGO from reversing the inhibitory effects of thalamostriatal optogenetic stimulation on sucrose self- administration. Overall, the study is of importance and tackles an important question. The authors test their hypothesis using an impressive set of interdisciplinary approaches. However, the authors make strong conclusions about their data that are not entirely supported by the present findings. Specifically, there are significant gaps that need to be addressed for the conclusions in their present form to be drawn, which include additional experiments and/or dialing back some of the interpretations and discussing caveats and alternative hypotheses. Below are some comments that will aid the authors in strengthening their manuscript and improving the cohesiveness of the study.  
+
+We want to thank Reviewer #1 for their excitement for our findings and very helpful feedback, which has aided us in improving the quality of our paper. Based on the concerns raised below, we have added 10 new experiments as well as new data analysis for existing imaging datasets. Furthermore, we have adjusted the language of the manuscript to ensure that caveats are addressed and more appropriately acknowledged.  
+
+1. There is a disconnect between the data across the different figures. The authors make strong claims that opioids acting through presynaptic receptors disrupt PVT inhibition, but this is largely decoupled from recordings of cell bodies and the effect of heroin presented earlier in the paper. Opioid receptors on terminals would inhibit terminal release
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+and any potential influence on cell body activity in thalamostriatal neurons would likely be an indirect consequence. That is, is there any relationship between what MORs are doing at terminals and what is happening in the cell bodies of thalamostriatal neurons? Moreover, it is unclear whether the exciting findings in Fig 5 are related to the findings with heroin. Additional studies linking the role of MORs on PVT terminals, activity of thalamostriatal cells, and how this is involved in heroin's effects would be needed to reach the interpretations presented, such as deleting MOR from thalamostriatal neurons and determining if that impacts the effect of heroin.  
+
+We agree with the Reviewer and have gone to great lengths to demonstrate that the findings across these figures (now Figure 4- 6) are related. First, we use two- photon imaging and slice electrophysiology to show that both heroin injection and DAMGO application (experiments 1,2) reduces the activity/excitability of \(\mathrm{PVT} \rightarrow \mathrm{NAc}\) somata and (experiment 3) reduces downstream synaptic inputs to NAC neurons in a manner that is prevented by Cre- dependent knockout of PVT \(\mu\) - ORs (see Figure 4, Figure 6, Supplementary Figure 9). Second, we now show that behavioral disinhibition caused by (experiments 4- 6) heroin injection or (experiments 7- 8) intra- NAC DAMGO infusion is reversed by Cre- dependent \(\mu\) - OR knockout in PVT, regardless of the behavioral suppressor administered (see Figure 5; Figure 6). These exciting new results lead us to conclude that \(\mu\) - ORs on PVT \(\rightarrow \mathrm{NAc}\) neurons (possibly both somatic and on axon terminals) are required for opioid- driven behavioral disinhibition. However, we cannot rule out that opioids could also act elsewhere to drive behavioral disinhibition, including post- synaptically in NAC. Thus, we have lightened the language of the manuscript and include additional discussion (see page 8):  
+
+"Despite our findings that PVT \(\mu\) - OR knockout prevents systemic heroin or intra- NAC DAMGO infusions from disinhibiting sucrose- seeking behaviors, caveats regarding the specificity of our results should be considered. First, we cannot dissociate whether opioid- driven \(\mu\) - OR activation on PVT somata or PVT \(\rightarrow \mathrm{NAc}\) axon terminals are required for our observed behavioral effects. Considering that heroin and DAMGO can dramatically reduce activity at both PVT \(\rightarrow \mathrm{NAc}\) somata and downstream synapses, it is possible that both mechanisms are involved. Second, additional mechanisms could also contribute to opioid- induced behavioral disinhibition, such as \(\mu\) - OR activation elsewhere in the brain and in other NAC cell types that express \(\mu\) - ORs<sup>54- 56</sup>. Overall, while our data suggests that opioid- induced inhibition of PVT \(\rightarrow \mathrm{NAc}\) neurons can disinhibit maladaptive behavioral actions, whether these effects are isolated to PVT \(\rightarrow \mathrm{NAc}\) somata and/or synapses has yet to be established."  
+
+2. The authors should discuss the caveat that IEM-1640 may be acting on other cell types not examined in the present paper or that CP-AMPARs may become engaged downstream of PVT inputs to NAC, including the PV neurons. The authors' interpretation should be much more cautious without additional evidence.  
+
+We agree. While our data showing that NAC CP- AMPAr blockade results in behavioral disinhibition led us to examine PV interneuron involvement, we do not definitively state that CP- AMPArs on PV- interneurons are mediating the effects of IEM- 1460. Furthermore, we now address caveats within the results (see page 5):
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+"While we therefore hypothesize that PVT \(\rightarrow\) NAc neurons act selectively at CP- AMPAr- enriched synapses on PV interneurons to suppress behavior, it should be noted that other synapses and NAc cell types may also be involved (see discussion)."  
+
+As well as the discussion (see page 7):  
+
+"While we used pharmacology and chemogenetics to target CP- AMPArs and PV interneurons, respectively, it is possible that these methods could have off- target effects. For example, non- PV cells within NAc could express CP- AMPArs, and thus CP- AMPAr antagonism may act on other FSIs or non- PV cell types."  
+
+3. Caveats associated with electrophysiological characterization of synaptic connections onto NAcc neurons should be discussed/addressed. A between subjects design comparing MSNs and PV neurons is subject to variability from ChR2 expression, and as such additional measures would be useful for determining synaptic strength differences between cell types. Moreover, since virus was used to label D2 cells the caveat that unlabeled cells may constitute putative D1 and unlabeled D2 cells should be addressed. That PV neurons display larger currents relative to MSNs is not surprising, since PV interneurons are broadly characterized by the presence of CP-AMPARs and enhanced excitatory synaptic strength in striatal structures and cortex. Furthermore, NAcc PV neurons been implicated in mediating aversive behavior independent of the PVT (see work by Morales group (NIDA) for example).  
+
+We agree and have lightened the language of the manuscript to reflect putative targeting of D1- MSNs. Furthermore, we have added the following paragraph to the discussion which addresses the design of this electrophysiological experiment (see page 7):  
+
+"Our electrophysiological data show that accumbal PV interneurons, as compared to putative D1- and D2- MSNs, receive elevated excitatory drive from PVT neurons, although there are potential caveats to our viral targeting techniques. First, we used D2- Cre and PV- Cre transgenic mice to target MSNs or PV interneurons, respectively, which could have led to variability in ChrinsonR expression between groups of animals. Second, in our D2- Cre transgenic mice, we classified non- fluorescent neurons as putative D1- MSNs, whereas these cells could have been unlabeled D2- MSNs or other cell populations. Despite these caveats, our findings are consistent with previous literature showing that accumbal fast- spiking interneurons (FSIs) receive greater excitatory input from PVT as compared with undefined MSNs using a within- subject design<sup>30</sup>. However, further studies comparing PVT synaptic input to each specific cell- type, including other subclasses of interneurons, within subjects could improve our understanding of PVT \(\rightarrow\) NAc circuit biology."  
+
+4. MOR activation may also inhibit glutamate release onto non-PV targets (e.g. MSNs), which may be contributing to the observed behavioral effects. The authors do not examine these synapses and make strong conclusions about PVT inputs to PV that warrant further
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+investigation. Furthermore, their physiology does not directly demonstrate presynaptic effects with the data presented, though it's implied.  
+
+This was an excellent point made by the Reviewer which we now address with new data and additional discussion. Using patch clamp electrophysiology, we confirm that DAMGO decreases PVT excitatory input to both accumbal PV- INs and MSNs in a manner that is blocked by Cre- dependent knockout of PVT \(\mu\) - ORs (see new Figure 6e- h). Thus, \(\mu\) - OR activation could be driving behavioral disinhibition through reduced PVT synaptic input to PV- INs, MSNs, or other cell types in NAc. However, our substantial data showing that PVT \(\rightarrow\) NAc<sup>PV- IN</sup> circuitry is necessary for the suppression of sucrose seeking suggests that this pathway is likely involved. Nonetheless, we have significantly lightened the language of the manuscript to ensure that one pathway over another is not assumed, and we add additional discussion in this regard (see page 7, text revisions in response to concern #1 above).  
+
+5. In fig 1, The authors describe a tonic decrease in ensemble 2 and 3 activity over the course of the sucrose SA session that is blocked by fear provoking stimuli and extinction and reappears upon reinstatement. However, the emphasis is placed on ensemble 3 activity aligned to lever pressing based on decoding accuracy. Further discussion on this would be useful for the reader.  
+
+We would like to thank the Reviewer for this comment, and we have therefore adjusted some of the paragraph describing these imaging results to ensure that each ensemble is appropriately addressed without bias on ensemble 3.  
+
+6. The authors state that ensemble 1 and 3 decoding accuracy is decreased after heroin due to decreased changes in calcium activity, which is consistent with 4H. However, 4G suggest that decoding accuracy in ensemble 1 is increased relative to saline day. The same mismatch appears for ensemble 2, but not as pronounced as ensemble 1.  
+
+This is a fantastic catch by the Reviewer that was due to some data analysis imperfections, which we have since found and fixed. Specifically, for the CDF plots we were calculating the decoding values for each neuron as the unshuffled decoding score minus the averaged shuffled decoding score across all neurons, such that chance decoding would equal 0. This is an issue as different neurons and clusters are going to have different shuffled values that deviate from one another due to the number of lever presses within a session (specifically, low presses generally result in higher shuffled decoding scores). The summarized heatmap therefore deviated from the results in the CDF plots as each cluster's decoding score was analyzed as its own unshuffled vs shuffled score within the heatmap (using a simple t-test). We therefore have addressed the above data analysis concerns in a few ways:  
+
+1) We have taken out the heatmap, which served as a poor descriptor of the data.  
+2) We use shuffled decoding values for each neuron for decoding score normalization.  
+3) We use single-cell tracking to analyze the same neurons between the two behavioral sessions, such that between-cell and between-cluster variability is accounted for (this is also in response to Reviewer #3, Concern #6).  
+4) We show that both the activated and inhibited ensembles display significant response adaptation following heroin injection (see Figure 4h).
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+5) We show both population and cluster-specific decoding within CDF plots (see new Figure 4i, j), revealing that decoding scores are significantly decreased at the population level and for cluster #3 (see new Supplementary Fig. 7c-e).  
+
+7. The switch to real time place preference evoked by opto PVT to NAcc stim with heroin injection is very interesting and in some ways may be relevant to the potentiation of sucrose self-administration that is observed in response to opto stim, TMT, and yohimbine in fig 4 and 5. This suggest that MOR activity may be unmasking a population of pro-sucrose administration neurons upon inhibition of MOR sensitive neurons.  
+
+We agree with the Reviewer about this interesting finding, and have added additional language in the results to highlight its importance (see page 5):  
+
+"Together, these data suggest that systemic opioids may be modulating PVT \(\rightarrow\) NAc neuronal activity, such that sucrose self-administration is promoted, rather than inhibited, in the presence of behavioral suppressors."  
+
+8. In supplementary figure s2, cluster 1 seems to contain a subset of cells that are briefly inhibited prior to lever press and are excited during reward delivery late in acquisition and inhibited during reinstatement, suggesting they are distinct from other cells in cluster 1. This cluster in some ways has both cluster 1 and 3 properties.  
+
+This was an excellent observation by the Reviewer. We have added and adjusted language in the discussion to include this point (see page 8):  
+
+"...though we cluster PVT \(\rightarrow\) NAc neurons by activity (see Fig. 1), each ensemble seems to contain subsets of cells with distinct responses, despite isolating PVT neurons by location (i.e., posterior) and connection (i.e., projections to NAc)."  
+
+9. The MOR statining of PVT neurons and their terminals could be due to chance alone as MOR-like immunoreactivity appears to be widespread across the image. In the zoom within the representative images it appears that background from adjacent regions was cropped or subtracted. Immunostaining for MOR with this in MOR IoxP mice would be useful for validating the antibody and genetic approach.  
+
+This is a fantastic point that we have addressed through several experiments and adjustments to the figures/text.  
+
+First, we show new example IHC that we used to validate Cre- dependent \(\mu\) - OR knockout in \(\mu\) - OR IoxP mice (New Supplementary Figure 8). This IHC experiment was performed using a different antibody, which was originally used by others to validate \(\mu\) - OR knockout in \(\mu\) - OR IoxP mice (Cui et al. 2014 PMID: 24413699). We therefore replaced the original images of PVT somata with those taken using the new antibody, and we do not perform the same background subtraction based on the Reviewer's excellent point. Furthermore, we do not include images of PVT \(\rightarrow\) NAc axon colocalized with \(\mu\) - OR IHC as overlap could be due to chance.  
+
+Second, we show example images and quantification from a new RNAscope experiment that was used to demonstrate Cre- dependent knockout of \(\mu\) - OR mRNA (see new Figure 5a- c).  
+
+Third, we show that PVT neuronal excitability and synaptic input to downstream NAc neurons are
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+inhibited by the \(\mu\) - OR agonist DAMGO, effects that are blocked by Cre- dependent knockout of PVT \(\mu\) - ORs (new Figure 6e- h; Supplementary Figure 9).  
+
+Altogether, these experiments justify the use of the antibody, \(\mu\) - OR loxP mice, and the point that \(\mu\) - ORs are expressed and functional within PVT \(\rightarrow\) NAc projection neurons.  
+
+10. NAcc PV cells are sparse relative to other interneuron populations. Moreover, PV-Cre mice used in the present study have incomplete genetic penetrance in the NAcc and may select for subtypes of NAcc fast-spiking interneurons in addition to non-PV cells with CP-AMPARs (see Adam Carter's work and Yan Dong). The authors should mention the incomplete penetrance to let the reader know that this may pertain a selective subpopulation of PV cells and overall broaden the discussion of how PV cells are limited in density in NAcc relative to striatum in the context of the present work.  
+
+We agree and have now added a paragraph to the discussion section to address these concerns (see page 7- 8):  
+
+"Our data support the idea that accumbal PV interneurons and CP- AMPRs are necessary for the suppression of sucrose self- administration. Previously, others have shown that accumbal PV interneurons, as well as other FSIs within NAc, can act as powerful regulators of local neuronal activity and behavior despite being sparsely distributed<sup>41- 45</sup>. While we used pharmacology and chemogenetics to target CP- AMPArs and PV interneurons, respectively, it is possible that these methods could have off- target effects. For example, non- PV cells within NAc could express CP- AMPArs, and thus CP- AMPAr antagonism may act on other FSIs or non- PV cell types. Furthermore, it is possible that our targeting of PV interneurons could have profound effects on downstream neurophysiology, and therefore may not be completely selective for our circuit- of- interest. Finally, our PV interneuron cell targeting is likely to select for only a subpopulation of PV- expressing neurons due to incomplete genetic penetrance of the PV- Cre transgenic mouse line<sup>46,47</sup>. Notably, our electrophysiological recordings suggest that our Cre- dependent targeting of PV interneurons at least selects for FSIs, as fluorescent cells within PV- Cre mice displayed fast- spiking properties. Additionally, we find that these cells are inwardly rectifying, suggesting the presence of CP- AMPArs. Despite these findings, future studies selectively targeting CP- AMPArs at PVT \(\rightarrow\) NAc<sup>PV- IN</sup> synapses could elucidate the precise role of these receptors for PVT \(\rightarrow\) NAc- dependent behavioral suppression."  
+
+## 11. The authors should show the lever press – aligned population activity during the fear-provoking stimuli in addition to the change across the session that is reported.  
+
+While we agree with the Reviewer that this would be interesting to show, we have not included this in our revision. First, our in vivo imaging heatmaps are generated by averaging neuronal activity across active lever presses and subjects. Neuronal activity is time- locked to the active lever press and, due to the low number of active lever presses observed on days where mice were exposed to behavioral suppressors (e.g., TMT, yohimbine, extinction learning), it is difficult not possible to generate a heatmap that accurately depicts lever pressing during behavioral
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+suppression. Thus, we decided the most accurate way to present changes in PVT \(\rightarrow\) NAc activity during behavioral suppression was by showing the change in fluorescence across sessions.  
+
+## 12. The intrinsic properties of the MSNs in the representative traces do not show the well-characterized regular spiking properties of healthy MSNs ex-vivo, raising the possibility the cells / preparation was not optimal.  
+
+We agree and would like to reassure the Reviewer that ex vivo cell preparation was fine. The lead PI has performed slice recordings for many years (including in striatum) and identical methodologies were used in this case. The issue is likely that the previous waveforms were from sweeps wherein a very large current pulse was administered ( \(>250\mathrm{pA}\) ; in attempt to show the very high frequency of firing in PV interneurons vs spike adaptation in MSNs). Thus, we have now adjusted the waveforms to show spiking in each cell type in response to a moderate current injection (50pA; see New Figure 3B).  
+
+13. The authors should cite relevant work on thalamostriatal circuitry where appropriate from the Penzo group at NIMH describing a role for thalamostriatal inputs in promoting avoidance and homeostatic feeding behavior.  
+
+Although we had cited Penzo et al., 2015, Beas et al. 2018, and Gao et al., 2020 (each from the Penzo group), we have now added Beas et al., 2020.  
+
+14. The ordering of the supplemental data makes it a little bit difficult to follow. I understand clumping all the inactive lever presses into one figure, but other figures are out of order relative to how the data are presented in the main text and figures.  
+
+We apologize for any confusion surrounding the Supplementary Figures. We have now distributed the inactive lever pressing graphs across Supplementary Figures, such that it aligns with how the data are presented within the main text.
+
+<--- Page Split --->
+
+# Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+Reviewer #2 (Recommendations for the authors):  
+
+Vollmer et al. report the results of an interesting series of experiments providing much needed new knowledge on the organization of the thalamostriatal pathway from paraventricular thalamus to nucleus accumbens in reward seeking behavior. The authors show, quite convincingly, that this circuit involves inputs onto parvalbumin neurons in the accumbens, is enriched in calcium permeable AMPA receptors, is opioid sensitive, and generally inhibits reward seeking across a variety of environmental conditions. These are all important knowledge gains.  
+
+There is a lot of work in this manuscript, some of it is very clever, all of it is very well done, and the manuscript is very well presented. It significantly extends the field. I think the manuscript will be of interest to many readers and that it adds critical new knowledge. I enjoyed reading it and I think others will too. I had only the following comments on the manuscript:  
+
+We would like to thank Reviewer #2 for their excellent suggestions and excitement for our study. We have responded to each concern below, through the addition of new experiments, analyses, discussion points, and consideration of alternative interpretations.  
+
+1. It is not clear to me that the manuscript speaks to "risky reward motivated behaviors". For example, the predator odor is presented in the home cages, not when mice are seeking sucrose. So, there is little to no 'risk' here. There is stress etc, but risk implies an adverse consequence of seeking sucrose. Moreover, the authors show that the same circuit mechanism suppress sucrose seeing after extinction training, which clearly has no risk at all. I am well persuaded by these data that this circuit is a general one for suppressing reward seeking behavior (as predicted by others), but I am less persuaded that the reader learns much about risk. I suggest that "risky" is removed from the title, abstract etc.  
+
+We agree and have changed the title according to the Reviewer's recommendation:  
+
+"An opioid- gated thalamoaccumbal circuit for the suppression of reward seeking".  
+
+Additionally, we have removed language from the manuscript wherein "risky" was used inappropriately.  
+
+2. "Goal-directed". In a number of places, the selective effects of manipulations on the 'active' lever and no effect on the 'inactive' lever are used to support the claim that behavior and the effects of manipulations are 'goal directed'. This is hard to evaluate because there is no actual test of the goal directedness of the behavior here (e.g., contingency degradation; outcome devaluation). Moreover, it appears the sucrose delivery tube was physically located under the 'active' lever (Figure 1a), which reduces the utility and relevance of the 'inactive' lever to the task. I think what the authors mean is that manipulations were lever or behaviorally specific. I am certainly persuaded of this. It may be more helpful to use these terms rather than imputing untested mental states to the animals.  
+
+We thank the Reviewer for bringing up these issues. We have changed the language within the manuscript to describe our manipulations as being lever- specific, rather than goal directed.
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+Furthermore, we apologize for any ambiguity within previous versions of Figure 1a. We have edited the figures so that the sucrose delivery tube appears to be directly in- between the active and inactive levers. We have also added language to clarify the sucrose spout position within the methods:  
+
+Under Head- fixed behavior, we report (page 24):  
+
+"Sucrose delivery spouts were arranged equally between both levers so that mice would not be biased toward either lever."  
+
+3. The decoder results are interesting, but these are based on only using the 1s before each lever press vs a random 1-s control. I wondered why only a 1s period was used and why this 1s was chosen over other possible durations? This could be better justified. Also, did the authors test a decoder using any other pre-lever press durations? If yes, how did they control for these exploratory analyses in their final statistical analyses?  
+
+This is an excellent question that deserves justification within the text, which we now add (see page 26):  
+
+"We used a 1- second epoch prior to the lever press based on: (1) a pre- lever press epoch would ensure that the decoding was not due to cue presentation, liquid delivery, or sucrose consumption and (2) past studies have demonstrated single- cell calcium events during a 1- second pre- reward trace interval can be used to accurately predict reward learning within a Pavlovian conditioning task<sup>6,26</sup>. Thus, a 1- second epoch immediately before lever pressing seemed most appropriate for our decoding analysis and was chosen beforehand such that only one analysis was performed."  
+
+4. The identification of these effects to PV interneurons is neat. PV neurons are sparse in the accumbens, so it is interesting that their manipulations here have such strong effects. However, the reader never learns how selective the hM4Di was to PV neurons in the accumbens in Figure 3i. Can the authors confirm that other cells in the accumbens did not express the DREADD?  
+
+We agree and have now added new data validating the use of the DIO- hM4Di- DREADD- mCherry for CNO- induced inhibition of NAc PV interneurons (see new Supplemental Figure 6). We specifically find that mCherry- expressing neurons display electrophysiological properties consistent with PV interneurons, and these cells are inhibited by bath application of CNO. In contrast, neighboring mCherry- negative neurons were not inhibited by CNO and show electrophysiological properties consistent with other NAc cells (in particular, medium spiny neurons).  
+
+## Minor:  
+
+1. I may have missed it, but do the authors state the number of cells in each cluster in Figure 1i or in Figure 4? These are worth clearly stating.  
+
+We agree and have now stated the number of cells in each cluster within the captions for Figures 1 and 4.
+
+<--- Page Split --->
+
+Vollmer, Green et al., 2022Nature Communications ResubmissionResponse to Reviewers  
+
+## 2. State dose of DAMGO used in the manuscript proper.  
+
+We agree and have stated the dose of DAMGO used for intracranial infusions within the manuscript proper.  
+
+## 3. Line 240: should refer to Figure 5H not 5F.  
+
+We thank the Reviewer for pointing this out and have corrected this mix- up.
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+Reviewer #3 (Recommendations for the authors):  
+
+The manuscript by Vollmer et al. describes the role of PVT- NAc projection in the suppression of sucrose seeking. Using the head- fixed sucrose seeking paradigm, authors performed two- photon imaging and optogenetic manipulation of PVT neurons projecting to NAc. Furthermore, the authors examine the role of the opioid receptors in this neuronal type. Overall, all experiments are well designed and performed. However, some interpretation is not fully convincing.  
+
+We would like to sincerely thank Reviewer #3 for their enthusiasm for the paper as well as their concerns about some of our data interpretation. In response, we both added experiments and dialed back some of the interpretation which we agree was over- extended.  
+
+1. It is difficult to connect the contribution of PVT projection to NAc PV neurons with behavioral changes in sucrose seeking. Especially, since several papers showed that PVT inputs preferentially modify D2-MSN in opioid-induced behavioral changes, it is not clear how the contribution of PVT projection is specifically modulating NAc PV neurons in sucrose seeking. Unless authors can image three different neurons (D1-MSN, D2-MSN, PV neurons) while modulating PVT inputs, this interpretation is not fully convincing. Moreover, since it is very likely that the chemogenetic inhibition of PV neurons (Figure 3k-o) will have massive effects on NAc circuitry already, the optogenetic stimulation of PVT together with PV chemogenetic manipulation may not be PV neuron dependent.  
+
+We would like to thank the Reviewer for this point, which we agree with overall and now address through easing of data interpretation and additional discussion. Specifically, while we predict that PVT \(\rightarrow\) NAc stimulation causes behavioral disinhibition through the activation of downstream PV- INs (considering the experimental evidence within our paper and supporting studies), we provide several critical caveats. Furthermore, while we conclude that opioids can act at \(\mu\) - opioid receptors at PVT \(\rightarrow\) NAc axons, somata, or both to drive behavioral disinhibition, we note that this could be due to a reduction in activity at a variety of PVT \(\rightarrow\) NAc synapses and downstream neurons, and not necessarily at PVT \(\rightarrow\) NAc<sup>PV-IN</sup> synapses specifically.  
+
+We do feel it is important to consider several other points. First, we show that intra- NAc infusion of a CP- AMPAR antagonist, but not D1 or D2 receptor antagonist, abolishes the suppression of sucrose seeking caused by PVT \(\rightarrow\) NAc stimulation. Considering the specificity of these receptors for FSIs/PV interneurons (Gittis et al., 2011 PMID: 22049415; Manz et al., 2020 PMID: 32726634) and PVT \(\rightarrow\) NAc<sup>PV-IN</sup> but not PVT \(\rightarrow\) NAc<sup>MSN</sup> synapses (Figure 3), it is likely that the PVT \(\rightarrow\) NAc<sup>PV-IN</sup> pathway is important for our behavioral findings. This interpretation is even further supported, if not confirmed, by our chemogenetics data showing that inhibition of PV- interneurons reverses the behavior- suppressing effects caused by PVT \(\rightarrow\) NAc stimulation, TMT exposure, and yohimbine exposure (Figure 3). It should be noted that these chemogenetic manipulations are less potent than optogenetic inhibition, as we see a modest and cell- type specific reduction in PV- IN excitability following CNO exposure (see new data in Supplementary Fig. 6c). Finally, as mentioned by Reviewer 1, PV interneurons represent a very sparse population of NAc neurons (~1- 2% of cells), and thus we feel that behavioral effects found via PV interneuron inhibition are quite fascinating. It should be noted that D1- MSNs and D2- MSNs, which also have robust collaterals and projections for local and distal inhibition, are far more common than PV- INs (~95% of NAc neurons) and are often targeted via optogenetics and chemogenetics to identify their function for behavioral control.
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+Despite these points, we fully agree that there may be off- target effects caused by our pharmacological or chemogenetic manipulations. Thus, we have added additional discussion to highlight the Reviewer's concerns (for example, see page 7- 8, text revisions in response to Reviewer #1, Concern #10 above):  
+
+"Our data support the idea that accummal PV interneurons and CP- AMPArs are necessary for the suppression of sucrose self- administration. Previously, others have shown that accummal PV interneurons, as well as other FSIs within NAC, can act as powerful regulators of local neuronal activity and behavior despite being sparsely distributed41- 45. While we used pharmacology and chemogenetics to target CP- AMPArs and PV interneurons, respectively, it is possible that these methods could have off- target effects. For example, non- PV cells within NAC could express CP- AMPArs, and thus CP- AMPAr antagonism may act on other FSIs or non- PV cell types. Furthermore, it is possible that our targeting of PV interneurons could have profound effects on downstream neurophysiology, and therefore may not be completely selective for our circuit- of- interest."  
+
+Next, while we agree with the Reviewer that being able to modulate PVT while monitoring D1- MSNs, D2- MSNs, and PV interneurons would aid in elucidating the precise mechanism in which PVT \(\rightarrow\) NAC neurons influence behavior, recording from 3 different cell types simultaneously during PVT stimulation is simply not feasible with current technologies available. As such, it is not possible for us to complete this experiment.  
+
+2. Authors found that CP-AMPARs are selectively located in PV neurons in NAC. However, the authors didn't show any effort to describe the potential roles of these CP-AMPARs on sucrose seeking. What is the role of PVT inputs for CP-AMPARs dependent PV neuronal activity?  
+
+We used a combination of optogenetics and neuropharmacology to show that CP- AMPARs are required for PVT \(\rightarrow\) NAC dependent suppression of sucrose seeking (Figure 3j). While CP- AMPARs are not the focus of the manuscript, these experiments provided initial support for the idea that PVT \(\rightarrow\) NAC PV interneuron synapses may be involved in the suppression of sucrose seeking. This idea was then strongly supported by the subsequent chemogenetics experiments, wherein we combine optogenetic manipulation of PVT \(\rightarrow\) NAC neurons with chemogenetic manipulation of PV interneurons. We have also added additional discussion highlighting the previous findings that PV interneurons within NAC can act as behavioral regulators (see page see page 7- 8, text revisions in response to Reviewer #1, Concern #10 above).  
+
+3. The effects of DAMGO application on synaptic transmission (Figure 5) can be due to the postsynaptic effects since NAC MSNs also express MORs. More extensive analysis of synaptic transmission is needed to fully validate the presynaptic roles of MORs in PVT axon fibers. Again, the behavioral effects of DAMGO (Figure 5h) can be also due to the postsynaptic effects due to the DAMGO-induced changes in postsynaptic MSNs.  
+
+This is an excellent point, which we now add experiments and discussion to address:  
+
+a) Synaptic physiology: We have added new synaptic physiology experiments. Using Cre-dependent knockout of \(\mu\) -ORs in PVT, we find that DAMGO no longer reduces
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+PVT \(\rightarrow\) NAc synaptic transmission (new Figure 6h). These data suggest that the effect of DAMGO on optically evoked EPSCs at PVT \(\rightarrow\) NAc synapses requires presynaptic \(\mu\) - ORs.  
+
+b) Behavioral findings: We have added 5 additional behavioral experiments to highlight the requirement for PVT \(\mu\) -ORs for opioid-induced behavioral disinhibition. While we confirm that heroin disinhibits sucrose seeking regardless of the behavioral suppressor presented (i.e., PVT \(\rightarrow\) NAc stimulation, TMT, or yohimbine), we show that knockout of PVT \(\mu\) -ORs prevents heroin-induced behavioral disinhibition (new Figure 5d-f). Furthermore, we show that PVT \(\mu\) -OR knockout prevents NAc DAMGO infusions from causing TMT- and yohimbine-induced behavioral disinhibition (previously we had only examined PVT \(\rightarrow\) NAc dependent behavioral disinhibition; new Figure 6k, I). Overall, these findings confirm that the effects of systemic heroin and intra-NAc DAMGO on behavioral disinhibition require PVT \(\mu\) -ORs.  
+
+c) Discussion: As rightly pointed out by the Reviewer, our experiments simply cannot rule out the possibility that DAMGO also has postsynaptic effects on PVT \(\rightarrow\) NAc neurophysiology and thus behavior. Thus, while we indicate that PVT \(\mu\) -ORs are indeed required for our observed findings, we dial back the interpretation throughout the manuscript and add discussion points to this effect (e.g., see page 8):  
+
+"...additional mechanisms could also contribute to opioid-induced behavioral disinhibition, such as \(\mu\) - OR activation elsewhere in the brain and in other NAc cell types that express \(\mu\) - ORs<sup>54-56</sup>. Overall, while our data suggests that opioid-induced inhibition of PVT \(\rightarrow\) NAc neurons can disinhibit maladaptive behavioral actions, whether these effects are isolated to PVT \(\rightarrow\) NAc somata and/or synapses has yet to be established."  
+
+4. The systemic injection of heroin can induce neural adaptation in other brain areas, not just PVT-NAc inputs. These other adaptations can change the efficacy of PVT-NAc stimulation with heroin injection. Thus, it is difficult to know how the heroin injection can reverse the effect of PVT-NAc stimulation clearly. Authors should show how PVT-NAc projections are specifically changed by the systemic heroin injection.  
+
+This is another outstanding point that we now address with several new experiments and discussion:  
+
+a) Using two-photon calcium imaging we demonstrate that PVT \(\rightarrow\) NAc neurons decrease activity and encoding of sucrose seeking following systemic heroin injection (Figure 4a-j).  
+b) Using patch-clamp electrophysiology we add new data showing that \(\mu\) -OR activation diminishes evoked activity in PVT \(\rightarrow\) NAc neurons, an effect that is prevented by the knockout of PVT \(\mu\) -ORs (new Supplementary Figure 9).  
+c) It remains true that the behavioral effects caused by heroin could be due to adaptations on or upstream of PVT \(\rightarrow\) NAc neurons. Thus, we knocked \(\mu\) -ORs out of PVT, and show that heroin-induced behavioral disinhibition completely absent (see new Figure 5). Thus, the behavioral and physiological effects of heroin described in our study require PVT \(\mu\) -ORs.
+
+<--- Page Split --->
+
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+d) Despite these findings, it remains possible that heroin could also be acting elsewhere to drive behavioral disinhibition and adaptations in neurophysiology, which we address through additions to the text such as that referenced above.  
+
+## 5. The effect of the optogenetic manipulation of PVT to NAc projection on TMT or Yohimbe mediated suppression of sucrose seeking was not examined.  
+
+In Figure 2, we bidirectionally manipulate PVT \(\rightarrow\) NAc neurons following TMT- (Figure 2f) and yohimbine- (Figure 2g) mediated suppression of sucrose seeking.  
+
+## 6. Authors can fully take advantage of two-photon imaging. The longitudinal imaging of single neurons can provide the trends in the adaptation of individual neurons.  
+
+This is an excellent point, and we have now added new data tracking PVT \(\rightarrow\) NAc neuronal activity across time. First, we track PVT \(\rightarrow\) NAc responses across early to late sucrose self- administration. We find that PVT \(\rightarrow\) NAc ensemble dynamics develop over learning, as ensembles 1 (activated) and 3 (inhibited) show a significant response adaptation over time (see new Supplementary Figure 2c- f). Next, we track PVT \(\rightarrow\) NAc neuronal activity from saline to heroin injection tests, and find that PVT \(\rightarrow\) NAc ensembles 1 and 3 have significant response attenuation during the heroin test such that the activity of these neurons can no longer be used to decode active lever pressing (see new Figure 4f- j; Supplementary Fig 7c- e).
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have adequately addressed my comments and concerns. Congratulations on the elegant and impactful study.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have made considerable revisions in response to my initial review. I thank them for this. I remain of the view that this is an interesting and important contribution to the literature on PVT function. I am less certain of the ultimate utility of simple lever and extinction type approaches for understanding core problems in reward seeking - and the field has largely moved on from these- but the kinds of insights generated by this manuscript are important and could lead to a revision of this view.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have been very responsive to my comments. They have added results from a number of experiments and also modified language to account for limitations.  
+
+I also feel that the authors have done a very thorough and careful job addressing concerns of other reviewers'.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a2e4cfd876998db6d4acbcedc4a2212a495eea77357accea0026d65cd7e6325f/supplementary_0_Peer Review File__a2e4cfd876998db6d4acbcedc4a2212a495eea77357accea0026d65cd7e6325f_det.mmd

@@ -0,0 +1,522 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 161, 875, 218]]<|/det|>
+An opioid- gated thalamoaccumbal circuit for the suppression of reward seeking in mice  
+
+<|ref|>image<|/ref|><|det|>[[95, 732, 262, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[271, 732, 880, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 325, 107]]<|/det|>
+## REVIEWER COMMENTS</B>  
+
+<|ref|>text<|/ref|><|det|>[[116, 129, 393, 145]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 163, 883, 675]]<|/det|>
+The present study by Vollmer, Green et al. test the hypothesis that a thalamostriatal circuit inhibits reward seeking during exposure to fear- provoking stimuli and that opioid receptor activation promotes risky behavior in the face of these stimuli by inhibiting this circuit via a presynaptic site of action. First the authors demonstrate that thalamostriatal PVT cells display heterogenous calcium responses to lever press activity (type 1 and 3 responses) associated with sucrose administration, and some ensembles are inhibited in a tonic manner. The authors demonstrate that optogenetic activation of PVT- NAcc pathway inhibited sucrose self- administration, while optogenetic inhibition of thalamostriatal neurons reversed TMT, yohimbine, and extinction- mediated reductions in sucrose self- administration. The authors subsequently demonstrated that thalamostriatal neurons innervate MSNs and PV interneurons, with the latter containing CP- AMPARs. The authors then show that the effects of optogenetic activation of PVT to NAcc afferents on sucrose self- administration is blocked by antagonism of CP- AMPARs in the NAcc and chemogenetic inhibition of NAcc PV neurons. The authors claim that thalamostriatal neurons and their terminals in the NAcc express MOR and that acute heroin administration decreases ensemble decoding of sucrose self- administration and blocks the suppressive effects of PVT to NAcc pathway stimulation and suppression induced by fear provoking stimuli. Lastly, the authors demonstrate that intra- NAcc DAMGO injections inhibit optogenetic -, TMT- and yohimbine- induced decreases in sucrose self- administration. They show that synapses from PVT to NAcc PV neurons are inhibited by DAMGO and that MOR deletion from PVT blocks the ability of intra- NAcc DAMGO from reversing the inhibitory effects of thalamostriatal optogenetic stimulation on sucrose self- administration. Overall, the study is of importance and tackles an important question. The authors test their hypothesis using an impressive set of interdisciplinary approaches. However, the authors make strong conclusions about their data that are not entirely supported by the present findings. Specifically, there are significant gaps that need to be addressed for the conclusions in their present form to be drawn, which include additional experiments and/or dialing back some of the interpretations and discussing caveats and alternative hypotheses. Below are some comments that will aid the authors in strengthening their manuscript and improving the cohesiveness of the study.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 695, 196, 710]]<|/det|>
+## Comments  
+
+<|ref|>text<|/ref|><|det|>[[114, 731, 881, 907]]<|/det|>
+- There is a disconnect between the data across the different figures. The authors make strong claims that opioids acting through presynaptic receptors disrupt PVT inhibition, but this is largely decoupled from recordings of cell bodies and the effect of heroin presented earlier in the paper. Opioid receptors on terminals would inhibit terminal release and any potential influence on cell body activity in thalamostriatal neurons would likely be an indirect consequence. That is, is there any relationship between what MORs are doing at terminals and what is happening in the cell bodies of thalamostriatal neurons? Moreover, it is unclear whether the exciting findings in Fig 5 are related to the findings with heroin. Additional studies linking the role of MORs on PVT terminals, activity of thalamostriatal cells, and how this is involved in heroin's effects would be needed to reach the interpretations presented, such as
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 821, 107]]<|/det|>
+deleting MOR from thalamostriatal neurons and determining if that impacts the effect of heroin.  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 877, 223]]<|/det|>
+- The authors should discuss the caveat that IEM-1640 may be acting on other cell types not examined in the present paper or that CP-AMPARs may become engaged downstream of PVT inputs to NAcc, including the PV neurons. The authors' interpretation should be much more cautious without additional evidence.  
+
+<|ref|>text<|/ref|><|det|>[[114, 244, 877, 418]]<|/det|>
+- Caveats associated with electrophysiological characterization of synaptic connections onto NAcc neurons should be discussed/addressed. A between subjects design comparing MSNs and PV neurons is subject to variability from ChR2 expression, and as such additional measures would be useful for determining synaptic strength differences between cell types. Moreover, since virus was used to label D2 cells the caveat that unlabeled cells may constitute putative D1 and unlabeled D2 cells should be addressed. That PV neurons display larger currents relative to MSNs is not surprising, since PV interneurons are broadly characterized by the presence of CP-AMPARs and enhanced excitatory synaptic strength in striatal structures and cortex. Furthermore, NAcc PV neurons been implicated in mediating aversive behavior independent of the PVT (see work by Morales group (NIDA) for example).  
+
+<|ref|>text<|/ref|><|det|>[[114, 439, 869, 534]]<|/det|>
+- MOR activation may also inhibit glutamate release onto non-PV targets (e.g. MSNs), which may be contributing to the observed behavioral effects. The authors do not examine these synapses and make strong conclusions about PVT inputs to PV that warrant further investigation. Furthermore, their physiology does not directly demonstrate presynaptic effects with the data presented, though it's implied.  
+
+<|ref|>text<|/ref|><|det|>[[114, 556, 870, 631]]<|/det|>
+- In fig 1, The authors describe a tonic decrease in ensemble 2 and 3 activity over the course of the sucrose SA session that is blocked by fear provoking stimuli and extinction and reappears upon reinstatement. However, the emphasis is placed on ensemble 3 activity aligned to lever pressing based on decoding accuracy. Further discussion on this would be useful for the reader.  
+
+<|ref|>text<|/ref|><|det|>[[114, 653, 882, 728]]<|/det|>
+- The authors state that ensemble 1 and 3 decoding accuracy is decreased after heroin due to decreased changes in calcium activity, which is consistent with 4H. However, 4G suggest that decoding accuracy in ensemble 1 is increased relative to saline day. The same mismatch appears for ensemble 2, but not as pronounced as ensemble 1.  
+
+<|ref|>text<|/ref|><|det|>[[114, 750, 881, 845]]<|/det|>
+- The switch to real time place preference evoked by opto PVT to NAcc stim with heroin injection is very interesting and in some ways may be relevant to the potentiation of sucrose self-administration that is observed in response to opto stim, TMT, and yohimbine in fig 4 and 5. This suggest that MOR activity may be unmasking a population of pro-sucrose administration neurons upon inhibition of MOR sensitive neurons.  
+
+<|ref|>text<|/ref|><|det|>[[114, 868, 869, 904]]<|/det|>
+- In supplementary figure s2, cluster 1 seems to contain a subset of cells that are briefly inhibited prior to lever press and are excited during reward delivery late in acquisition and inhibited during
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 860, 127]]<|/det|>
+reinstatement, suggesting they are distinct from other cells in cluster 1. This cluster in some ways has both cluster 1 and 3 properties.  
+
+<|ref|>text<|/ref|><|det|>[[115, 147, 866, 223]]<|/det|>
+- The MOR statining of PVT neurons and their terminals could be due to chance alone as MOR-like immunoreactivity appears to be widespread across the image. In the zoom within the representative images it appears that background from adjacent regions was cropped or subtracted. Immunostaining for MOR with this in MOR loxP mice would be useful for validating the antibody and genetic approach.  
+
+<|ref|>text<|/ref|><|det|>[[115, 243, 882, 360]]<|/det|>
+- NAcc PV cells are sparse relative to other interneuron populations. Moreover, PV-Cre mice used in the present study have incomplete genetic penetrance in the NAcc and may select for subtypes of NAcc fast-spiking interneurons in addition to non-PV cells with CP-AMPARs (see Adam Carter's work and Yan Dong). The authors should mention the incomplete penetrance to let the reader know that this may pertain a selective sub-population of PV cells and overall broaden the discussion of how PV cells are limited in density in NAcc relative to striatum in the context of the present work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 381, 880, 418]]<|/det|>
+- The authors should show the lever press – aligned population activity during the fear provoking stimuli in addition to the change across the session that is reported.  
+
+<|ref|>text<|/ref|><|det|>[[115, 439, 876, 496]]<|/det|>
+- The intrinsic properties of the MSNs in the representative traces do not show the well-characterized regular spiking properties of healthy MSNs ex-vivo, raising the possibility the cells / preparation was not optimal.  
+
+<|ref|>text<|/ref|><|det|>[[115, 517, 870, 574]]<|/det|>
+- The authors should cite relevant work on thalamostriatal circuitry where appropriate from the Penzo group at NIMH describing a role for thalamostriatal inputs in promoting avoidance and homeostatic feeding behavior.  
+
+<|ref|>text<|/ref|><|det|>[[115, 595, 875, 652]]<|/det|>
+- The ordering of the supplemental data makes it a little bit difficult to follow. I understand clumping all the inactive lever presses into one figure but other figures are out of order relative to how the data are presented in the main text and figures.  
+
+<|ref|>text<|/ref|><|det|>[[115, 714, 393, 730]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 752, 882, 866]]<|/det|>
+Vollmer et al. report the results of an interesting series of experiments providing much needed new knowledge on the organisation of the thalamostriatal pathway from paraventricular thalamus to nucleus accumbens in reward seeking behavior. The authors show, quite convincingly, that this circuit involves inputs onto parvalbumin neurons in the accumbens, is enriched in calcium permeable AMPA receptors, is opioid sensitive, and generally inhibits reward seeking across a variety of environmental conditions. These are all important knowledge gains.  
+
+<|ref|>text<|/ref|><|det|>[[115, 888, 836, 905]]<|/det|>
+There is a lot of work in this manuscript, some of it is very clever, all of it is very well done, and the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 872, 146]]<|/det|>
+manuscript is very well presented. It significantly extends the field. I think the manuscript will be of interest to many readers and that it adds critical new knowledge. I enjoyed reading it and I think others will too. I had only the following comments on the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[114, 167, 878, 302]]<|/det|>
+1. It is not clear to me that the manuscript speaks to "risky reward motivated behaviors". For example, the predator odor is presented in the home cages, not when mice are seeking sucrose. So, there is little to no 'risk' here. There is stress etc, but risk implies an adverse consequence of seeking sucrose. Moreover, the authors show that the same circuit mechanism suppress sucrose seeing after extinction training, which clearly has no risk at all. I am well persuaded by these data that this circuit is a general one for suppressing reward seeking behaviour (as predicted by others), but I am less persuaded that the reader learns much about risk. I suggest that "risky" is removed from the title, abstract etc.  
+
+<|ref|>text<|/ref|><|det|>[[114, 322, 877, 477]]<|/det|>
+2. "goal-directed'. In a number of places, the selective effects of manipulations on the 'active' lever and no effect on the 'inactive' lever are used to support the claim that behavior and the effects of manipulations are 'goal directed'. This is hard to evaluate because there is no actual test of the goal directedness of the behavior here (e.g., contingency degradation; outcome devaluation). Moreover, it appears the sucrose delivery tube was physically located under the 'active' lever (Figure 1a), which reduces the utility and relevance of the 'inactive' lever to the task. I think what the authors mean is that manipulations were lever or behaviorally specific. I am certainly persuaded of this. It may be more helpful to use these terms rather than imputing untested mental states to the animals.  
+
+<|ref|>text<|/ref|><|det|>[[114, 498, 874, 593]]<|/det|>
+3. The decoder results are interesting, but these are based on only using the 1s before each lever press vs a random 1-s control. I wondered why only a 1s period was used and why this 1s was chosen over other possible durations? This could be better justified. Also, did the authors test a decoder using any other pre-lever press durations? If yes, how did they control for these exploratory analyses in their final statistical analyses?  
+
+<|ref|>text<|/ref|><|det|>[[115, 615, 870, 691]]<|/det|>
+4. The identification of these effects to PV interneurons is neat. PV neurons are sparse in the accumbens, so it is interesting that their manipulations here have such strong effects. However, the reader never learns how selective the hM4Di was to PV neurons in the accumbens in Figure 3i. Can the authors confirm that other cells in the accumbens did not express the DREADD?  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 714, 163, 728]]<|/det|>
+## Minor  
+
+<|ref|>text<|/ref|><|det|>[[115, 732, 839, 808]]<|/det|>
+1. I may have missed it, but do the authors state the number of cells in each cluster in Figure 1i or I Figure 4? These are worth clearly stating. 
+2. State dose of DAMGO used in the manuscript proper. 
+3. Line 240: should refer to Figure 5H not 5F
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 878, 185]]<|/det|>
+The manuscript by Vollmer et al. describes the role of PVT- NAC projection in the suppression of sucrose seeking. Using the head- fixed sucrose seeking paradigm, authors performed two- photon imaging and optogenetic manipulation of PVT neurons projecting to NAC. Furthermore, the authors examine the role of the opioid receptors in this neuronal type. Overall, all experiments are well designed and performed. However, some interpretation is not fully convincing.  
+
+<|ref|>text<|/ref|><|det|>[[114, 206, 880, 360]]<|/det|>
+1. It is difficult to connect the contribution of PVT projection to NAC PV neurons with behavioral changes in sucrose seeking. Especially, since several papers showed that PVT inputs preferentially modify D2- MSN in opioid-induced behavioral changes, it is not clear how the contribution of PVT projection is specifically modulating NAC PV neurons in sucrose seeking. Unless authors can image three different neurons (D1-MSN, D2-MSN, PV neurons) while modulating PVT inputs, this interpretation is not fully convincing. Moreover, since it is very likely that the chemogenetic inhibition of PV neurons (Figure 3k-o) will have massive effects on NAC circuitry already, the optogenetic stimulation of PVT together with PV chemogenetic manipulation may not be PV neuron dependent.  
+
+<|ref|>text<|/ref|><|det|>[[114, 381, 866, 438]]<|/det|>
+2. Authors found that CP-AMPARs are selectively located in PV neurons in NAC. However, the authors didn't show any effort to describe the potential roles of these CP-AMPARs on sucrose seeking. What is the role of PVT inputs for CP-AMPARs dependent PV neuronal activity?  
+
+<|ref|>text<|/ref|><|det|>[[114, 459, 872, 555]]<|/det|>
+3. The effects of DAMGO application on synaptic transmission (Figure 5) can be due to the postsynaptic effects since NAC MSNs also express MORs. More extensive analysis of synaptic transmission is needed to fully validate the presynaptic roles of MORs in PVT axon fibers. Again, the behavioral effects of DAMGO (Figure 5h) can be also due to the postsynaptic effects due to the DAMGO-induced changes in postsynaptic MSNs.  
+
+<|ref|>text<|/ref|><|det|>[[114, 576, 866, 672]]<|/det|>
+4. The systemic injection of heroin can induce neural adaptation in other brain areas, not just PVT-NAC inputs. These other adaptations can change the efficacy of PVT-NAC stimulation with heroin injection. Thus, it is difficult to know how the heroin injection can reverse the effect of PVT-NAC stimulation clearly. Authors should show how PVT-NAC projections are specifically changed by the systemic heroin injection.  
+
+<|ref|>text<|/ref|><|det|>[[114, 693, 860, 730]]<|/det|>
+5. The effect of the optogenetic manipulation of PVT to NAC projection on TMT or Yohimbe mediated suppression of sucrose seeking was not examined.  
+
+<|ref|>text<|/ref|><|det|>[[114, 752, 866, 788]]<|/det|>
+6. Authors can fully take advantage of two-photon imaging. The longitudinal imaging of single neurons can provide the trends in the adaptation of individual neurons.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 44, 453, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 882, 190]]<|/det|>
+We would like to thank the Reviewers for their outstanding suggestions, which we have addressed point- by- point below (specific concerns are in BOLD text, followed by our responses and additions to the manuscript in ITALICIZED font). In response to these concerns and suggestions, we provide new data, analyses, and clarifications which we feel have significantly strengthened the manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 205, 536, 222]]<|/det|>
+## Reviewer #1 (Recommendations for the authors):  
+
+<|ref|>text<|/ref|><|det|>[[114, 237, 882, 718]]<|/det|>
+The present study by Vollmer, Green et al. test the hypothesis that a thalamostriatal circuit inhibits reward seeking during exposure to fear- provoking stimuli and that opioid receptor activation promotes risky behavior in the face of these stimuli by inhibiting this circuit via a presynaptic site of action. First the authors demonstrate that thalamostriatal PVT cells display heterogenous calcium responses to lever press activity (type 1 and 3 responses) associated with sucrose administration, and some ensembles are inhibited in a tonic manner. The authors demonstrate that optogenetic activation of PVT- NAcc pathway inhibited sucrose self- administration, while optogenetic inhibition of thalamostriatal neurons reversed TMT, yohimbine, and extinction- mediated reductions in sucrose self- administration. The authors subsequently demonstrated that thalamostriatal neurons innervate MSNs and PV interneurons, with the latter containing CP- AMPARs. The authors then show that the effects of optogenetic activation of PVT to NAcc afferents on sucrose self- administration is blocked by antagonism of CP- AMPARs in the NAcc and chemogenetic inhibition of NAcc PV neurons. The authors claim that thalamostriatal neurons and their terminals in the NAcc express MOR and that acute heroin administration decreases ensemble decoding of sucrose self- administration and blocks the suppressive effects of PVT to NAcc pathway stimulation and suppression induced by fear provoking stimuli. Lastly, the authors demonstrate that intra- NAcc DAMGO injections inhibit optogenetic -, TMT- and yohimbine- induced decreases in sucrose self- administration. They show that synapses from PVT to NAcc PV neurons are inhibited by DAMGO and that MOR deletion from PVT blocks the ability of intra- NAcc DAMGO from reversing the inhibitory effects of thalamostriatal optogenetic stimulation on sucrose self- administration. Overall, the study is of importance and tackles an important question. The authors test their hypothesis using an impressive set of interdisciplinary approaches. However, the authors make strong conclusions about their data that are not entirely supported by the present findings. Specifically, there are significant gaps that need to be addressed for the conclusions in their present form to be drawn, which include additional experiments and/or dialing back some of the interpretations and discussing caveats and alternative hypotheses. Below are some comments that will aid the authors in strengthening their manuscript and improving the cohesiveness of the study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 733, 881, 816]]<|/det|>
+We want to thank Reviewer #1 for their excitement for our findings and very helpful feedback, which has aided us in improving the quality of our paper. Based on the concerns raised below, we have added 10 new experiments as well as new data analysis for existing imaging datasets. Furthermore, we have adjusted the language of the manuscript to ensure that caveats are addressed and more appropriately acknowledged.  
+
+<|ref|>text<|/ref|><|det|>[[115, 832, 881, 898]]<|/det|>
+1. There is a disconnect between the data across the different figures. The authors make strong claims that opioids acting through presynaptic receptors disrupt PVT inhibition, but this is largely decoupled from recordings of cell bodies and the effect of heroin presented earlier in the paper. Opioid receptors on terminals would inhibit terminal release
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 44, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 882, 239]]<|/det|>
+and any potential influence on cell body activity in thalamostriatal neurons would likely be an indirect consequence. That is, is there any relationship between what MORs are doing at terminals and what is happening in the cell bodies of thalamostriatal neurons? Moreover, it is unclear whether the exciting findings in Fig 5 are related to the findings with heroin. Additional studies linking the role of MORs on PVT terminals, activity of thalamostriatal cells, and how this is involved in heroin's effects would be needed to reach the interpretations presented, such as deleting MOR from thalamostriatal neurons and determining if that impacts the effect of heroin.  
+
+<|ref|>text<|/ref|><|det|>[[115, 253, 882, 469]]<|/det|>
+We agree with the Reviewer and have gone to great lengths to demonstrate that the findings across these figures (now Figure 4- 6) are related. First, we use two- photon imaging and slice electrophysiology to show that both heroin injection and DAMGO application (experiments 1,2) reduces the activity/excitability of \(\mathrm{PVT} \rightarrow \mathrm{NAc}\) somata and (experiment 3) reduces downstream synaptic inputs to NAC neurons in a manner that is prevented by Cre- dependent knockout of PVT \(\mu\) - ORs (see Figure 4, Figure 6, Supplementary Figure 9). Second, we now show that behavioral disinhibition caused by (experiments 4- 6) heroin injection or (experiments 7- 8) intra- NAC DAMGO infusion is reversed by Cre- dependent \(\mu\) - OR knockout in PVT, regardless of the behavioral suppressor administered (see Figure 5; Figure 6). These exciting new results lead us to conclude that \(\mu\) - ORs on PVT \(\rightarrow \mathrm{NAc}\) neurons (possibly both somatic and on axon terminals) are required for opioid- driven behavioral disinhibition. However, we cannot rule out that opioids could also act elsewhere to drive behavioral disinhibition, including post- synaptically in NAC. Thus, we have lightened the language of the manuscript and include additional discussion (see page 8):  
+
+<|ref|>text<|/ref|><|det|>[[173, 484, 826, 722]]<|/det|>
+"Despite our findings that PVT \(\mu\) - OR knockout prevents systemic heroin or intra- NAC DAMGO infusions from disinhibiting sucrose- seeking behaviors, caveats regarding the specificity of our results should be considered. First, we cannot dissociate whether opioid- driven \(\mu\) - OR activation on PVT somata or PVT \(\rightarrow \mathrm{NAc}\) axon terminals are required for our observed behavioral effects. Considering that heroin and DAMGO can dramatically reduce activity at both PVT \(\rightarrow \mathrm{NAc}\) somata and downstream synapses, it is possible that both mechanisms are involved. Second, additional mechanisms could also contribute to opioid- induced behavioral disinhibition, such as \(\mu\) - OR activation elsewhere in the brain and in other NAC cell types that express \(\mu\) - ORs<sup>54- 56</sup>. Overall, while our data suggests that opioid- induced inhibition of PVT \(\rightarrow \mathrm{NAc}\) neurons can disinhibit maladaptive behavioral actions, whether these effects are isolated to PVT \(\rightarrow \mathrm{NAc}\) somata and/or synapses has yet to be established."  
+
+<|ref|>text<|/ref|><|det|>[[115, 757, 881, 823]]<|/det|>
+2. The authors should discuss the caveat that IEM-1640 may be acting on other cell types not examined in the present paper or that CP-AMPARs may become engaged downstream of PVT inputs to NAC, including the PV neurons. The authors' interpretation should be much more cautious without additional evidence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 839, 881, 906]]<|/det|>
+We agree. While our data showing that NAC CP- AMPAr blockade results in behavioral disinhibition led us to examine PV interneuron involvement, we do not definitively state that CP- AMPArs on PV- interneurons are mediating the effects of IEM- 1460. Furthermore, we now address caveats within the results (see page 5):
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 45, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[174, 124, 825, 191]]<|/det|>
+"While we therefore hypothesize that PVT \(\rightarrow\) NAc neurons act selectively at CP- AMPAr- enriched synapses on PV interneurons to suppress behavior, it should be noted that other synapses and NAc cell types may also be involved (see discussion)."  
+
+<|ref|>text<|/ref|><|det|>[[116, 206, 428, 222]]<|/det|>
+As well as the discussion (see page 7):  
+
+<|ref|>text<|/ref|><|det|>[[174, 239, 825, 304]]<|/det|>
+"While we used pharmacology and chemogenetics to target CP- AMPArs and PV interneurons, respectively, it is possible that these methods could have off- target effects. For example, non- PV cells within NAc could express CP- AMPArs, and thus CP- AMPAr antagonism may act on other FSIs or non- PV cell types."  
+
+<|ref|>text<|/ref|><|det|>[[114, 319, 883, 498]]<|/det|>
+3. Caveats associated with electrophysiological characterization of synaptic connections onto NAcc neurons should be discussed/addressed. A between subjects design comparing MSNs and PV neurons is subject to variability from ChR2 expression, and as such additional measures would be useful for determining synaptic strength differences between cell types. Moreover, since virus was used to label D2 cells the caveat that unlabeled cells may constitute putative D1 and unlabeled D2 cells should be addressed. That PV neurons display larger currents relative to MSNs is not surprising, since PV interneurons are broadly characterized by the presence of CP-AMPARs and enhanced excitatory synaptic strength in striatal structures and cortex. Furthermore, NAcc PV neurons been implicated in mediating aversive behavior independent of the PVT (see work by Morales group (NIDA) for example).  
+
+<|ref|>text<|/ref|><|det|>[[115, 511, 882, 562]]<|/det|>
+We agree and have lightened the language of the manuscript to reflect putative targeting of D1- MSNs. Furthermore, we have added the following paragraph to the discussion which addresses the design of this electrophysiological experiment (see page 7):  
+
+<|ref|>text<|/ref|><|det|>[[173, 575, 825, 830]]<|/det|>
+"Our electrophysiological data show that accumbal PV interneurons, as compared to putative D1- and D2- MSNs, receive elevated excitatory drive from PVT neurons, although there are potential caveats to our viral targeting techniques. First, we used D2- Cre and PV- Cre transgenic mice to target MSNs or PV interneurons, respectively, which could have led to variability in ChrinsonR expression between groups of animals. Second, in our D2- Cre transgenic mice, we classified non- fluorescent neurons as putative D1- MSNs, whereas these cells could have been unlabeled D2- MSNs or other cell populations. Despite these caveats, our findings are consistent with previous literature showing that accumbal fast- spiking interneurons (FSIs) receive greater excitatory input from PVT as compared with undefined MSNs using a within- subject design<sup>30</sup>. However, further studies comparing PVT synaptic input to each specific cell- type, including other subclasses of interneurons, within subjects could improve our understanding of PVT \(\rightarrow\) NAc circuit biology."  
+
+<|ref|>text<|/ref|><|det|>[[115, 850, 882, 900]]<|/det|>
+4. MOR activation may also inhibit glutamate release onto non-PV targets (e.g. MSNs), which may be contributing to the observed behavioral effects. The authors do not examine these synapses and make strong conclusions about PVT inputs to PV that warrant further
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 44, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 881, 142]]<|/det|>
+investigation. Furthermore, their physiology does not directly demonstrate presynaptic effects with the data presented, though it's implied.  
+
+<|ref|>text<|/ref|><|det|>[[115, 156, 882, 316]]<|/det|>
+This was an excellent point made by the Reviewer which we now address with new data and additional discussion. Using patch clamp electrophysiology, we confirm that DAMGO decreases PVT excitatory input to both accumbal PV- INs and MSNs in a manner that is blocked by Cre- dependent knockout of PVT \(\mu\) - ORs (see new Figure 6e- h). Thus, \(\mu\) - OR activation could be driving behavioral disinhibition through reduced PVT synaptic input to PV- INs, MSNs, or other cell types in NAc. However, our substantial data showing that PVT \(\rightarrow\) NAc<sup>PV- IN</sup> circuitry is necessary for the suppression of sucrose seeking suggests that this pathway is likely involved. Nonetheless, we have significantly lightened the language of the manuscript to ensure that one pathway over another is not assumed, and we add additional discussion in this regard (see page 7, text revisions in response to concern #1 above).  
+
+<|ref|>text<|/ref|><|det|>[[115, 331, 882, 413]]<|/det|>
+5. In fig 1, The authors describe a tonic decrease in ensemble 2 and 3 activity over the course of the sucrose SA session that is blocked by fear provoking stimuli and extinction and reappears upon reinstatement. However, the emphasis is placed on ensemble 3 activity aligned to lever pressing based on decoding accuracy. Further discussion on this would be useful for the reader.  
+
+<|ref|>text<|/ref|><|det|>[[115, 427, 882, 477]]<|/det|>
+We would like to thank the Reviewer for this comment, and we have therefore adjusted some of the paragraph describing these imaging results to ensure that each ensemble is appropriately addressed without bias on ensemble 3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 492, 882, 557]]<|/det|>
+6. The authors state that ensemble 1 and 3 decoding accuracy is decreased after heroin due to decreased changes in calcium activity, which is consistent with 4H. However, 4G suggest that decoding accuracy in ensemble 1 is increased relative to saline day. The same mismatch appears for ensemble 2, but not as pronounced as ensemble 1.  
+
+<|ref|>text<|/ref|><|det|>[[115, 572, 882, 732]]<|/det|>
+This is a fantastic catch by the Reviewer that was due to some data analysis imperfections, which we have since found and fixed. Specifically, for the CDF plots we were calculating the decoding values for each neuron as the unshuffled decoding score minus the averaged shuffled decoding score across all neurons, such that chance decoding would equal 0. This is an issue as different neurons and clusters are going to have different shuffled values that deviate from one another due to the number of lever presses within a session (specifically, low presses generally result in higher shuffled decoding scores). The summarized heatmap therefore deviated from the results in the CDF plots as each cluster's decoding score was analyzed as its own unshuffled vs shuffled score within the heatmap (using a simple t-test). We therefore have addressed the above data analysis concerns in a few ways:  
+
+<|ref|>text<|/ref|><|det|>[[144, 747, 882, 860]]<|/det|>
+1) We have taken out the heatmap, which served as a poor descriptor of the data.  
+2) We use shuffled decoding values for each neuron for decoding score normalization.  
+3) We use single-cell tracking to analyze the same neurons between the two behavioral sessions, such that between-cell and between-cluster variability is accounted for (this is also in response to Reviewer #3, Concern #6).  
+4) We show that both the activated and inhibited ensembles display significant response adaptation following heroin injection (see Figure 4h).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[114, 44, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[144, 108, 882, 158]]<|/det|>
+5) We show both population and cluster-specific decoding within CDF plots (see new Figure 4i, j), revealing that decoding scores are significantly decreased at the population level and for cluster #3 (see new Supplementary Fig. 7c-e).  
+
+<|ref|>text<|/ref|><|det|>[[114, 172, 882, 254]]<|/det|>
+7. The switch to real time place preference evoked by opto PVT to NAcc stim with heroin injection is very interesting and in some ways may be relevant to the potentiation of sucrose self-administration that is observed in response to opto stim, TMT, and yohimbine in fig 4 and 5. This suggest that MOR activity may be unmasking a population of pro-sucrose administration neurons upon inhibition of MOR sensitive neurons.  
+
+<|ref|>text<|/ref|><|det|>[[113, 268, 881, 301]]<|/det|>
+We agree with the Reviewer about this interesting finding, and have added additional language in the results to highlight its importance (see page 5):  
+
+<|ref|>text<|/ref|><|det|>[[173, 316, 825, 364]]<|/det|>
+"Together, these data suggest that systemic opioids may be modulating PVT \(\rightarrow\) NAc neuronal activity, such that sucrose self-administration is promoted, rather than inhibited, in the presence of behavioral suppressors."  
+
+<|ref|>text<|/ref|><|det|>[[114, 379, 882, 444]]<|/det|>
+8. In supplementary figure s2, cluster 1 seems to contain a subset of cells that are briefly inhibited prior to lever press and are excited during reward delivery late in acquisition and inhibited during reinstatement, suggesting they are distinct from other cells in cluster 1. This cluster in some ways has both cluster 1 and 3 properties.  
+
+<|ref|>text<|/ref|><|det|>[[113, 458, 882, 492]]<|/det|>
+This was an excellent observation by the Reviewer. We have added and adjusted language in the discussion to include this point (see page 8):  
+
+<|ref|>text<|/ref|><|det|>[[173, 506, 825, 555]]<|/det|>
+"...though we cluster PVT \(\rightarrow\) NAc neurons by activity (see Fig. 1), each ensemble seems to contain subsets of cells with distinct responses, despite isolating PVT neurons by location (i.e., posterior) and connection (i.e., projections to NAc)."  
+
+<|ref|>text<|/ref|><|det|>[[114, 569, 882, 652]]<|/det|>
+9. The MOR statining of PVT neurons and their terminals could be due to chance alone as MOR-like immunoreactivity appears to be widespread across the image. In the zoom within the representative images it appears that background from adjacent regions was cropped or subtracted. Immunostaining for MOR with this in MOR IoxP mice would be useful for validating the antibody and genetic approach.  
+
+<|ref|>text<|/ref|><|det|>[[114, 666, 882, 699]]<|/det|>
+This is a fantastic point that we have addressed through several experiments and adjustments to the figures/text.  
+
+<|ref|>text<|/ref|><|det|>[[114, 714, 882, 826]]<|/det|>
+First, we show new example IHC that we used to validate Cre- dependent \(\mu\) - OR knockout in \(\mu\) - OR IoxP mice (New Supplementary Figure 8). This IHC experiment was performed using a different antibody, which was originally used by others to validate \(\mu\) - OR knockout in \(\mu\) - OR IoxP mice (Cui et al. 2014 PMID: 24413699). We therefore replaced the original images of PVT somata with those taken using the new antibody, and we do not perform the same background subtraction based on the Reviewer's excellent point. Furthermore, we do not include images of PVT \(\rightarrow\) NAc axon colocalized with \(\mu\) - OR IHC as overlap could be due to chance.  
+
+<|ref|>text<|/ref|><|det|>[[113, 841, 881, 875]]<|/det|>
+Second, we show example images and quantification from a new RNAscope experiment that was used to demonstrate Cre- dependent knockout of \(\mu\) - OR mRNA (see new Figure 5a- c).  
+
+<|ref|>text<|/ref|><|det|>[[113, 889, 881, 907]]<|/det|>
+Third, we show that PVT neuronal excitability and synaptic input to downstream NAc neurons are
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 44, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 883, 143]]<|/det|>
+inhibited by the \(\mu\) - OR agonist DAMGO, effects that are blocked by Cre- dependent knockout of PVT \(\mu\) - ORs (new Figure 6e- h; Supplementary Figure 9).  
+
+<|ref|>text<|/ref|><|det|>[[115, 156, 883, 190]]<|/det|>
+Altogether, these experiments justify the use of the antibody, \(\mu\) - OR loxP mice, and the point that \(\mu\) - ORs are expressed and functional within PVT \(\rightarrow\) NAc projection neurons.  
+
+<|ref|>text<|/ref|><|det|>[[115, 204, 882, 317]]<|/det|>
+10. NAcc PV cells are sparse relative to other interneuron populations. Moreover, PV-Cre mice used in the present study have incomplete genetic penetrance in the NAcc and may select for subtypes of NAcc fast-spiking interneurons in addition to non-PV cells with CP-AMPARs (see Adam Carter's work and Yan Dong). The authors should mention the incomplete penetrance to let the reader know that this may pertain a selective subpopulation of PV cells and overall broaden the discussion of how PV cells are limited in density in NAcc relative to striatum in the context of the present work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 331, 883, 365]]<|/det|>
+We agree and have now added a paragraph to the discussion section to address these concerns (see page 7- 8):  
+
+<|ref|>text<|/ref|><|det|>[[173, 379, 825, 746]]<|/det|>
+"Our data support the idea that accumbal PV interneurons and CP- AMPRs are necessary for the suppression of sucrose self- administration. Previously, others have shown that accumbal PV interneurons, as well as other FSIs within NAc, can act as powerful regulators of local neuronal activity and behavior despite being sparsely distributed<sup>41- 45</sup>. While we used pharmacology and chemogenetics to target CP- AMPArs and PV interneurons, respectively, it is possible that these methods could have off- target effects. For example, non- PV cells within NAc could express CP- AMPArs, and thus CP- AMPAr antagonism may act on other FSIs or non- PV cell types. Furthermore, it is possible that our targeting of PV interneurons could have profound effects on downstream neurophysiology, and therefore may not be completely selective for our circuit- of- interest. Finally, our PV interneuron cell targeting is likely to select for only a subpopulation of PV- expressing neurons due to incomplete genetic penetrance of the PV- Cre transgenic mouse line<sup>46,47</sup>. Notably, our electrophysiological recordings suggest that our Cre- dependent targeting of PV interneurons at least selects for FSIs, as fluorescent cells within PV- Cre mice displayed fast- spiking properties. Additionally, we find that these cells are inwardly rectifying, suggesting the presence of CP- AMPArs. Despite these findings, future studies selectively targeting CP- AMPArs at PVT \(\rightarrow\) NAc<sup>PV- IN</sup> synapses could elucidate the precise role of these receptors for PVT \(\rightarrow\) NAc- dependent behavioral suppression."  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 763, 880, 797]]<|/det|>
+## 11. The authors should show the lever press – aligned population activity during the fear-provoking stimuli in addition to the change across the session that is reported.  
+
+<|ref|>text<|/ref|><|det|>[[115, 811, 882, 908]]<|/det|>
+While we agree with the Reviewer that this would be interesting to show, we have not included this in our revision. First, our in vivo imaging heatmaps are generated by averaging neuronal activity across active lever presses and subjects. Neuronal activity is time- locked to the active lever press and, due to the low number of active lever presses observed on days where mice were exposed to behavioral suppressors (e.g., TMT, yohimbine, extinction learning), it is difficult not possible to generate a heatmap that accurately depicts lever pressing during behavioral
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 44, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 883, 142]]<|/det|>
+suppression. Thus, we decided the most accurate way to present changes in PVT \(\rightarrow\) NAc activity during behavioral suppression was by showing the change in fluorescence across sessions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 156, 881, 205]]<|/det|>
+## 12. The intrinsic properties of the MSNs in the representative traces do not show the well-characterized regular spiking properties of healthy MSNs ex-vivo, raising the possibility the cells / preparation was not optimal.  
+
+<|ref|>text<|/ref|><|det|>[[115, 220, 882, 333]]<|/det|>
+We agree and would like to reassure the Reviewer that ex vivo cell preparation was fine. The lead PI has performed slice recordings for many years (including in striatum) and identical methodologies were used in this case. The issue is likely that the previous waveforms were from sweeps wherein a very large current pulse was administered ( \(>250\mathrm{pA}\) ; in attempt to show the very high frequency of firing in PV interneurons vs spike adaptation in MSNs). Thus, we have now adjusted the waveforms to show spiking in each cell type in response to a moderate current injection (50pA; see New Figure 3B).  
+
+<|ref|>text<|/ref|><|det|>[[115, 347, 882, 397]]<|/det|>
+13. The authors should cite relevant work on thalamostriatal circuitry where appropriate from the Penzo group at NIMH describing a role for thalamostriatal inputs in promoting avoidance and homeostatic feeding behavior.  
+
+<|ref|>text<|/ref|><|det|>[[115, 411, 881, 444]]<|/det|>
+Although we had cited Penzo et al., 2015, Beas et al. 2018, and Gao et al., 2020 (each from the Penzo group), we have now added Beas et al., 2020.  
+
+<|ref|>text<|/ref|><|det|>[[115, 458, 882, 508]]<|/det|>
+14. The ordering of the supplemental data makes it a little bit difficult to follow. I understand clumping all the inactive lever presses into one figure, but other figures are out of order relative to how the data are presented in the main text and figures.  
+
+<|ref|>text<|/ref|><|det|>[[115, 522, 882, 571]]<|/det|>
+We apologize for any confusion surrounding the Supplementary Figures. We have now distributed the inactive lever pressing graphs across Supplementary Figures, such that it aligns with how the data are presented within the main text.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[115, 45, 454, 93]]<|/det|>
+# Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 536, 125]]<|/det|>
+Reviewer #2 (Recommendations for the authors):  
+
+<|ref|>text<|/ref|><|det|>[[115, 142, 882, 255]]<|/det|>
+Vollmer et al. report the results of an interesting series of experiments providing much needed new knowledge on the organization of the thalamostriatal pathway from paraventricular thalamus to nucleus accumbens in reward seeking behavior. The authors show, quite convincingly, that this circuit involves inputs onto parvalbumin neurons in the accumbens, is enriched in calcium permeable AMPA receptors, is opioid sensitive, and generally inhibits reward seeking across a variety of environmental conditions. These are all important knowledge gains.  
+
+<|ref|>text<|/ref|><|det|>[[115, 269, 882, 350]]<|/det|>
+There is a lot of work in this manuscript, some of it is very clever, all of it is very well done, and the manuscript is very well presented. It significantly extends the field. I think the manuscript will be of interest to many readers and that it adds critical new knowledge. I enjoyed reading it and I think others will too. I had only the following comments on the manuscript:  
+
+<|ref|>text<|/ref|><|det|>[[115, 365, 881, 415]]<|/det|>
+We would like to thank Reviewer #2 for their excellent suggestions and excitement for our study. We have responded to each concern below, through the addition of new experiments, analyses, discussion points, and consideration of alternative interpretations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 432, 882, 562]]<|/det|>
+1. It is not clear to me that the manuscript speaks to "risky reward motivated behaviors". For example, the predator odor is presented in the home cages, not when mice are seeking sucrose. So, there is little to no 'risk' here. There is stress etc, but risk implies an adverse consequence of seeking sucrose. Moreover, the authors show that the same circuit mechanism suppress sucrose seeing after extinction training, which clearly has no risk at all. I am well persuaded by these data that this circuit is a general one for suppressing reward seeking behavior (as predicted by others), but I am less persuaded that the reader learns much about risk. I suggest that "risky" is removed from the title, abstract etc.  
+
+<|ref|>text<|/ref|><|det|>[[115, 576, 780, 593]]<|/det|>
+We agree and have changed the title according to the Reviewer's recommendation:  
+
+<|ref|>text<|/ref|><|det|>[[172, 608, 820, 625]]<|/det|>
+"An opioid- gated thalamoaccumbal circuit for the suppression of reward seeking".  
+
+<|ref|>text<|/ref|><|det|>[[115, 641, 882, 674]]<|/det|>
+Additionally, we have removed language from the manuscript wherein "risky" was used inappropriately.  
+
+<|ref|>text<|/ref|><|det|>[[115, 689, 882, 850]]<|/det|>
+2. "Goal-directed". In a number of places, the selective effects of manipulations on the 'active' lever and no effect on the 'inactive' lever are used to support the claim that behavior and the effects of manipulations are 'goal directed'. This is hard to evaluate because there is no actual test of the goal directedness of the behavior here (e.g., contingency degradation; outcome devaluation). Moreover, it appears the sucrose delivery tube was physically located under the 'active' lever (Figure 1a), which reduces the utility and relevance of the 'inactive' lever to the task. I think what the authors mean is that manipulations were lever or behaviorally specific. I am certainly persuaded of this. It may be more helpful to use these terms rather than imputing untested mental states to the animals.  
+
+<|ref|>text<|/ref|><|det|>[[115, 867, 882, 900]]<|/det|>
+We thank the Reviewer for bringing up these issues. We have changed the language within the manuscript to describe our manipulations as being lever- specific, rather than goal directed.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 45, 453, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 882, 177]]<|/det|>
+Furthermore, we apologize for any ambiguity within previous versions of Figure 1a. We have edited the figures so that the sucrose delivery tube appears to be directly in- between the active and inactive levers. We have also added language to clarify the sucrose spout position within the methods:  
+
+<|ref|>text<|/ref|><|det|>[[115, 192, 504, 210]]<|/det|>
+Under Head- fixed behavior, we report (page 24):  
+
+<|ref|>text<|/ref|><|det|>[[172, 225, 825, 259]]<|/det|>
+"Sucrose delivery spouts were arranged equally between both levers so that mice would not be biased toward either lever."  
+
+<|ref|>text<|/ref|><|det|>[[115, 274, 882, 357]]<|/det|>
+3. The decoder results are interesting, but these are based on only using the 1s before each lever press vs a random 1-s control. I wondered why only a 1s period was used and why this 1s was chosen over other possible durations? This could be better justified. Also, did the authors test a decoder using any other pre-lever press durations? If yes, how did they control for these exploratory analyses in their final statistical analyses?  
+
+<|ref|>text<|/ref|><|det|>[[115, 371, 882, 405]]<|/det|>
+This is an excellent question that deserves justification within the text, which we now add (see page 26):  
+
+<|ref|>text<|/ref|><|det|>[[173, 418, 825, 548]]<|/det|>
+"We used a 1- second epoch prior to the lever press based on: (1) a pre- lever press epoch would ensure that the decoding was not due to cue presentation, liquid delivery, or sucrose consumption and (2) past studies have demonstrated single- cell calcium events during a 1- second pre- reward trace interval can be used to accurately predict reward learning within a Pavlovian conditioning task<sup>6,26</sup>. Thus, a 1- second epoch immediately before lever pressing seemed most appropriate for our decoding analysis and was chosen beforehand such that only one analysis was performed."  
+
+<|ref|>text<|/ref|><|det|>[[115, 563, 882, 645]]<|/det|>
+4. The identification of these effects to PV interneurons is neat. PV neurons are sparse in the accumbens, so it is interesting that their manipulations here have such strong effects. However, the reader never learns how selective the hM4Di was to PV neurons in the accumbens in Figure 3i. Can the authors confirm that other cells in the accumbens did not express the DREADD?  
+
+<|ref|>text<|/ref|><|det|>[[115, 660, 882, 779]]<|/det|>
+We agree and have now added new data validating the use of the DIO- hM4Di- DREADD- mCherry for CNO- induced inhibition of NAc PV interneurons (see new Supplemental Figure 6). We specifically find that mCherry- expressing neurons display electrophysiological properties consistent with PV interneurons, and these cells are inhibited by bath application of CNO. In contrast, neighboring mCherry- negative neurons were not inhibited by CNO and show electrophysiological properties consistent with other NAc cells (in particular, medium spiny neurons).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 795, 172, 811]]<|/det|>
+## Minor:  
+
+<|ref|>text<|/ref|><|det|>[[115, 826, 882, 860]]<|/det|>
+1. I may have missed it, but do the authors state the number of cells in each cluster in Figure 1i or in Figure 4? These are worth clearly stating.  
+
+<|ref|>text<|/ref|><|det|>[[115, 874, 882, 907]]<|/det|>
+We agree and have now stated the number of cells in each cluster within the captions for Figures 1 and 4.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[114, 45, 454, 95]]<|/det|>
+Vollmer, Green et al., 2022Nature Communications ResubmissionResponse to Reviewers  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 124, 591, 142]]<|/det|>
+## 2. State dose of DAMGO used in the manuscript proper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 157, 883, 192]]<|/det|>
+We agree and have stated the dose of DAMGO used for intracranial infusions within the manuscript proper.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 207, 497, 224]]<|/det|>
+## 3. Line 240: should refer to Figure 5H not 5F.  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 714, 257]]<|/det|>
+We thank the Reviewer for pointing this out and have corrected this mix- up.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 45, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 536, 126]]<|/det|>
+Reviewer #3 (Recommendations for the authors):  
+
+<|ref|>text<|/ref|><|det|>[[115, 142, 882, 239]]<|/det|>
+The manuscript by Vollmer et al. describes the role of PVT- NAc projection in the suppression of sucrose seeking. Using the head- fixed sucrose seeking paradigm, authors performed two- photon imaging and optogenetic manipulation of PVT neurons projecting to NAc. Furthermore, the authors examine the role of the opioid receptors in this neuronal type. Overall, all experiments are well designed and performed. However, some interpretation is not fully convincing.  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 882, 304]]<|/det|>
+We would like to sincerely thank Reviewer #3 for their enthusiasm for the paper as well as their concerns about some of our data interpretation. In response, we both added experiments and dialed back some of the interpretation which we agree was over- extended.  
+
+<|ref|>text<|/ref|><|det|>[[115, 320, 882, 466]]<|/det|>
+1. It is difficult to connect the contribution of PVT projection to NAc PV neurons with behavioral changes in sucrose seeking. Especially, since several papers showed that PVT inputs preferentially modify D2-MSN in opioid-induced behavioral changes, it is not clear how the contribution of PVT projection is specifically modulating NAc PV neurons in sucrose seeking. Unless authors can image three different neurons (D1-MSN, D2-MSN, PV neurons) while modulating PVT inputs, this interpretation is not fully convincing. Moreover, since it is very likely that the chemogenetic inhibition of PV neurons (Figure 3k-o) will have massive effects on NAc circuitry already, the optogenetic stimulation of PVT together with PV chemogenetic manipulation may not be PV neuron dependent.  
+
+<|ref|>text<|/ref|><|det|>[[115, 481, 882, 610]]<|/det|>
+We would like to thank the Reviewer for this point, which we agree with overall and now address through easing of data interpretation and additional discussion. Specifically, while we predict that PVT \(\rightarrow\) NAc stimulation causes behavioral disinhibition through the activation of downstream PV- INs (considering the experimental evidence within our paper and supporting studies), we provide several critical caveats. Furthermore, while we conclude that opioids can act at \(\mu\) - opioid receptors at PVT \(\rightarrow\) NAc axons, somata, or both to drive behavioral disinhibition, we note that this could be due to a reduction in activity at a variety of PVT \(\rightarrow\) NAc synapses and downstream neurons, and not necessarily at PVT \(\rightarrow\) NAc<sup>PV-IN</sup> synapses specifically.  
+
+<|ref|>text<|/ref|><|det|>[[115, 624, 882, 898]]<|/det|>
+We do feel it is important to consider several other points. First, we show that intra- NAc infusion of a CP- AMPAR antagonist, but not D1 or D2 receptor antagonist, abolishes the suppression of sucrose seeking caused by PVT \(\rightarrow\) NAc stimulation. Considering the specificity of these receptors for FSIs/PV interneurons (Gittis et al., 2011 PMID: 22049415; Manz et al., 2020 PMID: 32726634) and PVT \(\rightarrow\) NAc<sup>PV-IN</sup> but not PVT \(\rightarrow\) NAc<sup>MSN</sup> synapses (Figure 3), it is likely that the PVT \(\rightarrow\) NAc<sup>PV-IN</sup> pathway is important for our behavioral findings. This interpretation is even further supported, if not confirmed, by our chemogenetics data showing that inhibition of PV- interneurons reverses the behavior- suppressing effects caused by PVT \(\rightarrow\) NAc stimulation, TMT exposure, and yohimbine exposure (Figure 3). It should be noted that these chemogenetic manipulations are less potent than optogenetic inhibition, as we see a modest and cell- type specific reduction in PV- IN excitability following CNO exposure (see new data in Supplementary Fig. 6c). Finally, as mentioned by Reviewer 1, PV interneurons represent a very sparse population of NAc neurons (~1- 2% of cells), and thus we feel that behavioral effects found via PV interneuron inhibition are quite fascinating. It should be noted that D1- MSNs and D2- MSNs, which also have robust collaterals and projections for local and distal inhibition, are far more common than PV- INs (~95% of NAc neurons) and are often targeted via optogenetics and chemogenetics to identify their function for behavioral control.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 44, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[115, 123, 882, 190]]<|/det|>
+Despite these points, we fully agree that there may be off- target effects caused by our pharmacological or chemogenetic manipulations. Thus, we have added additional discussion to highlight the Reviewer's concerns (for example, see page 7- 8, text revisions in response to Reviewer #1, Concern #10 above):  
+
+<|ref|>text<|/ref|><|det|>[[173, 204, 825, 381]]<|/det|>
+"Our data support the idea that accummal PV interneurons and CP- AMPArs are necessary for the suppression of sucrose self- administration. Previously, others have shown that accummal PV interneurons, as well as other FSIs within NAC, can act as powerful regulators of local neuronal activity and behavior despite being sparsely distributed41- 45. While we used pharmacology and chemogenetics to target CP- AMPArs and PV interneurons, respectively, it is possible that these methods could have off- target effects. For example, non- PV cells within NAC could express CP- AMPArs, and thus CP- AMPAr antagonism may act on other FSIs or non- PV cell types. Furthermore, it is possible that our targeting of PV interneurons could have profound effects on downstream neurophysiology, and therefore may not be completely selective for our circuit- of- interest."  
+
+<|ref|>text<|/ref|><|det|>[[115, 395, 882, 476]]<|/det|>
+Next, while we agree with the Reviewer that being able to modulate PVT while monitoring D1- MSNs, D2- MSNs, and PV interneurons would aid in elucidating the precise mechanism in which PVT \(\rightarrow\) NAC neurons influence behavior, recording from 3 different cell types simultaneously during PVT stimulation is simply not feasible with current technologies available. As such, it is not possible for us to complete this experiment.  
+
+<|ref|>text<|/ref|><|det|>[[115, 491, 882, 556]]<|/det|>
+2. Authors found that CP-AMPARs are selectively located in PV neurons in NAC. However, the authors didn't show any effort to describe the potential roles of these CP-AMPARs on sucrose seeking. What is the role of PVT inputs for CP-AMPARs dependent PV neuronal activity?  
+
+<|ref|>text<|/ref|><|det|>[[115, 571, 882, 715]]<|/det|>
+We used a combination of optogenetics and neuropharmacology to show that CP- AMPARs are required for PVT \(\rightarrow\) NAC dependent suppression of sucrose seeking (Figure 3j). While CP- AMPARs are not the focus of the manuscript, these experiments provided initial support for the idea that PVT \(\rightarrow\) NAC PV interneuron synapses may be involved in the suppression of sucrose seeking. This idea was then strongly supported by the subsequent chemogenetics experiments, wherein we combine optogenetic manipulation of PVT \(\rightarrow\) NAC neurons with chemogenetic manipulation of PV interneurons. We have also added additional discussion highlighting the previous findings that PV interneurons within NAC can act as behavioral regulators (see page see page 7- 8, text revisions in response to Reviewer #1, Concern #10 above).  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 882, 811]]<|/det|>
+3. The effects of DAMGO application on synaptic transmission (Figure 5) can be due to the postsynaptic effects since NAC MSNs also express MORs. More extensive analysis of synaptic transmission is needed to fully validate the presynaptic roles of MORs in PVT axon fibers. Again, the behavioral effects of DAMGO (Figure 5h) can be also due to the postsynaptic effects due to the DAMGO-induced changes in postsynaptic MSNs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 825, 790, 843]]<|/det|>
+This is an excellent point, which we now add experiments and discussion to address:  
+
+<|ref|>text<|/ref|><|det|>[[141, 858, 882, 891]]<|/det|>
+a) Synaptic physiology: We have added new synaptic physiology experiments. Using Cre-dependent knockout of \(\mu\) -ORs in PVT, we find that DAMGO no longer reduces
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 44, 454, 94]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[172, 109, 883, 143]]<|/det|>
+PVT \(\rightarrow\) NAc synaptic transmission (new Figure 6h). These data suggest that the effect of DAMGO on optically evoked EPSCs at PVT \(\rightarrow\) NAc synapses requires presynaptic \(\mu\) - ORs.  
+
+<|ref|>text<|/ref|><|det|>[[144, 157, 883, 318]]<|/det|>
+b) Behavioral findings: We have added 5 additional behavioral experiments to highlight the requirement for PVT \(\mu\) -ORs for opioid-induced behavioral disinhibition. While we confirm that heroin disinhibits sucrose seeking regardless of the behavioral suppressor presented (i.e., PVT \(\rightarrow\) NAc stimulation, TMT, or yohimbine), we show that knockout of PVT \(\mu\) -ORs prevents heroin-induced behavioral disinhibition (new Figure 5d-f). Furthermore, we show that PVT \(\mu\) -OR knockout prevents NAc DAMGO infusions from causing TMT- and yohimbine-induced behavioral disinhibition (previously we had only examined PVT \(\rightarrow\) NAc dependent behavioral disinhibition; new Figure 6k, I). Overall, these findings confirm that the effects of systemic heroin and intra-NAc DAMGO on behavioral disinhibition require PVT \(\mu\) -ORs.  
+
+<|ref|>text<|/ref|><|det|>[[144, 332, 882, 414]]<|/det|>
+c) Discussion: As rightly pointed out by the Reviewer, our experiments simply cannot rule out the possibility that DAMGO also has postsynaptic effects on PVT \(\rightarrow\) NAc neurophysiology and thus behavior. Thus, while we indicate that PVT \(\mu\) -ORs are indeed required for our observed findings, we dial back the interpretation throughout the manuscript and add discussion points to this effect (e.g., see page 8):  
+
+<|ref|>text<|/ref|><|det|>[[231, 428, 826, 538]]<|/det|>
+"...additional mechanisms could also contribute to opioid-induced behavioral disinhibition, such as \(\mu\) - OR activation elsewhere in the brain and in other NAc cell types that express \(\mu\) - ORs<sup>54-56</sup>. Overall, while our data suggests that opioid-induced inhibition of PVT \(\rightarrow\) NAc neurons can disinhibit maladaptive behavioral actions, whether these effects are isolated to PVT \(\rightarrow\) NAc somata and/or synapses has yet to be established."  
+
+<|ref|>text<|/ref|><|det|>[[115, 570, 882, 653]]<|/det|>
+4. The systemic injection of heroin can induce neural adaptation in other brain areas, not just PVT-NAc inputs. These other adaptations can change the efficacy of PVT-NAc stimulation with heroin injection. Thus, it is difficult to know how the heroin injection can reverse the effect of PVT-NAc stimulation clearly. Authors should show how PVT-NAc projections are specifically changed by the systemic heroin injection.  
+
+<|ref|>text<|/ref|><|det|>[[115, 667, 882, 700]]<|/det|>
+This is another outstanding point that we now address with several new experiments and discussion:  
+
+<|ref|>text<|/ref|><|det|>[[142, 714, 883, 907]]<|/det|>
+a) Using two-photon calcium imaging we demonstrate that PVT \(\rightarrow\) NAc neurons decrease activity and encoding of sucrose seeking following systemic heroin injection (Figure 4a-j).  
+b) Using patch-clamp electrophysiology we add new data showing that \(\mu\) -OR activation diminishes evoked activity in PVT \(\rightarrow\) NAc neurons, an effect that is prevented by the knockout of PVT \(\mu\) -ORs (new Supplementary Figure 9).  
+c) It remains true that the behavioral effects caused by heroin could be due to adaptations on or upstream of PVT \(\rightarrow\) NAc neurons. Thus, we knocked \(\mu\) -ORs out of PVT, and show that heroin-induced behavioral disinhibition completely absent (see new Figure 5). Thus, the behavioral and physiological effects of heroin described in our study require PVT \(\mu\) -ORs.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 45, 454, 95]]<|/det|>
+## Vollmer, Green et al., 2022 Nature Communications Resubmission Response to Reviewers  
+
+<|ref|>text<|/ref|><|det|>[[144, 123, 883, 174]]<|/det|>
+d) Despite these findings, it remains possible that heroin could also be acting elsewhere to drive behavioral disinhibition and adaptations in neurophysiology, which we address through additions to the text such as that referenced above.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 187, 881, 222]]<|/det|>
+## 5. The effect of the optogenetic manipulation of PVT to NAc projection on TMT or Yohimbe mediated suppression of sucrose seeking was not examined.  
+
+<|ref|>text<|/ref|><|det|>[[115, 235, 883, 270]]<|/det|>
+In Figure 2, we bidirectionally manipulate PVT \(\rightarrow\) NAc neurons following TMT- (Figure 2f) and yohimbine- (Figure 2g) mediated suppression of sucrose seeking.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 283, 883, 317]]<|/det|>
+## 6. Authors can fully take advantage of two-photon imaging. The longitudinal imaging of single neurons can provide the trends in the adaptation of individual neurons.  
+
+<|ref|>text<|/ref|><|det|>[[115, 330, 883, 462]]<|/det|>
+This is an excellent point, and we have now added new data tracking PVT \(\rightarrow\) NAc neuronal activity across time. First, we track PVT \(\rightarrow\) NAc responses across early to late sucrose self- administration. We find that PVT \(\rightarrow\) NAc ensemble dynamics develop over learning, as ensembles 1 (activated) and 3 (inhibited) show a significant response adaptation over time (see new Supplementary Figure 2c- f). Next, we track PVT \(\rightarrow\) NAc neuronal activity from saline to heroin injection tests, and find that PVT \(\rightarrow\) NAc ensembles 1 and 3 have significant response attenuation during the heroin test such that the activity of these neurons can no longer be used to decode active lever pressing (see new Figure 4f- j; Supplementary Fig 7c- e).
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 304, 107]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 129, 393, 146]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 167, 857, 204]]<|/det|>
+The authors have adequately addressed my comments and concerns. Congratulations on the elegant and impactful study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 245, 393, 262]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 284, 879, 380]]<|/det|>
+The authors have made considerable revisions in response to my initial review. I thank them for this. I remain of the view that this is an interesting and important contribution to the literature on PVT function. I am less certain of the ultimate utility of simple lever and extinction type approaches for understanding core problems in reward seeking - and the field has largely moved on from these- but the kinds of insights generated by this manuscript are important and could lead to a revision of this view.  
+
+<|ref|>text<|/ref|><|det|>[[115, 420, 393, 437]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 459, 852, 496]]<|/det|>
+The authors have been very responsive to my comments. They have added results from a number of experiments and also modified language to account for limitations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 517, 841, 555]]<|/det|>
+I also feel that the authors have done a very thorough and careful job addressing concerns of other reviewers'.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a300f42c6c7076d8e9c786cc4f268e362383d3d808eb255900d0a008046f1069/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__a300f42c6c7076d8e9c786cc4f268e362383d3d808eb255900d0a008046f1069/supplementary_0_Peer Review File__a300f42c6c7076d8e9c786cc4f268e362383d3d808eb255900d0a008046f1069.mmd

@@ -0,0 +1,271 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Untangling competition between epitaxial strain and growth stress through examination of variations in local oxidation  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS</B>  
+
+Reviewer #1 (Remarks to the Author):  
+
+Detailed TEM examination was carried out on two TEM specimens lifted out of two oxidized Zr grains. It is a long way to expect that these provide representatives for electrochemical corrosion of all alloys in all harsh environments. There are some good insights however presentation could be much better. There appears to be confusion between preferential growth of some oxide grow directions and transformations between oxide types and Zr metal. A major effort to explain the research would be useful. The present version is extremely difficult to understand, and maybe even incomprehensible to a highly trained researcher in the field. The abstract should present main findings and put them into context. There are no conclusions.  
+
+Incomprehensible, needs to be translated into language that is understandable to the trained practitioner in oxidation studies  
+
+a. Please number pages!  
+
+1. The problem definition is vague. Harsh environment. Is this molten salt at 1000 C, salt water, or CO2 + H2S at 500 C. Please be a little more specific.  
+
+2. Is this electrochemical corrosion in an aqueous NaCl solution or a molten salt?  
+
+3. It would be useful to have a clear statement of research aims to replace the long discussion at the end of the Main Text.  
+
+4. Full chemical composition should be given.  
+
+5. Justification is needed that weight gain is a useful method to determine passive film thickness, and that there is no oxide loss during corrosion.  
+
+6. Please explain all acronyms. What is SPD?  
+
+7. This work is based on a \(2 \text{mm} \times 2 \text{mm}\) specimen that was oxidised, had most of the oxide removed by grinding, from which two TEM specimens were produced and examined. How representative are these? Justification is needed.  
+
+8. These oxides are thick compared with passive films, 1-2 nm in thickness.  
+
+9. Oxide growth direction is undefined.  
+
+10. the transformation from Zr to ZrO2 is presumably by oxidation, If so please state this. What is the
+
+<--- Page Split --->
+
+evidence for the transformation of tetragonal ZrO2 to monoclic ZrO2, and how does this occur?  
+
+11. Or are you trying to say that Zr oxidised to both oxide variants independently? Or are you trying to say that one oxide variant grows on top of the other? What is the evidence for any of this?  
+
+12. What is precession diffraction? Please explain. Presumably the black areas are where there is lack of data?  
+
+13. It would be good to indicate the orientations of the oxide grains and of the Zr grains in Fig. 4(a)  
+
+14. The majority of the gains in Fig. 5b are black which presumably means unidentified  
+
+15. The model in Fig. 7 is neat. However, please indicate the confidence level, as most of grains were black in Fig. 5.  
+
+16. The presented discussion about water splitting can only be relevant until the first oxide monolayer forms.  
+
+17. What is the orientation of a Zr grain with a split basal orientation? There appears to be a mixup of macroscopic and gain focused conceptions.  
+
+18. There is also confusing terminology about oxide transformation, when the model indicates growth of grains of particular orientations on other grains.  
+
+19. The atomic relations in Fig. 8 are instructive. However, the split-basal orientation needs to be included.  
+
+20. Are the strains in Table 1 relevant. If so please provide TEM plane spacings of these oxides in support  
+
+Reviewer #2 (Remarks to the Author):  
+
+The study investigates the mechanisms contributing to zirconium oxide phase formation during corrosion of zirconium. The research method utilized orientation mapping techniques in scanning electron microscope and transmission electron microscope to determine the texture in parent and oxide phases. The experimental data and the theoretical calculations orientation relationship in the phases reveal the energetic favor of the tetragonal oxide phase nucleation on zirconium substrate during corrosion. This is the first time that a study demonstrates the possibility of tetragonal oxide nucleation in zirconium though such theories have been discussed. The manuscript argues that the orientation of the zirconium substrate would affect the oxide nucleation and growth process. Thus, the research work
+
+<--- Page Split --->
+
+argues the influence of epitaxial strain and growth stress in the formation of oxide phases and their texture in zirconium.  
+
+The following are the areas where more clarity and information are needed.  
+
+1. The authors' conclusion on the nucleation of the tetragonal oxide phases is based on the post-corrosion analysis of the microstructure and assumed orientation relationship calculations between Zr, tetragonal oxide, and monoclinic oxide. The analyzed sample has more than \(95\%\) transformed monoclinic oxide phases. Indeed, the results are high-quality. However, the authors may consider using in situ TEM observation of such tetragonal oxide nucleation can provide direct and more convincing evidence. Another option is to have an oxidation condition under which the tetragonal phase is a major phase. There are several papers regarding using precession electron diffraction to study the oxidation of Zr or zircaloy and the tetragonal phase is popular (J. Nucl. Mater. 556 (2021) 153196; Scr Mater. 145 (2018) 95).  
+
+2. In region 2, the number of tetragonal oxide grains analyzed is \(\sim 530\) , which is 3-4 times lesser than that in region 1. The pole figures in region 2 show close to random texture. How representative is the pole figure calculated for the phase in region 2 from a smaller number of grains?  
+
+3. Since each oxide grain has one orientation data point in the SEM-EBSD study, it is critical to comment on the correctness of the indexing.  
+
+4. There is a considerable difference in the orientation measurements between the EBSD and SPED techniques.  
+
+5. Ma et al. APL 106 (2015) 101603 shows non-equilibrium oxidation states, Zr1+, Zr2+, and Zr3+ for suboxide (ZrOx) rather than just ZrO.  
+
+6. Will the initial oxidation product of Zr be amorphous and then form nanocrystal grains?  
+
+7. Could the authors calculate the lattice mismatch and interfacial energy for different orientations? In reality, \(< 9\%\) lattice mismatch is needed for obtaining epitaxy. "epitaxial relationship" could be misleading.  
+
+8. Some typos: line 210: Fig 2C, line 349: 10 mm etc.
+
+<--- Page Split --->
+
+We would like to thank the reviewers for their useful feedback, which have been carefully considered and addressed in the revised manuscript. Please see below responses to individual comments, where the relevant part in the manuscript has been given via the line numbers in the revised document. The changes in the manuscript have been highlighted.  
+
+## Reviewer #1 remarks and authors' responses  
+
+"Detailed TEM examination was carried out on two TEM specimens lifted out of two oxidized Zr grains. It is a long way to expect that these provide representatives for electrochemical corrosion of all alloys in all harsh environments. There are some good insights however presentation could be much better. There appears to be confusion between preferential growth of some oxide grow directions and transformations between oxide types and Zr metal. A major effort to explain the research would be useful. The present version is extremely difficult to understand, and maybe even incomprehensible to a highly trained researcher in the field. The abstract should present main findings and put them into context. There are no conclusions.  
+
+Incomprehensible, needs to be translated into language that is understandable to the trained practitioner in oxidation studies."  
+
+We have provided a clearer explanation of the research and made it more accessible by rewriting the abstract and the introductory section of the main text as well as making additions to the results, discussion, and methods sections. In particular, the importance of and the difficulties in studying complex oxide microstructures to understand corrosion properties have been pointed out in the first paragraph in lines 38- 52. Furthermore, extensive details of the oxidation of Zr alloys are given in lines 102- 120, which explain the mechanism of formation of a protective oxide film. Some of the initial small oxide grains with a particular crystallographic orientation grow preferentially so that the growth stresses in the oxide film are minimised. During this process, the local conditions that stabilised the metastable phases change, e.g. there is a decrease in the stress and the grain size increases, and so a phase transformation to the stable oxide phase is observed.  
+
+Edits to the abstract highlight the main aims and findings of the work, namely the ability to study a complex metal- oxide microstructure, such as that of the oxide layer that forms on Zr alloys during waterside corrosion, and to gain a new understanding of the competing mechanisms, the epitaxial strain and the growth stress, that control the formation of a protective oxide film and hence oxidation. These findings could be used to create tailored metal crystallographic textures so that more protective oxides are formed.  
+
+While the journal format does not include a "conclusions" section, concluding remarks that summarise the main findings of this manuscript have been extended. We have shown that in alloys which experience inward corrosion and a Pilling- Bedworth ratio of more than one, the epitaxial strain forms a less protective and disordered oxide grain morphology, whereas the growth stress drives a well- ordered protective oxide microstructure, and therefore better corrosion performance. Therefore, it is important to consider local substrate orientations even for materials considered to be highly textured as minor changes in orientation can still lead to significant changes to oxide protectiveness. These findings demonstrate the potential of optimising the processing routes of engineering alloys, based on a crystallographic texture, which is optimum for the corrosion properties. These concluding remarks can be found in lines 578- 587.
+
+<--- Page Split --->
+
+"a. Please number pages!"  
+
+Page numbers have been added to the manuscript.  
+
+"1. The problem definition is vague. Harsh environment. Is this molten salt at 1000 C, salt water, or \(CO_2 + H_2S\) at 500 C. Please be a little more specific.  
+
+2. Is this electrochemical corrosion in an aqueous NaCl solution or a molten salt?"  
+
+Details about the corrosion environment used in the present study have been added to the introductory section (lines 126- 127) and expanded in the methods section (lines 593- 596): "The coupon was subjected to corrosion testing for 46 days at \(350^{\circ}C \pm 0.5^{\circ}C\) in 316H stainless steel autoclaves in simulated pressurised water reactor (PWR) chemistry at a raised pH level as part of a large- scale testing program \(^{69}\) , forming an average oxide thickness of \(\sim 1.2 \mu m\) as estimated from weight gain data".  
+
+"3. It would be useful to have a clear statement of research aims to replace the long discussion at the end of the Main Text."  
+
+The research aims have been made clearer in lines 121- 135, to reflect the main aim of demonstrating how our multi- scale analysis can be used to provide a new insight into the oxidation of zirconium alloys in typical pressurised water reactor chemistry by studying the substrate and oxide microstructure and the main driving mechanisms, the epitaxial strain and the growth stress.  
+
+"4. Full chemical composition should be given."  
+
+The full composition has been added in the methods section and materials selection subsection, in particular 1.5 wt%. Sn, 0.14 wt% Fe, 0.1 wt% Cr and 0.06 wt% Ni, where the balance is Zr (please see lines 591- 593).  
+
+"5. Justification is needed that weight gain is a useful method to determine passive film thickness, and that there is no oxide loss during corrosion."  
+
+The explanation for using the weight gain as a valid method for determining the oxide film thickness has been added in in the methods section and materials selection subsection, including a reference: "Zirconium forms a protective adherent oxide film, where all the oxygen in the chemical reaction produces zirconium oxide, therefore, the weight gain of the specimens is widely used as a direct measure of the oxide film thickness \(^{68, 71}\) . "(please see lines 596- 599).  
+
+"6. Please explain all acronyms. What is SPD?"  
+
+The 'SPED' acronym ('SPD' acronym is not present in the manuscript) and all other acronyms have been defined at the first instance of their use, e.g. line 123. Furthermore, the acronyms have been reexplained in new sections to help with understanding and remove any possible confusion, for instance lines 47, 122, 141, 270, 313, 326, 379, 639.  
+
+"7. This work is based on a 2 mm x 2 mm specimen that was oxidised, had most of the oxide removed by grinding, from which two TEM specimens were produced and examined. How representative are these? Justification is needed."  
+
+The electron backscatter diffraction (EBSD) analysis of the metal confirms that the metal grains 1 and 1' have crystallographic orientations within the expected crystallographic texture measured in single- phase Zr alloys such as the alloy used in this study, Zircaloy- 2. Metal grains 2 and 2' are representative of a smaller but significant (about 20%) fraction of grains that these alloys
+
+<--- Page Split --->
+
+contain. (Please see added discussion in lines 174- 182). The EBSD analysis of the oxide is based on a very large number of grains (approximately 560,000) and confirms numerous studies<sup>54, 64</sup>, which show monoclinic oxide texture {10<sup>7</sup>} (where \(l = 2\) , 3, 4) and tetragonal oxide texture {001} in the metal grain with typical texture (i.e. grains 1 and 1'). So, although TEM specimens are small, the results agree very well with the EBSD analysis from a much larger number of grains, which agree well with previous studies of this alloy, and so we have high confidence in our results. (Please see lines 147- 157, 497- 508). The importance of understanding the relationship of very localised measurements provided by state- of- the- art techniques to the overall corrosion behaviour is in- fact an important point this work is trying to highlight.  
+
+Furthermore, we have considered the texture of an oxide formed from a large number of metal grains in a typical sample with split- basal texture. We have shown that the modelled oxide macrotexture using the derived orientation- relationships agrees well with macrotexture measurements in the literature. Please see Fig. 9 and the discussion in lines 561- 577.  
+
+"8. These oxides are thick compared with passive films, 1- 2 nm in thickness."  
+
+Further explanations have been added regarding the zirconium oxide film thickness and its protectiveness (lines 102- 120), which is also summarised here. Zirconium is a very reactive metal which forms a semi- passivating oxide layer and exhibits cyclic corrosion kinetics. The oxide layer is protective up to a thickness of about 2 microns, at which point it becomes unstable and a breakdown in the protectiveness of the oxide is observed. This leads to the rapid growth of a new fresh protective oxide layer and so the process repeats in a cyclic manner. In this study, we are focused on the first \(1.2 \mu \mathrm{m}\) , and so only on the protective pre- transition oxide, which has been clarified in lines 142- 144.  
+
+"9. Oxide growth direction is undefined."  
+
+The oxide growth direction has now been marked on several figures in order avoid any confusion: Fig. 1c, Fig. 4a, Fig. 5a, Fig. 6a and Fig. 7. The inward oxide growth has been explained and discussed in lines 86- 108.  
+
+"10. The transformation from Zr to ZrO2 is presumably by oxidation, if so please state this. What is the evidence for the transformation of tetragonal ZrO2 to monoclinic ZrO2, and how does this occur?"  
+
+Yes, the transformation of Zr to ZrO2 is by oxidation, which has been made clearer in lines 102- 110. The tetragonal ZrO2 is meta- stable at typical reactor conditions (please see lines 119- 122 and 157- 163). Further explanation and extensive literature evidence for the tetragonal ZrO2 to monoclinic ZrO2 transformations have been added in lines 110- 120. The tetragonal phase is stabilised by a combination of small grain size, compressive stress and oxygen vacancies, all present near the metal- oxide interface. As the oxide layer grows inwards, the older tetragonal grains get further away from the metal- oxide interfaces where the compressive stress decreases, and they grow larger, and so the conditions that stabilised them are no longer present, and they transform to the stable monoclinic phase.  
+
+"11. Or are you trying to say that Zr oxidised to both oxide variants independently? Or are you trying to say that one oxide variant grows on top of the other? What is the evidence for any of this?"  
+
+There has been extensive evidence that Zr oxidises to a mixture of multiple oxide phases (hexagonal ZrO, tetragonal ZrO2 and monoclinic ZrO2), i.e. with different crystal structure, as discussed in lines 102- 120. As some of these oxide phases are metastable (stabilised by local factors
+
+<--- Page Split --->
+
+near the metal- oxide interface), as the oxidation progresses and the stabilising factors change, they transform to the stable oxide phase (monoclinic). In this study, for the first time we show evidence that during the oxidation of Zr metal grain with an orientation from the typical alloy texture the tetragonal \(\mathrm{ZrO_2}\) phase forms first, which then transforms to monoclinic \(\mathrm{ZrO_2}\) as the oxide thickens and is found to be beneficial for the protectiveness of the oxide (lines 218- 231, 448- 464, 497- 512). The evidence for this comes from the measured orientation data for the two phases using electron backscatter diffraction in Fig. 3. There has been much debate in the literature as to whether tetragonal oxide is always formed as a precursor to monoclinic, or if both can form independently and this study is the first to show direct evidence of this. As discussed in lines 218- 230, our model of possible theoretical orientation relationships shows that the experimentally measured monoclinic texture variants could only be formed if they had formed as tetragonal \(\mathrm{ZrO_2}\) first and then transformed to monoclinic based on the identified orientation relationship. Furthermore, an explanation has been added in lines 235- 237 to clarify what a crystallographic symmetry variant is: it refers to a crystal with the same crystal structure as another variant, but with different orientation with respect to the parent crystal from which it formed during a phase transformation.  
+
+"12. What is precession diffraction? Please explain. Presumably the black areas are where there is lack of data?"  
+
+Please see explanation added in lines 655- 659, including a reference for further information, and further specifications regarding the scanning precession electron diffraction technique, including orientation and phase reliability index criteria for coloured and black areas in the maps in lines 663 and 664. The reliability index criteria are also given in the relevant figure captions (Fig. 4 and 5).  
+
+"13. It would be good to indicate the orientations of the oxide grains and of the Zr grains in Fig. 4(a)"  
+
+The orientation of the metal grain has been denoted by superimposing the Zr unit cell in Figs. 4a and 5a. The monoclinic oxide orientations are given in Fig. 4c and d.  
+
+"14. The majority of the gains in Fig. 5b are black which presumably means unidentified."  
+
+The procedure for identifying grains as indexed (coloured) or non- indexed ('black') is described in the methods section, the scanning precession electron diffraction subsection and appropriate references for this established technique are given in lines 663 and 664. The reliability index is chosen so that we have a very high confidence in the grains that are indexed even if there is a high fraction of non- indexed grains, since we are using the electron backscatter diffraction results from a much larger number of oxide grains to compare and verify with these from the scanning precession electron diffraction.  
+
+"15. The model in Fig. 7 is neat. However, please indicate the confidence level, as most of grains were black in Fig. 5."  
+
+There is a high level of confidence in the model presented in Fig. 7, because although as you point out there is a large number of nonindexed grains using the SPED technique, we have achieved a very high indexing confidence in the EBSD results, which allowed us to sample more than 5 million oxide grains.  
+
+"16. The presented discussion about water splitting can only be relevant until the first oxide monolayer forms."
+
+<--- Page Split --->
+
+The difference in the energetics of the interactions of oxygen and hydrogen with different zirconium metal orientations is relevant for the oxidation process beyond the first oxide monolayer, because the oxide layer grows inwards by diffusion of oxygen into the metal after it diffuses through the existing oxide. Zr has a high O solubility and so there is a layer of oxygen saturated metal followed by layer of a combination of suboxide and metastable oxide phase, which are more likely to form in the presence of higher fraction of oxygen vacancies. And there would be a higher fraction of oxygen vacancies for metal orientations with faster and deeper oxygen penetration, as already described in lines 438- 447.  
+
+"17. What is the orientation of a Zr grain with a split basal orientation? There appears to be a mixup of macroscopic and gain focused conceptions."  
+
+That is a Zr grain with the basal pole \(< 0002>\) of the hexagonal close packed crystal inclined at an angle with respect to the normal direction, i.e. the direction normal to the metal- oxide interface, within the range for a 'split- basal' texture in single- phase Zr alloys. The "split- basal' orientation" was used as a short way to describe these grains, but this has now been removed and explained better. Please see lines 429- 431 and 449- 450. As described in lines 408- 412- 442, "in a typical metal grain in split- basal textured single- phase Zr alloys, the c- axis of the hcp crystal is positioned at \(20^{\circ}\) to \(40^{\circ}\) away from the outer surface normal, and so pyramidal planes with Miller indices {h0il} are close to parallel to the outer surface (Fig. 7a)"  
+
+"18. There is also confusing terminology about oxide transformation, when the model indicates growth of grains of particular orientations on other grains."  
+
+There is "growth of grains of particular orientation on other grains", which is the effect of an epitaxial strain, or in other words, lattice matching between two crystal structures. We show evidence for the epitaxial strain as a mechanism in both the metal- oxide transformation and the transformation between oxide phases (lines 218- 230, 246- 257, 363- 376). New explanations have been added in the introductory part as mentioned previously to make clearer the subject of the different types of transformations and why they occur (lines 102- 120).  
+
+"19. The atomic relations in Fig. 8 are instructive. However, the split- basal orientation needs to be included."  
+
+As mentioned in the response to comment 17, the 'split- basal' orientation refers to the orientation of the Zr substrate grain with the basal pole at an angle with respect to the cladding tube surface. On the other hand, Fig. 8 shows the atomic interfaces formed between the different crystal phases and so it demonstrates the relative atom positions. Fig. 8 has been extended to include schematics of the oxide microstructure and the normal direction in order to make clearer the relative orientation between the atomic interfaces and the global orientations.  
+
+"20. Are the strains in Table 1 relevant. If so please provide TEM plane spacings of these oxides in support"  
+
+The quantities shown in Table 1 have been renamed from 'strains' to 'mismatch' to reflect the fact that these represent theoretical differences in the Zr- Zr distances in the corresponding crystal structure. These are used to as an approximate measure of the epitaxial strains in order to compare qualitatively the lattice restrictions during the phase transformations in both regions, as explained in lines 465- 471.
+
+<--- Page Split --->
+
+## Reviewer #2 remarks and authors' responses  
+
+"The study investigates the mechanisms contributing to zirconium oxide phase formation during corrosion of zirconium. The research method utilized orientation mapping techniques in scanning electron microscope and transmission electron microscope to determine the texture in parent and oxide phases. The experimental data and the theoretical calculations orientation relationship in the phases reveal the energetic favor of the tetragonal oxide phase nucleation on zirconium substrate during corrosion. This is the first time that a study demonstrates the possibility of tetragonal oxide nucleation in zirconium though such theories have been discussed. The manuscript argues that the orientation of the zirconium substrate would affect the oxide nucleation and growth process. Thus, the research work argues the influence of epitaxial strain and growth stress in the formation of oxide phases and their texture in zirconium.  
+
+The following are the areas where more clarity and information are needed.  
+
+1. The authors' conclusion on the nucleation of the tetragonal oxide phases is based on the post-corrosion analysis of the microstructure and assumed orientation relationship calculations between Zr, tetragonal oxide, and monoclinic oxide. The analyzed sample has more than \(95\%\) transformed monoclinic oxide phases. Indeed, the results are high-quality. However, the authors may consider using in situ TEM observation of such tetragonal oxide nucleation can provide direct and more convincing evidence. Another option is to have an oxidation condition under which the tetragonal phase is a major phase. There are several papers regarding using precession electron diffraction to study the oxidation of Zr or zircaloy and the tetragonal phase is popular (J. Nucl. Mater. 556 (2021) 153196; Scr Mater. 145 (2018) 95)."  
+
+We thank the reviewer for this comment on the quality of the results. We would like to emphasise the importance of the compressive stresses in the oxide layer in stabilising the tetragonal phase, as discussed in lines 102- 120. Therefore, as described in the manuscript (lines 47- 52), TEM- based techniques are performed on electron transparent samples, which alter the microstructure by the partial release of these stresses and the transformation of the stress- stabilised tetragonal grains. For that reason, we do not consider in- situ TEM as a viable option to give insights into the effect of the compressive stresses on the tetragonal oxide nucleation. As Harlow et al. points out (Scr Mater. 145 (2018) 95) significant care must be taken when interpreting TEM results for the tetragonal phase, as some stress release is always present. We also note that we used scanning precession diffraction (SPED) in the TEM to study the oxide thickness, grain morphology and to verify the monoclinic oxide texture that was measured using EBSD, where the stress state is maintained. There are useful applications of using SPED for the study of the tetragonal phase, as the reviewer has pointed out, such as when it is chemically stabilised in the case of J. Nucl. Mater. 556 (2021) 153196, or when it is stabilised by the small grain size. However, there are numerous results from non- destructive techniques such as XRD \(^{48 - 51}\) showing that the monoclinic phase is the major phase in standard autoclave- formed oxides, although some transformation of the minor tetragonal phase is expected during TEM preparation.  
+
+"2. In region 2, the number of tetragonal oxide grains analyzed is \(\sim 530\) , which is 3-4 times lesser than that in region 1. The pole figures in region 2 show close to random texture. How representative is the pole figure calculated for the phase in region 2 from a smaller number of grains?"  
+
+Although there are a smaller number of tetragonal grains in region 2, we consider the result of a random texture reliable. We have confirmed the results for both regions 1 and 2, via comparison with regions 1' and 2', which contain 258 and 285 tetragonal grains respectively, as
+
+<--- Page Split --->
+
+shown in the Supplementary material. As the effect of the epitaxial strain on the tetragonal texture has been captured with only 258 grains in region 1', we consider the 531 tetragonal grains in region 2 to be reliable for the determination of a random texture. The manuscript has been edited to include this additional argument in lines 199- 202.  
+
+"3. Since each oxide grain has one orientation data point in the SEM-EBSD study, it is critical to comment on the correctness of the indexing."  
+
+Due to the fine grain size of the oxide, EBSD must be performed using a high- resolution SEM instrument. This retains sufficient brightness in a small beam footprint, which reduces overlapping of diffraction patterns from neighbouring grains and so improves indexing. It is thought that the relatively low indexing rate in the oxide here ( \(\sim 60\%\) ) results from overlapping patterns. It is therefore likely that when a pattern is indexed, it is from a position where the beam coincides towards the centre of an oxide grain, reducing the effect of overlapping patterns from neighbouring grains. We therefore have high confidence in the indexing in these regions, both in terms of orientation accuracy (Average mean angular deviation (MAD) of 1.07 and 1.45 for the monoclinic and tetragonal phases respectively), and in the phase accuracy, as the monoclinic and tetragonal phases have significantly different crystal structures, and so misindexing is unlikely. The accuracy of the indexing using bulk EBSD is reflected by the similarities in observed textures to those observed by non- destructive XRD measurements in the literature \(^{48 - 51}\) . The discussion has been added to the manuscript in lines 147- 157.  
+
+"4. There is a considerable difference in the orientation measurements between the EBSD and SPED techniques."  
+
+The difference in the EBSD and SPED monoclinic oxide texture measurements have been carefully considered and the discussion has been expanded in lines 287- 298 and 353- 358. Although there is a difference between the EBSD and SPED monoclinic oxide texture measurements (15° and 5° difference in the peak misorientation angle away from the {10- 6} monoclinic pole and the {11- 2} monoclinic pole in region 1 and region 2, respectively), that difference does not change the proposed mechanisms for monoclinic oxide texture formation. It can be clearly seen in Figs. 4d and 5d that the SPED measured {10- 6} and the {11- 2} pole figures in regions 1 and 2 confirm the main patterns shown in the EBSD measured pole figures in Fig. 3. Moreover, the difference between the maximum in the misorientation angle distributions and the theoretical value is about 13° for both the SPED and the EBSD in region 1, and 4° and 7° for, respectively, SPED and EBSD in region 2. These values are within the expected uncertainties, and further confirm a very good agreement between the two techniques and between the experimental measurements and the theoretical model. The main reasons for the differences between the two techniques include the large difference between the sampling statistics in the two techniques, misalignment between the samples used in each technique, internal misorientations within the metal grain, misindexing of some SPED data and relative surface orientation before/after oxide removal. Additionally, there are differences in the measured grain population between the two techniques with EBSD biased towards larger grains. For these reasons, it is important to combine different techniques when studying such complex oxide systems.  
+
+"5. Ma et al. APL 106 (2015) 101603 shows non-equilibrium oxidation states, Zr1+, Zr2+, and Zr3+ for suboxide (ZrOx) rather than just ZrO."  
+
+We thank the reviewer for the insightful reference, which finds that a gradual change in the stoichiometry through the formation of suboxides is energetically more favourable compared to an
+
+<--- Page Split --->
+
+abrupt change from Zr to ZrO2 in the initial stages of the oxidation. This supports the findings in the current manuscript, that the metastable phases provide a more gradual route for the formation of monoclinic ZrO2 and therefore a more protective oxide layer. Discussion and reference have been added in lines 520- 528, 532- 533.  
+
+"6. Will the initial oxidation product of Zr be amorphous and then form nanocrystal grains?"  
+
+There have been some limited suggestions in the literature for the presence of amorphous grains in the initial oxidation stages of Zr (e.g. Ploc and Zhou et al.), however, the general opinion is that if they form, they only form in the very initial stage of the oxidation and certainly would be of limited number. Furthermore, the effects of the epitaxial strains that we found are strong enough to support the idea that the amorphous phase might only form in very limited oxide grains and on very short time and length scales in single- phase Zr alloys.  
+
+Ploc R. A. Transmission electron microscopy of thin (<2000 Å) thermally formed ZrO2 films. J Nucl Mater 1968; 28:48- 60.  
+
+Zhou BX, Li Q, Yao MY, Liu WQ, Chu YL. Effect of water chemistry and composition on microstructural evolution of oxide on Zr- alloys. J ASTM Int 2009:360.  
+
+"7. Could the authors calculate the lattice mismatch and interfacial energy for different orientations? In reality, <9% lattice mismatch is needed for obtaining epitaxy. "epitaxial relationship" could be misleading."  
+
+The theoretical lattice mismatch has been calculated in Table 1, which are based on the Zr- Zr distances in the corresponding lattices. To improve clarity, Table 1 and the relevant discussion in lines 466- 471 have been updated. These measures are only used to qualitatively compare the mismatch in the transformations in the two regions, as in reality there will be atom relaxations, specifically in the normal directions and so the mismatch would be lower, so it would fall within the quoted number of 9%. The calculation of the interfacial energies is beyond the aims of the current manuscript, although it would be an interesting study. Our aim was to show that our crystallographic texture measurements clearly showed the presence of orientation relationships between the metal and oxide phases, and we have defined these by modelling the theoretical orientation relationships.  
+
+"8. Some typos: line 210: Fig 2C, line 349: 10 mm etc."  
+
+The typos have been corrected.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+nice work, publish, reviewers' points dealt with  
+
+Reviewer #3 (Remarks to the Author):  
+
+I have read the revised manuscript and I am satisfied with the changes and how the authors addressed the points raised.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a300f42c6c7076d8e9c786cc4f268e362383d3d808eb255900d0a008046f1069/supplementary_0_Peer Review File__a300f42c6c7076d8e9c786cc4f268e362383d3d808eb255900d0a008046f1069_det.mmd

@@ -0,0 +1,378 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 162, 866, 219]]<|/det|>
+Untangling competition between epitaxial strain and growth stress through examination of variations in local oxidation  
+
+<|ref|>image<|/ref|><|det|>[[95, 732, 262, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[271, 732, 880, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 326, 107]]<|/det|>
+## REVIEWER COMMENTS</B>  
+
+<|ref|>text<|/ref|><|det|>[[115, 129, 393, 145]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 166, 876, 321]]<|/det|>
+Detailed TEM examination was carried out on two TEM specimens lifted out of two oxidized Zr grains. It is a long way to expect that these provide representatives for electrochemical corrosion of all alloys in all harsh environments. There are some good insights however presentation could be much better. There appears to be confusion between preferential growth of some oxide grow directions and transformations between oxide types and Zr metal. A major effort to explain the research would be useful. The present version is extremely difficult to understand, and maybe even incomprehensible to a highly trained researcher in the field. The abstract should present main findings and put them into context. There are no conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 342, 800, 379]]<|/det|>
+Incomprehensible, needs to be translated into language that is understandable to the trained practitioner in oxidation studies  
+
+<|ref|>text<|/ref|><|det|>[[115, 401, 296, 418]]<|/det|>
+a. Please number pages!  
+
+<|ref|>text<|/ref|><|det|>[[115, 439, 880, 476]]<|/det|>
+1. The problem definition is vague. Harsh environment. Is this molten salt at 1000 C, salt water, or CO2 + H2S at 500 C. Please be a little more specific.  
+
+<|ref|>text<|/ref|><|det|>[[115, 497, 706, 514]]<|/det|>
+2. Is this electrochemical corrosion in an aqueous NaCl solution or a molten salt?  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 881, 573]]<|/det|>
+3. It would be useful to have a clear statement of research aims to replace the long discussion at the end of the Main Text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 595, 448, 612]]<|/det|>
+4. Full chemical composition should be given.  
+
+<|ref|>text<|/ref|><|det|>[[115, 634, 860, 671]]<|/det|>
+5. Justification is needed that weight gain is a useful method to determine passive film thickness, and that there is no oxide loss during corrosion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 692, 440, 709]]<|/det|>
+6. Please explain all acronyms. What is SPD?  
+
+<|ref|>text<|/ref|><|det|>[[115, 731, 876, 787]]<|/det|>
+7. This work is based on a \(2 \text{mm} \times 2 \text{mm}\) specimen that was oxidised, had most of the oxide removed by grinding, from which two TEM specimens were produced and examined. How representative are these? Justification is needed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 810, 661, 827]]<|/det|>
+8. These oxides are thick compared with passive films, 1-2 nm in thickness.  
+
+<|ref|>text<|/ref|><|det|>[[115, 849, 402, 865]]<|/det|>
+9. Oxide growth direction is undefined.  
+
+<|ref|>text<|/ref|><|det|>[[112, 887, 856, 905]]<|/det|>
+10. the transformation from Zr to ZrO2 is presumably by oxidation, If so please state this. What is the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 90, 815, 107]]<|/det|>
+evidence for the transformation of tetragonal ZrO2 to monoclic ZrO2, and how does this occur?  
+
+<|ref|>text<|/ref|><|det|>[[113, 129, 866, 166]]<|/det|>
+11. Or are you trying to say that Zr oxidised to both oxide variants independently? Or are you trying to say that one oxide variant grows on top of the other? What is the evidence for any of this?  
+
+<|ref|>text<|/ref|><|det|>[[113, 187, 876, 223]]<|/det|>
+12. What is precession diffraction? Please explain. Presumably the black areas are where there is lack of data?  
+
+<|ref|>text<|/ref|><|det|>[[113, 245, 839, 263]]<|/det|>
+13. It would be good to indicate the orientations of the oxide grains and of the Zr grains in Fig. 4(a)  
+
+<|ref|>text<|/ref|><|det|>[[113, 285, 750, 302]]<|/det|>
+14. The majority of the gains in Fig. 5b are black which presumably means unidentified  
+
+<|ref|>text<|/ref|><|det|>[[113, 323, 857, 360]]<|/det|>
+15. The model in Fig. 7 is neat. However, please indicate the confidence level, as most of grains were black in Fig. 5.  
+
+<|ref|>text<|/ref|><|det|>[[113, 381, 866, 417]]<|/det|>
+16. The presented discussion about water splitting can only be relevant until the first oxide monolayer forms.  
+
+<|ref|>text<|/ref|><|det|>[[113, 420, 864, 457]]<|/det|>
+17. What is the orientation of a Zr grain with a split basal orientation? There appears to be a mixup of macroscopic and gain focused conceptions.  
+
+<|ref|>text<|/ref|><|det|>[[113, 479, 883, 515]]<|/det|>
+18. There is also confusing terminology about oxide transformation, when the model indicates growth of grains of particular orientations on other grains.  
+
+<|ref|>text<|/ref|><|det|>[[113, 537, 828, 574]]<|/det|>
+19. The atomic relations in Fig. 8 are instructive. However, the split-basal orientation needs to be included.  
+
+<|ref|>text<|/ref|><|det|>[[113, 596, 881, 614]]<|/det|>
+20. Are the strains in Table 1 relevant. If so please provide TEM plane spacings of these oxides in support  
+
+<|ref|>text<|/ref|><|det|>[[115, 714, 393, 730]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 752, 881, 906]]<|/det|>
+The study investigates the mechanisms contributing to zirconium oxide phase formation during corrosion of zirconium. The research method utilized orientation mapping techniques in scanning electron microscope and transmission electron microscope to determine the texture in parent and oxide phases. The experimental data and the theoretical calculations orientation relationship in the phases reveal the energetic favor of the tetragonal oxide phase nucleation on zirconium substrate during corrosion. This is the first time that a study demonstrates the possibility of tetragonal oxide nucleation in zirconium though such theories have been discussed. The manuscript argues that the orientation of the zirconium substrate would affect the oxide nucleation and growth process. Thus, the research work
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 844, 127]]<|/det|>
+argues the influence of epitaxial strain and growth stress in the formation of oxide phases and their texture in zirconium.  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 671, 166]]<|/det|>
+The following are the areas where more clarity and information are needed.  
+
+<|ref|>text<|/ref|><|det|>[[113, 186, 880, 360]]<|/det|>
+1. The authors' conclusion on the nucleation of the tetragonal oxide phases is based on the post-corrosion analysis of the microstructure and assumed orientation relationship calculations between Zr, tetragonal oxide, and monoclinic oxide. The analyzed sample has more than \(95\%\) transformed monoclinic oxide phases. Indeed, the results are high-quality. However, the authors may consider using in situ TEM observation of such tetragonal oxide nucleation can provide direct and more convincing evidence. Another option is to have an oxidation condition under which the tetragonal phase is a major phase. There are several papers regarding using precession electron diffraction to study the oxidation of Zr or zircaloy and the tetragonal phase is popular (J. Nucl. Mater. 556 (2021) 153196; Scr Mater. 145 (2018) 95).  
+
+<|ref|>text<|/ref|><|det|>[[113, 362, 880, 420]]<|/det|>
+2. In region 2, the number of tetragonal oxide grains analyzed is \(\sim 530\) , which is 3-4 times lesser than that in region 1. The pole figures in region 2 show close to random texture. How representative is the pole figure calculated for the phase in region 2 from a smaller number of grains?  
+
+<|ref|>text<|/ref|><|det|>[[113, 421, 875, 459]]<|/det|>
+3. Since each oxide grain has one orientation data point in the SEM-EBSD study, it is critical to comment on the correctness of the indexing.  
+
+<|ref|>text<|/ref|><|det|>[[113, 460, 844, 497]]<|/det|>
+4. There is a considerable difference in the orientation measurements between the EBSD and SPED techniques.  
+
+<|ref|>text<|/ref|><|det|>[[113, 499, 848, 536]]<|/det|>
+5. Ma et al. APL 106 (2015) 101603 shows non-equilibrium oxidation states, Zr1+, Zr2+, and Zr3+ for suboxide (ZrOx) rather than just ZrO.  
+
+<|ref|>text<|/ref|><|det|>[[115, 538, 775, 556]]<|/det|>
+6. Will the initial oxidation product of Zr be amorphous and then form nanocrystal grains?  
+
+<|ref|>text<|/ref|><|det|>[[115, 558, 866, 615]]<|/det|>
+7. Could the authors calculate the lattice mismatch and interfacial energy for different orientations? In reality, \(< 9\%\) lattice mismatch is needed for obtaining epitaxy. "epitaxial relationship" could be misleading.  
+
+<|ref|>text<|/ref|><|det|>[[115, 617, 497, 634]]<|/det|>
+8. Some typos: line 210: Fig 2C, line 349: 10 mm etc.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 848, 152]]<|/det|>
+We would like to thank the reviewers for their useful feedback, which have been carefully considered and addressed in the revised manuscript. Please see below responses to individual comments, where the relevant part in the manuscript has been given via the line numbers in the revised document. The changes in the manuscript have been highlighted.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 163, 463, 178]]<|/det|>
+## Reviewer #1 remarks and authors' responses  
+
+<|ref|>text<|/ref|><|det|>[[117, 189, 866, 325]]<|/det|>
+"Detailed TEM examination was carried out on two TEM specimens lifted out of two oxidized Zr grains. It is a long way to expect that these provide representatives for electrochemical corrosion of all alloys in all harsh environments. There are some good insights however presentation could be much better. There appears to be confusion between preferential growth of some oxide grow directions and transformations between oxide types and Zr metal. A major effort to explain the research would be useful. The present version is extremely difficult to understand, and maybe even incomprehensible to a highly trained researcher in the field. The abstract should present main findings and put them into context. There are no conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 336, 819, 369]]<|/det|>
+Incomprehensible, needs to be translated into language that is understandable to the trained practitioner in oxidation studies."  
+
+<|ref|>text<|/ref|><|det|>[[117, 380, 878, 551]]<|/det|>
+We have provided a clearer explanation of the research and made it more accessible by rewriting the abstract and the introductory section of the main text as well as making additions to the results, discussion, and methods sections. In particular, the importance of and the difficulties in studying complex oxide microstructures to understand corrosion properties have been pointed out in the first paragraph in lines 38- 52. Furthermore, extensive details of the oxidation of Zr alloys are given in lines 102- 120, which explain the mechanism of formation of a protective oxide film. Some of the initial small oxide grains with a particular crystallographic orientation grow preferentially so that the growth stresses in the oxide film are minimised. During this process, the local conditions that stabilised the metastable phases change, e.g. there is a decrease in the stress and the grain size increases, and so a phase transformation to the stable oxide phase is observed.  
+
+<|ref|>text<|/ref|><|det|>[[117, 561, 866, 664]]<|/det|>
+Edits to the abstract highlight the main aims and findings of the work, namely the ability to study a complex metal- oxide microstructure, such as that of the oxide layer that forms on Zr alloys during waterside corrosion, and to gain a new understanding of the competing mechanisms, the epitaxial strain and the growth stress, that control the formation of a protective oxide film and hence oxidation. These findings could be used to create tailored metal crystallographic textures so that more protective oxides are formed.  
+
+<|ref|>text<|/ref|><|det|>[[117, 675, 870, 846]]<|/det|>
+While the journal format does not include a "conclusions" section, concluding remarks that summarise the main findings of this manuscript have been extended. We have shown that in alloys which experience inward corrosion and a Pilling- Bedworth ratio of more than one, the epitaxial strain forms a less protective and disordered oxide grain morphology, whereas the growth stress drives a well- ordered protective oxide microstructure, and therefore better corrosion performance. Therefore, it is important to consider local substrate orientations even for materials considered to be highly textured as minor changes in orientation can still lead to significant changes to oxide protectiveness. These findings demonstrate the potential of optimising the processing routes of engineering alloys, based on a crystallographic texture, which is optimum for the corrosion properties. These concluding remarks can be found in lines 578- 587.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[120, 85, 320, 100]]<|/det|>
+"a. Please number pages!"  
+
+<|ref|>text<|/ref|><|det|>[[177, 110, 564, 125]]<|/det|>
+Page numbers have been added to the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 136, 860, 170]]<|/det|>
+"1. The problem definition is vague. Harsh environment. Is this molten salt at 1000 C, salt water, or \(CO_2 + H_2S\) at 500 C. Please be a little more specific.  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 730, 197]]<|/det|>
+2. Is this electrochemical corrosion in an aqueous NaCl solution or a molten salt?"  
+
+<|ref|>text<|/ref|><|det|>[[118, 207, 866, 309]]<|/det|>
+Details about the corrosion environment used in the present study have been added to the introductory section (lines 126- 127) and expanded in the methods section (lines 593- 596): "The coupon was subjected to corrosion testing for 46 days at \(350^{\circ}C \pm 0.5^{\circ}C\) in 316H stainless steel autoclaves in simulated pressurised water reactor (PWR) chemistry at a raised pH level as part of a large- scale testing program \(^{69}\) , forming an average oxide thickness of \(\sim 1.2 \mu m\) as estimated from weight gain data".  
+
+<|ref|>text<|/ref|><|det|>[[118, 320, 874, 353]]<|/det|>
+"3. It would be useful to have a clear statement of research aims to replace the long discussion at the end of the Main Text."  
+
+<|ref|>text<|/ref|><|det|>[[118, 364, 866, 432]]<|/det|>
+The research aims have been made clearer in lines 121- 135, to reflect the main aim of demonstrating how our multi- scale analysis can be used to provide a new insight into the oxidation of zirconium alloys in typical pressurised water reactor chemistry by studying the substrate and oxide microstructure and the main driving mechanisms, the epitaxial strain and the growth stress.  
+
+<|ref|>text<|/ref|><|det|>[[120, 442, 473, 458]]<|/det|>
+"4. Full chemical composition should be given."  
+
+<|ref|>text<|/ref|><|det|>[[118, 469, 872, 520]]<|/det|>
+The full composition has been added in the methods section and materials selection subsection, in particular 1.5 wt%. Sn, 0.14 wt% Fe, 0.1 wt% Cr and 0.06 wt% Ni, where the balance is Zr (please see lines 591- 593).  
+
+<|ref|>text<|/ref|><|det|>[[118, 530, 852, 564]]<|/det|>
+"5. Justification is needed that weight gain is a useful method to determine passive film thickness, and that there is no oxide loss during corrosion."  
+
+<|ref|>text<|/ref|><|det|>[[118, 574, 864, 659]]<|/det|>
+The explanation for using the weight gain as a valid method for determining the oxide film thickness has been added in in the methods section and materials selection subsection, including a reference: "Zirconium forms a protective adherent oxide film, where all the oxygen in the chemical reaction produces zirconium oxide, therefore, the weight gain of the specimens is widely used as a direct measure of the oxide film thickness \(^{68, 71}\) . "(please see lines 596- 599).  
+
+<|ref|>text<|/ref|><|det|>[[120, 670, 468, 685]]<|/det|>
+"6. Please explain all acronyms. What is SPD?"  
+
+<|ref|>text<|/ref|><|det|>[[118, 696, 878, 764]]<|/det|>
+The 'SPED' acronym ('SPD' acronym is not present in the manuscript) and all other acronyms have been defined at the first instance of their use, e.g. line 123. Furthermore, the acronyms have been reexplained in new sections to help with understanding and remove any possible confusion, for instance lines 47, 122, 141, 270, 313, 326, 379, 639.  
+
+<|ref|>text<|/ref|><|det|>[[118, 775, 876, 825]]<|/det|>
+"7. This work is based on a 2 mm x 2 mm specimen that was oxidised, had most of the oxide removed by grinding, from which two TEM specimens were produced and examined. How representative are these? Justification is needed."  
+
+<|ref|>text<|/ref|><|det|>[[118, 836, 873, 904]]<|/det|>
+The electron backscatter diffraction (EBSD) analysis of the metal confirms that the metal grains 1 and 1' have crystallographic orientations within the expected crystallographic texture measured in single- phase Zr alloys such as the alloy used in this study, Zircaloy- 2. Metal grains 2 and 2' are representative of a smaller but significant (about 20%) fraction of grains that these alloys
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 877, 242]]<|/det|>
+contain. (Please see added discussion in lines 174- 182). The EBSD analysis of the oxide is based on a very large number of grains (approximately 560,000) and confirms numerous studies<sup>54, 64</sup>, which show monoclinic oxide texture {10<sup>7</sup>} (where \(l = 2\) , 3, 4) and tetragonal oxide texture {001} in the metal grain with typical texture (i.e. grains 1 and 1'). So, although TEM specimens are small, the results agree very well with the EBSD analysis from a much larger number of grains, which agree well with previous studies of this alloy, and so we have high confidence in our results. (Please see lines 147- 157, 497- 508). The importance of understanding the relationship of very localised measurements provided by state- of- the- art techniques to the overall corrosion behaviour is in- fact an important point this work is trying to highlight.  
+
+<|ref|>text<|/ref|><|det|>[[118, 251, 853, 319]]<|/det|>
+Furthermore, we have considered the texture of an oxide formed from a large number of metal grains in a typical sample with split- basal texture. We have shown that the modelled oxide macrotexture using the derived orientation- relationships agrees well with macrotexture measurements in the literature. Please see Fig. 9 and the discussion in lines 561- 577.  
+
+<|ref|>text<|/ref|><|det|>[[120, 330, 691, 346]]<|/det|>
+"8. These oxides are thick compared with passive films, 1- 2 nm in thickness."  
+
+<|ref|>text<|/ref|><|det|>[[118, 357, 862, 494]]<|/det|>
+Further explanations have been added regarding the zirconium oxide film thickness and its protectiveness (lines 102- 120), which is also summarised here. Zirconium is a very reactive metal which forms a semi- passivating oxide layer and exhibits cyclic corrosion kinetics. The oxide layer is protective up to a thickness of about 2 microns, at which point it becomes unstable and a breakdown in the protectiveness of the oxide is observed. This leads to the rapid growth of a new fresh protective oxide layer and so the process repeats in a cyclic manner. In this study, we are focused on the first \(1.2 \mu \mathrm{m}\) , and so only on the protective pre- transition oxide, which has been clarified in lines 142- 144.  
+
+<|ref|>text<|/ref|><|det|>[[120, 504, 425, 520]]<|/det|>
+"9. Oxide growth direction is undefined."  
+
+<|ref|>text<|/ref|><|det|>[[118, 531, 878, 581]]<|/det|>
+The oxide growth direction has now been marked on several figures in order avoid any confusion: Fig. 1c, Fig. 4a, Fig. 5a, Fig. 6a and Fig. 7. The inward oxide growth has been explained and discussed in lines 86- 108.  
+
+<|ref|>text<|/ref|><|det|>[[118, 592, 857, 643]]<|/det|>
+"10. The transformation from Zr to ZrO2 is presumably by oxidation, if so please state this. What is the evidence for the transformation of tetragonal ZrO2 to monoclinic ZrO2, and how does this occur?"  
+
+<|ref|>text<|/ref|><|det|>[[117, 654, 872, 807]]<|/det|>
+Yes, the transformation of Zr to ZrO2 is by oxidation, which has been made clearer in lines 102- 110. The tetragonal ZrO2 is meta- stable at typical reactor conditions (please see lines 119- 122 and 157- 163). Further explanation and extensive literature evidence for the tetragonal ZrO2 to monoclinic ZrO2 transformations have been added in lines 110- 120. The tetragonal phase is stabilised by a combination of small grain size, compressive stress and oxygen vacancies, all present near the metal- oxide interface. As the oxide layer grows inwards, the older tetragonal grains get further away from the metal- oxide interfaces where the compressive stress decreases, and they grow larger, and so the conditions that stabilised them are no longer present, and they transform to the stable monoclinic phase.  
+
+<|ref|>text<|/ref|><|det|>[[118, 818, 870, 851]]<|/det|>
+"11. Or are you trying to say that Zr oxidised to both oxide variants independently? Or are you trying to say that one oxide variant grows on top of the other? What is the evidence for any of this?"  
+
+<|ref|>text<|/ref|><|det|>[[118, 862, 870, 912]]<|/det|>
+There has been extensive evidence that Zr oxidises to a mixture of multiple oxide phases (hexagonal ZrO, tetragonal ZrO2 and monoclinic ZrO2), i.e. with different crystal structure, as discussed in lines 102- 120. As some of these oxide phases are metastable (stabilised by local factors
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 875, 343]]<|/det|>
+near the metal- oxide interface), as the oxidation progresses and the stabilising factors change, they transform to the stable oxide phase (monoclinic). In this study, for the first time we show evidence that during the oxidation of Zr metal grain with an orientation from the typical alloy texture the tetragonal \(\mathrm{ZrO_2}\) phase forms first, which then transforms to monoclinic \(\mathrm{ZrO_2}\) as the oxide thickens and is found to be beneficial for the protectiveness of the oxide (lines 218- 231, 448- 464, 497- 512). The evidence for this comes from the measured orientation data for the two phases using electron backscatter diffraction in Fig. 3. There has been much debate in the literature as to whether tetragonal oxide is always formed as a precursor to monoclinic, or if both can form independently and this study is the first to show direct evidence of this. As discussed in lines 218- 230, our model of possible theoretical orientation relationships shows that the experimentally measured monoclinic texture variants could only be formed if they had formed as tetragonal \(\mathrm{ZrO_2}\) first and then transformed to monoclinic based on the identified orientation relationship. Furthermore, an explanation has been added in lines 235- 237 to clarify what a crystallographic symmetry variant is: it refers to a crystal with the same crystal structure as another variant, but with different orientation with respect to the parent crystal from which it formed during a phase transformation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 378, 852, 412]]<|/det|>
+"12. What is precession diffraction? Please explain. Presumably the black areas are where there is lack of data?"  
+
+<|ref|>text<|/ref|><|det|>[[118, 422, 879, 491]]<|/det|>
+Please see explanation added in lines 655- 659, including a reference for further information, and further specifications regarding the scanning precession electron diffraction technique, including orientation and phase reliability index criteria for coloured and black areas in the maps in lines 663 and 664. The reliability index criteria are also given in the relevant figure captions (Fig. 4 and 5).  
+
+<|ref|>text<|/ref|><|det|>[[118, 501, 874, 518]]<|/det|>
+"13. It would be good to indicate the orientations of the oxide grains and of the Zr grains in Fig. 4(a)"  
+
+<|ref|>text<|/ref|><|det|>[[118, 528, 844, 562]]<|/det|>
+The orientation of the metal grain has been denoted by superimposing the Zr unit cell in Figs. 4a and 5a. The monoclinic oxide orientations are given in Fig. 4c and d.  
+
+<|ref|>text<|/ref|><|det|>[[120, 572, 787, 589]]<|/det|>
+"14. The majority of the gains in Fig. 5b are black which presumably means unidentified."  
+
+<|ref|>text<|/ref|><|det|>[[118, 598, 874, 718]]<|/det|>
+The procedure for identifying grains as indexed (coloured) or non- indexed ('black') is described in the methods section, the scanning precession electron diffraction subsection and appropriate references for this established technique are given in lines 663 and 664. The reliability index is chosen so that we have a very high confidence in the grains that are indexed even if there is a high fraction of non- indexed grains, since we are using the electron backscatter diffraction results from a much larger number of oxide grains to compare and verify with these from the scanning precession electron diffraction.  
+
+<|ref|>text<|/ref|><|det|>[[118, 728, 877, 761]]<|/det|>
+"15. The model in Fig. 7 is neat. However, please indicate the confidence level, as most of grains were black in Fig. 5."  
+
+<|ref|>text<|/ref|><|det|>[[118, 772, 874, 840]]<|/det|>
+There is a high level of confidence in the model presented in Fig. 7, because although as you point out there is a large number of nonindexed grains using the SPED technique, we have achieved a very high indexing confidence in the EBSD results, which allowed us to sample more than 5 million oxide grains.  
+
+<|ref|>text<|/ref|><|det|>[[118, 851, 803, 884]]<|/det|>
+"16. The presented discussion about water splitting can only be relevant until the first oxide monolayer forms."
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 878, 221]]<|/det|>
+The difference in the energetics of the interactions of oxygen and hydrogen with different zirconium metal orientations is relevant for the oxidation process beyond the first oxide monolayer, because the oxide layer grows inwards by diffusion of oxygen into the metal after it diffuses through the existing oxide. Zr has a high O solubility and so there is a layer of oxygen saturated metal followed by layer of a combination of suboxide and metastable oxide phase, which are more likely to form in the presence of higher fraction of oxygen vacancies. And there would be a higher fraction of oxygen vacancies for metal orientations with faster and deeper oxygen penetration, as already described in lines 438- 447.  
+
+<|ref|>text<|/ref|><|det|>[[118, 232, 871, 265]]<|/det|>
+"17. What is the orientation of a Zr grain with a split basal orientation? There appears to be a mixup of macroscopic and gain focused conceptions."  
+
+<|ref|>text<|/ref|><|det|>[[118, 275, 879, 412]]<|/det|>
+That is a Zr grain with the basal pole \(< 0002>\) of the hexagonal close packed crystal inclined at an angle with respect to the normal direction, i.e. the direction normal to the metal- oxide interface, within the range for a 'split- basal' texture in single- phase Zr alloys. The "split- basal' orientation" was used as a short way to describe these grains, but this has now been removed and explained better. Please see lines 429- 431 and 449- 450. As described in lines 408- 412- 442, "in a typical metal grain in split- basal textured single- phase Zr alloys, the c- axis of the hcp crystal is positioned at \(20^{\circ}\) to \(40^{\circ}\) away from the outer surface normal, and so pyramidal planes with Miller indices {h0il} are close to parallel to the outer surface (Fig. 7a)"  
+
+<|ref|>text<|/ref|><|det|>[[118, 422, 833, 456]]<|/det|>
+"18. There is also confusing terminology about oxide transformation, when the model indicates growth of grains of particular orientations on other grains."  
+
+<|ref|>text<|/ref|><|det|>[[118, 466, 875, 569]]<|/det|>
+There is "growth of grains of particular orientation on other grains", which is the effect of an epitaxial strain, or in other words, lattice matching between two crystal structures. We show evidence for the epitaxial strain as a mechanism in both the metal- oxide transformation and the transformation between oxide phases (lines 218- 230, 246- 257, 363- 376). New explanations have been added in the introductory part as mentioned previously to make clearer the subject of the different types of transformations and why they occur (lines 102- 120).  
+
+<|ref|>text<|/ref|><|det|>[[118, 579, 852, 612]]<|/det|>
+"19. The atomic relations in Fig. 8 are instructive. However, the split- basal orientation needs to be included."  
+
+<|ref|>text<|/ref|><|det|>[[118, 623, 877, 725]]<|/det|>
+As mentioned in the response to comment 17, the 'split- basal' orientation refers to the orientation of the Zr substrate grain with the basal pole at an angle with respect to the cladding tube surface. On the other hand, Fig. 8 shows the atomic interfaces formed between the different crystal phases and so it demonstrates the relative atom positions. Fig. 8 has been extended to include schematics of the oxide microstructure and the normal direction in order to make clearer the relative orientation between the atomic interfaces and the global orientations.  
+
+<|ref|>text<|/ref|><|det|>[[118, 736, 847, 769]]<|/det|>
+"20. Are the strains in Table 1 relevant. If so please provide TEM plane spacings of these oxides in support"  
+
+<|ref|>text<|/ref|><|det|>[[118, 780, 860, 864]]<|/det|>
+The quantities shown in Table 1 have been renamed from 'strains' to 'mismatch' to reflect the fact that these represent theoretical differences in the Zr- Zr distances in the corresponding crystal structure. These are used to as an approximate measure of the epitaxial strains in order to compare qualitatively the lattice restrictions during the phase transformations in both regions, as explained in lines 465- 471.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[120, 85, 463, 100]]<|/det|>
+## Reviewer #2 remarks and authors' responses  
+
+<|ref|>text<|/ref|><|det|>[[118, 111, 874, 281]]<|/det|>
+"The study investigates the mechanisms contributing to zirconium oxide phase formation during corrosion of zirconium. The research method utilized orientation mapping techniques in scanning electron microscope and transmission electron microscope to determine the texture in parent and oxide phases. The experimental data and the theoretical calculations orientation relationship in the phases reveal the energetic favor of the tetragonal oxide phase nucleation on zirconium substrate during corrosion. This is the first time that a study demonstrates the possibility of tetragonal oxide nucleation in zirconium though such theories have been discussed. The manuscript argues that the orientation of the zirconium substrate would affect the oxide nucleation and growth process. Thus, the research work argues the influence of epitaxial strain and growth stress in the formation of oxide phases and their texture in zirconium.  
+
+<|ref|>text<|/ref|><|det|>[[120, 292, 686, 308]]<|/det|>
+The following are the areas where more clarity and information are needed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 319, 864, 473]]<|/det|>
+1. The authors' conclusion on the nucleation of the tetragonal oxide phases is based on the post-corrosion analysis of the microstructure and assumed orientation relationship calculations between Zr, tetragonal oxide, and monoclinic oxide. The analyzed sample has more than \(95\%\) transformed monoclinic oxide phases. Indeed, the results are high-quality. However, the authors may consider using in situ TEM observation of such tetragonal oxide nucleation can provide direct and more convincing evidence. Another option is to have an oxidation condition under which the tetragonal phase is a major phase. There are several papers regarding using precession electron diffraction to study the oxidation of Zr or zircaloy and the tetragonal phase is popular (J. Nucl. Mater. 556 (2021) 153196; Scr Mater. 145 (2018) 95)."  
+
+<|ref|>text<|/ref|><|det|>[[117, 484, 878, 775]]<|/det|>
+We thank the reviewer for this comment on the quality of the results. We would like to emphasise the importance of the compressive stresses in the oxide layer in stabilising the tetragonal phase, as discussed in lines 102- 120. Therefore, as described in the manuscript (lines 47- 52), TEM- based techniques are performed on electron transparent samples, which alter the microstructure by the partial release of these stresses and the transformation of the stress- stabilised tetragonal grains. For that reason, we do not consider in- situ TEM as a viable option to give insights into the effect of the compressive stresses on the tetragonal oxide nucleation. As Harlow et al. points out (Scr Mater. 145 (2018) 95) significant care must be taken when interpreting TEM results for the tetragonal phase, as some stress release is always present. We also note that we used scanning precession diffraction (SPED) in the TEM to study the oxide thickness, grain morphology and to verify the monoclinic oxide texture that was measured using EBSD, where the stress state is maintained. There are useful applications of using SPED for the study of the tetragonal phase, as the reviewer has pointed out, such as when it is chemically stabilised in the case of J. Nucl. Mater. 556 (2021) 153196, or when it is stabilised by the small grain size. However, there are numerous results from non- destructive techniques such as XRD \(^{48 - 51}\) showing that the monoclinic phase is the major phase in standard autoclave- formed oxides, although some transformation of the minor tetragonal phase is expected during TEM preparation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 785, 875, 836]]<|/det|>
+"2. In region 2, the number of tetragonal oxide grains analyzed is \(\sim 530\) , which is 3-4 times lesser than that in region 1. The pole figures in region 2 show close to random texture. How representative is the pole figure calculated for the phase in region 2 from a smaller number of grains?"  
+
+<|ref|>text<|/ref|><|det|>[[118, 847, 872, 898]]<|/det|>
+Although there are a smaller number of tetragonal grains in region 2, we consider the result of a random texture reliable. We have confirmed the results for both regions 1 and 2, via comparison with regions 1' and 2', which contain 258 and 285 tetragonal grains respectively, as
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 870, 152]]<|/det|>
+shown in the Supplementary material. As the effect of the epitaxial strain on the tetragonal texture has been captured with only 258 grains in region 1', we consider the 531 tetragonal grains in region 2 to be reliable for the determination of a random texture. The manuscript has been edited to include this additional argument in lines 199- 202.  
+
+<|ref|>text<|/ref|><|det|>[[118, 163, 830, 197]]<|/det|>
+"3. Since each oxide grain has one orientation data point in the SEM-EBSD study, it is critical to comment on the correctness of the indexing."  
+
+<|ref|>text<|/ref|><|det|>[[117, 207, 877, 428]]<|/det|>
+Due to the fine grain size of the oxide, EBSD must be performed using a high- resolution SEM instrument. This retains sufficient brightness in a small beam footprint, which reduces overlapping of diffraction patterns from neighbouring grains and so improves indexing. It is thought that the relatively low indexing rate in the oxide here ( \(\sim 60\%\) ) results from overlapping patterns. It is therefore likely that when a pattern is indexed, it is from a position where the beam coincides towards the centre of an oxide grain, reducing the effect of overlapping patterns from neighbouring grains. We therefore have high confidence in the indexing in these regions, both in terms of orientation accuracy (Average mean angular deviation (MAD) of 1.07 and 1.45 for the monoclinic and tetragonal phases respectively), and in the phase accuracy, as the monoclinic and tetragonal phases have significantly different crystal structures, and so misindexing is unlikely. The accuracy of the indexing using bulk EBSD is reflected by the similarities in observed textures to those observed by non- destructive XRD measurements in the literature \(^{48 - 51}\) . The discussion has been added to the manuscript in lines 147- 157.  
+
+<|ref|>text<|/ref|><|det|>[[118, 439, 864, 472]]<|/det|>
+"4. There is a considerable difference in the orientation measurements between the EBSD and SPED techniques."  
+
+<|ref|>text<|/ref|><|det|>[[117, 483, 875, 809]]<|/det|>
+The difference in the EBSD and SPED monoclinic oxide texture measurements have been carefully considered and the discussion has been expanded in lines 287- 298 and 353- 358. Although there is a difference between the EBSD and SPED monoclinic oxide texture measurements (15° and 5° difference in the peak misorientation angle away from the {10- 6} monoclinic pole and the {11- 2} monoclinic pole in region 1 and region 2, respectively), that difference does not change the proposed mechanisms for monoclinic oxide texture formation. It can be clearly seen in Figs. 4d and 5d that the SPED measured {10- 6} and the {11- 2} pole figures in regions 1 and 2 confirm the main patterns shown in the EBSD measured pole figures in Fig. 3. Moreover, the difference between the maximum in the misorientation angle distributions and the theoretical value is about 13° for both the SPED and the EBSD in region 1, and 4° and 7° for, respectively, SPED and EBSD in region 2. These values are within the expected uncertainties, and further confirm a very good agreement between the two techniques and between the experimental measurements and the theoretical model. The main reasons for the differences between the two techniques include the large difference between the sampling statistics in the two techniques, misalignment between the samples used in each technique, internal misorientations within the metal grain, misindexing of some SPED data and relative surface orientation before/after oxide removal. Additionally, there are differences in the measured grain population between the two techniques with EBSD biased towards larger grains. For these reasons, it is important to combine different techniques when studying such complex oxide systems.  
+
+<|ref|>text<|/ref|><|det|>[[118, 820, 874, 853]]<|/det|>
+"5. Ma et al. APL 106 (2015) 101603 shows non-equilibrium oxidation states, Zr1+, Zr2+, and Zr3+ for suboxide (ZrOx) rather than just ZrO."  
+
+<|ref|>text<|/ref|><|det|>[[118, 864, 870, 897]]<|/det|>
+We thank the reviewer for the insightful reference, which finds that a gradual change in the stoichiometry through the formation of suboxides is energetically more favourable compared to an
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 866, 151]]<|/det|>
+abrupt change from Zr to ZrO2 in the initial stages of the oxidation. This supports the findings in the current manuscript, that the metastable phases provide a more gradual route for the formation of monoclinic ZrO2 and therefore a more protective oxide layer. Discussion and reference have been added in lines 520- 528, 532- 533.  
+
+<|ref|>text<|/ref|><|det|>[[120, 162, 808, 179]]<|/det|>
+"6. Will the initial oxidation product of Zr be amorphous and then form nanocrystal grains?"  
+
+<|ref|>text<|/ref|><|det|>[[118, 189, 875, 291]]<|/det|>
+There have been some limited suggestions in the literature for the presence of amorphous grains in the initial oxidation stages of Zr (e.g. Ploc and Zhou et al.), however, the general opinion is that if they form, they only form in the very initial stage of the oxidation and certainly would be of limited number. Furthermore, the effects of the epitaxial strains that we found are strong enough to support the idea that the amorphous phase might only form in very limited oxide grains and on very short time and length scales in single- phase Zr alloys.  
+
+<|ref|>text<|/ref|><|det|>[[118, 301, 846, 335]]<|/det|>
+Ploc R. A. Transmission electron microscopy of thin (<2000 Å) thermally formed ZrO2 films. J Nucl Mater 1968; 28:48- 60.  
+
+<|ref|>text<|/ref|><|det|>[[118, 345, 875, 380]]<|/det|>
+Zhou BX, Li Q, Yao MY, Liu WQ, Chu YL. Effect of water chemistry and composition on microstructural evolution of oxide on Zr- alloys. J ASTM Int 2009:360.  
+
+<|ref|>text<|/ref|><|det|>[[118, 389, 875, 441]]<|/det|>
+"7. Could the authors calculate the lattice mismatch and interfacial energy for different orientations? In reality, <9% lattice mismatch is needed for obtaining epitaxy. "epitaxial relationship" could be misleading."  
+
+<|ref|>text<|/ref|><|det|>[[118, 451, 879, 605]]<|/det|>
+The theoretical lattice mismatch has been calculated in Table 1, which are based on the Zr- Zr distances in the corresponding lattices. To improve clarity, Table 1 and the relevant discussion in lines 466- 471 have been updated. These measures are only used to qualitatively compare the mismatch in the transformations in the two regions, as in reality there will be atom relaxations, specifically in the normal directions and so the mismatch would be lower, so it would fall within the quoted number of 9%. The calculation of the interfacial energies is beyond the aims of the current manuscript, although it would be an interesting study. Our aim was to show that our crystallographic texture measurements clearly showed the presence of orientation relationships between the metal and oxide phases, and we have defined these by modelling the theoretical orientation relationships.  
+
+<|ref|>text<|/ref|><|det|>[[120, 615, 525, 631]]<|/det|>
+"8. Some typos: line 210: Fig 2C, line 349: 10 mm etc."  
+
+<|ref|>text<|/ref|><|det|>[[179, 643, 417, 658]]<|/det|>
+The typos have been corrected.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 304, 106]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 130, 395, 146]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 168, 461, 184]]<|/det|>
+nice work, publish, reviewers' points dealt with  
+
+<|ref|>text<|/ref|><|det|>[[115, 226, 395, 242]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 265, 867, 301]]<|/det|>
+I have read the revised manuscript and I am satisfied with the changes and how the authors addressed the points raised.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a34463a0497b4d378c75c004251d861299ce28e08f5c4c35853a9d5a3569cb12/images_list.json

@@ -0,0 +1,154 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure R1: Current-voltage curves of \\(200\\mu m\\) MAPbI \\(_3\\) photoconductor in the dark and under X-ray.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure R2: Performance of thick perovskite film devices. a, Current-voltage curves in terms of DCS electrode voltages and drain currents of \\(80\\mu m\\) and \\(200\\mu m\\) MAPbI\\(_3\\) /SnO\\(_2\\) /In\\(_2\\) O\\(_3\\)",
+    "footnote": [],
+    "bbox": [
+      [
+        194,
+        353,
+        785,
+        842
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure R1: a, Sectional view of a single-pixel in the DCS array. The DCS detector is in series with a switching TFT to control the on/off of a single-pixel. b, Sectional view of a single-pixel in the a-Se array.",
+    "footnote": [],
+    "bbox": [
+      [
+        202,
+        180,
+        768,
+        571
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure R2: Current-voltage curves of \\(200\\mu m\\) MAPbI3 photoconductor in the dark and under X-ray.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Figure R3: Performance of thick perovskite film devices. a, Current-voltage curves in terms of DCS electrode voltages and drain currents of \\(80\\mu m\\) and \\(200\\mu m\\) MAPbI3/SnO2/In2O3 TFT in the dark. b, Current-voltage curves in terms of DCS electrode voltages and drain currents of \\(80\\mu m\\) and \\(200\\mu m\\) MAPbI3/SnO2/In2O3 TFT under \\(21.25\\mu Gy_{air}s^{-1}\\) . c, Pulse-train response of \\(200\\mu m\\) MAPbI3/SnO2/In2O3 TFT device under \\(21.25\\mu Gy_{air}s^{-1}\\) . The DCS electrode is biased with CV. d, Pulse-train response of \\(200\\mu m\\) MAPbI3/SnO2/In2O3 TFT device under \\(21.25\\mu Gy_{air}s^{-1}\\) . The DCS electrode is disabled",
+    "footnote": [],
+    "bbox": [
+      [
+        192,
+        102,
+        784,
+        592
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Figure R1: Current-voltage curves in terms of DCS electrode voltages and drain currents of devices with \\(1\\mu m MAPbBr_3\\) , MAPbI3 and \\(\\mathrm{FA}_{0.92}\\mathrm{Cs}_{0.04}\\mathrm{MA}_{0.04}\\mathrm{PbI}_3\\) on \\(\\mathrm{In}_2\\mathrm{O}_3\\) and IGZO conduction channel in the dark and under X-ray. a, Current-voltage curves of \\(1\\mu m MAPbBr_3\\) , MAPbI3 and \\(\\mathrm{FA}_{0.92}\\mathrm{Cs}_{0.04}\\mathrm{MA}_{0.04}\\mathrm{PbI}_3\\) on \\(\\mathrm{C}_{60} / \\mathrm{In}_2\\mathrm{O}_3\\) TFT in the dark. b, Current-voltage curves of \\(1\\mu m\\) MAPbBr3, MAPbI3 and \\(\\mathrm{FA}_{0.92}\\mathrm{Cs}_{0.04}\\mathrm{MA}_{0.04}\\mathrm{PbI}_3\\) on \\(\\mathrm{C}_{60} / \\mathrm{IGZO}\\) TFT in the dark. c, Current-voltage curves of \\(1\\mu m\\) MAPbBr3, MAPbI3 and \\(\\mathrm{FA}_{0.92}\\mathrm{Cs}_{0.04}\\mathrm{MA}_{0.04}\\mathrm{PbI}_3\\) on \\(\\mathrm{C}_{60} / \\mathrm{In}_2\\mathrm{O}_3\\) TFT under 21.25 \\(\\mu \\mathrm{Gy}_{\\mathrm{air}}\\mathrm{s}^{-1}\\) . d, Current-voltage curves of \\(1\\mu m\\) MAPbBr3, MAPbI3 and \\(\\mathrm{FA}_{0.92}\\mathrm{Cs}_{0.04}\\mathrm{MA}_{0.04}\\mathrm{PbI}_3\\) on \\(\\mathrm{C}_{60} / \\mathrm{IGZO}\\) TFT under 21.25 \\(\\mu \\mathrm{Gy}_{\\mathrm{air}}\\mathrm{s}^{-1}\\) .",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Figure R2: Pulse-train response of different perovskite and conduction channel materials based devices. The DCS electrode are all biased with CV. a, Pulse-train response of MAPbBr3/C60/In2O3 TFT under 21.25 μGyair s-1. b, Pulse-train response of MAPbI3/C60/In2O3 TFT under 21.25 μGyair s-1. c, Pulse-train response of MAPbBr3/C60/IGZO TFT under 21.25 μGyair s-1. d, Pulse-train response of MAPbI3/C60/IGZO TFT under 21.25 μGyair s-1, e, Pulse-train response of FA0.92Cs0.04MA0.04PbI3/C60/IGZO TFT under 21.25 μGyair s-1.",
+    "footnote": [],
+    "bbox": [
+      [
+        191,
+        92,
+        800,
+        465
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Figure R3: a, Transfer curve of \\(\\mathsf{In}_2\\mathsf{O}_3\\) transistor with Ni/Au and b, Al source/drain electrodes.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Figure R4: a, Current-voltage curves of \\(1\\mu m\\) MAPbBr3, MAPbI3 and \\(\\mathsf{FA}_{0.92}\\mathsf{Cs}_{0.04}\\mathsf{MA}_{0.04}\\mathsf{PbI}_3\\) on",
+    "footnote": [],
+    "bbox": [
+      [
+        195,
+        700,
+        770,
+        867
+      ]
+    ],
+    "page_idx": 14
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Figure R1. TRPL of \\(\\mathrm{In}_{2}\\mathrm{O}_{3}\\) film. Its average carrier lifetime is calculated as 2.041 ms.",
+    "footnote": [],
+    "bbox": [
+      [
+        315,
+        536,
+        651,
+        742
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Figure R2. Laser beam pulse response of the device. The rise and fall time are 23 ms and 31 ms, respectively.",
+    "footnote": [],
+    "bbox": [
+      [
+        328,
+        377,
+        647,
+        593
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Figure R3. Sensitivity of devices with different thicknesses.",
+    "footnote": [],
+    "bbox": [
+      [
+        333,
+        181,
+        636,
+        375
+      ]
+    ],
+    "page_idx": 18
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a34463a0497b4d378c75c004251d861299ce28e08f5c4c35853a9d5a3569cb12/supplementary_0_Peer Review File__a34463a0497b4d378c75c004251d861299ce28e08f5c4c35853a9d5a3569cb12.mmd

@@ -0,0 +1,518 @@
+
+# natureportfolio  
+
+Peer Review File  
+
+Realizing Nearly- Zero Dark Current and Ultrahigh Signal- to- Noise Ratio Perovskite X- ray Detector and Image Array by Dark- Current- Shunting Strategy  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS</B>  
+
+Reviewer #1 (Remarks to the Author):  
+
+Peng Jin, Yingjie Tang et al. report about a dark- current shunting method for pixelated X- ray imagers based on Metal halide perovskites as absorbing layer.  
+
+The manuscript is well prepared and of interest for researchers working in the specific field of X- Ray detection.  
+
+The title is appropriate and reflects in a good way the work described in the article. Also, the abstract reflects the content of the article.  
+
+The introduction clearly describes the state of the art and application fields of the presented topics and problem being investigated. The reduction of the dark current is indeed one of the major challenges for perovskite- based X- Ray detectors.  
+
+Figures are accurately placed and easy to understand and the references are placed accurately.  
+
+Results are novel and quite remarkable, however the perovskite layer thickness of \(2\mu m\) precludes its use in medical X- ray imagers. The limited X- ray absorption (few percent) for \(2\mu m\) thick perovskites is too low to be used in industrial applications. This makes the impact of the manuscript rather limited. For the pixel architecture described in this work, and its working principle, it is not obvious to have the same performance of dark current suppression by retaining at the same time high photocurrent collection, for perovskite thicknesses of few hundreds of \(\mu m\) . If the authors can demonstrate that a similar behavior is obtainable for perovskite thicknesses of \(100\mu m\) and above, then I would recommend the publication in this journal otherwise a journal with lower impact factor should be better suited.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The manuscript investigates a new device structure strategy for MHP x- ray detectors at the purpose to decrease the dark current in metal halide perovskite x- ray detectors. The strategy is based on the use of a planar geometry for the detection of photoelectrons induced by the x- ray irradiation and by a third electrode that works as a blanking unit for the noise. The manuscript is generally well written and the author show some nice results in their configuration. I have few comments that the authors should try to insert in their discussion to allow the reader to better understand the potential of their work.  
+
+1- The device structure that they propose is substantially more complex than the structure used for other direct conversion devices such the one based on a- Se. They should comment on the consequence of this more complicated structure for the read- out electronics. 
+2- The energy of the used x- rays should be stated in the experimental section. 
+3- In the experimental resukts the active layer is 2micrometers, however, the thickness ideally should be higher. The authors should comment on the effect of the active layer thickness on their device performance and voltages that would need to be apply. 
+4- The authors mentioned that in their opinion this is not a transistor structure, but as they have a
+
+<--- Page Split --->
+
+substantial PMMA layer this is certainly working as capacitor. I think the authors should better explain their point by comparing with phototransistors structures eventually reported in literature if there are any made with MHP.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors report a perovskite based x- ray detector with a new device structure using dark shunting electrodes to suppress dark current and obtained a high SNR. The dark current values obtained are probably the lowest reported so far in literature. Another significant result is that the dark current is stable which is essential for practical devices. They also demonstrate x- ray imaging with detector array.  
+
+Although the results obtained are impressive, the working mechanism or the device physics part was not clear to me. For example, it was not clear why the "dark electrons" flow from the source through ETL and perovskite to the dark shunting electrode but the photogenerated electrons flow in the opposite direction. The authors mention that the photogenerated electrons in the perovskite drift at the ETL interface, so wouldn't there be a barrier for dark electrons flowing in the opposite direction? Perhaps an energy band diagram can be shown?  
+
+Also, I was not convinced if this device structure is generally applicable. Would it still work for another perovskite composition and another conducting channel?  
+
+Figures could be improved. Some figures have a lot of text in small font.  
+
+## Other comments  
+
+Pg 3 "...the drain only receives X- ray- generated electrons but rejects dark electrons..." This sentence seems misleading. Could the authors clarify how the drain rejects dark electrons?  
+
+Pg 3 Fig 1. The photo- induced electrons and holes are enclosed in a dashed oval. Could the authors explain what it means?  
+
+Pg 5 "...those photo electrons are drifted when interfaced with the electron transport layer (ETL) and sensitizes the lateral conduction channel" It is not clear what the authors mean by sensitizes the lateral channel.  
+
+Perhaps the authors could use an energy band diagram in SI?  
+
+Also I would expect the effective field at the interface to be not strong and solely relying on the ETL interface to extract electrons seems like an inefficient strategy.  
+
+Pg 6 "The primary functional layers of this device, perovskite and In2O3" Could the authors explain what
+
+<--- Page Split --->
+
+kind of material In2O3 is why they chose it? This is also the first time they mention In2O3. Perhaps they could mention somewhere in the beginning that they used In2O3 as the conductive channel because I was under the impression that it was all perovskite.  
+
+Pg 8. Fig 3e. Could the authors comment on why the sensitivity for low dose is higher than for high dose?  
+
+Pg. 10 "The detection limit of the DCS detector (SNR=3) is as low as 7.84 nGyair s- 1" Could the authors show a pulse for 7.84 nGyair s- 1?  
+
+Pg 12. "of dark current drafting" Typo  
+
+Pg. 14 Fig 5d Could the authors comment on the mobility values? 1.55 cm2/V/s seems low.  
+
+Could the authors mention what the variation in the critical voltage for different devices were?
+
+<--- Page Split --->
+
+Firstly, we very much appreciated the constructive comments and valuable suggestions from reviewers 1, 2 and 3. Now all the questions have been thoroughly considered and carefully answered. And hopefully, our responses can release your concerns.  
+
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*  
+
+The author's answers to Reviewer #1's comment:  
+
+Dear reviewer:  
+
+Thank you very much for your supportive comments and constructive suggestions. We have carefully read your reviews and noticed your concerns. Your question was answered as below. The manuscript has been revised with fully consideration of your comments.  
+
+Reviewer's comments: Peng Jin, Yingjie Tang et al. report about a dark- current shunting method for pixelated X- ray imagers based on Metal halide perovskites as absorbing layer. The manuscript is well prepared and of interest for researchers working in the specific field of X- Ray detection.  
+
+The title is appropriate and reflects in a good way the work described in the article. Also, the abstract reflects the content of the article.  
+
+The introduction clearly describes the state of the art and application fields of the presented topics and problem being investigated. The reduction of the dark current is indeed one of the major challenges for perovskite- based X- Ray detectors.  
+
+Figures are accurately placed and easy to understand and the references are placed accurately.  
+
+A0: Thanks so much for your comments that you feel this work is novel and important.  
+
+Q1: Results are novel and quite remarkable, however the perovskite layer thickness of \(2\mu \mathrm{m}\) precludes its use in medical X- ray imagers. The limited X- ray absorption (few percent) for 2 \(\mu \mathrm{m}\) thick perovskites is too low to be used in industrial applications. This makes the impact of the manuscript rather limited. For the pixel architecture described in this work, and its working principle, it is not obvious to have the same performance of dark current suppression by retaining at the same time high photocurrent collection, for perovskite thicknesses of few hundreds of \(\mu \mathrm{m}\) . If the authors can demonstrate that a similar behavior is obtainable for perovskite thicknesses of \(100\mu \mathrm{m}\) and above, then I would recommend the publication in this
+
+<--- Page Split --->
+
+journal otherwise a journal with lower impact factor should be better suited.  
+
+A1: We appreciate it very much that your feel those results are novel and remarkable. We certainly understand your concerns. The main purpose of our work is to demonstrate a common device strategy method to suppress dark current and obtain a high signal- to- noise ratio, there is certainly quite a lot of rooms to further improve the device performance by material engineering, and we are working on it certainly. Actually in my points of view, depositing high- quality thick film perovskite is still quite challenging, since there is a solubility limit for the perovskite precursor solution.  
+
+Our device strategy is very universal and can be applied in various material systems with different thicknesses, and compositions, we have demonstrated this in our revised manuscript (page 12, Figure S8). Thick X- ray sensitive materials can obtain stronger absorption of incident X- ray. But the synthesis of high- quality thick perovskite films is still a big challenge around the world. We have used the nearly saturated perovskite precursor solution in the first version of the manuscript, the 2um perovskite film is the thickest we can fabricate by the typical precursor- based deposition. In this round, we referred to some previous works on thick perovskite films and added experiments to verify our DCS method in thick films, which was fabricated by the polymer binding method. [Adv. Eng. Mater., 18: 1189- 1199] [IEEE Sensors Journal, vol. 7, no. 6, pp. 925- 930, June 2007] [Adv. Funct. Mater. 2022, 32, 2110729] The 200 μm and 80 μm MAPbI₃/SnO₂/In₂O₃ TFT devices are shown below. The devices are fabricated by blade- coating the MAPbI₃ and PMMA binder (dissolved in PMMA/Toluene solvent, MAPbI₃:PMMA = 1:2) on SnO₂/In₂O₃ TFT. Because the Toluene solvent will largely dissolve the C₆₀ layer, we replace it with SnO₂, which is also an electron transport layer (ETL) for perovskite.
+
+<--- Page Split --->
+![](images/Figure_unknown_1.jpg)
+  
+
+The perovskite film consists of MAPbI \(_3\) and PMMA polymer binder, PMMA is an insulator, thus, the conductivity and sensitivity of such devices are relatively low. We measured the I- V curves of the \(200\mu m\) MAPbI \(_3\) and PMMA polymer binder film- based photoconductor (Au/MAPbI3 and PMMA polymer mixture film/Au) in the dark and under X- ray, both dark current and photocurrent are very low. At the voltage of 0.5V, the photocurrent is 25 pA (Figure R1).  
+
+![](images/Figure_unknown_2.jpg)
+
+<center>Figure R1: Current-voltage curves of \(200\mu m\) MAPbI \(_3\) photoconductor in the dark and under X-ray. </center>  
+
+We measured the current- voltage curves in terms of DCS electrode voltages and drain
+
+<--- Page Split --->
+
+currents of these thick devices in the dark and under X- ray. For all the \(80\mu m\) and \(200\mu m\) devices we fabricated, all of the devices' dark currents can be suppressed to 0 A. We found that the Critical Voltage increased with the rise of the thickness of X- ray sensitive materials (Figure R2a below). The dark current of \(200\mu m\) device can also be suppressed to nearly zero with DCS method, the dark current baseline is stable and signal- to- noise ratio can also be very high (Figure R2c,d below). We have added those new results in our revised manuscript (page 12).  
+
+Overall, we believe our device strategy proposed in this study is very universal and can be applied in various material systems with different thicknesses.  
+
+![](images/Figure_unknown_3.jpg)
+
+<center>Figure R2: Performance of thick perovskite film devices. a, Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80\mu m\) and \(200\mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) </center>
+
+<--- Page Split --->
+
+TFT in the dark. b, Current- voltage curves in terms of DCS electrode voltages and drain currents of \(80 \mu m\) and \(200 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT under 21.25 \(\mu\) Gy\(_air\) s\(^{- 1}\) . c, Pulse- train response of \(200 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT device under 21.25 \(\mu\) Gy\(_air\) s\(^{- 1}\) . The DCS electrode is biased with CV. d, Pulse- train response of \(200 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT device under 21.25 \(\mu\) Gy\(_air\) s\(^{- 1}\) . The DCS electrode is disabled  
+
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\* \*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*
+
+The author's answers to Reviewer #2's comment:  
+
+Dear reviewer:  
+
+We really appreciate your reviews and valuable pieces of advice for our paper. We have carefully read your comments and noticed your concerns. All of your questions were answered below.  
+
+The manuscript has been revised with full consideration of your concerns and suggestions.  
+
+Reviewer's comments: The manuscript investigates a new device structure strategy for MHP x- ray detectors at the purpose to decrease the dark current in metal halide perovskite x- ray detectors. The strategy is based on the use of a planar geometry for the detection of photoelectrons induced by the x- ray irradiation and by a third electrode that works as a blanking unit for the noise. The manuscript is generally well written and the author show some nice results in their configuration. I have few comments that the authors should try to insert in their discussion to allow the reader to better understand the potential of their work.  
+
+Q1: The device structure that they propose is substantially more complex than the structure used for other direct conversion devices such the one based on a- Se. They should comment on the consequence of this more complicated structure for the read- out electronics.  
+
+A1: Thank you for your careful review and recognition. And thanks for this great question! We fully understand your concerns and we'd like to illustrate this more clearly in the revised manuscript, and we have added related comments in our revised manuscript (page 16). Our device structure is shown in Supplementary Information Figure S13, and is shown below
+
+<--- Page Split --->
+
+(Figure R1a). It is true that the single photoconductive detector is very simple, but when they are integrated with the back panel, the TFT device is also needed. The general structure of the pixelized a- Se detection unit is shown in Figure R1b below [Proc. SPIE 5368, Medical Imaging 2004: Physics of Medical Imaging, (6 May 2004)]. As you can see, the traditional a- Se array is fabricated by coupling the a- Se material in the surface of back panel, which is prepared ahead typically with a- Si, and then evaporating the top common contact electrode. In our structure, the \(\mathsf{In}_2\mathsf{O}_3\) TFT array- based back panel is also prepared ahead, the latter process is actually quite similar. Given that we have the In2O3 TFT back panel in hand, what we do next is to deposit the X- ray sensitive material (perovskite) onto the \(\mathsf{In}_2\mathsf{O}_3\) TFT array back panel and then evaporate the top common DCS electrode. The difference in the fabrication process is mainly the manufacture of the back panel. If the \(\mathsf{In}_2\mathsf{O}_3\) (or IGZO) TFT array is fully developed and well- manufactured by the company, there won't be too much difficulty in making our DCS method- based device and arrays. As we can know the IGZO- based backplane is starting to take over the a- Si TFT. As for the signal readout, the a- Se detector directly readout the charge (current) signal from Drain electrode of the switching MOSFET (as shown below, Figure R1b), our device also readout the signal from Drain electrode of the switching TFT (Figure R1a), the analog data processing units including the preamplifier, Analog to Digital (AD) conversion can be shared with the conventional a- Si sensors, actually, we are using the same ROIC chips purchased from ADI in our study for the X- ray image demonstration.  
+
+Another difference might be related to the pixel size, with the industrial facilities, the typical pixel size of a photoconductive a- Se detector is \(\sim 100 \mu \mathrm{m}\) , in this study, we have demonstrated the X- ray imager with a pixel size of \(\sim 300 \mu \mathrm{m}\) , with our in- house lab facilities, compare with the commercial X- ray detector's pixel size (100 \(\mu \mathrm{m}\) ), we think it is not very difficult for companies to use our DCS method to produce 100 \(\mu \mathrm{m}\) pixel based X- ray detector array in the future.  
+
+One particular good thing of our device is that, for the typical X- ray detector array, we usually have to purposely design a big storage capacitance ( \(C_{\mathrm{st}}\) ) in each pixel, which is parallel connected with the built- in capacitance (naturally formed by the parallel electrodes), to beat the dark current. Otherwise, the capacitance is too small that they can be easily
+
+<--- Page Split --->
+
+filled up by the dark current. In our DCS detector, the dark current can be suppressed to extremely low, thus, in principle, we don't need to put an extra parallel storage capacitance in each pixel, which would simplify the pixel circuit design.  
+
+![](images/Figure_unknown_4.jpg)
+
+<center>Figure R1: a, Sectional view of a single-pixel in the DCS array. The DCS detector is in series with a switching TFT to control the on/off of a single-pixel. b, Sectional view of a single-pixel in the a-Se array. </center>  
+
+Q2: The energy of the used x- rays should be stated in the experimental section.  
+
+A2: Thank you for your advice, we have added the energy of the used x- rays (tube voltage: 40 keV and 50 keV) in our experimental section. (page 23) The used output spectrum at different voltages is shown below, the figure was added in the SI. (Figure S17)
+
+<--- Page Split --->
+![](images/Figure_unknown_5.jpg)
+  
+
+Q3: In the experimental results the active layer is 2micrometers, however, the thickness ideally should be higher. The authors should comment on the effect of the active layer thickness on their device performance and voltages that would need to be apply.  
+
+A3: We appreciate it very much for your advice. We certainly understand your concerns. The main purpose of our work is to demonstrate a common device strategy method to suppress dark current and obtain a high signal- to- noise ratio, there is certainly quite a lot of rooms to further improve the device performance by material engineering, and we are working on it certainly. Actually in my points of view, depositing high- quality thick film perovskite is still quite challenging, since there is a solubility limit for the perovskite precursor solution.  
+
+Our device strategy is very universal and can be applied in various material systems with different thicknesses, and compositions, we have demonstrated this in our revised manuscript (page 12, Figure S8). Thick X- ray sensitive materials can obtain stronger absorption of incident X- ray. But the synthesis of high- quality thick perovskite films is still a big challenge around the world. We have used the nearly saturated perovskite precursor solution in the first version of the manuscript, the \(2\mu m\) perovskite film is the thickest we can fabricate by the typical precursor- based deposition. In this round, we referred to some previous works on thick perovskite films and added experiments to verify our DCS method in thick films, which was fabricated by the polymer binding method. [Adv. Eng. Mater., 18: 1189- 1199] [IEEE Sensors Journal, vol. 7, no. 6, pp. 925- 930, June 2007] [Adv. Funct. Mater.
+
+<--- Page Split --->
+
+2022, 32, 2110729] The \(200 \mu m\) and \(80 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT devices are shown below. The devices are fabricated by blade-coating the MAPbI\(_3\) and PMMA binder (dissolved in PMMA/Toluene solvent, MAPbI\(_3\) :PMMA = 1:2) on SnO\(_2\) /In\(_2\) O\(_3\) TFT. Because the Toluene solvent will largely dissolve the C\(_{60}\) layer, we replace it with SnO\(_2\), which is also an electron transport layer (ETL) for perovskite.  
+
+![](images/Figure_unknown_6.jpg)
+  
+
+The perovskite film consists of MAPbI\(_3\) and PMMA polymer binder, PMMA is an insulator, thus, the conductivity and sensitivity of such devices are relatively low. We measured the I- V curves of the \(200 \mu m\) MAPbI\(_3\) and PMMA polymer binder film- based photoconductor (Au/MAPbI3 and PMMA polymer mixture film/Au) in the dark and under X- ray, both dark current and photocurrent are very low. At the voltage of 0.5V, the photocurrent is 25 pA (Figure R2).
+
+<--- Page Split --->
+![](images/Figure_unknown_7.jpg)
+
+<center>Figure R2: Current-voltage curves of \(200\mu m\) MAPbI3 photoconductor in the dark and under X-ray. </center>  
+
+We measured the current- voltage curves in terms of DCS electrode voltages and drain currents of these thick devices in the dark and under X- ray. For all the \(80\mu m\) and \(200\mu m\) devices we fabricated, all of the devices' dark currents can be suppressed to 0 A. We found that the Critical Voltage increased with the rise of the thickness of X- ray sensitive materials (Figure R3a below). The dark current of \(200\mu m\) device can also be suppressed to nearly zero with DCS method, the dark current baseline is stable and signal- to- noise ratio can also be very high (Figure R3c,d below). We have added those new results in our revised manuscript (page 12).  
+
+Overall, we believe our device strategy proposed in this study is very universal and can be applied in various material systems with different thicknesses.
+
+<--- Page Split --->
+![](images/Figure_unknown_8.jpg)
+
+<center>Figure R3: Performance of thick perovskite film devices. a, Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80\mu m\) and \(200\mu m\) MAPbI3/SnO2/In2O3 TFT in the dark. b, Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80\mu m\) and \(200\mu m\) MAPbI3/SnO2/In2O3 TFT under \(21.25\mu Gy_{air}s^{-1}\) . c, Pulse-train response of \(200\mu m\) MAPbI3/SnO2/In2O3 TFT device under \(21.25\mu Gy_{air}s^{-1}\) . The DCS electrode is biased with CV. d, Pulse-train response of \(200\mu m\) MAPbI3/SnO2/In2O3 TFT device under \(21.25\mu Gy_{air}s^{-1}\) . The DCS electrode is disabled </center>  
+
+Q4: The authors mentioned that in their opinion this is not a transistor structure, but as they have a substantial PMMA layer this is certainly working as capacitor. I think the authors
+
+<--- Page Split --->
+
+should better explain their point by comparing with phototransistors structures eventually reported in literature if there are any made with MHP.  
+
+A4: Thanks for this wonderful question, we totally understand your concerns, and we have added more discussions in our revised manuscript (page 9). We'd like to prompt that in our device, even without the PMMA layer, a similar device behavior is observed, and there is the dark- current- shunting effect as well (Supplementary Information Figure S6). We believe the very thin PMMA layer only acts as a protection layer and we don't find a capacitor effect in the photocurrent and dark current. Some of the devices may use the thick PMMA layer as a dielectric layer, but their PMMA layer is usually very thick. Our PMMA layer is only \(\sim 50 \text{nm}\) , it can hardly act as a dielectric layer. Besides, comparing the MHP transistor's transfer curves [Nat Electron 5, 78- 83 (2022)] to our device (Supplementary Information Figure S5), there is no on/off state in our device when adjusting the DCS electrode's voltage.  
+
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*  
+
+The author's answers to Reviewer #3's comment:  
+
+Dear reviewer:  
+
+Thank you very much for your reviews and valuable suggestions for our paper. These suggestions really help us to refine our manuscript and make it more understandable for readers. We have carefully read your comments and noticed your concerns. All of your questions were answered below. And the manuscript has been revised with the entire consideration of your suggestions and concerns.  
+
+Reviewer's comments: The authors report a perovskite based x- ray detector with a new device structure using dark shunting electrodes to suppress dark current and obtained a high SNR. The dark current values obtained are probably the lowest reported so far in literature. Another significant result is that the dark current is stable which is essential for practical devices. They also demonstrate x- ray imaging with detector array.  
+
+A0: Thanks so much for your comments that you recognize the work's result about the low and stable dark current as well as the imaging with detector array.
+
+<--- Page Split --->
+
+Q1: Although the results obtained are impressive, the working mechanism or the device physics part was not clear to me. For example, it was not clear why the "dark electrons" flow from the source through ETL and perovskite to the dark shunting electrode but the photogenerated electrons flow in the opposite direction. The authors mention that the photogenerated electrons in the perovskite drift at the ETL interface, so wouldn't there be a barrier for dark electrons flowing in the opposite direction? Perhaps an energy band diagram can be shown?  
+
+A1: Thank you very much for your positive comments and very careful review. We understand your concerns and revised the working mechanism (Fig.1 in the manuscript). Your comments really promote us to think deeper and in more detail about the device's working principles, we have added a band diagram below, and put more discussion on the working mechanism, the energy levels come from our previous work measured with UPS [Small Methods 2022, 6, 2200500].  
+
+Firstly, with our device design, the X- ray photocurrent is actually decreased under DCS mode, compared with the control device without DCS voltages. But the actual conduction channel is the high- mobility \(\mathsf{In}_{2}\mathsf{O}_{3}\) , not the perovskite, and there is a photoconductive gain of the \(\mathsf{In}_{2}\mathsf{O}_{3}\) channel, therefore the decrease of X- ray photocurrent is not much, and the overall SNR is enhanced with several orders of magnitudes. This sort of gain is observed in many hybrid device structure. For the typical two- terminal photoconductor or photoresistor detector made of perovskite, the conduction channel is always perovskite and is usually without gain (unless there is trapped- induced photoconductive gain), we believe the photocurrent will be cut more if we put a similar DCS electrode.  
+
+Let us get back to our device structure, under the dark conditions, most of the dark electrons can be collected by the DCS electrode (positive), or at least they will no longer be received by the drain, which gives the zero- dark current. (In the manuscript page 4, Figure 1b; Page 10, Figure 3b; Page 14, Figure 4a). We believe this observation is solid.  
+
+We admit that this DCS bias could also affect the X- ray photocurrent since we indeed find a reduced photocurrent. (In the manuscript Page 14, Figure 4a) In order to better illustrate the detailed working mechanism under X- ray, we divided the device in two regions
+
+<--- Page Split --->
+
+separated by the red dash line. Those two regions have different external electrical fields, the source is grounded and the DCS and drain are both positive, (in our cases they are 0.56 and 0.5 respectively), the left region has much larger external electric field than the right region.  
+
+![](images/Figure_unknown_9.jpg)
+  
+
+On the left of the red dash line: As shown in Figure b, between the Source electrode and DCS electrode, there is an externally applied voltage when the device is working, this external voltage will generate an electric field from DCS electrode to Source. The photogenerated electrons in this region move in the same direction with shunted dark electrons. Thus, the photocurrent decreases when the device is working, as you can see in Fig. 3e in the manuscript. But with the great photoconductive gain effect of the \(\mathrm{In}_2\mathrm{O}_3\) conduction channel (high mobility), the sensitivity can still reach a high value.  
+
+![](images/Figure_unknown_10.jpg)
+  
+
+On the right of the red dash line: The external electric field is much weaker in this region
+
+<--- Page Split --->
+
+(the DCS and Drains are both positively biased, and in our cases they are 0.56 and 0.5 respectively), and the built- in electric field is more favorable for pushing electron downward to the conduction channel. (Figure b). But with the built- in electric field between perovskite and \(C_{60}\) due to their energy band type- II alignment, you can see it from figure b, it will be more easier for the photogenerated electrons to be drifted from the perovskite towards the \(\mathrm{In}_2\mathrm{O}_3\) . Those X- ray photo- charges mainly contribute to the observed drain photo- current with the help of the fast conduction channel.  
+
+Q2: Also, I was not convinced if this device structure is generally applicable. Would it still work for another perovskite composition and another conducting channel?  
+
+A2: We totally understand your concerns and fabricated the devices with different perovskite materials and other metal oxide conduction channels to demonstrate the generality of DCS method. The devices are shown below, we used MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{In}_2\mathrm{O}_3\) and IGZO conduction channel.  
+
+![](images/Figure_unknown_11.jpg)
+  
+
+We measured the current- voltage curves in terms of DCS electrode voltages and drain currents of these devices in the dark and under X- ray (Figure R1). It can be seen that the Critical Voltage (CV) of these devices almost the same (0.35 – 0.4 V). All of the devices' dark current can be suppressed to 0 A by applying the DCS method.
+
+<--- Page Split --->
+![PLACEHOLDER_19_0]
+
+<center>Figure R1: Current-voltage curves in terms of DCS electrode voltages and drain currents of devices with \(1\mu m MAPbBr_3\) , MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{In}_2\mathrm{O}_3\) and IGZO conduction channel in the dark and under X-ray. a, Current-voltage curves of \(1\mu m MAPbBr_3\) , MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{In}_2\mathrm{O}_3\) TFT in the dark. b, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{IGZO}\) TFT in the dark. c, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{In}_2\mathrm{O}_3\) TFT under 21.25 \(\mu \mathrm{Gy}_{\mathrm{air}}\mathrm{s}^{-1}\) . d, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{IGZO}\) TFT under 21.25 \(\mu \mathrm{Gy}_{\mathrm{air}}\mathrm{s}^{-1}\) . </center>  
+
+We measured these devices' pulse train response when the DCS electrode is biased with CV (Figure R2). All of the devices demonstrate stable current baselines and high signal- to- noise ratio (pulse- train response of \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3 / \mathrm{C}_{60} / \mathrm{In}_2\mathrm{O}_3\) have already demonstrated in the manuscript). It can be seen that the DCS method is very general and can be applied in various devices with different material systems. We have added this comment in our revised manuscript (page 12).
+
+<--- Page Split --->
+![PLACEHOLDER_20_0]
+
+<center>Figure R2: Pulse-train response of different perovskite and conduction channel materials based devices. The DCS electrode are all biased with CV. a, Pulse-train response of MAPbBr3/C60/In2O3 TFT under 21.25 μGyair s-1. b, Pulse-train response of MAPbI3/C60/In2O3 TFT under 21.25 μGyair s-1. c, Pulse-train response of MAPbBr3/C60/IGZO TFT under 21.25 μGyair s-1. d, Pulse-train response of MAPbI3/C60/IGZO TFT under 21.25 μGyair s-1, e, Pulse-train response of FA0.92Cs0.04MA0.04PbI3/C60/IGZO TFT under 21.25 μGyair s-1. </center>  
+
+Instead of using the DCS method to solution- processed devices, we also tried to apply it in sputtered IGZO- based devices. We are not the experts on sputtering IGZO, but we did try to do it here with our own sputtering, honestly, the TFT device is not much better than our solution- processed devices. As shown below, the dark current of the MAPbBr3/SnO2/IGZO (Sputtered) TFT can also be suppressed to 0 A with the DCS method.
+
+<--- Page Split --->
+![PLACEHOLDER_21_0]
+  
+
+Q3: Figures could be improved. Some figures have a lot of text in small font.  
+
+A3: Thank you very much for your advice. The small font has been improved in the figures.  
+
+Other comments  
+
+Q4: Pg 3 "...the drain only receives X- ray- generated electrons but rejects dark electrons..." This sentence seems misleading. Could the authors clarify how the drain rejects dark electrons?  
+
+A4: Thank you very much for your careful review and noting it. This is a wrong statement and we have corrected the "rejects" to "will not receive" in the manuscript. "Not receive" means that there will no dark electrons reach the drain electrode and be collected.  
+
+Q5: Pg 3 Fig 1. The photo- induced electrons and holes are enclosed in a dashed oval. Could the authors explain what it means?  
+
+A5: We totally understand your concerns. The photo- induced electrons and holes are enclosed in a dashed oval means an electron- hole pair generated together under the X- ray illumination, and the electrons and holes in a dashed oval are used as examples to illustrate the motion of whole electrons in the material.  
+
+Q6: Pg 5 "...those photo electrons are drifted when interfaced with the electron transport
+
+<--- Page Split --->
+
+layer (ETL) and sensitizes the lateral conduction channel" It is not clear what the authors mean by sensitizes the lateral channel.  
+
+Perhaps the authors could use an energy band diagram in SI?  
+
+A6: We totally understand your concerns and we'd like to make them clear in our revised manuscript (page 6). Sensitization is the process of generating charge carriers in the semiconductors who cannot absorb X- ray and generate hole- electron pairs themselves (Such as \(\mathrm{In}_2\mathrm{O}_3\) ). These semiconductors may not able to generate charges under X- ray but have superior mobility to transport and recirculate charges and greatly amplify the photocurrent. Thus, we use a sensitizer (Perovskite) to help generate charges under the light (X- ray) and extract them to these semiconductors \(\mathrm{(In_2O_3)}\) . The photosensitive material generates an abundance of charges, these charges then being captured by a transport layer with superior mobility (we call it conduction channel). Then, a much higher photoconductive gain can be obtained. This method allows us to use separate layers for charge photogeneration and transport to enhance photoconductive gain. Many researchers have used such sensitizing method to obtain great high photosensitivity of photodetectors. [ACS Appl. Mater. Interfaces 2019, 11, 36880- 36885] [Adv. Mater. 2015, 27, 6885- 6891] [Adv. Mater. 2021, 33, 2101717]. In addition, we also added a band diagram in SI.  
+
+![PLACEHOLDER_22_0]
+  
+
+Q7: Also I would expect the effective field at the interface to be not strong and solely relying on the ETL interface to extract electrons seems like an inefficient strategy.  
+
+A7: Indeed, we understand your concerns. In this kind of lateral device structure, the
+
+<--- Page Split --->
+
+electrons extraction seems inefficient, but the core advantage of such hybrid lateral structure is the conduction channel has a great photoconductive gain, the captured carriers by conduction channel can transport rapidly in the conduction channel and reinject and recirculate swift between metal contacts. The amount of charges passing through the cross section of a conductor per unit of time is much greater. Therefore, these carriers in the conduction channel can offer much stronger photocurrent than in the X- ray- sensitive materials. This gain effect can greatly amplify the photocurrent signal with superior mobility of the conduction channel. Even the electrons extraction is inefficient, the conduction channel can amplify the photocurrent signal thousands or ten thousands of times with the photocurrent gain effect [Adv. Funct. Mater. 2020, 30, 1903907]. Thus, the finally obtained photocurrent can be much higher than the device only with photosensitive materials. What's more, the addition of ETL like \(\mathsf{C}_{60}\) can be an effective way to help extract electrons from perovskite, and this has been demonstrated in other research [ACS Appl. Mater. Interfaces 2018, 10, 50, 44144- 44151]. Our finally obtained photocurrent sensitivity still reaches high, as you can see that our detector's sensitivity can obtain \(2 \times 10^{4} \mu \text{C Gy}_{\text{air}}^{- 1} \text{cm}^{- 2}\) . Only when the DCS electrode is working does the sensitivity decrease a little.  
+
+Q8: Pg 6 "The primary functional layers of this device, perovskite and In2O3" Could the authors explain what kind of material In2O3 is why they chose it? This is also the first time they mention In2O3. Perhaps they could mention somewhere in the beginning that they used In2O3 as the conductive channel because I was under the impression that it was all perovskite.  
+
+A8: We appreciate your comments and we have added the illustration in our revised manuscript (page 7). \(\mathsf{In}_2\mathsf{O}_3\) is a kind of n- type semiconductor material which has been widely used as the channel material for thin- film transistors [Appl. Phys. Rev. 3, 021303 (2016)]. And it merits of low- temperature solution- processability. In this work, we used \(\mathsf{In}_2\mathsf{O}_3\) and perovskite as transport layer and sensitizer layer, respectively, to construct the hybrid photodetector. Owing to the superior mobility of \(\mathsf{In}_2\mathsf{O}_3\) thin film, it can accelerate the transport speed of photo- induced carriers generated in the perovskite. In addition, the trapped carriers on the interface can alter the conductivity of the \(\mathsf{In}_2\mathsf{O}_3\) channel through
+
+<--- Page Split --->
+
+capacitive coupling and higher photoconductive gain can be obtained which have been well demonstrated in our prior work [Small Methods 6, 2200500 (2022)] [Adv Mater 27, 6885- 6891 (2015)]. And the hybrid structure can also be effective for perovskite/graphene and perovskite/IGZO and other different material combinations [Adv Mater 27, 41- 46 (2015)] [Adv Mater 32, e1907527 (2020)].  
+
+Q9: Pg 8. Fig 3e. Could the authors comment on why the sensitivity for low dose is higher than for high dose?  
+
+A9: Yes, we'd like to illustrate it in our revised manuscript (page 8). It is actually quite a common phenomenon in many reported perovskite X- ray detectors. As the X- ray dose rate is increased, the photocurrent demonstrates a sublinear dependence on it. This reduction in sensitivity can be explained in terms of trap states present either in \(\mathsf{In}_2\mathsf{O}_3\) or at the interface between the \(\mathsf{In}_2\mathsf{O}_3\) and the underlying \(\mathsf{SiO}_2\) layer. The photocurrent mainly comes from the photo- induced electrons filling the trap states in metal oxide and changing the metal oxide's conductivity, under high illumination intensities the density of available trap states is reduced, resulting in saturation of the photoresponse. [Nature Nanotechnology 8, 497- 501 (2013)] [Adv. Funct. Mater. 2019, 29, 1808182]  
+
+Q10: Pg. 10 "The detection limit of the DCS detector (SNR=3) is as low as 7.84 nGyair s- 1" Could the authors show a pulse for 7.84 nGyair s- 1?  
+
+A10: We understand your concerns, but we are very sorry that because the 7.84 nGyair s- 1 is extremely low, in our experiment condition and with the limitation of our equipment, we can not demonstrate a pulse for 7.84 nGyair s- 1. But you can see that our detector under 83 nGyair s- 1 (the lowest dose that we believe we can reliably measure with our ion chamber.) shows a much higher photocurrent than \(3 \times \mathsf{I}_{\mathsf{noise}}\) (152 fA), and the SNR also much higher than 3. In addition, this method to calculate the detection limit has been widely used before. [Nat Commun 12, 5258 (2021)] [Adv. Mater. 2021, 33, 2101717]. We agree with reviewers that it is not very solid to use the reverse extension line to estimate the lowest detection limit, but we have tried our best and give the pulse response at 83 nGyair s- 1 showing SNR
+
+<--- Page Split --->
+
+of honestly,  
+
+Q11: Pg 12. "of dark current drafting" Typo  
+
+A11: Thank you for your careful review and mention the mistake. We have corrected it to "of dark current drifting".  
+
+Q12: Pg. 14 Fig 5d Could the authors comment on the mobility values? 1.55 cm2/V/s seems low.  
+
+A12: We understand your concerns. The \(\ln_2O_3\) possesses a high intrinsic electron concentration and high mobility features [Journal of Physics and Chemistry of Solids 38, 819- 824 (1977)] [Adv Mater 27, 7168- 7175 (2015)]. In this work, we used solution- processed method to fabricate the \(\ln_2O_3\) thin film transistor as the switching backplane for addressing the photodetector sensors. And the linear mobility of the \(\ln_2O_3\) transistor was calculated according to the transfer curve by the following equation:  
+
+\[\mu_{lin} = \frac{dI_D}{dV_{GS}}\times \frac{L}{W}\times \frac{1}{C_{ox}\times V_{DS}}\]  
+
+However, the high contact resistance ( \(R_{contact}\) ) between the Ni/Au source/drain (S/D) electrodes and \(\ln_2O_3\) thin film will hinder the extracted mobility value based on the transfer curves [IEEE Transactions on Electron Devices 66, 5166- 5169 (2019)]. To illustrate this point, we fabricated two \(\ln_2O_3\) transistors with Ni/Au and Al source/drain electrodes, respectively, for comparison as shown in the below figure R3a and b. The extracted linear mobility of the \(\ln_2O_3\) transistor with Al S/D electrodes was \(8.23cm^2 /V\cdot s\) which is higher than that with Ni/Au S/D electrodes ( \(1.53cm^2 /V\cdot s\) ).  
+
+Although the higher \(R_{contact}\) affect the extracted mobility value, it can be conducive to reduce transistor off current ( \(I_{off}\) ) which is important for reducing the signal crosstalk among sensing units. In addition, using gold as the electrode can keep more stable device performance, because aluminum is easy to react with perovskite and cause device damage.
+
+<--- Page Split --->
+![PLACEHOLDER_26_0]
+
+<center>Figure R3: a, Transfer curve of \(\mathsf{In}_2\mathsf{O}_3\) transistor with Ni/Au and b, Al source/drain electrodes. </center>  
+
+Q13: Could the authors mention what the variation in the critical voltage for different devices were?  
+
+A13: Thank you very much for your advice. Yes, we'd like to mention this and have added the comments in our revised manuscript (page 12). With the results of current- voltage curves in terms of DCS electrode voltages and drain currents of same thick devices with different X- ray sensitive layer (contribute to most of the thickness of the device) and different thick devices with same X- ray sensitive material (Figure R4). The critical voltage (CV) seems will not be influenced by different compositions of materials, but will increase with greater distance with the conduction channel, which is also the thickness of X- ray sensitive material. Thus, we propose with the increase of the distance with conduction channel, the applied electric field of DCS electrode can be weaker around the conduction channel, and thus have weaker force to attract the dark electrons. It need stronger electric field to attract the dark electrons to DCS electrode and suppress the dark current to 0 A.  
+
+![PLACEHOLDER_26_1]
+
+<center>Figure R4: a, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathsf{FA}_{0.92}\mathsf{Cs}_{0.04}\mathsf{MA}_{0.04}\mathsf{PbI}_3\) on </center>
+
+<--- Page Split --->
+
+\(\mathsf{C}_{60} / \mathsf{In}_{2}\mathsf{O}_{3}\) TFT in the dark. \(\mathbf{b}\) , Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80 \mu \mathrm{m}\) and \(200 \mu \mathrm{m}\) MAPbI\(_3\) (PMMA binder)/SnO\(_2 / \mathsf{In}_{2}\mathsf{O}_{3}\) TFT in the dark.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS</B>  
+
+Reviewer #1 (Remarks to the Author):  
+
+Peng Jin, Yingjie Tang et al. improved the quality of the manuscript substantially and fully addressed my main concern related to the thickness of the absorbing perovskite layer. In the revised manuscript the thickness has been increased two orders of magnitudes up to \(200\mu m\) which is already significant, considering e.g., X- ray absorption in mammography application. The authors demonstrated the validity of their concept of dark current reduction for thick absorbing layers (200um) in a similar way as reported earlier for thin layers (2um). The signal to noise ration clearly suffers from increasing the absorbing layer thickness, however it should not preclude publication. A trade- off between X- ray absorption, dark current and signal height must be found accurately if industrial applications are envisioned. I suggest the publication of the revised manuscript as is.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have made a very professional and careful work in answering the questions posed by the reviewers and in amending the manuscript. I'm of the opinion that the current version is of high quality and should be therefore published in Nature Communication.  
+
+Reviewer #3 (Remarks to the Author):  
+
+I am satisfied with most of the authors' answers. I have three other concerns:  
+
+First, in the revised version the authors mention "great photoconductive gain" several times. Can they get a rough estimation of the gain factor? If they have mobility and lifetime of the channel, then they can calculate this. The disadvantage of having photoconductive gain is lower speed. Can the authors estimate the pulse rise/fall time?  
+
+Second the authors mention that the perovskite photoconductive layer has "low mobility" and the In2O3 channel has "superior mobility". For In2O3 they measured a mobility of \(1.55 \text{cm}^2 /\text{V} /\text{s}\). In the literature they cite (Adv Mater 27, 7168- 7175 (2015) 8 cm2/V/s is reported. But the typical mobility values for polycrystalline perovskite reported in lit (0.1- 10 cm2/V/s) are in the same range.  
+
+My third concern is about the thickness. I understand that the novelty here is the device structure and they have not optimized the thickness, interlayers, etc however do they even need 2 microns? The authors mention that the xray generated carriers drift under a built- in electric field between X- ray
+
+<--- Page Split --->
+
+sensitive material and electron transport layer. In my opinion, the electric field at the ETL interface is not strong and Im not sure if the depletion region extends 2 microns to the top DSC electrode. I would expect the devices to be limited by charge carrier extraction. There is not much benefit of absorbing more xrays by making thicker films if they cant extract the carriers generated. Moreover, I would expect thicker devices to be worse. Since xray absorption decreases exponentially, for thicker devices most of the carriers generated close to the top surface from where the xray is illuminated might not reach the interface where extraction is more efficient.  
+
+Other minor comments:  
+
+Line 128 and 133 CV has been defined earlier. The authors could avoid using acronyms for terms that have not been used many times such as Wv, PCD. They are not commonly used acronyms and makes it difficult to read  
+
+Line 172 "Our PMMA layer is only \(\sim 50 \text{nm}\) , it cannot act as a dielectric layer." Thinner layers have higher capacitance than thicker layer.  
+
+Line 216 "The detection limit of the DCS detector 217 (SNR = 3) is as low as 7.84 nGyair s- 1(Fig. 4g)," The authors should mention how they obtained this number in the manuscript (used reverse extension line to estimate the lowest detection....?)  
+
+Line 242 "To preliminarily investigate its spatial resolvability in scanning- based X- ray imaging, we measured its modulation transfer function(MTF) by scanning the object (Figure S11a). The used line pair mask plate is shown in Figure S11b, the minimum resolved line pair is 5.5 lp mm- 1." Could the authors show the scanned x- ray image of the object and include it in SI? Was this a single pixel device? MTF for the detector array would be more relevant. Did they measure it?  
+
+Line 255 "of dark current drafting" Typo  
+
+Figure still have a lot of text in small font. They could put only the most relevant/important text in the figure and the description could go in the caption or in the main text. For eg, they could remove or reduce in fig 2. direction of dark electron motion, photoconductive gain. In fig 4 almost no shift of dark current drifting Fig 3a text too small
+
+<--- Page Split --->
+
+As for the gain factor, the total gain produced by Heterojunction X- ray Phototransistors can be calculated as [Adv. Mater. 2021, 33, 2101717] [Adv. Funct. Mater. 2019, 29, 1900234]:  
+
+\[G = I_{\mathrm{s}}E_{\mathrm{e - h}} / \epsilon Dm e\]  
+
+Where \(I_{s}\) is the X- ray signal current (4 nA), \(E_{\mathrm{e - h}}\) is the EHP creation energy given by an empirical mode, [Nucl. Instrum. Methods Phys. Res., Sect. A 2006, 565, 637. ] \(\mathrm{E}_{\mathrm{e - h}} = 1.43\) \(+2E_{\mathrm{g}}\) ( \(\mathrm{E}_{\mathrm{g}}\) is the bandgap of perovskite, 1.55 eV), \(\epsilon\) is the fraction of absorbed photons (0.1 at \(40\mathrm{keV}\) X- ray), \(D\) is the dose rate (233 \(\mu \mathrm{Gy} / \mathrm{air} \mathrm{s}^{- 1}\) ), m is the mass of perovskite ( \(2.4 \times 10^{- 8}\) g), and \(\epsilon\) is the elementary charge. Note that the total gain is composed of impact ionization gain and photoconductive gain, in which the former is determined by the ratio of X- ray photon energy \(\mathrm{E}_{\mathrm{ph}}\) to \(\mathrm{E}_{\mathrm{e - h}}\) .  
+
+Under the \(40\mathrm{keV}\) X- ray. The overall gain factor (including ionization and photoconductive) is calculated as \(\approx 3.24 \times 10^{- 4}\) .  
+
+If using the mobility and lifetime of the channel to calculate the photoconductive gain, the gain factor can be calculated as:  
+
+\[G_{P} = \frac{t_{r}}{t_{t}} = \frac{t_{r}}{L^{2}}\mu V_{\mathrm{DS}}\]  
+
+where \(\mu\) is mobility (1.55 cm2/V s) and \(\mathrm{V}_{\mathrm{DS}}\) is applied drain- source voltage (0.5 V), L is the channel length (20 \(\mu \mathrm{m}\) ) and \(\mathrm{t}_{r}\) is the carrier lifetime of the channel (2.041 ms). The carrier lifetime of the channel was measured by time- resolution photoluminescence spectrum (TRPL):  
+
+![PLACEHOLDER_30_0]
+
+<center>Figure R1. TRPL of \(\mathrm{In}_{2}\mathrm{O}_{3}\) film. Its average carrier lifetime is calculated as 2.041 ms. </center>  
+
+As you can see, the metal oxide's lifetime is relatively long. The unique oxygen- sensitized photoconduction mechanism has allowed the photoconductive gain of metal oxide to reach an extremely high level, which is several orders of magnitude higher than the conventional thin film detectors. According to the widely adopted photoconduction model, the long carrier lifetime of metal oxide is proposed to be the origin of high- gain transport. [Nanoscale, 2013, 5, 6867- 6873] However, the very long carrier lifetime of metal oxide has aroused some
+
+<--- Page Split --->
+
+unpleasant characteristics, such as slow response time, especially the persistence photocurrent (PPC) effect, which has been frequently founded in metal oxide- based phototransistors. [Adv. Mater. 2015, 27, 6885- 6891]  
+
+The Gain factor is calculated as \(\approx 400\) . As for the perovskite, its lifetime is estimated as 100 ns the Gain factor of \(2\mu m\) perovskite photoconductive detector (under the same applied electric field) with the same mobility is calculated as \(\approx 0.2\) , which is much smaller than our type of devices.  
+
+As for your next concern, we totally agree with your words that "The disadvantage of having photoconductive gain is lower speed." The slower recombination of the carriers that took part in the transportation in the channel, the higher gain. Thus, the high gain and fast speed generally cannot be obtained simultaneously. We measured the rise and fall time of our device under the laser beam pulse. The DCS electrode is applied with CV. The rise and fall time are 23 ms and 31 ms, respectively:  
+
+![PLACEHOLDER_31_0]
+
+<center>Figure R2. Laser beam pulse response of the device. The rise and fall time are 23 ms and 31 ms, respectively. </center>  
+
+Such a response speed can meet the general \(30 \text{Hz}\) dynamic monitoring application. But as for X- ray detection, the commercially used X- ray detectors, such as a- Se detector, are fabricated in a very thick geometry, to obtain stronger absorption to X- ray. But such a thick film will definitely sacrifice the response speed to the X- ray. And in most practical cases, a- Se detector can only be used in static X- ray imaging. Even for the potential dynamic imaging applications, \(30 \text{Hz}\) can be satisfied in most situations. In our opinion, at least at this moment, for the direct perovskite X- ray detector, the dynamic response is not the first priority, since the competing technology (a- Se) is not used for dynamic X- ray imaging as well. Additionally, for those reported perovskite X- ray detectors with typical vertical geometry, the response time is also not much fast than our devices [Nature 550, 87- 91 (2017)] [Nat. Electron. 4, 681- 688 (2021)], and I believe there are also considerable photoconductive gains due to the traps of perovskite films.
+
+<--- Page Split --->
+
+But as for the dark current, it directly refers to whether the detector can or cannot be used no matter if it is static or dynamic imaging applications. High dark current can quickly fill up the storage capacitance of TFT or CMOS pixels prior to X- ray illumination, if the capacitance is filled with charges at the dark, there won't be any response from the detector when the X- ray comes in. Thus, compared with the response speed, the high dark current issue is much more urgent to be solved, at least in this early stage of perovskite X- ray detectors  
+
+Q2: Second the authors mention that the perovskite photoconductive layer has "low mobility" and the In2O3 channel has "superior mobility". For In2O3 they measured a mobility of 1.55 cm2/V/s. In the literature they cite (Adv Mater 27, 7168- 7175 (2015) 8 cm2/V/s is reported. But the typical mobility values for polycrystalline perovskite reported in lit (0.1- 10 cm2/V/s) are in the same range.  
+
+A2: We totally understand your concerns. We agree with your opinion about the mobility between \(\mathsf{In}_2\mathsf{O}_3\) and Perovskite is quite reasonable. Thus, we revised our manuscript that the perovskite has "lower mobility", and the conduction channel, such as \(\mathsf{In}_2\mathsf{O}_3\) , has "higher mobility". In fact, realizing the 3D polycrystalline perovskite with such high carrier mobility as our solution- processed polycrystalline \(\mathsf{In}_2\mathsf{O}_3\) is still a very challenging work. [Science Advances, 7, 18, (2021)] The major difficulty is the ion migration, which causes a partial screening of the applied field, yielding a very low room temperature \(\mu_{\mathrm{FET}}\) of \(10^{- 4} \mathrm{cm}^2 /\mathrm{Vs}\) in thin films of perovskite, such as MAPbI3. [Nat. Commun. 6, 7383 (2015).] Even in single- crystal perovskite- based devices, strong hysteresis and moderate mobilities of \(< 10^{- 3} \mathrm{cm}^2 /\mathrm{Vs}\) were reported at room temperature. [Nat. Commun. 7, 11330 (2016).] Various instabilities, such as the mechanism of carrier scattering and trapping, the role of ion migration, the origin of hysteresis in the device characteristics, and the electronic structure of grain boundaries in perovskites have made them reveal unusually lower mobility than in the ideal state.  
+
+But as for the \(\mathsf{In}_2\mathsf{O}_3\) , their mobility is very high, and stable and is very easy to be reproduced, the calculated electron mobility \(\mu_e\) [cm2/(Vs)] can be up to 270 - 274, the experiment value can be 7.81 - 190. [Applied Physics Reviews 9, 011315 (2022)] They can easily achieve high mobility in various fabrication processes, such as ALD [J. Phys. Chem. C 115, 15384- 15389 (2011a)] (84 cm2/(Vs)), spin- coating, PLD [Appl. Phys. Lett. 62, 2332- 2334 (1993)] (50 cm2/(Vs)), MOCVD [Vacuum 167, 1- 5 (2019)] (42 cm2/(Vs)), spray pyrolysis [J. Cryst. Growth 240, 142- 151 (2002)] (42.6 cm2/(Vs)), and dc magnetron sputtering. The \(\mathsf{In}_2\mathsf{O}_3\) film sputtered at room temperature without post- annealing results in layers with reasonably high mobility of \(51.3 \mathrm{cm}^2 /\mathrm{Vs}\) . [Nat. Commun. 6, 8932 (2015)]  
+
+The \(\mathsf{In}_2\mathsf{O}_3\) fabricated by spin- coating might have the relatively low mobility, but such a time- saving and low- cost method can be widely used in low- cost large- area applications. What's more, the electron mobility of our \(\mathsf{In}_2\mathsf{O}_3\) transistor with aluminum Source/Drain electrodes was \(8.23 \mathrm{cm}^2 /\mathrm{V}\cdot \mathrm{s}\) which is higher than that with Ni/Au Source/Drain electrodes (1.55 \(\mathrm{cm}^2 /\mathrm{V}\cdot \mathrm{s}\) ). But unfortunately, the perovskite will have a chemical reaction with aluminum, which may deteriorate the device's stability and performance. We have to use Ni/Au
+
+<--- Page Split --->
+
+Source/Drain electrodes as sacrificing the mobility of \(\mathsf{In}_2\mathsf{O}_3\) to gain stability of the device.  
+
+Q3: My third concern is about the thickness. I understand that the novelty here is the device structure and they have not optimized the thickness, interlayers, etc however do they even need 2 microns? The authors mention that the xray generated carriers drift under a built- in electric field between X- ray sensitive material and electron transport layer. In my opinion, the electric field at the ETL interface is not strong and Im not sure if the depletion region extends 2 microns to the top DSC electrode. I would expect the devices to be limited by charge carrier extraction. There is not much benefit of absorbing more xrays by making thicker films if they cant extract the carriers generated. Moreover, I would expect thicker devices to be worse. Since xray absorption decreases exponentially, for thicker devices most of the carriers generated close to the top surface from where the xray is illuminated might not reach the interface where extraction is more efficient.  
+
+A3: Genius perspectives! We appreciate your meticulous reading of our manuscript and your smart views. In perovskite, most of the carriers are indeed generated close to the top surface, if the electric field at the ETL interface is not strong enough to extract the surface photoinduced electrons in perovskite, the thicker geometry might result in poor sensitivity. To figure out this question, we made devices with \(2\mu m\) , \(1\mu m\) and \(500 \text{nm}\) perovskite, the devices were fabricated in the same conditions.  
+
+We measured their sensitivities when their DCS electrode is applied with CV, and found that \(2\mu m\) device has the highest sensitivity \((7560\mu \mathrm{Gy}_{\mathrm{air}}^{- 1}\mathrm{cm}^{- 2})\) , the \(1\mu m\) device owns a sensitivity of \(4060\mu \mathrm{Gy}_{\mathrm{air}}^{- 1}\mathrm{cm}^{- 2}\) , the \(500 \text{nm}\) device reveals the lowest \((2580\mu \mathrm{Gy}_{\mathrm{air}}^{- 1}\mathrm{cm}^{- 2})\) (Figure R3). The results seem to point out that in such a relatively thin geometry, the device with thicker perovskite film performs better in collecting photo- induced electrons, the electric field at the ETL interface might strong enough to extract the surface photo- induced electrons in perovskite. Again the relatively high sensitivity should be partially related to the photoconductive gains. Personally, when we talk about the word "extract" we are actually talking about the "drift", but we should also not forget about the "diffusion" especially given the high diffusion length of perovskite crystals. As the drifted charges are collected, it also gives additional driving force for diffusion.  
+
+We definitely believe your perspective is generally right, thicker films can generate more photo- induced electrons, thin devices might have better capacity in extracting the majority of the photo- induced electrons at the top of the perovskite. There must be a trade- off between the thickness of perovskite and the device's performance, similar to what Reviewer #1 said, "A trade- off between X- ray absorption, dark current and signal height must be found accurately if industrial applications are envisioned". The \(2\mu m\) perovskite in such a device structure might be a relatively thin geometry and the electric field at the ETL interface seems able to extract the photo- induced electrons efficiently. But based on our laboratory experiment condition, we are very sorry we cannot demonstrate the performance of the device with high- quality pure perovskite film thicker than \(2\mu m\) (we have shown the thick perovskite- PMMA composite films in the last round revision), because the precursor solution
+
+<--- Page Split --->
+
+of perovskite we used is already saturated. We hope the other researchers who have expertise in fabricating thick perovskite films can solve this optimization question in the future. We have added the comments in our revised manuscript (page 12), and thank you very much for your constructive advice.  
+
+![PLACEHOLDER_34_0]
+
+<center>Figure R3. Sensitivity of devices with different thicknesses. </center>  
+
+Other minor comments:  
+
+Q4: Line 128 and 133 CV has been defined earlier. The authors could avoid using acronyms for terms that have not been used many times such as WV, PCD. They are not commonly used acronyms and makes it difficult to read.  
+
+A4: Thank you very much for your careful review, we have deleted the WV and PCD in the manuscript, and added an annotation in the caption.  
+
+Q5: Line 172 "Our PMMA layer is only \(\sim 50 \text{nm}\) , it cannot act as a dielectric layer." Thinner layers have higher capacitance than thicker layer.  
+
+A5: Thank you very much for your notification. The PMMA layer is not insulative but may introduce a capacitance between the DCS electrode and perovskite. From the device's output signal, the capacitance seems will not affect the final signal, which is collected in the Drain electrode on the other side of the perovskite. We have added some comments on the capacitance of the PMMA in our manuscript (page 9).  
+
+Q6: Line 216 "The detection limit of the DCS detector 217 (SNR = 3) is as low as 7.84 nGyair s- 1(Fig. 4g),"  
+
+The authors should mention how they obtained this number in the manuscript (used reverse extension line to estimate the lowest detection....?)  
+
+A6: Thank you for your suggestions. We have added a comment" The lowest detection limit was estimated by the reverse extension line to where the \(\mathsf{SNR} = 3\) " in the revised manuscript.
+
+<--- Page Split --->
+
+(page 15)  
+
+Q7: Line 242 "To preliminarily investigate its spatial resolvability in scanning- based X- ray imaging, we measured its modulation transfer function(MTF) by scanning the object (Figure S11a). The used line pair mask plate is shown in Figure S11b, the minimum resolved line pair is \(5.5 \text{lp mm}^{- 1}\) ."  
+
+Could the authors show the scanned x- ray image of the object and include it in SI? Was this a single pixel device? MTF for the detector array would be more relevant. Did they measure it?  
+
+A7: Thank you for your suggestions. Yes, we calculated the MTF with a single- pixel device, we preliminarily calculated the MTF by scanning the object with a stepping motor, and recording the current value of a single- pixel device. We added this comment in our revised manuscript to avoid misleading the readers (page 12, 13). Indeed, the MTF was sometimes calculated by scanning the whole X- ray image of the object, but as you can see, every scanning line to this object is the same, when calculating the MTF, only need to scan one of the lines can preliminarily calculate the MTF. As for MTF for our imaging array, we are really sorry that we didn't measure it. Even the world's most state- of- the- art perovskite imaging array fabricated by Samsung \((1,428 \times 1,428 \text{ pixels in } 10 \text{ cm} \times 10 \text{ cm})\) with a pixel pitch of \(70 \mu \text{m}\) [Nature 550, 87- 91 (2017)], their MTF for the detector array was just \(3.1 \text{ lp mm}^{- 1}\) . In our preliminarily demonstrated detector array, there are only \(64 \times 64 \text{ pixels in } 2 \text{cm} \times 2 \text{cm}\) with a pixel pitch of \(300 \mu \text{m}\) . Thus, MTF for our detector array can be negligible. We aim to demonstrate a prototype to reveal that this method can be successfully used in an imaging array and hope to promote the method rather than the detailed performance factors of the array.  
+
+Q8: Line 255 "of dark current drafting" Typo  
+
+A8: Thank you for your notification. We have revised the "drafting" to "drifting" in Line 255.  
+
+Q9: Figure still have a lot of text in small font. They could put only the most relevant/important text in the figure and the description could go in the caption or in the main text.  
+
+For eg, they could remove or reduce in fig 2. direction of dark electron motion, photoconductive gain. In fig 4 almost no shift of dark current drifting  
+
+Fig 3a text too small  
+
+A9: Thank you for your advice. We have improved the small font and deleted some irrelevant texts in the figure.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a34463a0497b4d378c75c004251d861299ce28e08f5c4c35853a9d5a3569cb12/supplementary_0_Peer Review File__a34463a0497b4d378c75c004251d861299ce28e08f5c4c35853a9d5a3569cb12_det.mmd

@@ -0,0 +1,669 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 506, 90]]<|/det|>
+# natureportfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[106, 161, 838, 250]]<|/det|>
+Realizing Nearly- Zero Dark Current and Ultrahigh Signal- to- Noise Ratio Perovskite X- ray Detector and Image Array by Dark- Current- Shunting Strategy  
+
+<|ref|>image<|/ref|><|det|>[[95, 732, 261, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[270, 732, 880, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 325, 107]]<|/det|>
+## REVIEWER COMMENTS</B>  
+
+<|ref|>text<|/ref|><|det|>[[116, 129, 393, 146]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 167, 856, 204]]<|/det|>
+Peng Jin, Yingjie Tang et al. report about a dark- current shunting method for pixelated X- ray imagers based on Metal halide perovskites as absorbing layer.  
+
+<|ref|>text<|/ref|><|det|>[[116, 207, 850, 242]]<|/det|>
+The manuscript is well prepared and of interest for researchers working in the specific field of X- Ray detection.  
+
+<|ref|>text<|/ref|><|det|>[[116, 245, 860, 281]]<|/det|>
+The title is appropriate and reflects in a good way the work described in the article. Also, the abstract reflects the content of the article.  
+
+<|ref|>text<|/ref|><|det|>[[115, 284, 875, 340]]<|/det|>
+The introduction clearly describes the state of the art and application fields of the presented topics and problem being investigated. The reduction of the dark current is indeed one of the major challenges for perovskite- based X- Ray detectors.  
+
+<|ref|>text<|/ref|><|det|>[[115, 342, 812, 360]]<|/det|>
+Figures are accurately placed and easy to understand and the references are placed accurately.  
+
+<|ref|>text<|/ref|><|det|>[[114, 361, 881, 516]]<|/det|>
+Results are novel and quite remarkable, however the perovskite layer thickness of \(2\mu m\) precludes its use in medical X- ray imagers. The limited X- ray absorption (few percent) for \(2\mu m\) thick perovskites is too low to be used in industrial applications. This makes the impact of the manuscript rather limited. For the pixel architecture described in this work, and its working principle, it is not obvious to have the same performance of dark current suppression by retaining at the same time high photocurrent collection, for perovskite thicknesses of few hundreds of \(\mu m\) . If the authors can demonstrate that a similar behavior is obtainable for perovskite thicknesses of \(100\mu m\) and above, then I would recommend the publication in this journal otherwise a journal with lower impact factor should be better suited.  
+
+<|ref|>text<|/ref|><|det|>[[116, 577, 393, 593]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 615, 878, 731]]<|/det|>
+The manuscript investigates a new device structure strategy for MHP x- ray detectors at the purpose to decrease the dark current in metal halide perovskite x- ray detectors. The strategy is based on the use of a planar geometry for the detection of photoelectrons induced by the x- ray irradiation and by a third electrode that works as a blanking unit for the noise. The manuscript is generally well written and the author show some nice results in their configuration. I have few comments that the authors should try to insert in their discussion to allow the reader to better understand the potential of their work.  
+
+<|ref|>text<|/ref|><|det|>[[113, 752, 880, 907]]<|/det|>
+1- The device structure that they propose is substantially more complex than the structure used for other direct conversion devices such the one based on a- Se. They should comment on the consequence of this more complicated structure for the read- out electronics. 
+2- The energy of the used x- rays should be stated in the experimental section. 
+3- In the experimental resukts the active layer is 2micrometers, however, the thickness ideally should be higher. The authors should comment on the effect of the active layer thickness on their device performance and voltages that would need to be apply. 
+4- The authors mentioned that in their opinion this is not a transistor structure, but as they have a
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 864, 146]]<|/det|>
+substantial PMMA layer this is certainly working as capacitor. I think the authors should better explain their point by comparing with phototransistors structures eventually reported in literature if there are any made with MHP.  
+
+<|ref|>text<|/ref|><|det|>[[116, 207, 393, 223]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 245, 872, 322]]<|/det|>
+The authors report a perovskite based x- ray detector with a new device structure using dark shunting electrodes to suppress dark current and obtained a high SNR. The dark current values obtained are probably the lowest reported so far in literature. Another significant result is that the dark current is stable which is essential for practical devices. They also demonstrate x- ray imaging with detector array.  
+
+<|ref|>text<|/ref|><|det|>[[114, 343, 882, 457]]<|/det|>
+Although the results obtained are impressive, the working mechanism or the device physics part was not clear to me. For example, it was not clear why the "dark electrons" flow from the source through ETL and perovskite to the dark shunting electrode but the photogenerated electrons flow in the opposite direction. The authors mention that the photogenerated electrons in the perovskite drift at the ETL interface, so wouldn't there be a barrier for dark electrons flowing in the opposite direction? Perhaps an energy band diagram can be shown?  
+
+<|ref|>text<|/ref|><|det|>[[115, 479, 864, 516]]<|/det|>
+Also, I was not convinced if this device structure is generally applicable. Would it still work for another perovskite composition and another conducting channel?  
+
+<|ref|>text<|/ref|><|det|>[[115, 537, 640, 555]]<|/det|>
+Figures could be improved. Some figures have a lot of text in small font.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 597, 242, 613]]<|/det|>
+## Other comments  
+
+<|ref|>text<|/ref|><|det|>[[115, 635, 857, 672]]<|/det|>
+Pg 3 "...the drain only receives X- ray- generated electrons but rejects dark electrons..." This sentence seems misleading. Could the authors clarify how the drain rejects dark electrons?  
+
+<|ref|>text<|/ref|><|det|>[[115, 693, 841, 730]]<|/det|>
+Pg 3 Fig 1. The photo- induced electrons and holes are enclosed in a dashed oval. Could the authors explain what it means?  
+
+<|ref|>text<|/ref|><|det|>[[115, 751, 872, 807]]<|/det|>
+Pg 5 "...those photo electrons are drifted when interfaced with the electron transport layer (ETL) and sensitizes the lateral conduction channel" It is not clear what the authors mean by sensitizes the lateral channel.  
+
+<|ref|>text<|/ref|><|det|>[[115, 811, 564, 828]]<|/det|>
+Perhaps the authors could use an energy band diagram in SI?  
+
+<|ref|>text<|/ref|><|det|>[[115, 830, 844, 867]]<|/det|>
+Also I would expect the effective field at the interface to be not strong and solely relying on the ETL interface to extract electrons seems like an inefficient strategy.  
+
+<|ref|>text<|/ref|><|det|>[[112, 888, 879, 907]]<|/det|>
+Pg 6 "The primary functional layers of this device, perovskite and In2O3" Could the authors explain what
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 874, 146]]<|/det|>
+kind of material In2O3 is why they chose it? This is also the first time they mention In2O3. Perhaps they could mention somewhere in the beginning that they used In2O3 as the conductive channel because I was under the impression that it was all perovskite.  
+
+<|ref|>text<|/ref|><|det|>[[115, 167, 840, 203]]<|/det|>
+Pg 8. Fig 3e. Could the authors comment on why the sensitivity for low dose is higher than for high dose?  
+
+<|ref|>text<|/ref|><|det|>[[115, 225, 865, 262]]<|/det|>
+Pg. 10 "The detection limit of the DCS detector (SNR=3) is as low as 7.84 nGyair s- 1" Could the authors show a pulse for 7.84 nGyair s- 1?  
+
+<|ref|>text<|/ref|><|det|>[[115, 283, 394, 301]]<|/det|>
+Pg 12. "of dark current drafting" Typo  
+
+<|ref|>text<|/ref|><|det|>[[115, 322, 780, 340]]<|/det|>
+Pg. 14 Fig 5d Could the authors comment on the mobility values? 1.55 cm2/V/s seems low.  
+
+<|ref|>text<|/ref|><|det|>[[115, 361, 810, 379]]<|/det|>
+Could the authors mention what the variation in the critical voltage for different devices were?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 90, 821, 161]]<|/det|>
+Firstly, we very much appreciated the constructive comments and valuable suggestions from reviewers 1, 2 and 3. Now all the questions have been thoroughly considered and carefully answered. And hopefully, our responses can release your concerns.  
+
+<|ref|>text<|/ref|><|det|>[[178, 174, 815, 186]]<|/det|>
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*  
+
+<|ref|>text<|/ref|><|det|>[[178, 203, 528, 218]]<|/det|>
+The author's answers to Reviewer #1's comment:  
+
+<|ref|>text<|/ref|><|det|>[[178, 231, 284, 245]]<|/det|>
+Dear reviewer:  
+
+<|ref|>text<|/ref|><|det|>[[178, 257, 824, 329]]<|/det|>
+Thank you very much for your supportive comments and constructive suggestions. We have carefully read your reviews and noticed your concerns. Your question was answered as below. The manuscript has been revised with fully consideration of your comments.  
+
+<|ref|>text<|/ref|><|det|>[[178, 367, 821, 467]]<|/det|>
+Reviewer's comments: Peng Jin, Yingjie Tang et al. report about a dark- current shunting method for pixelated X- ray imagers based on Metal halide perovskites as absorbing layer. The manuscript is well prepared and of interest for researchers working in the specific field of X- Ray detection.  
+
+<|ref|>text<|/ref|><|det|>[[178, 479, 820, 522]]<|/det|>
+The title is appropriate and reflects in a good way the work described in the article. Also, the abstract reflects the content of the article.  
+
+<|ref|>text<|/ref|><|det|>[[178, 533, 821, 605]]<|/det|>
+The introduction clearly describes the state of the art and application fields of the presented topics and problem being investigated. The reduction of the dark current is indeed one of the major challenges for perovskite- based X- Ray detectors.  
+
+<|ref|>text<|/ref|><|det|>[[178, 617, 820, 660]]<|/det|>
+Figures are accurately placed and easy to understand and the references are placed accurately.  
+
+<|ref|>text<|/ref|><|det|>[[178, 672, 790, 688]]<|/det|>
+A0: Thanks so much for your comments that you feel this work is novel and important.  
+
+<|ref|>text<|/ref|><|det|>[[178, 700, 822, 911]]<|/det|>
+Q1: Results are novel and quite remarkable, however the perovskite layer thickness of \(2\mu \mathrm{m}\) precludes its use in medical X- ray imagers. The limited X- ray absorption (few percent) for 2 \(\mu \mathrm{m}\) thick perovskites is too low to be used in industrial applications. This makes the impact of the manuscript rather limited. For the pixel architecture described in this work, and its working principle, it is not obvious to have the same performance of dark current suppression by retaining at the same time high photocurrent collection, for perovskite thicknesses of few hundreds of \(\mu \mathrm{m}\) . If the authors can demonstrate that a similar behavior is obtainable for perovskite thicknesses of \(100\mu \mathrm{m}\) and above, then I would recommend the publication in this
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 90, 712, 106]]<|/det|>
+journal otherwise a journal with lower impact factor should be better suited.  
+
+<|ref|>text<|/ref|><|det|>[[177, 117, 822, 301]]<|/det|>
+A1: We appreciate it very much that your feel those results are novel and remarkable. We certainly understand your concerns. The main purpose of our work is to demonstrate a common device strategy method to suppress dark current and obtain a high signal- to- noise ratio, there is certainly quite a lot of rooms to further improve the device performance by material engineering, and we are working on it certainly. Actually in my points of view, depositing high- quality thick film perovskite is still quite challenging, since there is a solubility limit for the perovskite precursor solution.  
+
+<|ref|>text<|/ref|><|det|>[[176, 312, 823, 718]]<|/det|>
+Our device strategy is very universal and can be applied in various material systems with different thicknesses, and compositions, we have demonstrated this in our revised manuscript (page 12, Figure S8). Thick X- ray sensitive materials can obtain stronger absorption of incident X- ray. But the synthesis of high- quality thick perovskite films is still a big challenge around the world. We have used the nearly saturated perovskite precursor solution in the first version of the manuscript, the 2um perovskite film is the thickest we can fabricate by the typical precursor- based deposition. In this round, we referred to some previous works on thick perovskite films and added experiments to verify our DCS method in thick films, which was fabricated by the polymer binding method. [Adv. Eng. Mater., 18: 1189- 1199] [IEEE Sensors Journal, vol. 7, no. 6, pp. 925- 930, June 2007] [Adv. Funct. Mater. 2022, 32, 2110729] The 200 μm and 80 μm MAPbI₃/SnO₂/In₂O₃ TFT devices are shown below. The devices are fabricated by blade- coating the MAPbI₃ and PMMA binder (dissolved in PMMA/Toluene solvent, MAPbI₃:PMMA = 1:2) on SnO₂/In₂O₃ TFT. Because the Toluene solvent will largely dissolve the C₆₀ layer, we replace it with SnO₂, which is also an electron transport layer (ETL) for perovskite.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[179, 90, 816, 300]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[177, 312, 822, 467]]<|/det|>
+The perovskite film consists of MAPbI \(_3\) and PMMA polymer binder, PMMA is an insulator, thus, the conductivity and sensitivity of such devices are relatively low. We measured the I- V curves of the \(200\mu m\) MAPbI \(_3\) and PMMA polymer binder film- based photoconductor (Au/MAPbI3 and PMMA polymer mixture film/Au) in the dark and under X- ray, both dark current and photocurrent are very low. At the voltage of 0.5V, the photocurrent is 25 pA (Figure R1).  
+
+<|ref|>image<|/ref|><|det|>[[283, 533, 675, 780]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 802, 820, 846]]<|/det|>
+<center>Figure R1: Current-voltage curves of \(200\mu m\) MAPbI \(_3\) photoconductor in the dark and under X-ray. </center>  
+
+<|ref|>text<|/ref|><|det|>[[177, 875, 819, 893]]<|/det|>
+We measured the current- voltage curves in terms of DCS electrode voltages and drain
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 88, 822, 273]]<|/det|>
+currents of these thick devices in the dark and under X- ray. For all the \(80\mu m\) and \(200\mu m\) devices we fabricated, all of the devices' dark currents can be suppressed to 0 A. We found that the Critical Voltage increased with the rise of the thickness of X- ray sensitive materials (Figure R2a below). The dark current of \(200\mu m\) device can also be suppressed to nearly zero with DCS method, the dark current baseline is stable and signal- to- noise ratio can also be very high (Figure R2c,d below). We have added those new results in our revised manuscript (page 12).  
+
+<|ref|>text<|/ref|><|det|>[[178, 284, 820, 328]]<|/det|>
+Overall, we believe our device strategy proposed in this study is very universal and can be applied in various material systems with different thicknesses.  
+
+<|ref|>image<|/ref|><|det|>[[194, 353, 785, 842]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 857, 820, 902]]<|/det|>
+<center>Figure R2: Performance of thick perovskite film devices. a, Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80\mu m\) and \(200\mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 89, 821, 216]]<|/det|>
+TFT in the dark. b, Current- voltage curves in terms of DCS electrode voltages and drain currents of \(80 \mu m\) and \(200 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT under 21.25 \(\mu\) Gy\(_air\) s\(^{- 1}\) . c, Pulse- train response of \(200 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT device under 21.25 \(\mu\) Gy\(_air\) s\(^{- 1}\) . The DCS electrode is biased with CV. d, Pulse- train response of \(200 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT device under 21.25 \(\mu\) Gy\(_air\) s\(^{- 1}\) . The DCS electrode is disabled  
+
+<|ref|>text<|/ref|><|det|>[[178, 226, 815, 238]]<|/det|>
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\* \*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*
+
+<|ref|>text<|/ref|><|det|>[[178, 285, 528, 300]]<|/det|>
+The author's answers to Reviewer #2's comment:  
+
+<|ref|>text<|/ref|><|det|>[[178, 313, 284, 327]]<|/det|>
+Dear reviewer:  
+
+<|ref|>text<|/ref|><|det|>[[178, 340, 820, 410]]<|/det|>
+We really appreciate your reviews and valuable pieces of advice for our paper. We have carefully read your comments and noticed your concerns. All of your questions were answered below.  
+
+<|ref|>text<|/ref|><|det|>[[178, 422, 812, 438]]<|/det|>
+The manuscript has been revised with full consideration of your concerns and suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[177, 478, 821, 710]]<|/det|>
+Reviewer's comments: The manuscript investigates a new device structure strategy for MHP x- ray detectors at the purpose to decrease the dark current in metal halide perovskite x- ray detectors. The strategy is based on the use of a planar geometry for the detection of photoelectrons induced by the x- ray irradiation and by a third electrode that works as a blanking unit for the noise. The manuscript is generally well written and the author show some nice results in their configuration. I have few comments that the authors should try to insert in their discussion to allow the reader to better understand the potential of their work.  
+
+<|ref|>text<|/ref|><|det|>[[178, 723, 820, 799]]<|/det|>
+Q1: The device structure that they propose is substantially more complex than the structure used for other direct conversion devices such the one based on a- Se. They should comment on the consequence of this more complicated structure for the read- out electronics.  
+
+<|ref|>text<|/ref|><|det|>[[178, 811, 820, 911]]<|/det|>
+A1: Thank you for your careful review and recognition. And thanks for this great question! We fully understand your concerns and we'd like to illustrate this more clearly in the revised manuscript, and we have added related comments in our revised manuscript (page 16). Our device structure is shown in Supplementary Information Figure S13, and is shown below
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[176, 87, 822, 636]]<|/det|>
+(Figure R1a). It is true that the single photoconductive detector is very simple, but when they are integrated with the back panel, the TFT device is also needed. The general structure of the pixelized a- Se detection unit is shown in Figure R1b below [Proc. SPIE 5368, Medical Imaging 2004: Physics of Medical Imaging, (6 May 2004)]. As you can see, the traditional a- Se array is fabricated by coupling the a- Se material in the surface of back panel, which is prepared ahead typically with a- Si, and then evaporating the top common contact electrode. In our structure, the \(\mathsf{In}_2\mathsf{O}_3\) TFT array- based back panel is also prepared ahead, the latter process is actually quite similar. Given that we have the In2O3 TFT back panel in hand, what we do next is to deposit the X- ray sensitive material (perovskite) onto the \(\mathsf{In}_2\mathsf{O}_3\) TFT array back panel and then evaporate the top common DCS electrode. The difference in the fabrication process is mainly the manufacture of the back panel. If the \(\mathsf{In}_2\mathsf{O}_3\) (or IGZO) TFT array is fully developed and well- manufactured by the company, there won't be too much difficulty in making our DCS method- based device and arrays. As we can know the IGZO- based backplane is starting to take over the a- Si TFT. As for the signal readout, the a- Se detector directly readout the charge (current) signal from Drain electrode of the switching MOSFET (as shown below, Figure R1b), our device also readout the signal from Drain electrode of the switching TFT (Figure R1a), the analog data processing units including the preamplifier, Analog to Digital (AD) conversion can be shared with the conventional a- Si sensors, actually, we are using the same ROIC chips purchased from ADI in our study for the X- ray image demonstration.  
+
+<|ref|>text<|/ref|><|det|>[[177, 644, 822, 800]]<|/det|>
+Another difference might be related to the pixel size, with the industrial facilities, the typical pixel size of a photoconductive a- Se detector is \(\sim 100 \mu \mathrm{m}\) , in this study, we have demonstrated the X- ray imager with a pixel size of \(\sim 300 \mu \mathrm{m}\) , with our in- house lab facilities, compare with the commercial X- ray detector's pixel size (100 \(\mu \mathrm{m}\) ), we think it is not very difficult for companies to use our DCS method to produce 100 \(\mu \mathrm{m}\) pixel based X- ray detector array in the future.  
+
+<|ref|>text<|/ref|><|det|>[[177, 812, 821, 912]]<|/det|>
+One particular good thing of our device is that, for the typical X- ray detector array, we usually have to purposely design a big storage capacitance ( \(C_{\mathrm{st}}\) ) in each pixel, which is parallel connected with the built- in capacitance (naturally formed by the parallel electrodes), to beat the dark current. Otherwise, the capacitance is too small that they can be easily
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 90, 821, 162]]<|/det|>
+filled up by the dark current. In our DCS detector, the dark current can be suppressed to extremely low, thus, in principle, we don't need to put an extra parallel storage capacitance in each pixel, which would simplify the pixel circuit design.  
+
+<|ref|>image<|/ref|><|det|>[[202, 180, 768, 571]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 580, 821, 653]]<|/det|>
+<center>Figure R1: a, Sectional view of a single-pixel in the DCS array. The DCS detector is in series with a switching TFT to control the on/off of a single-pixel. b, Sectional view of a single-pixel in the a-Se array. </center>  
+
+<|ref|>text<|/ref|><|det|>[[177, 692, 741, 708]]<|/det|>
+Q2: The energy of the used x- rays should be stated in the experimental section.  
+
+<|ref|>text<|/ref|><|det|>[[177, 719, 820, 791]]<|/det|>
+A2: Thank you for your advice, we have added the energy of the used x- rays (tube voltage: 40 keV and 50 keV) in our experimental section. (page 23) The used output spectrum at different voltages is shown below, the figure was added in the SI. (Figure S17)
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[312, 87, 684, 284]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[178, 349, 821, 420]]<|/det|>
+Q3: In the experimental results the active layer is 2micrometers, however, the thickness ideally should be higher. The authors should comment on the effect of the active layer thickness on their device performance and voltages that would need to be apply.  
+
+<|ref|>text<|/ref|><|det|>[[177, 431, 822, 615]]<|/det|>
+A3: We appreciate it very much for your advice. We certainly understand your concerns. The main purpose of our work is to demonstrate a common device strategy method to suppress dark current and obtain a high signal- to- noise ratio, there is certainly quite a lot of rooms to further improve the device performance by material engineering, and we are working on it certainly. Actually in my points of view, depositing high- quality thick film perovskite is still quite challenging, since there is a solubility limit for the perovskite precursor solution.  
+
+<|ref|>text<|/ref|><|det|>[[177, 626, 823, 893]]<|/det|>
+Our device strategy is very universal and can be applied in various material systems with different thicknesses, and compositions, we have demonstrated this in our revised manuscript (page 12, Figure S8). Thick X- ray sensitive materials can obtain stronger absorption of incident X- ray. But the synthesis of high- quality thick perovskite films is still a big challenge around the world. We have used the nearly saturated perovskite precursor solution in the first version of the manuscript, the \(2\mu m\) perovskite film is the thickest we can fabricate by the typical precursor- based deposition. In this round, we referred to some previous works on thick perovskite films and added experiments to verify our DCS method in thick films, which was fabricated by the polymer binding method. [Adv. Eng. Mater., 18: 1189- 1199] [IEEE Sensors Journal, vol. 7, no. 6, pp. 925- 930, June 2007] [Adv. Funct. Mater.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 88, 822, 219]]<|/det|>
+2022, 32, 2110729] The \(200 \mu m\) and \(80 \mu m\) MAPbI\(_3\) /SnO\(_2\) /In\(_2\) O\(_3\) TFT devices are shown below. The devices are fabricated by blade-coating the MAPbI\(_3\) and PMMA binder (dissolved in PMMA/Toluene solvent, MAPbI\(_3\) :PMMA = 1:2) on SnO\(_2\) /In\(_2\) O\(_3\) TFT. Because the Toluene solvent will largely dissolve the C\(_{60}\) layer, we replace it with SnO\(_2\), which is also an electron transport layer (ETL) for perovskite.  
+
+<|ref|>image<|/ref|><|det|>[[179, 230, 816, 440]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[177, 450, 822, 606]]<|/det|>
+The perovskite film consists of MAPbI\(_3\) and PMMA polymer binder, PMMA is an insulator, thus, the conductivity and sensitivity of such devices are relatively low. We measured the I- V curves of the \(200 \mu m\) MAPbI\(_3\) and PMMA polymer binder film- based photoconductor (Au/MAPbI3 and PMMA polymer mixture film/Au) in the dark and under X- ray, both dark current and photocurrent are very low. At the voltage of 0.5V, the photocurrent is 25 pA (Figure R2).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[283, 115, 672, 365]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 386, 820, 430]]<|/det|>
+<center>Figure R2: Current-voltage curves of \(200\mu m\) MAPbI3 photoconductor in the dark and under X-ray. </center>  
+
+<|ref|>text<|/ref|><|det|>[[177, 459, 822, 671]]<|/det|>
+We measured the current- voltage curves in terms of DCS electrode voltages and drain currents of these thick devices in the dark and under X- ray. For all the \(80\mu m\) and \(200\mu m\) devices we fabricated, all of the devices' dark currents can be suppressed to 0 A. We found that the Critical Voltage increased with the rise of the thickness of X- ray sensitive materials (Figure R3a below). The dark current of \(200\mu m\) device can also be suppressed to nearly zero with DCS method, the dark current baseline is stable and signal- to- noise ratio can also be very high (Figure R3c,d below). We have added those new results in our revised manuscript (page 12).  
+
+<|ref|>text<|/ref|><|det|>[[178, 682, 820, 727]]<|/det|>
+Overall, we believe our device strategy proposed in this study is very universal and can be applied in various material systems with different thicknesses.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[192, 102, 784, 592]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 607, 821, 792]]<|/det|>
+<center>Figure R3: Performance of thick perovskite film devices. a, Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80\mu m\) and \(200\mu m\) MAPbI3/SnO2/In2O3 TFT in the dark. b, Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80\mu m\) and \(200\mu m\) MAPbI3/SnO2/In2O3 TFT under \(21.25\mu Gy_{air}s^{-1}\) . c, Pulse-train response of \(200\mu m\) MAPbI3/SnO2/In2O3 TFT device under \(21.25\mu Gy_{air}s^{-1}\) . The DCS electrode is biased with CV. d, Pulse-train response of \(200\mu m\) MAPbI3/SnO2/In2O3 TFT device under \(21.25\mu Gy_{air}s^{-1}\) . The DCS electrode is disabled </center>  
+
+<|ref|>text<|/ref|><|det|>[[178, 857, 820, 902]]<|/det|>
+Q4: The authors mentioned that in their opinion this is not a transistor structure, but as they have a substantial PMMA layer this is certainly working as capacitor. I think the authors
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 90, 820, 134]]<|/det|>
+should better explain their point by comparing with phototransistors structures eventually reported in literature if there are any made with MHP.  
+
+<|ref|>text<|/ref|><|det|>[[176, 145, 822, 411]]<|/det|>
+A4: Thanks for this wonderful question, we totally understand your concerns, and we have added more discussions in our revised manuscript (page 9). We'd like to prompt that in our device, even without the PMMA layer, a similar device behavior is observed, and there is the dark- current- shunting effect as well (Supplementary Information Figure S6). We believe the very thin PMMA layer only acts as a protection layer and we don't find a capacitor effect in the photocurrent and dark current. Some of the devices may use the thick PMMA layer as a dielectric layer, but their PMMA layer is usually very thick. Our PMMA layer is only \(\sim 50 \text{nm}\) , it can hardly act as a dielectric layer. Besides, comparing the MHP transistor's transfer curves [Nat Electron 5, 78- 83 (2022)] to our device (Supplementary Information Figure S5), there is no on/off state in our device when adjusting the DCS electrode's voltage.  
+
+<|ref|>text<|/ref|><|det|>[[178, 420, 816, 430]]<|/det|>
+\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*  
+
+<|ref|>text<|/ref|><|det|>[[178, 506, 527, 522]]<|/det|>
+The author's answers to Reviewer #3's comment:  
+
+<|ref|>text<|/ref|><|det|>[[178, 535, 284, 549]]<|/det|>
+Dear reviewer:  
+
+<|ref|>text<|/ref|><|det|>[[177, 561, 821, 689]]<|/det|>
+Thank you very much for your reviews and valuable suggestions for our paper. These suggestions really help us to refine our manuscript and make it more understandable for readers. We have carefully read your comments and noticed your concerns. All of your questions were answered below. And the manuscript has been revised with the entire consideration of your suggestions and concerns.  
+
+<|ref|>text<|/ref|><|det|>[[177, 728, 821, 857]]<|/det|>
+Reviewer's comments: The authors report a perovskite based x- ray detector with a new device structure using dark shunting electrodes to suppress dark current and obtained a high SNR. The dark current values obtained are probably the lowest reported so far in literature. Another significant result is that the dark current is stable which is essential for practical devices. They also demonstrate x- ray imaging with detector array.  
+
+<|ref|>text<|/ref|><|det|>[[178, 868, 820, 911]]<|/det|>
+A0: Thanks so much for your comments that you recognize the work's result about the low and stable dark current as well as the imaging with detector array.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 90, 822, 273]]<|/det|>
+Q1: Although the results obtained are impressive, the working mechanism or the device physics part was not clear to me. For example, it was not clear why the "dark electrons" flow from the source through ETL and perovskite to the dark shunting electrode but the photogenerated electrons flow in the opposite direction. The authors mention that the photogenerated electrons in the perovskite drift at the ETL interface, so wouldn't there be a barrier for dark electrons flowing in the opposite direction? Perhaps an energy band diagram can be shown?  
+
+<|ref|>text<|/ref|><|det|>[[177, 284, 822, 440]]<|/det|>
+A1: Thank you very much for your positive comments and very careful review. We understand your concerns and revised the working mechanism (Fig.1 in the manuscript). Your comments really promote us to think deeper and in more detail about the device's working principles, we have added a band diagram below, and put more discussion on the working mechanism, the energy levels come from our previous work measured with UPS [Small Methods 2022, 6, 2200500].  
+
+<|ref|>text<|/ref|><|det|>[[177, 451, 822, 690]]<|/det|>
+Firstly, with our device design, the X- ray photocurrent is actually decreased under DCS mode, compared with the control device without DCS voltages. But the actual conduction channel is the high- mobility \(\mathsf{In}_{2}\mathsf{O}_{3}\) , not the perovskite, and there is a photoconductive gain of the \(\mathsf{In}_{2}\mathsf{O}_{3}\) channel, therefore the decrease of X- ray photocurrent is not much, and the overall SNR is enhanced with several orders of magnitudes. This sort of gain is observed in many hybrid device structure. For the typical two- terminal photoconductor or photoresistor detector made of perovskite, the conduction channel is always perovskite and is usually without gain (unless there is trapped- induced photoconductive gain), we believe the photocurrent will be cut more if we put a similar DCS electrode.  
+
+<|ref|>text<|/ref|><|det|>[[178, 728, 821, 828]]<|/det|>
+Let us get back to our device structure, under the dark conditions, most of the dark electrons can be collected by the DCS electrode (positive), or at least they will no longer be received by the drain, which gives the zero- dark current. (In the manuscript page 4, Figure 1b; Page 10, Figure 3b; Page 14, Figure 4a). We believe this observation is solid.  
+
+<|ref|>text<|/ref|><|det|>[[178, 840, 821, 912]]<|/det|>
+We admit that this DCS bias could also affect the X- ray photocurrent since we indeed find a reduced photocurrent. (In the manuscript Page 14, Figure 4a) In order to better illustrate the detailed working mechanism under X- ray, we divided the device in two regions
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 90, 822, 190]]<|/det|>
+separated by the red dash line. Those two regions have different external electrical fields, the source is grounded and the DCS and drain are both positive, (in our cases they are 0.56 and 0.5 respectively), the left region has much larger external electric field than the right region.  
+
+<|ref|>image<|/ref|><|det|>[[184, 202, 884, 424]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[176, 441, 824, 626]]<|/det|>
+On the left of the red dash line: As shown in Figure b, between the Source electrode and DCS electrode, there is an externally applied voltage when the device is working, this external voltage will generate an electric field from DCS electrode to Source. The photogenerated electrons in this region move in the same direction with shunted dark electrons. Thus, the photocurrent decreases when the device is working, as you can see in Fig. 3e in the manuscript. But with the great photoconductive gain effect of the \(\mathrm{In}_2\mathrm{O}_3\) conduction channel (high mobility), the sensitivity can still reach a high value.  
+
+<|ref|>image<|/ref|><|det|>[[187, 634, 884, 860]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[176, 877, 820, 893]]<|/det|>
+On the right of the red dash line: The external electric field is much weaker in this region
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 88, 822, 274]]<|/det|>
+(the DCS and Drains are both positively biased, and in our cases they are 0.56 and 0.5 respectively), and the built- in electric field is more favorable for pushing electron downward to the conduction channel. (Figure b). But with the built- in electric field between perovskite and \(C_{60}\) due to their energy band type- II alignment, you can see it from figure b, it will be more easier for the photogenerated electrons to be drifted from the perovskite towards the \(\mathrm{In}_2\mathrm{O}_3\) . Those X- ray photo- charges mainly contribute to the observed drain photo- current with the help of the fast conduction channel.  
+
+<|ref|>text<|/ref|><|det|>[[177, 339, 820, 385]]<|/det|>
+Q2: Also, I was not convinced if this device structure is generally applicable. Would it still work for another perovskite composition and another conducting channel?  
+
+<|ref|>text<|/ref|><|det|>[[177, 395, 821, 496]]<|/det|>
+A2: We totally understand your concerns and fabricated the devices with different perovskite materials and other metal oxide conduction channels to demonstrate the generality of DCS method. The devices are shown below, we used MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{In}_2\mathrm{O}_3\) and IGZO conduction channel.  
+
+<|ref|>image<|/ref|><|det|>[[189, 505, 818, 753]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[177, 765, 821, 866]]<|/det|>
+We measured the current- voltage curves in terms of DCS electrode voltages and drain currents of these devices in the dark and under X- ray (Figure R1). It can be seen that the Critical Voltage (CV) of these devices almost the same (0.35 – 0.4 V). All of the devices' dark current can be suppressed to 0 A by applying the DCS method.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[232, 108, 750, 456]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[176, 477, 822, 690]]<|/det|>
+<center>Figure R1: Current-voltage curves in terms of DCS electrode voltages and drain currents of devices with \(1\mu m MAPbBr_3\) , MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{In}_2\mathrm{O}_3\) and IGZO conduction channel in the dark and under X-ray. a, Current-voltage curves of \(1\mu m MAPbBr_3\) , MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{In}_2\mathrm{O}_3\) TFT in the dark. b, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{IGZO}\) TFT in the dark. c, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{In}_2\mathrm{O}_3\) TFT under 21.25 \(\mu \mathrm{Gy}_{\mathrm{air}}\mathrm{s}^{-1}\) . d, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3\) on \(\mathrm{C}_{60} / \mathrm{IGZO}\) TFT under 21.25 \(\mu \mathrm{Gy}_{\mathrm{air}}\mathrm{s}^{-1}\) . </center>  
+
+<|ref|>text<|/ref|><|det|>[[177, 718, 822, 875]]<|/det|>
+We measured these devices' pulse train response when the DCS electrode is biased with CV (Figure R2). All of the devices demonstrate stable current baselines and high signal- to- noise ratio (pulse- train response of \(\mathrm{FA}_{0.92}\mathrm{Cs}_{0.04}\mathrm{MA}_{0.04}\mathrm{PbI}_3 / \mathrm{C}_{60} / \mathrm{In}_2\mathrm{O}_3\) have already demonstrated in the manuscript). It can be seen that the DCS method is very general and can be applied in various devices with different material systems. We have added this comment in our revised manuscript (page 12).
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[191, 92, 800, 465]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[176, 479, 821, 636]]<|/det|>
+<center>Figure R2: Pulse-train response of different perovskite and conduction channel materials based devices. The DCS electrode are all biased with CV. a, Pulse-train response of MAPbBr3/C60/In2O3 TFT under 21.25 μGyair s-1. b, Pulse-train response of MAPbI3/C60/In2O3 TFT under 21.25 μGyair s-1. c, Pulse-train response of MAPbBr3/C60/IGZO TFT under 21.25 μGyair s-1. d, Pulse-train response of MAPbI3/C60/IGZO TFT under 21.25 μGyair s-1, e, Pulse-train response of FA0.92Cs0.04MA0.04PbI3/C60/IGZO TFT under 21.25 μGyair s-1. </center>  
+
+<|ref|>text<|/ref|><|det|>[[176, 672, 822, 802]]<|/det|>
+Instead of using the DCS method to solution- processed devices, we also tried to apply it in sputtered IGZO- based devices. We are not the experts on sputtering IGZO, but we did try to do it here with our own sputtering, honestly, the TFT device is not much better than our solution- processed devices. As shown below, the dark current of the MAPbBr3/SnO2/IGZO (Sputtered) TFT can also be suppressed to 0 A with the DCS method.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[296, 92, 700, 340]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[177, 377, 718, 393]]<|/det|>
+Q3: Figures could be improved. Some figures have a lot of text in small font.  
+
+<|ref|>text<|/ref|><|det|>[[177, 404, 817, 421]]<|/det|>
+A3: Thank you very much for your advice. The small font has been improved in the figures.  
+
+<|ref|>text<|/ref|><|det|>[[178, 460, 298, 475]]<|/det|>
+Other comments  
+
+<|ref|>text<|/ref|><|det|>[[177, 488, 821, 560]]<|/det|>
+Q4: Pg 3 "...the drain only receives X- ray- generated electrons but rejects dark electrons..." This sentence seems misleading. Could the authors clarify how the drain rejects dark electrons?  
+
+<|ref|>text<|/ref|><|det|>[[177, 571, 821, 642]]<|/det|>
+A4: Thank you very much for your careful review and noting it. This is a wrong statement and we have corrected the "rejects" to "will not receive" in the manuscript. "Not receive" means that there will no dark electrons reach the drain electrode and be collected.  
+
+<|ref|>text<|/ref|><|det|>[[177, 682, 820, 725]]<|/det|>
+Q5: Pg 3 Fig 1. The photo- induced electrons and holes are enclosed in a dashed oval. Could the authors explain what it means?  
+
+<|ref|>text<|/ref|><|det|>[[177, 737, 821, 837]]<|/det|>
+A5: We totally understand your concerns. The photo- induced electrons and holes are enclosed in a dashed oval means an electron- hole pair generated together under the X- ray illumination, and the electrons and holes in a dashed oval are used as examples to illustrate the motion of whole electrons in the material.  
+
+<|ref|>text<|/ref|><|det|>[[175, 876, 820, 893]]<|/det|>
+Q6: Pg 5 "...those photo electrons are drifted when interfaced with the electron transport
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 90, 819, 134]]<|/det|>
+layer (ETL) and sensitizes the lateral conduction channel" It is not clear what the authors mean by sensitizes the lateral channel.  
+
+<|ref|>text<|/ref|><|det|>[[178, 145, 612, 161]]<|/det|>
+Perhaps the authors could use an energy band diagram in SI?  
+
+<|ref|>text<|/ref|><|det|>[[176, 172, 822, 551]]<|/det|>
+A6: We totally understand your concerns and we'd like to make them clear in our revised manuscript (page 6). Sensitization is the process of generating charge carriers in the semiconductors who cannot absorb X- ray and generate hole- electron pairs themselves (Such as \(\mathrm{In}_2\mathrm{O}_3\) ). These semiconductors may not able to generate charges under X- ray but have superior mobility to transport and recirculate charges and greatly amplify the photocurrent. Thus, we use a sensitizer (Perovskite) to help generate charges under the light (X- ray) and extract them to these semiconductors \(\mathrm{(In_2O_3)}\) . The photosensitive material generates an abundance of charges, these charges then being captured by a transport layer with superior mobility (we call it conduction channel). Then, a much higher photoconductive gain can be obtained. This method allows us to use separate layers for charge photogeneration and transport to enhance photoconductive gain. Many researchers have used such sensitizing method to obtain great high photosensitivity of photodetectors. [ACS Appl. Mater. Interfaces 2019, 11, 36880- 36885] [Adv. Mater. 2015, 27, 6885- 6891] [Adv. Mater. 2021, 33, 2101717]. In addition, we also added a band diagram in SI.  
+
+<|ref|>image<|/ref|><|det|>[[184, 565, 800, 787]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[177, 829, 820, 875]]<|/det|>
+Q7: Also I would expect the effective field at the interface to be not strong and solely relying on the ETL interface to extract electrons seems like an inefficient strategy.  
+
+<|ref|>text<|/ref|><|det|>[[176, 885, 819, 902]]<|/det|>
+A7: Indeed, we understand your concerns. In this kind of lateral device structure, the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[176, 88, 822, 525]]<|/det|>
+electrons extraction seems inefficient, but the core advantage of such hybrid lateral structure is the conduction channel has a great photoconductive gain, the captured carriers by conduction channel can transport rapidly in the conduction channel and reinject and recirculate swift between metal contacts. The amount of charges passing through the cross section of a conductor per unit of time is much greater. Therefore, these carriers in the conduction channel can offer much stronger photocurrent than in the X- ray- sensitive materials. This gain effect can greatly amplify the photocurrent signal with superior mobility of the conduction channel. Even the electrons extraction is inefficient, the conduction channel can amplify the photocurrent signal thousands or ten thousands of times with the photocurrent gain effect [Adv. Funct. Mater. 2020, 30, 1903907]. Thus, the finally obtained photocurrent can be much higher than the device only with photosensitive materials. What's more, the addition of ETL like \(\mathsf{C}_{60}\) can be an effective way to help extract electrons from perovskite, and this has been demonstrated in other research [ACS Appl. Mater. Interfaces 2018, 10, 50, 44144- 44151]. Our finally obtained photocurrent sensitivity still reaches high, as you can see that our detector's sensitivity can obtain \(2 \times 10^{4} \mu \text{C Gy}_{\text{air}}^{- 1} \text{cm}^{- 2}\) . Only when the DCS electrode is working does the sensitivity decrease a little.  
+
+<|ref|>text<|/ref|><|det|>[[177, 562, 821, 689]]<|/det|>
+Q8: Pg 6 "The primary functional layers of this device, perovskite and In2O3" Could the authors explain what kind of material In2O3 is why they chose it? This is also the first time they mention In2O3. Perhaps they could mention somewhere in the beginning that they used In2O3 as the conductive channel because I was under the impression that it was all perovskite.  
+
+<|ref|>text<|/ref|><|det|>[[177, 701, 821, 912]]<|/det|>
+A8: We appreciate your comments and we have added the illustration in our revised manuscript (page 7). \(\mathsf{In}_2\mathsf{O}_3\) is a kind of n- type semiconductor material which has been widely used as the channel material for thin- film transistors [Appl. Phys. Rev. 3, 021303 (2016)]. And it merits of low- temperature solution- processability. In this work, we used \(\mathsf{In}_2\mathsf{O}_3\) and perovskite as transport layer and sensitizer layer, respectively, to construct the hybrid photodetector. Owing to the superior mobility of \(\mathsf{In}_2\mathsf{O}_3\) thin film, it can accelerate the transport speed of photo- induced carriers generated in the perovskite. In addition, the trapped carriers on the interface can alter the conductivity of the \(\mathsf{In}_2\mathsf{O}_3\) channel through
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 89, 821, 218]]<|/det|>
+capacitive coupling and higher photoconductive gain can be obtained which have been well demonstrated in our prior work [Small Methods 6, 2200500 (2022)] [Adv Mater 27, 6885- 6891 (2015)]. And the hybrid structure can also be effective for perovskite/graphene and perovskite/IGZO and other different material combinations [Adv Mater 27, 41- 46 (2015)] [Adv Mater 32, e1907527 (2020)].  
+
+<|ref|>text<|/ref|><|det|>[[177, 256, 820, 301]]<|/det|>
+Q9: Pg 8. Fig 3e. Could the authors comment on why the sensitivity for low dose is higher than for high dose?  
+
+<|ref|>text<|/ref|><|det|>[[176, 311, 824, 551]]<|/det|>
+A9: Yes, we'd like to illustrate it in our revised manuscript (page 8). It is actually quite a common phenomenon in many reported perovskite X- ray detectors. As the X- ray dose rate is increased, the photocurrent demonstrates a sublinear dependence on it. This reduction in sensitivity can be explained in terms of trap states present either in \(\mathsf{In}_2\mathsf{O}_3\) or at the interface between the \(\mathsf{In}_2\mathsf{O}_3\) and the underlying \(\mathsf{SiO}_2\) layer. The photocurrent mainly comes from the photo- induced electrons filling the trap states in metal oxide and changing the metal oxide's conductivity, under high illumination intensities the density of available trap states is reduced, resulting in saturation of the photoresponse. [Nature Nanotechnology 8, 497- 501 (2013)] [Adv. Funct. Mater. 2019, 29, 1808182]  
+
+<|ref|>text<|/ref|><|det|>[[177, 618, 820, 662]]<|/det|>
+Q10: Pg. 10 "The detection limit of the DCS detector (SNR=3) is as low as 7.84 nGyair s- 1" Could the authors show a pulse for 7.84 nGyair s- 1?  
+
+<|ref|>text<|/ref|><|det|>[[176, 673, 822, 913]]<|/det|>
+A10: We understand your concerns, but we are very sorry that because the 7.84 nGyair s- 1 is extremely low, in our experiment condition and with the limitation of our equipment, we can not demonstrate a pulse for 7.84 nGyair s- 1. But you can see that our detector under 83 nGyair s- 1 (the lowest dose that we believe we can reliably measure with our ion chamber.) shows a much higher photocurrent than \(3 \times \mathsf{I}_{\mathsf{noise}}\) (152 fA), and the SNR also much higher than 3. In addition, this method to calculate the detection limit has been widely used before. [Nat Commun 12, 5258 (2021)] [Adv. Mater. 2021, 33, 2101717]. We agree with reviewers that it is not very solid to use the reverse extension line to estimate the lowest detection limit, but we have tried our best and give the pulse response at 83 nGyair s- 1 showing SNR
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 91, 261, 105]]<|/det|>
+of honestly,  
+
+<|ref|>text<|/ref|><|det|>[[177, 145, 486, 161]]<|/det|>
+Q11: Pg 12. "of dark current drafting" Typo  
+
+<|ref|>text<|/ref|><|det|>[[177, 173, 820, 216]]<|/det|>
+A11: Thank you for your careful review and mention the mistake. We have corrected it to "of dark current drifting".  
+
+<|ref|>text<|/ref|><|det|>[[177, 255, 820, 300]]<|/det|>
+Q12: Pg. 14 Fig 5d Could the authors comment on the mobility values? 1.55 cm2/V/s seems low.  
+
+<|ref|>text<|/ref|><|det|>[[177, 310, 822, 468]]<|/det|>
+A12: We understand your concerns. The \(\ln_2O_3\) possesses a high intrinsic electron concentration and high mobility features [Journal of Physics and Chemistry of Solids 38, 819- 824 (1977)] [Adv Mater 27, 7168- 7175 (2015)]. In this work, we used solution- processed method to fabricate the \(\ln_2O_3\) thin film transistor as the switching backplane for addressing the photodetector sensors. And the linear mobility of the \(\ln_2O_3\) transistor was calculated according to the transfer curve by the following equation:  
+
+<|ref|>equation<|/ref|><|det|>[[411, 475, 586, 507]]<|/det|>
+\[\mu_{lin} = \frac{dI_D}{dV_{GS}}\times \frac{L}{W}\times \frac{1}{C_{ox}\times V_{DS}}\]  
+
+<|ref|>text<|/ref|><|det|>[[177, 514, 824, 700]]<|/det|>
+However, the high contact resistance ( \(R_{contact}\) ) between the Ni/Au source/drain (S/D) electrodes and \(\ln_2O_3\) thin film will hinder the extracted mobility value based on the transfer curves [IEEE Transactions on Electron Devices 66, 5166- 5169 (2019)]. To illustrate this point, we fabricated two \(\ln_2O_3\) transistors with Ni/Au and Al source/drain electrodes, respectively, for comparison as shown in the below figure R3a and b. The extracted linear mobility of the \(\ln_2O_3\) transistor with Al S/D electrodes was \(8.23cm^2 /V\cdot s\) which is higher than that with Ni/Au S/D electrodes ( \(1.53cm^2 /V\cdot s\) ).  
+
+<|ref|>text<|/ref|><|det|>[[177, 709, 821, 811]]<|/det|>
+Although the higher \(R_{contact}\) affect the extracted mobility value, it can be conducive to reduce transistor off current ( \(I_{off}\) ) which is important for reducing the signal crosstalk among sensing units. In addition, using gold as the electrode can keep more stable device performance, because aluminum is easy to react with perovskite and cause device damage.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[198, 85, 808, 263]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 275, 818, 291]]<|/det|>
+<center>Figure R3: a, Transfer curve of \(\mathsf{In}_2\mathsf{O}_3\) transistor with Ni/Au and b, Al source/drain electrodes. </center>  
+
+<|ref|>text<|/ref|><|det|>[[178, 331, 820, 374]]<|/det|>
+Q13: Could the authors mention what the variation in the critical voltage for different devices were?  
+
+<|ref|>text<|/ref|><|det|>[[177, 385, 822, 680]]<|/det|>
+A13: Thank you very much for your advice. Yes, we'd like to mention this and have added the comments in our revised manuscript (page 12). With the results of current- voltage curves in terms of DCS electrode voltages and drain currents of same thick devices with different X- ray sensitive layer (contribute to most of the thickness of the device) and different thick devices with same X- ray sensitive material (Figure R4). The critical voltage (CV) seems will not be influenced by different compositions of materials, but will increase with greater distance with the conduction channel, which is also the thickness of X- ray sensitive material. Thus, we propose with the increase of the distance with conduction channel, the applied electric field of DCS electrode can be weaker around the conduction channel, and thus have weaker force to attract the dark electrons. It need stronger electric field to attract the dark electrons to DCS electrode and suppress the dark current to 0 A.  
+
+<|ref|>image<|/ref|><|det|>[[195, 700, 770, 867]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 878, 820, 894]]<|/det|>
+<center>Figure R4: a, Current-voltage curves of \(1\mu m\) MAPbBr3, MAPbI3 and \(\mathsf{FA}_{0.92}\mathsf{Cs}_{0.04}\mathsf{MA}_{0.04}\mathsf{PbI}_3\) on </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 90, 819, 134]]<|/det|>
+\(\mathsf{C}_{60} / \mathsf{In}_{2}\mathsf{O}_{3}\) TFT in the dark. \(\mathbf{b}\) , Current-voltage curves in terms of DCS electrode voltages and drain currents of \(80 \mu \mathrm{m}\) and \(200 \mu \mathrm{m}\) MAPbI\(_3\) (PMMA binder)/SnO\(_2 / \mathsf{In}_{2}\mathsf{O}_{3}\) TFT in the dark.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 325, 107]]<|/det|>
+## REVIEWER COMMENTS</B>  
+
+<|ref|>text<|/ref|><|det|>[[116, 129, 393, 145]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 166, 875, 341]]<|/det|>
+Peng Jin, Yingjie Tang et al. improved the quality of the manuscript substantially and fully addressed my main concern related to the thickness of the absorbing perovskite layer. In the revised manuscript the thickness has been increased two orders of magnitudes up to \(200\mu m\) which is already significant, considering e.g., X- ray absorption in mammography application. The authors demonstrated the validity of their concept of dark current reduction for thick absorbing layers (200um) in a similar way as reported earlier for thin layers (2um). The signal to noise ration clearly suffers from increasing the absorbing layer thickness, however it should not preclude publication. A trade- off between X- ray absorption, dark current and signal height must be found accurately if industrial applications are envisioned. I suggest the publication of the revised manuscript as is.  
+
+<|ref|>text<|/ref|><|det|>[[116, 421, 393, 438]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 459, 871, 517]]<|/det|>
+The authors have made a very professional and careful work in answering the questions posed by the reviewers and in amending the manuscript. I'm of the opinion that the current version is of high quality and should be therefore published in Nature Communication.  
+
+<|ref|>text<|/ref|><|det|>[[116, 557, 393, 574]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 596, 682, 613]]<|/det|>
+I am satisfied with most of the authors' answers. I have three other concerns:  
+
+<|ref|>text<|/ref|><|det|>[[114, 635, 864, 732]]<|/det|>
+First, in the revised version the authors mention "great photoconductive gain" several times. Can they get a rough estimation of the gain factor? If they have mobility and lifetime of the channel, then they can calculate this. The disadvantage of having photoconductive gain is lower speed. Can the authors estimate the pulse rise/fall time?  
+
+<|ref|>text<|/ref|><|det|>[[114, 752, 845, 829]]<|/det|>
+Second the authors mention that the perovskite photoconductive layer has "low mobility" and the In2O3 channel has "superior mobility". For In2O3 they measured a mobility of \(1.55 \text{cm}^2 /\text{V} /\text{s}\). In the literature they cite (Adv Mater 27, 7168- 7175 (2015) 8 cm2/V/s is reported. But the typical mobility values for polycrystalline perovskite reported in lit (0.1- 10 cm2/V/s) are in the same range.  
+
+<|ref|>text<|/ref|><|det|>[[114, 850, 865, 907]]<|/det|>
+My third concern is about the thickness. I understand that the novelty here is the device structure and they have not optimized the thickness, interlayers, etc however do they even need 2 microns? The authors mention that the xray generated carriers drift under a built- in electric field between X- ray
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 881, 224]]<|/det|>
+sensitive material and electron transport layer. In my opinion, the electric field at the ETL interface is not strong and Im not sure if the depletion region extends 2 microns to the top DSC electrode. I would expect the devices to be limited by charge carrier extraction. There is not much benefit of absorbing more xrays by making thicker films if they cant extract the carriers generated. Moreover, I would expect thicker devices to be worse. Since xray absorption decreases exponentially, for thicker devices most of the carriers generated close to the top surface from where the xray is illuminated might not reach the interface where extraction is more efficient.  
+
+<|ref|>text<|/ref|><|det|>[[115, 265, 293, 281]]<|/det|>
+Other minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 303, 872, 360]]<|/det|>
+Line 128 and 133 CV has been defined earlier. The authors could avoid using acronyms for terms that have not been used many times such as Wv, PCD. They are not commonly used acronyms and makes it difficult to read  
+
+<|ref|>text<|/ref|><|det|>[[115, 381, 678, 418]]<|/det|>
+Line 172 "Our PMMA layer is only \(\sim 50 \text{nm}\) , it cannot act as a dielectric layer." Thinner layers have higher capacitance than thicker layer.  
+
+<|ref|>text<|/ref|><|det|>[[115, 439, 872, 496]]<|/det|>
+Line 216 "The detection limit of the DCS detector 217 (SNR = 3) is as low as 7.84 nGyair s- 1(Fig. 4g)," The authors should mention how they obtained this number in the manuscript (used reverse extension line to estimate the lowest detection....?)  
+
+<|ref|>text<|/ref|><|det|>[[115, 517, 880, 613]]<|/det|>
+Line 242 "To preliminarily investigate its spatial resolvability in scanning- based X- ray imaging, we measured its modulation transfer function(MTF) by scanning the object (Figure S11a). The used line pair mask plate is shown in Figure S11b, the minimum resolved line pair is 5.5 lp mm- 1." Could the authors show the scanned x- ray image of the object and include it in SI? Was this a single pixel device? MTF for the detector array would be more relevant. Did they measure it?  
+
+<|ref|>text<|/ref|><|det|>[[115, 635, 410, 652]]<|/det|>
+Line 255 "of dark current drafting" Typo  
+
+<|ref|>text<|/ref|><|det|>[[115, 673, 876, 770]]<|/det|>
+Figure still have a lot of text in small font. They could put only the most relevant/important text in the figure and the description could go in the caption or in the main text. For eg, they could remove or reduce in fig 2. direction of dark electron motion, photoconductive gain. In fig 4 almost no shift of dark current drifting Fig 3a text too small
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 103, 820, 138]]<|/det|>
+As for the gain factor, the total gain produced by Heterojunction X- ray Phototransistors can be calculated as [Adv. Mater. 2021, 33, 2101717] [Adv. Funct. Mater. 2019, 29, 1900234]:  
+
+<|ref|>equation<|/ref|><|det|>[[432, 142, 564, 158]]<|/det|>
+\[G = I_{\mathrm{s}}E_{\mathrm{e - h}} / \epsilon Dm e\]  
+
+<|ref|>text<|/ref|><|det|>[[177, 159, 821, 288]]<|/det|>
+Where \(I_{s}\) is the X- ray signal current (4 nA), \(E_{\mathrm{e - h}}\) is the EHP creation energy given by an empirical mode, [Nucl. Instrum. Methods Phys. Res., Sect. A 2006, 565, 637. ] \(\mathrm{E}_{\mathrm{e - h}} = 1.43\) \(+2E_{\mathrm{g}}\) ( \(\mathrm{E}_{\mathrm{g}}\) is the bandgap of perovskite, 1.55 eV), \(\epsilon\) is the fraction of absorbed photons (0.1 at \(40\mathrm{keV}\) X- ray), \(D\) is the dose rate (233 \(\mu \mathrm{Gy} / \mathrm{air} \mathrm{s}^{- 1}\) ), m is the mass of perovskite ( \(2.4 \times 10^{- 8}\) g), and \(\epsilon\) is the elementary charge. Note that the total gain is composed of impact ionization gain and photoconductive gain, in which the former is determined by the ratio of X- ray photon energy \(\mathrm{E}_{\mathrm{ph}}\) to \(\mathrm{E}_{\mathrm{e - h}}\) .  
+
+<|ref|>text<|/ref|><|det|>[[177, 308, 820, 342]]<|/det|>
+Under the \(40\mathrm{keV}\) X- ray. The overall gain factor (including ionization and photoconductive) is calculated as \(\approx 3.24 \times 10^{- 4}\) .  
+
+<|ref|>text<|/ref|><|det|>[[177, 364, 820, 398]]<|/det|>
+If using the mobility and lifetime of the channel to calculate the photoconductive gain, the gain factor can be calculated as:  
+
+<|ref|>equation<|/ref|><|det|>[[434, 401, 562, 432]]<|/det|>
+\[G_{P} = \frac{t_{r}}{t_{t}} = \frac{t_{r}}{L^{2}}\mu V_{\mathrm{DS}}\]  
+
+<|ref|>text<|/ref|><|det|>[[177, 437, 821, 510]]<|/det|>
+where \(\mu\) is mobility (1.55 cm2/V s) and \(\mathrm{V}_{\mathrm{DS}}\) is applied drain- source voltage (0.5 V), L is the channel length (20 \(\mu \mathrm{m}\) ) and \(\mathrm{t}_{r}\) is the carrier lifetime of the channel (2.041 ms). The carrier lifetime of the channel was measured by time- resolution photoluminescence spectrum (TRPL):  
+
+<|ref|>image<|/ref|><|det|>[[315, 536, 651, 742]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 752, 762, 768]]<|/det|>
+<center>Figure R1. TRPL of \(\mathrm{In}_{2}\mathrm{O}_{3}\) film. Its average carrier lifetime is calculated as 2.041 ms. </center>  
+
+<|ref|>text<|/ref|><|det|>[[177, 789, 821, 898]]<|/det|>
+As you can see, the metal oxide's lifetime is relatively long. The unique oxygen- sensitized photoconduction mechanism has allowed the photoconductive gain of metal oxide to reach an extremely high level, which is several orders of magnitude higher than the conventional thin film detectors. According to the widely adopted photoconduction model, the long carrier lifetime of metal oxide is proposed to be the origin of high- gain transport. [Nanoscale, 2013, 5, 6867- 6873] However, the very long carrier lifetime of metal oxide has aroused some
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 85, 820, 139]]<|/det|>
+unpleasant characteristics, such as slow response time, especially the persistence photocurrent (PPC) effect, which has been frequently founded in metal oxide- based phototransistors. [Adv. Mater. 2015, 27, 6885- 6891]  
+
+<|ref|>text<|/ref|><|det|>[[178, 159, 820, 231]]<|/det|>
+The Gain factor is calculated as \(\approx 400\) . As for the perovskite, its lifetime is estimated as 100 ns the Gain factor of \(2\mu m\) perovskite photoconductive detector (under the same applied electric field) with the same mobility is calculated as \(\approx 0.2\) , which is much smaller than our type of devices.  
+
+<|ref|>text<|/ref|><|det|>[[178, 252, 821, 362]]<|/det|>
+As for your next concern, we totally agree with your words that "The disadvantage of having photoconductive gain is lower speed." The slower recombination of the carriers that took part in the transportation in the channel, the higher gain. Thus, the high gain and fast speed generally cannot be obtained simultaneously. We measured the rise and fall time of our device under the laser beam pulse. The DCS electrode is applied with CV. The rise and fall time are 23 ms and 31 ms, respectively:  
+
+<|ref|>image<|/ref|><|det|>[[328, 377, 647, 593]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[177, 604, 820, 639]]<|/det|>
+<center>Figure R2. Laser beam pulse response of the device. The rise and fall time are 23 ms and 31 ms, respectively. </center>  
+
+<|ref|>text<|/ref|><|det|>[[177, 659, 822, 879]]<|/det|>
+Such a response speed can meet the general \(30 \text{Hz}\) dynamic monitoring application. But as for X- ray detection, the commercially used X- ray detectors, such as a- Se detector, are fabricated in a very thick geometry, to obtain stronger absorption to X- ray. But such a thick film will definitely sacrifice the response speed to the X- ray. And in most practical cases, a- Se detector can only be used in static X- ray imaging. Even for the potential dynamic imaging applications, \(30 \text{Hz}\) can be satisfied in most situations. In our opinion, at least at this moment, for the direct perovskite X- ray detector, the dynamic response is not the first priority, since the competing technology (a- Se) is not used for dynamic X- ray imaging as well. Additionally, for those reported perovskite X- ray detectors with typical vertical geometry, the response time is also not much fast than our devices [Nature 550, 87- 91 (2017)] [Nat. Electron. 4, 681- 688 (2021)], and I believe there are also considerable photoconductive gains due to the traps of perovskite films.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 103, 821, 213]]<|/det|>
+But as for the dark current, it directly refers to whether the detector can or cannot be used no matter if it is static or dynamic imaging applications. High dark current can quickly fill up the storage capacitance of TFT or CMOS pixels prior to X- ray illumination, if the capacitance is filled with charges at the dark, there won't be any response from the detector when the X- ray comes in. Thus, compared with the response speed, the high dark current issue is much more urgent to be solved, at least in this early stage of perovskite X- ray detectors  
+
+<|ref|>text<|/ref|><|det|>[[177, 233, 825, 323]]<|/det|>
+Q2: Second the authors mention that the perovskite photoconductive layer has "low mobility" and the In2O3 channel has "superior mobility". For In2O3 they measured a mobility of 1.55 cm2/V/s. In the literature they cite (Adv Mater 27, 7168- 7175 (2015) 8 cm2/V/s is reported. But the typical mobility values for polycrystalline perovskite reported in lit (0.1- 10 cm2/V/s) are in the same range.  
+
+<|ref|>text<|/ref|><|det|>[[177, 326, 821, 583]]<|/det|>
+A2: We totally understand your concerns. We agree with your opinion about the mobility between \(\mathsf{In}_2\mathsf{O}_3\) and Perovskite is quite reasonable. Thus, we revised our manuscript that the perovskite has "lower mobility", and the conduction channel, such as \(\mathsf{In}_2\mathsf{O}_3\) , has "higher mobility". In fact, realizing the 3D polycrystalline perovskite with such high carrier mobility as our solution- processed polycrystalline \(\mathsf{In}_2\mathsf{O}_3\) is still a very challenging work. [Science Advances, 7, 18, (2021)] The major difficulty is the ion migration, which causes a partial screening of the applied field, yielding a very low room temperature \(\mu_{\mathrm{FET}}\) of \(10^{- 4} \mathrm{cm}^2 /\mathrm{Vs}\) in thin films of perovskite, such as MAPbI3. [Nat. Commun. 6, 7383 (2015).] Even in single- crystal perovskite- based devices, strong hysteresis and moderate mobilities of \(< 10^{- 3} \mathrm{cm}^2 /\mathrm{Vs}\) were reported at room temperature. [Nat. Commun. 7, 11330 (2016).] Various instabilities, such as the mechanism of carrier scattering and trapping, the role of ion migration, the origin of hysteresis in the device characteristics, and the electronic structure of grain boundaries in perovskites have made them reveal unusually lower mobility than in the ideal state.  
+
+<|ref|>text<|/ref|><|det|>[[177, 603, 821, 768]]<|/det|>
+But as for the \(\mathsf{In}_2\mathsf{O}_3\) , their mobility is very high, and stable and is very easy to be reproduced, the calculated electron mobility \(\mu_e\) [cm2/(Vs)] can be up to 270 - 274, the experiment value can be 7.81 - 190. [Applied Physics Reviews 9, 011315 (2022)] They can easily achieve high mobility in various fabrication processes, such as ALD [J. Phys. Chem. C 115, 15384- 15389 (2011a)] (84 cm2/(Vs)), spin- coating, PLD [Appl. Phys. Lett. 62, 2332- 2334 (1993)] (50 cm2/(Vs)), MOCVD [Vacuum 167, 1- 5 (2019)] (42 cm2/(Vs)), spray pyrolysis [J. Cryst. Growth 240, 142- 151 (2002)] (42.6 cm2/(Vs)), and dc magnetron sputtering. The \(\mathsf{In}_2\mathsf{O}_3\) film sputtered at room temperature without post- annealing results in layers with reasonably high mobility of \(51.3 \mathrm{cm}^2 /\mathrm{Vs}\) . [Nat. Commun. 6, 8932 (2015)]  
+
+<|ref|>text<|/ref|><|det|>[[177, 789, 820, 897]]<|/det|>
+The \(\mathsf{In}_2\mathsf{O}_3\) fabricated by spin- coating might have the relatively low mobility, but such a time- saving and low- cost method can be widely used in low- cost large- area applications. What's more, the electron mobility of our \(\mathsf{In}_2\mathsf{O}_3\) transistor with aluminum Source/Drain electrodes was \(8.23 \mathrm{cm}^2 /\mathrm{V}\cdot \mathrm{s}\) which is higher than that with Ni/Au Source/Drain electrodes (1.55 \(\mathrm{cm}^2 /\mathrm{V}\cdot \mathrm{s}\) ). But unfortunately, the perovskite will have a chemical reaction with aluminum, which may deteriorate the device's stability and performance. We have to use Ni/Au
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 85, 800, 101]]<|/det|>
+Source/Drain electrodes as sacrificing the mobility of \(\mathsf{In}_2\mathsf{O}_3\) to gain stability of the device.  
+
+<|ref|>text<|/ref|><|det|>[[177, 140, 821, 342]]<|/det|>
+Q3: My third concern is about the thickness. I understand that the novelty here is the device structure and they have not optimized the thickness, interlayers, etc however do they even need 2 microns? The authors mention that the xray generated carriers drift under a built- in electric field between X- ray sensitive material and electron transport layer. In my opinion, the electric field at the ETL interface is not strong and Im not sure if the depletion region extends 2 microns to the top DSC electrode. I would expect the devices to be limited by charge carrier extraction. There is not much benefit of absorbing more xrays by making thicker films if they cant extract the carriers generated. Moreover, I would expect thicker devices to be worse. Since xray absorption decreases exponentially, for thicker devices most of the carriers generated close to the top surface from where the xray is illuminated might not reach the interface where extraction is more efficient.  
+
+<|ref|>text<|/ref|><|det|>[[177, 344, 822, 453]]<|/det|>
+A3: Genius perspectives! We appreciate your meticulous reading of our manuscript and your smart views. In perovskite, most of the carriers are indeed generated close to the top surface, if the electric field at the ETL interface is not strong enough to extract the surface photoinduced electrons in perovskite, the thicker geometry might result in poor sensitivity. To figure out this question, we made devices with \(2\mu m\) , \(1\mu m\) and \(500 \text{nm}\) perovskite, the devices were fabricated in the same conditions.  
+
+<|ref|>text<|/ref|><|det|>[[177, 474, 821, 675]]<|/det|>
+We measured their sensitivities when their DCS electrode is applied with CV, and found that \(2\mu m\) device has the highest sensitivity \((7560\mu \mathrm{Gy}_{\mathrm{air}}^{- 1}\mathrm{cm}^{- 2})\) , the \(1\mu m\) device owns a sensitivity of \(4060\mu \mathrm{Gy}_{\mathrm{air}}^{- 1}\mathrm{cm}^{- 2}\) , the \(500 \text{nm}\) device reveals the lowest \((2580\mu \mathrm{Gy}_{\mathrm{air}}^{- 1}\mathrm{cm}^{- 2})\) (Figure R3). The results seem to point out that in such a relatively thin geometry, the device with thicker perovskite film performs better in collecting photo- induced electrons, the electric field at the ETL interface might strong enough to extract the surface photo- induced electrons in perovskite. Again the relatively high sensitivity should be partially related to the photoconductive gains. Personally, when we talk about the word "extract" we are actually talking about the "drift", but we should also not forget about the "diffusion" especially given the high diffusion length of perovskite crystals. As the drifted charges are collected, it also gives additional driving force for diffusion.  
+
+<|ref|>text<|/ref|><|det|>[[177, 696, 821, 897]]<|/det|>
+We definitely believe your perspective is generally right, thicker films can generate more photo- induced electrons, thin devices might have better capacity in extracting the majority of the photo- induced electrons at the top of the perovskite. There must be a trade- off between the thickness of perovskite and the device's performance, similar to what Reviewer #1 said, "A trade- off between X- ray absorption, dark current and signal height must be found accurately if industrial applications are envisioned". The \(2\mu m\) perovskite in such a device structure might be a relatively thin geometry and the electric field at the ETL interface seems able to extract the photo- induced electrons efficiently. But based on our laboratory experiment condition, we are very sorry we cannot demonstrate the performance of the device with high- quality pure perovskite film thicker than \(2\mu m\) (we have shown the thick perovskite- PMMA composite films in the last round revision), because the precursor solution
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 85, 821, 157]]<|/det|>
+of perovskite we used is already saturated. We hope the other researchers who have expertise in fabricating thick perovskite films can solve this optimization question in the future. We have added the comments in our revised manuscript (page 12), and thank you very much for your constructive advice.  
+
+<|ref|>image<|/ref|><|det|>[[333, 181, 636, 375]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[290, 381, 704, 397]]<|/det|>
+<center>Figure R3. Sensitivity of devices with different thicknesses. </center>  
+
+<|ref|>text<|/ref|><|det|>[[178, 457, 347, 471]]<|/det|>
+Other minor comments:  
+
+<|ref|>text<|/ref|><|det|>[[177, 492, 821, 545]]<|/det|>
+Q4: Line 128 and 133 CV has been defined earlier. The authors could avoid using acronyms for terms that have not been used many times such as WV, PCD. They are not commonly used acronyms and makes it difficult to read.  
+
+<|ref|>text<|/ref|><|det|>[[177, 548, 820, 582]]<|/det|>
+A4: Thank you very much for your careful review, we have deleted the WV and PCD in the manuscript, and added an annotation in the caption.  
+
+<|ref|>text<|/ref|><|det|>[[177, 621, 750, 656]]<|/det|>
+Q5: Line 172 "Our PMMA layer is only \(\sim 50 \text{nm}\) , it cannot act as a dielectric layer." Thinner layers have higher capacitance than thicker layer.  
+
+<|ref|>text<|/ref|><|det|>[[177, 659, 821, 749]]<|/det|>
+A5: Thank you very much for your notification. The PMMA layer is not insulative but may introduce a capacitance between the DCS electrode and perovskite. From the device's output signal, the capacitance seems will not affect the final signal, which is collected in the Drain electrode on the other side of the perovskite. We have added some comments on the capacitance of the PMMA in our manuscript (page 9).  
+
+<|ref|>text<|/ref|><|det|>[[177, 789, 820, 823]]<|/det|>
+Q6: Line 216 "The detection limit of the DCS detector 217 (SNR = 3) is as low as 7.84 nGyair s- 1(Fig. 4g),"  
+
+<|ref|>text<|/ref|><|det|>[[177, 826, 820, 860]]<|/det|>
+The authors should mention how they obtained this number in the manuscript (used reverse extension line to estimate the lowest detection....?)  
+
+<|ref|>text<|/ref|><|det|>[[177, 863, 821, 897]]<|/det|>
+A6: Thank you for your suggestions. We have added a comment" The lowest detection limit was estimated by the reverse extension line to where the \(\mathsf{SNR} = 3\) " in the revised manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[178, 86, 250, 101]]<|/det|>
+(page 15)  
+
+<|ref|>text<|/ref|><|det|>[[178, 122, 820, 194]]<|/det|>
+Q7: Line 242 "To preliminarily investigate its spatial resolvability in scanning- based X- ray imaging, we measured its modulation transfer function(MTF) by scanning the object (Figure S11a). The used line pair mask plate is shown in Figure S11b, the minimum resolved line pair is \(5.5 \text{lp mm}^{- 1}\) ."  
+
+<|ref|>text<|/ref|><|det|>[[178, 197, 820, 249]]<|/det|>
+Could the authors show the scanned x- ray image of the object and include it in SI? Was this a single pixel device? MTF for the detector array would be more relevant. Did they measure it?  
+
+<|ref|>text<|/ref|><|det|>[[177, 252, 821, 529]]<|/det|>
+A7: Thank you for your suggestions. Yes, we calculated the MTF with a single- pixel device, we preliminarily calculated the MTF by scanning the object with a stepping motor, and recording the current value of a single- pixel device. We added this comment in our revised manuscript to avoid misleading the readers (page 12, 13). Indeed, the MTF was sometimes calculated by scanning the whole X- ray image of the object, but as you can see, every scanning line to this object is the same, when calculating the MTF, only need to scan one of the lines can preliminarily calculate the MTF. As for MTF for our imaging array, we are really sorry that we didn't measure it. Even the world's most state- of- the- art perovskite imaging array fabricated by Samsung \((1,428 \times 1,428 \text{ pixels in } 10 \text{ cm} \times 10 \text{ cm})\) with a pixel pitch of \(70 \mu \text{m}\) [Nature 550, 87- 91 (2017)], their MTF for the detector array was just \(3.1 \text{ lp mm}^{- 1}\) . In our preliminarily demonstrated detector array, there are only \(64 \times 64 \text{ pixels in } 2 \text{cm} \times 2 \text{cm}\) with a pixel pitch of \(300 \mu \text{m}\) . Thus, MTF for our detector array can be negligible. We aim to demonstrate a prototype to reveal that this method can be successfully used in an imaging array and hope to promote the method rather than the detailed performance factors of the array.  
+
+<|ref|>text<|/ref|><|det|>[[178, 566, 491, 581]]<|/det|>
+Q8: Line 255 "of dark current drafting" Typo  
+
+<|ref|>text<|/ref|><|det|>[[177, 585, 818, 601]]<|/det|>
+A8: Thank you for your notification. We have revised the "drafting" to "drifting" in Line 255.  
+
+<|ref|>text<|/ref|><|det|>[[178, 640, 820, 692]]<|/det|>
+Q9: Figure still have a lot of text in small font. They could put only the most relevant/important text in the figure and the description could go in the caption or in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[178, 696, 820, 730]]<|/det|>
+For eg, they could remove or reduce in fig 2. direction of dark electron motion, photoconductive gain. In fig 4 almost no shift of dark current drifting  
+
+<|ref|>text<|/ref|><|det|>[[178, 733, 325, 747]]<|/det|>
+Fig 3a text too small  
+
+<|ref|>text<|/ref|><|det|>[[178, 751, 820, 785]]<|/det|>
+A9: Thank you for your advice. We have improved the small font and deleted some irrelevant texts in the figure.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a36fd9cfce3a65fdcd8e373d981fd6fd59760913c0bc25c5198938cae748b313/images_list.json

@@ -0,0 +1,111 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "supplementary information.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Response letter figure 2.4",
+    "footnote": [],
+    "bbox": [
+      [
+        265,
+        100,
+        872,
+        333
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Response letter figure 3.1",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Response letter figure 4.2",
+    "footnote": [],
+    "bbox": [
+      [
+        270,
+        710,
+        780,
+        824
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Response letter figure 5.1",
+    "footnote": [],
+    "bbox": [
+      [
+        272,
+        170,
+        732,
+        440
+      ]
+    ],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_9.jpg",
+    "caption": "Response letter figure 9.2",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 11
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_12.jpg",
+    "caption": "Response letter figure 12.2",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "14 The bulk RNA-seq t-test comparison essentially assumes that MC9 is roughly the same size in all patients. A bulk analysis that takes into account the size of MC9 in each patient might be more sensitive at detecting genes of interest.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_20.jpg",
+    "caption": "Supplementary Figure 20",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_19.jpg",
+    "caption": "Supplementary figure 19",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Supplementary_Figure_5.jpg",
+    "caption": "Supplementary Figure 5",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 15
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a36fd9cfce3a65fdcd8e373d981fd6fd59760913c0bc25c5198938cae748b313/supplementary_0_Peer Review File__a36fd9cfce3a65fdcd8e373d981fd6fd59760913c0bc25c5198938cae748b313.mmd

@@ -0,0 +1,730 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Early response evaluation by single cell signaling profiling in acute myeloid leukemia  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+In this manuscript, the authors used mass cytometry to measure the abundance of 36 extracellular and intracellular proteins in the peripheral blood of 32 AML patients at 0, 4, and 24 hours after treatment with \(7 + 3\) chemotherapy. Using LASSO Cox regression, p- ERK1/2 and p- p38 abundances in one myeloid cell population (called here "MC9") at one time point (24 hr) were found to be significantly correlated with survival. This paper is well written and clear. The overall result is statistically significant and interesting, and suggests that this proteomic approach may become very important to the field. That result does not appear to be extremely robust given the arbitrary nature of the clustering analysis, and the lack of correlation with chemotherapy response, ELN risk status, transplantation status, or mutational status. As the authors note, this may simply be due to the size of the cohort, which is small relative to the heterogeneity and complexity of this disease. The authors also demonstrated that p- ERK1/2 and p- p38 abundance in flow- sorted blasts is correlated with survival, which strengthens the conclusions.  
+
+Specific comments:  
+
+- MC9 is surprisingly heterogeneous with respect to CD34 expression. Because clustering is arbitrary, and clusters do not represent an absolute "ground truth," this heterogeneity warrants further investigation – for instance, given the heterogeneity of this important marker, can the authors justify calling this one cluster? In general, MC9 should be better characterized, for example, by highlighting each patient on the tSNE plot, and by showing the expression of each analyte, particularly CD34 and p-ERK1/2, on the tSNE plot, so that we can better understand heterogeneity within that cluster.- What proportion of malignant cells does MC9 represent in each patient? Does this proportion correlate with survival? Does change in the size of MC9 or time relate to survival? Does this depend on the extent of CD34 expression within MC9? One limitation of the proteomic approach is that it is difficult to know exactly which cells are malignant, because no genetic/mutational information is available, although sorting and targeted sequencing for somatic mutations could be performed to address this further.  
+
+- Further discussion of the specific relevance of MC9 to survival would be helpful. For example, why is protein expression in this cluster more relevant to survival than, say, the size or characteristics of the blast cluster, MC1? Is change in the relative size of MC1 and MC9 (over time) correlated with survival?  
+
+- High and low p-ERK1/2 patients were defined relative to each other, depending on whether they were above or below the median in this cohort. Is there an absolute protein abundance that can be used to determine whether new patients are "high" or "low?" If not, how would new patients be classified with respect to p-ERK1/2 abundance? The result would be stronger if there were an additional validation cohort that could be independently classified with respect to p-ERK1/2 abd p-38 abundance.  
+
+- Mutation correlations were very anecdotal. Obviously, this analysis would benefit from more patients and exome or WGS sequencing.  
+
+- The confirmatory flow cytometry analysis on pp. 14-15 actually gave a more significant result than the original mass cytometry clustering analysis. This strengthens the overall claim of the paper, but raises some interesting questions. How do the immature cells defined by this blast gate correspond to the FlowSOM clusters shown earlier? Are these flow-gated blasts in MC9 or MC1? The authors could reanalyze the cytometry data in a manner analogous to the flow sorting – this might strengthen the result, or would at least add clarity by connecting these two analyses to each other.  
+
+- For the bulk RNA-seq analysis, a more unbiased analysis could be useful. What other genes were differentially expressed between the p-ERK1/2 low and high patients? Moreover, if the samples are hierarchically clustered based on the bulk RNA-seq data alone, do they naturally cluster into the same groups (p-ERK1/2 low and high)?  
+
+- The bulk RNA-seq t-test comparison essentially assumes that MC9 is roughly the same size in all patients. A bulk analysis that takes into account the size of MC9 in each patient might be more sensitive at detecting genes of interest.
+
+<--- Page Split --->
+
+- The drug sensitivity matrix should also be hierarchically clustered by patient and compared to the proteomic data to discover other potential correlates of drug sensitivity in an unbiased fashion.- The authors repeatedly mention that their proteomic data indicates the presence of "clonal heterogeneity," but this is traditionally defined by genetics, not proteomics. For instance, clustering of somatic mutations by variant allele frequency would reveal clonal heterogeneity.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+The authors used 36- dimensional mass cytometry data to identify genes sensitive to the chemotherapy response. These were further validated using RNAseq and mass spec. The reason of this study is well explained and further validated using DSRT.  
+
+1. The tSNE plot in Figure 1 can be colored with ID to see if clustering is not influenced by batch effect.  
+
+2. Many clusters including MC1,2,3 are scattered. 1) Isn't it because of batch effect? coloring based on patient ID will answer. 2) Why they are scattered? Is it due to the assignment to the existing clusters? How then they can be assigned to the same cluster? For instance, MC1 is scattered into to clusters? How these are defined as MC1?  
+
+3. MC9 has been extensively used for the analysis. Why MC9 (the myeloid cluster) could become most informative cluster among other MCs? Any speculation on it?  
+
+4. Besides 24h-p-ERK1/2, p-p38 at 24, p-Rb at 24h were reported. The combination of several markers could lead a better performance. This has to be tested and discussed. In the same way, can the information obtained from RNAseq and Mass spec be used to in designing combination of markers?  
+
+Reviewer #3:  
+
+Remarks to the Author:  
+
+In the present manuscript the authors assessed the correlation between single cell signaling and clinical response to induction chemotherapy in 32 patients with AML. Samples were collected at diagnosis and during the course of induction chemotherapy, and analyzed by mass cytometry, RNAseq and mass spectrometry proteomics in order to assess 1/ the chemotherapy- induced phospho- signaling modulations in the myeloid compartment and 2/ the prognostic significance of phospho- protein expression levels in myeloid cells 24h after treatment initiation. Overall, the manuscript is well written and the mass cytometry data are presented in a clear way. However some issues detailed below prevent publication of this manuscript in its present form.  
+
+Graphical abstract :  
+
+1/ The legend is constructed as a standard legend; this legend should rather focus on the summary of the key points and the main conclusions of the article.  
+
+2/ The risk stratification by bone marrow tumor load is mentioned in the discussion section, but has not been displayed in the results section or in the supplementary data. The overall survival according to the ELN classification only appears in the graphical abstract as well, and should be integrated to the body of the article.  
+
+Results section  
+
+1/ Identification of an independent prognostic factor in AML should be confirmed using a Cox
+
+<--- Page Split --->
+
+regression including the validated prognostic factors in order to exclude potential confounding factors. In the present work, this Cox regression should include 24h- pERK1/2, age, ELN, WBC count at diagnosis, as well as allogeneic stem cell transplantation, ideally as a time- dependent covariate. Of note, the poor survival in the ELN favorable group, especially taking into account the fact that most patients are young patients, underlies that there might be a bias in the present cohort and probably precludes proper Cox regression for a formal identification of a potential prognostic factor.  
+
+2/ The MRD status is a cornerstone for prognostication of AML. An analysis with respect to the MDR would therefore be essential before concluding about the potential usefulness of measuring pERK1/2 to refine the therapeutic decision algorithm in AML.  
+
+3/ The authors claim in the abstract that they questioned whether the signaling response to therapy was more informative than analysis at time of diagnosis.  
+
+In fig 2d, the basal level of pERK1/2 seems low in the low- 24h- pERK1/2 group and high in the high- 24h- pERK1/2 group. Did the authors assess the impact of baseline expression in the metacluster 9 on clinical outcome for a direct comparison? This information is important in order to appreciate the improvement of the quality of prediction using samples drawn during induction chemotherapy rather than baseline samples.  
+
+4/ Why did the authors choose to include patients with MDS and B- ALL in this manuscript? At least the authors should retain only the 32 patients included in the study in the patient characteristics table.  
+
+5/ In supplementary fig 3, the cluster of NK cells is located at 2 different places on the tSNE plot. Some cells that were annotated as NK cells clusterize in the region of myeloid cells, and might be leukemic blasts that express CD56. This would also explain the low CD45 mean expression of this cluster compared with CD45 expression in the clusters of CD4 and CD8 T cells. The authors should confirm by manual gating the exact nature of these cells.  
+
+The authors should provide expression of the different markers projected on the tSNE plots in supplementary 3 in order to enable appreciation of the homogeneity of the different clusters. Minor: in supplementary fig 3, the lack of heatmap annotation in the lower panel makes the figure difficult to read.  
+
+6/ In Fig 3, the authors detail the manual gating strategy of leukemic blasts in order to confirm the results obtained with machine learning algorithms and assess the transferability of the results in a routine setting. However, the metacluster used to generate the initial survival curves according to pERK1/2 expression was MC9, which was considered as a myeloid cluster, by contrast with the MC1 and MC2 that were considered as the AML blast clusters (line 198). Could the authors clarify this discrepancy in the selection of the cell population they chose for manual confirmation of the results?  
+
+7/ In fig 6, results are presented for 16 and 9 drugs in the low 24h - p- ERK1/2 group and the high 24h - p- ERK1/2 group, respectively. The lack of consistence between panel b and c makes it difficult to compare the graphs. In addition, the drugs that are displayed in figure 6 should be listed in the materials and methods section ("MEK inhibitor", "DNA inhibitor" or "anti- oxidant" being too evasive). The barplots should be annotated so as the drug class appears as "Protein X inhibitor" rather than "Protein X".  
+
+The aim of this figure is to correlate the intracellular signaling profiles detected by mass cytometry with ex vivo drug sensitivity (line 573). The way the results are presented does not enable a direct comparison between the two groups of patients.  
+
+In supplementary figure 13, several drugs are designated as "MEK inhibitor", "BCL2 inhibitor", etc; in figure 6, do the names of the therapeutic classes refer to the same drugs in the upper and the lower panel?  
+
+Overall, the conclusions of the authors related to the results of fig 6 are not supported by the results, notably their conclusions regarding the RAS mutational status and the benefit from MEK inhibitors
+
+<--- Page Split --->
+
+(line 608). The low number of patients included in this part of the work as well as the high heterogeneity of patients in terms of mutational status precludes any solid conclusion based on these results.  
+
+Minor: In supplementary figure 13, the scale should be identical for all patients.  
+
+## Supplementary material  
+
+1/ A 2D plot of phosphoprotein expression by group (in particular p- ERK1/2 in the high- vs the low- 24h- p- ERK1/2 group) is lacking in the supplementary data. This would enable to appreciate the quality of the staining, and would enable the readers to be more confident with the data currently presented as arcsinh transformed 90th percentile of p- ERK1/2.  
+
+2/ In supplementary table 1, the 24h- pERK1/2 status would be relevant information to include, which would enable to appreciate the repartition of the confounding factors (age, ELN, WBC count, and allogeneic stem cell transplantation and MRD) according to pERK1/2 expression.  
+
+## Conclusion  
+
+The authors claim that "early single cell signaling response to chemotherapy provided precise prognostic information independent of stratification by genetics". Taking into account the limitations of this work ie the small sample size \((N = 32)\) , the absence of effect on survival of validated prediction tools in this cohort (ELN), as well as the absence of comparison with the prediction based on the MRD, the authors may moderate this conclusion.  
+
+Minor comments and typos Line 115: and overall poor overall survival Line 618: the 18 patients that was analyzed  
+
+Reviewer #4:  
+
+Remarks to the Author:  
+
+The manuscript by Tislevoll et al. demonstrates an elegant new approach to predicting patient outcome based on molecular changes detected early in the treatment phase, rather than relying on genetic parameters only. The potential impact of such discoveries is vast, as it allows much more efficient treatment strategies to be determined and/or altered along the way if needed, thereby saving actual lives. Therefore I commend the authors for undertaking this work, which I expect will inspire many follow- up studies pursuing similar early treatment response markers across a range of diseases.  
+
+The manuscript is well written, the analyses conducted were thorough and nicely presented, and the combination of a range of technologies (and modalities) significantly strengthened the work as well in my opinion. The authors demonstrate appropriate use of the various technologies, and I especially liked the barcoding and reference sample spike in approach for their CyTOF analyses. Seeing a combination of single cell proteomics (on more than 35 million individual cells!), bulk proteomics and RNAseq, and targeted sequencing, in a clinical setting and all integrated through machine learning, is not a mean feat, and the results presented speak for themselves.  
+
+Thus, in my opinion, the work should be accepted for publication, and I have only a few minor comments laid out below:  
+
+1. On pg. 7, the authors declare the cells in MC1 and MC2 to be CD34+ blasts due to their high expression of CD34 and CD117. As these are also classical stem cell markers, could the authors clarify
+
+<--- Page Split --->
+
+why they are deemed differentiated blasts rather than a more primitive cell type?  
+
+2. Fig2b - is this average expression across the single cells measured within cluster MC9? Was any sub-segregation possible to determine the critical cell type in which response needs to be measured? In other words, given the single cell nature of their analyses, was there anything more specifically known about which cell types had high or low ERK? Or is it truly a homogeneous cell effect?  
+
+3. Regarding the RNAseq vs MS (Super-SILAC) results - > were none of the RNA observations confirmed in proteomics? I.e. were those candidates not detected in MS or did they show different results? A bit more discussion on the value of bulk MS vs the single cell CyTOF would also help strengthen the reasoning for including both in the study, and help follow-up work to decide for one or the other vs both.  
+
+4. I would like to request more detail on the in vitro drug screening experiments. E.g. what media was used, any growth factors or stromal cells, etc). Can they also comment on overall viability of the patient samples once put in vitro? As primary AML is notoriously difficult to culture in vitro while maintaining their hierarchical nature, it would be nice have this explored in a bit more detail. Their results were very encouraging, and could suggest potential alternative treatment strategies for those patients not responding to standard-of-care.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1  
+
+## expertise in AML genomics (Remarks to the Author):  
+
+In this manuscript, the authors used mass cytometry to measure the abundance of 36 extracellular and intracellular proteins in the peripheral blood of 32 AML patients at 0, 4, and 24 hours after treatment with 7+3 chemotherapy. Using LASSO Cox regression, p- ERK1/2 and p- p38 abundances in one myeloid cell population (called here "MC9") at one time point (24 hr) were found to be significantly correlated with survival. This paper is well written and clear. The overall result is statistically significant and interesting, and suggests that this proteomic approach may become very important to the field. That result does not appear to be extremely robust given the arbitrary nature of the clustering analysis, and the lack of correlation with chemotherapy response, ELN risk status, transplantation status, or mutational status. As the authors note, this may simply be due to the size of the cohort, which is small relative to the heterogeneity and complexity of this disease. The authors also demonstrated that p- ERK1/2 and p- p38 abundance in flow- sorted blasts is correlated with survival, which strengthens the conclusions.  
+
+Specific comments:  
+
+1 MC9 is surprisingly heterogeneous with respect to CD34 expression. Because clustering is arbitrary, and clusters do not represent an absolute "ground truth," this heterogeneity warrants further investigation for instance, given the heterogeneity of this important marker, can the authors justify calling this one cluster?  
+
+1.1 This is an important point and should always be a concern when analyzing CyTOF data. The different clustering algorithms available will assess the combination of surface markers and their expression levels to cluster similar cells together. The clustering will always depend on the combination of antibodies in the panel, and as the antibody number is limited, clusters will always be more or less heterogeneous also in respect to expression of surface proteins not included in the
+
+<--- Page Split --->
+
+panel. Furthermore, most algorithms are affected by manual input of different analysis parameters. For example, FlowSOM uses a manually set number of output clusters. Thus, a prerequisite for applying this algorithm is to have extensive biological knowledge of the sample analyzed, so one can assess approximately how many clusters one would expect to get out with the specific panel used. However, the heterogeneity of AML makes it difficult to decide on a specific number of clusters. One could choose a high number of clusters, which we have experimented with. This resulted in a high number of unique clusters represented by few patients, which is a challenge when performing cross- sectional analyses of an entire cohort. However, what we realized when experimenting with cluster number was that the (presumably) healthy cell subsets in the AML samples (i.e. lymphoid cells) always ended up in quite homogeneous clusters, whereas the myeloid cells formed heterogeneous clusters (Supplementary figure 2) This is likely due to aberrant expression levels of surface proteins that are more or less functional in the myeloid lineage, as opposed to healthy cells where the combination and expression levels of the surface proteins is tightly regulated and reflects cell function. We took advantage of this feature by selecting a cluster number that was sufficient to identify the expected healthy cell subsets (guided by healthy PB and BM samples) while clustering the remainder of the (presumably immature) cells in as few clusters as possible (in essence "under- clustering" the patient samples). Using this approach, we ended up with two blast clusters, one myeloid cluster and one stem cell cluster, where all patients were represented, which allowed a cross- cohort analysis of the signaling parameters.  
+
+The points above aptly illustrate, as rightly pointed out by the reviewer, that no cluster will ever represent an "absolute ground truth". Clustering of single cell data is always a compromise. But based on this reasoning, we believe that we can justify calling MC9 one cluster. However, we certainly agree that MC9 is a heterogeneous cluster. Therefore, we have taken the concerns of the reviewer into consideration, and included a further characterization of the immunophenotype of MC9 in the revised manuscript (as more thoroughly described in the answer to question 2 below). We have also included a justification of the (slightly unconventional) analytical approach in the results section of the revised manuscript. (line 342)  
+
+2 In general, MC9 should be better characterized, for example, by highlighting each patient on the tSNE plot, and by showing the expression of each analyte, particularly CD34 and p- ERK1/2, on the tSNE plot, so that we can better understand heterogeneity within that cluster.  
+
+2.1 We thank the reviewer for this comment and have done a further characterization of MC9 in the revised manuscript, section "Immunophenotypical characterization of metacluster(MC)9"), new supplementary figures 7 and 8. We exported MC9 as identified by FlowSOM for all patients and did a new FlowSOM with only MC9. (Supplementary Fig. 9) We chose the same analytical approach as the original pipeline, with 10 metaclusters. These were named sub-clusters (Sub-MC1-10). As expected, the majority of cells clustered together (sub-C3 61%). We also performed a new LASSO Cox regression analysis, as described originally, using the 10 sub-Cs identified within MC9. This analysis confirmed the prognostic significance of pERK1/2 at 24h, but within a small sub-cluster; sub-MC7 (3.24% of total). This cluster was present, but small, in all patient samples. p-p38 was found to be of prognostic significance at 24h within a separate small sub-cluster; sub-MC2 (4.78% of total).
+
+<--- Page Split --->
+![](images/Figure_2.jpg)
+  
+
+2.2 We also manually gated the pERK1/2 positive and negative cells in MC9 separately in each patient, to investigate the immunophenotype of these cells. (Supplementary Fig.8) This analysis has been added to the result section of the manuscript under the paragraph "Immunophenotypic characterization of metacluster (MC) 9". Both positive and negative pERK1/2 cells had a heterogenous immunophenotype, but we found the expression of AXL(p=0.0002), CD90 (p=4.7E-05), CD56 (p=0.0006) and CD34 (p=0.002) to be significantly higher in pERK1/2 positive cells.
+
+<--- Page Split --->
+![](images/Figure_3.jpg)
+  
+
+2.3 We have made new tSNE plots that highlight the distribution of each of the 32 AML patients (identified by color) on the tSNE plot of all patients(grey) in MC9 Supplementary Fig. 22). These tSNE plots show only MC9. Supplementary figure 22a a show all patients by color, b shows the 10 different sub- metaclusters as described in supplementary Fig 8b. And c shows the Individuals for each barcodepool, to make the distribution per patient more visible and to be able to assess if there are any batch effects. This figure has been added to the
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>supplementary information. </center>  
+
+2.4 Furthermore, we also made new tSNE plots showing the expression of pERK1/2 and CD34 in the different sub-clusters of MC9 (This figure is not included in the revised manuscript, but shown below (Response letter figure 2.4a). These plots contain all 32 AML patients. Response letter figure 2.4b shows all surface markers in our antibody panel in MC9 at timepoint pre-treatment for all 32 AML patients. Notably, The FlowSOM analysis (described in 2.1) was performed on all cells in MC9 from all patients without down-sampling. The t-SNE plots (described in 2.1 and 2.4) are down-sampled equally per timepoint, showing 20,000 cells per plot.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Response letter figure 2.4 </center>  
+
+2.5 To estimate if the CD34 cells contributed to the prognostic effect of \(p\) -ERK1/2 in MC9, we exported MC9 for each patient and manually excluded/gated out the CD34 positive cells (response letter figure 2.5 a). Next, we assessed only the CD34 negative cells in MC9 and stratified the patients based on the \(90^{\text{th}}\) percentile 24h pERK1/2 value and divided the patients into two groups by the median 24h pERK1/2 value. There was still a significant difference in survival between the two groups (Response letter figure 2.5 b). (Log- rank (Mantel- Cox) test, \(p\) -value =0.0104)  
+
+Response letter figure 2.5  
+
+![](images/Figure_9.jpg)
+  
+
+## 3 What proportion of malignant cells does MC9 represent in each patient?  
+
+3.1 The proportion of MC9 per patient is shown in Supplementary Figure 3 (red). In this figure, the proportion of this cluster is also shown for peripheral blood (yellow) and bone marrow (green) from healthy donors. MC9 is expanded in most of the AML patients compared to the 7 healthy donors. MC9 is more abundant in the three bone marrow samples than in peripheral blood, which indicates a more immature origin of these cells. We have assumed that this cluster is likely malignant due to the expansion in the AML patients and the immature myeloid phenotype. As described in 2.1 and shown in supplementary figure 8, we performed a new FlowSOM analysis on only MC9, identifying 10 sub-clusters (sub- C1- 10). The immunophenotype of these sub- Cs for the 32 AML patients at diagnosis are shown in supplementary figure 8b. The largest sub- C, sub- C3
+
+<--- Page Split --->
+
+contained \(61.07\%\) of the cells in MC9, this cluster had a normal myeloid phenotype with C064, C038 and C033. Three clusters (sub- C 1, sub- C4 and sub- C2) had expression of C034 and were therefore classified as malignant cells (23.81%). The other sub- Cs had myeloid markers (C038, 033, C064) with aberrant lymphoid markers (C03, C04, C08, C020) and were therefore classified as aberrant clusters (15,12%). In the figure below, response letter figure 3.1, the malignant sub- clusters sub- C1, 2 and 4 was summarized and are shown as \(0\%\) of total MC9 cell population per patient.  
+
+![](images/Figure_12.jpg)
+
+<center>Response letter figure 3.1 </center>  
+
+## 4 Does this proportion correlate with survival?  
+
+4.1 The size of each metacluster was included as a feature in the LASSO Cox regression analysis and was not predictive of survival. We also performed a new LASSO cox regression analysis with only MC9 and the 10 sub-clusters. The size of neither of the sub- clusters was predictive of survival.  
+
+4.2 To investigate if the proportion of malignant cells correlated with survival, we stratified the patients based on \(0\%\) of malignant cells in MC9 and divided them into two groups by the median value (16 patients in each group). Response letter figure 4.2 (not included in the revised manuscript) shows the Kaplan Meier curve of the two groups. There was no significant difference between patients with a high proportion of malignant cells in MC9 and the ones with a low proportion (not significant, Log- rank (Mantel- Cox) test.  
+
+![](images/Figure_unknown_1.jpg)
+
+<center>Response letter figure 4.2 </center>  
+
+## 5 Does change in the size of MC9 or time relate to survival?  
+
+5.1 No, the change (ratio 24h/0h) or the delta (24h- 0h) of MC9 does not relate to survival. A simple linear regression showed that neither the delta nor the ratio of
+
+<--- Page Split --->
+
+MC9 was significantly correlated to survival. When we divided the patients into two groups (16 patients in each) and did a Kaplan Meire survival analysis, there was no significant difference. (Response letter figure 5.1. Not included in the revised manuscript).  
+
+![](images/Supplementary_Figure_20.jpg)
+
+<center>Response letter figure 5.1 </center>  
+
+## 6 Does this depend on the extent of CD34 expression within MC9?  
+
+6 Does this depend on the extent of CD34 expression within MC9?6.1 Only 10 of the 32 patients have CD34 expression in MC9, this is shown in supplementary figure 10b. Most of these patients (7/10) were in the high 24h- p- ERK1/2 group. The median level of CD34 was not significantly higher in high 24h- pERK1/2 group than in low group (grouped students t- test) and not when used as a categorical variable with 3 patients in low group and 7 patients in high group (Fishers exact t- test). The Kaplan- Meier survival analysis above (response letter figure 4.2) shows that when patients were stratified into two groups based on the proportion of CD34 cells in MC9, there was not a significant difference in survival between patients with high and low proportion of CD34+ cells in MC9. (Log- Rank Mantel cox test).  
+
+7 One limitation of the proteomic approach is that it is difficult to know exactly which cells are malignant, because no genetic/mutational information is available, although sorting and targeted sequencing for somatic mutations could be performed to address this further.  
+
+7.1 We agree that the lack of genetic information is a limitation for both the bulk proteomics and single cell CyTOF analyses. Sorting and targeted sequencing could certainly have been performed, although in our opinion this would not provide additional information that would be useful for assessing the results from proteomic analyses, as these are still from a bulk sample. It could, however, provide useful additional information for the CyTOF analyses. To our knowledge, no direct correlation between genotype and immunophenotype has been identified in AML, and such experiments could provide additional information to elucidate this connection. Furthermore, novel techniques are also available that can provide both genetic- and proteomic information from the same single cell analysis, such as CITE-seq (Stoeckius et al. Nature Methods 2017). Perhaps this could be a fitting
+
+<--- Page Split --->
+
+future approach to investigate this issue. However, this was outside the scope of the current work, and we did not have additional patient material available for such analyses.  
+
+7.2 Although we don't have genetic information on the specific cell subsets identified by CyTOF, we still believe that identification of malignant cells is possible. Hematological diagnostics is heavily based on flow cytometry, where expansion of immature cells and/or cells with aberrant immunophenotypes is routinely identified as (likely) malignant. Thus, we think it is reasonable to assume that such cells identified by CyTOF are likely malignant as well. By including PB and BM samples from several healthy donors in our analyses, we ensure that cell populations with immature or abnormal phenotypes can be easily detected in the patient samples. Furthermore, the results from this study show that therapy response is not merely within the malignant subpopulation, and that presumably healthy subsets can also be involved in therapy response  
+
+7.3 Indirectly, cells in MC9 are more abundant in the three bone marrow samples than in peripheral blood, which indicates a more immature origin of these cells. We have assumed that this cluster is malignant due to the expansion in the AML patients and the immature myeloid phenotype. See also 3.1 for more details.  
+
+8 Further discussion of the specific relevance of MC9 to survival would be helpful. For example, why is protein expression in this cluster more relevant to survival than, say, the size or characteristics of the blast cluster, MC1?  
+
+8 Further discussion of the specific relevance of MC9 to survival would be helpful. For example, why is protein expression in this cluster more relevant to survival than, say, the size or characteristics of the blast cluster, MC1?8.1 The relevance of altered protein expression in MC9 is identified through the Cox Lasso regression model. The model also included MC1 (and all other MCs) but did not identify significant alterations in these MCs following treatment that correlated to survival. As MC1 was a particularly interesting cluster due to the expression of CD34, this cluster has been further characterized in supplementary figure 10 and in the section "Signaling in CD34+ MC1 AML blast cluster" in the manuscript (line 385). Notably, pERK1/2 did not confer prognostic information in MC1. Why MC9 is relevant and not the clusters with typical blast immunophenotype, such as MC1, is a very interesting question. However, based on our data, we can merely speculate. As shown by Levine et al (Cell 2015) the immunophenotype is not necessarily correlated to the signaling and function of the malignant cells. Levine et al showed that the immunophenotype and signaling in healthy donors were tightly coupled. However, in the AML samples the stratification of primitive and mature signaling had no association with the CD34 expression. Surface markers might not be a reliable proxy for the cell function and the expression of surface markers might be more fluent than what we normally acknowledge. We have added some thoughts around this in the discussion section of the revised manuscript (line 670)  
+
+## 9 Is change in the relative size of MC1 and MC9 (over time) correlated with survival?  
+
+9 Is change in the relative size of MC1 and MC9 (over time) correlated with survival?9.1 We have investigated the correlation between the size of MC9 at all timepoints as shown in response letter figure 9.1 below. The correlation to survival was not significant when the cluster size was used as a continuous variable as shown by the simple linear regression. In the original LASSO cox regression analysis, the metacluster size was also used as a continuous variable, and no metacluster size was predictive of survival. However, when patients were stratified by MC9 size into high and low group (split by median) with 16 patients in each group. (3 patients did not have samples at 4 hours, therefore we used 14 vs 15 patients for the 4h analysis). Patients with more cells in MC9 at 24h had a significantly poorer survival
+
+<--- Page Split --->
+
+than the patients with less cells in MC9 at 24h. This was not significant at Oh or 4h timepoints. This underscores the negative prognostic value of MC9 at 24h after start of treatment.  
+
+Response letter figure 9.1  
+
+![](images/Supplementary_Figure_19.jpg)
+  
+
+9.2 There is no correlation between the size of MC1 at any timepoint or the change in MC1 from Oh to 24h and patient survival, as shown in response letter figure 9.2 below. The correlation was analyzed by a simple linear regression, and was not significant for any of them. Patients were also stratified according to MC1 size at all timepoints and analyzed by Kaplan- Meier survival analysis, as described above. There was no statistical significance in survival between patients with high and low levels of MC1 at any timepoint.  
+
+![](images/Supplementary_Figure_5.jpg)
+
+<center>Response letter figure 9.2 </center>  
+
+10 High and low p- ERK1/2 patients were defined relative to each other, depending on whether they were above or below the median in this cohort. Is there an absolute protein abundance that can be used to determine whether new patients are "high" or "low?"  
+
+10.1 The level of p- ERK1/2 in the healthy donors might be used to determine if the patients are high or low in p- ERK1/2. In Supplementary Fig. 5b, the level of pERK1/2 in patients in high 24h- pERK1/2 group is shown next to the 7 healthy donors in
+
+<--- Page Split --->
+
+MC9 (bone marrow in green, peripheral blood in yellow). The level of p- ERK1/2 in MC9 In the high 24h- pERK1/2 group at 24 hours were significantly higher than the levels in the 7 healthy donors. The median value of pERK1/2 in the 7 healthy donors were 0.49 (range 0.42- 0.63). Thereby any value above this might indicate a poor prognosis. Kaplan- meier survival curve with a cut off \(< 0.05>\) gives 10 patients in low group and 22 patients in high group: p- value of: 0.0004. (Kaplan Meier survival analysis, Log- Rank Mantel cox test)  
+
+Response letter figure 10.1  
+
+![PLACEHOLDER_16_0]
+  
+
+If not, how would new patients be classified with respect to p- ERK1/2 abundance?  
+
+10.2 Based on future analyses of healthy donors and new AML cohorts, it is likely that we can create a normal range for p- ERK1/2 and p- p38. However, we question if not the response based on therapy (p- ERK1/2 ratio) may be a more robust marker for response. We search for new cohorts in clinical trials that can be analyzed prospectively. The multiparameter analysis of CyTOF and spectral flow cytometry should allow standard samples and calibration beads could be added.  
+
+The result would be stronger if there were an additional validation cohort that could be independently classified with respect to p- ERK1/2 and p- p38 abundance.  
+
+10.3 We appreciate this suggestion and are planning future control cohort experiments, ideally in a controlled phase III trial. This is outside of the scope of this proof- of- principle study. The material in this analysis is based on full blood samples added a special fixation before red cell lysis and processing. Such material is not available in other cohort to our knowledge. Cryopreserved material will not allow a similar analysis of signaling at baseline and 24 h after start of chemotherapy. However, several lines of evidence support our observation. A) Control analyses of other cellular metaclusters do not reproduce the same prognostic information. B) Our results from similar collections in chronic myeloid leukemia (CML) with targeted kinase inhibitor therapy suggest that the concept is robust. (Gullaksen et al. Haematologica 2017)  
+
+11 Mutation correlations were very anecdotal. Obviously, this analysis would benefit from more patients and exome or WGS sequencing.  
+
+11.1 This is a very timely issue, and the prognostic classification of AML based on genomics is based on thousands of patients, e.g. ELN 2017 genetic risk classification (Dohner et al. Blood 2017, Herold et al. 2020). Our analyses indicate stratification with higher precision. This may have important implications for better precision medicine in AML prognostication. This needs to be addressed in larger trials, e.g. Loweberg et al. Blood Adv 2021. However, a challenge is that
+
+<--- Page Split --->
+
+specially prepared material is required. Your comment motivates us to test your suggestion in a larger prospective AML trial in the HOVON/SAKK network.  
+
+11.2 Although we agree that the mutation correlations are anecdotal, the clinical data provided is contemporary diagnostics in most academic centers in Europe, focusing on use in ELN 2017 genetic risk classification including ongoing validation and sub stratification. We hypothesized that some mutations should have essential impact on signaling early during chemotherapy, more like indicated in previous works using ex vivo stimulation (Irish et al. Cell 2004). However, we found a tendency for patients with the same mutations to have a similar signaling pattern, and this might be useful in future studies. As stated in the manuscript, more robust analyses of the correlation between mutational status and signaling would require a much larger cohort.  
+
+The confirmatory flow cytometry analysis on pp. 14- 15 actually gave a more significant result than the original mass cytometry clustering analysis. This strengthens the overall claim of the paper, but raises some interesting questions. How do the immature cells defined by this blast gate correspond to the FlowSOM clusters shown earlier?  
+
+11.3 For the validation by manual gating, our strategy was to exclude lymphoid cell subsets (CD45 high) and granulocytes (CD66b high). We agree that this gating strategy does not represent the same cells as in MC9, as also the CD34 positive cells are also CD45 low, CD66low. However, the main objective for performing this analysis was to investigate whether the significant functional signaling response could be identified in a bulk sample analogous to a crude blast gate used by conventional flow cytometry. There were several reasons for this, including to bridge our CyTOF analyses with conventional flow cytometry, which is routinely applied in the clinic. Furthermore, these results are also more comparable to our validation experiments by proteomics and RNAseq, as these are bulk analyses performed on lymfoprepped patient samples (where neutrophils are removed). The DSRT analyses are also performed using lymfoprepped samples.  
+
+11.4 Are these flow- gated blasts in MC9 or MC1? As stated above, the manual biaxial gate likely includes both MC9 and MC1. To be sure that the prognostic pERK1/2 signal did not come from the CD34 positive cells in the manually gated CD45, CD66low population, we also did a manual gating of the bi- axial gated cells where we gated away the CD34 positive cells to see if the prognostic information of pERK1/2 was still there. The gating strategy is shown in Response letter figure 12.2a below, b shows the \(90^{\text{th}}\) percentile arcsinh transformed pERK1/2 values at all timepoints for these cells. The prognostic value of pERK1/2 in this cell population was still significant, as shown by the Kaplan- meier curve (Response letter figure 12.2 c, Log- rank (Mantel- Cox) test: p- value 0.0048. )
+
+<--- Page Split --->
+![PLACEHOLDER_18_0]
+
+<center>Response letter figure 12.2 </center>  
+
+12 The authors could reanalyze the cytometry data in a manner analogous to the flow sorting - this might strengthen the result, or would at least add clarity by connecting these two analyses to each other.  
+
+12.1 We apologize for any unclarities in the original text, but the manual biaxial gating of immature cells (CD45low/CD66b-) is indeed a re-analysis of the original CyTOF data. We performed these analyses on the original unclustered data, as an alternative approach to unsupervised clustering using FlowSOM. The gating strategy was aimed at capturing all malignant cells in the samples, to investigate whether the significant signaling response could be identified in an analysis of something comparable to a bulk (lymfopped) sample, as described above. We have reviewed the text of the original manuscript, and made edits to the results section (line 400) to make this more clear.  
+
+13 For the bulk RNA-seq analysis, a more unbiased analysis could be useful.  
+
+13.1 What other genes were differentially expressed between the p-ERK1/2 low and high patients? To address this question, we did a student's t-test with FDR cutoff at \(< 0.05\) between the low vs high 24h-pERK1/2 groups (all timepoints). This analysis was performed on all genes in our RNAseq bulk dataset ( \(n = 50.668\) genes). When these results were clustered by hierarchical clustering (Euclidean distance) the high and low 24h-pERK1/2 groups formed two separate clusters. The three patients with FLT3- ITD mutations formed a separate cluster within the high 24h-pERK1/2 group. The result was 76 significant genes. (Supplementary figure 11) 40 of these genes were high in high 24h-pERK1/2 group, there among HOXA9, HOXA10, HOXA10- AS. HOXA9 have been reported by others to be a poor prognostic factor in normal karyotype AML (Collins et al. Oncogene 2016) The myeloid oncogene TRIB1 is a pseudokinase that interacts with MEK1 to enhance its phosphorylation of ERK1/2. Overexpression of Trib1 enhances HOXA9 induced
+
+<--- Page Split --->
+
+leukemogenesis (Takashi et al, Blood 2010). This figure has been added to the supplementary information as supplementary figure 11.  
+
+![PLACEHOLDER_19_0]
+  
+
+13.2 Moreover, if the samples are hierarchically clustered based on the bulk RNA- seq data alone, do they naturally cluster into the same groups (p- ERK1/2 low and high)? To assess this question, we did an unsupervised hierarchical clustering (Euclidean distance) of all genes present in more than 3 samples (n=50.490) at all timepoints in all patients (n=14 patients). The heatmap is shown in the figure below (response letter figure 14.2a). There is no obvious clustering of the patients into high and low pERK1/2 group. All patients except for P16 made their own sub cluster with the three samples from each patient in the same cluster. This confirms the heterogeneity of the patients. Response letter figure 14.2b shows the unsupervised clustering of the pERK1/2 and pp38 inducible genes. Here the three patients with Inv(16) clustered together.
+
+<--- Page Split --->
+![PLACEHOLDER_20_0]
+
+<center>14 The bulk RNA-seq t-test comparison essentially assumes that MC9 is roughly the same size in all patients. A bulk analysis that takes into account the size of MC9 in each patient might be more sensitive at detecting genes of interest. </center>  
+
+14.1 The bulk analysis is a weighted sum of all the metaclusters. So even if we correct for the size of MC9, we would still be left with different mixes of all other clusters. And we cannot correct for all of them. Notably, the RNAseq samples are lymfoprepped, the granulocytes are not included. The bulk RNA-seq t-test was not correlating with the size of MC9. We interpret this as gene expression in dominating genes, not defined gene expression.  
+
+14.2 However, in an attempt to answer this question, we have performed an analysis where we weighted for the genes highly expressed in patients with a high proportion om MC9, rather than to weight the analysis based on MC9 size. We did a student's t-test between the two patients with most cells in MC9, namely P14 and P15 (over \(80\%\) of the cells in MC9) and the two patients with the lowest number of cells in MC9; P9 and P20 (with less than \(20\%\) of the cells in MC9). All of these patients were in high 24h-pERK1/2 group. We selected the genes that were significantly higher in P14 and P15 ( \(n = 3637\) , \(p\) -value \(< 0.05\) ). Next, we did a grouped students t-test between the patients in high 24h pERK1/2 group vs the ones in low 24h-pERK1/2 group on these selected genes ( \(p\) -value cut of \(< 0.05\) ). P14, P15, P20 and P9 were excluded from this analysis. The significant genes are shown in response letter figure 15.2d. Patients in high and low group clustered into separate clusters, with hierarchical clustering (Euclidean distance). We found several genes of interest among the genes that were significantly higher in high 24h-pERK1/2 group. There among MAPK11, the beta isoform of P38, and three genes known to activate c-Jun (MAP3K13, MAP4K3 and MAP4K5). RUNX1 and MEIS1 was also significantly higher in high 24h-pERK1/2 group. Response letter figure 15.2 shows the results from this analysis. 15.2a shows the size of MC9 in each patients (patients with RNAseq data have a black border). Figure 15.2b
+
+<--- Page Split --->
+
+shows the pERK1/2 level of the four patients used for the first t- test to enrich for genes more abundant in P14 and P15. 15.2c shows the unsupervised clustering of these genes. 15.2d shows a heatmap of the genes that were significantly different between the low and high 24h- pERK1/2 group.  
+
+![PLACEHOLDER_21_0]
+  
+
+15 The drug sensitivity matrix should also be hierarchically clustered by patient and compared to the proteomic data to discover other potential correlates of drug sensitivity in an unbiased fashion.  
+
+15.1 The analysis has been performed, and a new figure has been added to the revised manuscript (Supplementary Fig. 15). The selective drug sensitivity score (sDSS) for all patients were clustered by hierarchical clustering (Euclidean distance), we applied a sDSS cut- off at \(+ / - 5\) , drugs that did not have values above or below 5 were not included (Supplementary figure 15a).15.2 We also performed a grouped students t- test between the patients in high versus low Oh and 24h pERK1/2 group on the sDSS (cut- off \(+ / - 5\) ). As the sDSS data were only performed at time of diagnosis, we chose to compare it with the pERK1/2 value at time of diagnosis. Two drugs were significantly different between the two groups, namely the HSP90 inhibitor Tanespimycin and the
+
+<--- Page Split --->
+
+hypomethylating agent Azacitidine/Vidaza). Both had high sDSS in high Oh pERK1/2 group. The most significant drug that had a high sDSS in patients with high pERK1/2 at Oh, was Azacitidine (Vidaza)( \(p = 0.0128\) ). When we compared the low vs high 24h- pERK1/2 group, the most significant drug with high sDSS in high pERK1/2 group was the Akt inhibitor MK- 2206 2HCl ( \(p = 0.026\) ). Supplementary Fig.15b show the results of the students t- test for the Oh pERK1/2 groups, patients are sorted by the pERK1/2 value in MC9 at Oh. Supplementary Fig.15c show the correlation between sDSS for Azacitidine vs pERK1/2 in MC9 at Oh. (Simple linear regression analysis, \(p\) - value \(= 0.0159\) ). This figure have been added to the supplementary information and described in the Paragraph "Drug sensitivity data show sensitivity for HSP90,mTOR, BCL- 2 and MEK inhibitors".  
+
+![PLACEHOLDER_22_0]
+  
+
+16 The authors repeatedly mention that their proteomic data indicates the presence of "clonal heterogeneity," but this is traditionally defined by genetics, not proteomics. For instance, clustering of somatic mutations by variant allele frequency would reveal clonal heterogeneity.  
+
+16.1 It is true that the link between genotype and immunophenotype is not well explored, and it is currently not known whether immunophenotypic heterogeneity in AML is related to clonal heterogeneity, although it is known that the disease is characterized by both. Thus, we agree that this term has been improperly used in the manuscript. This statement has now been removed throughout the revised manuscript. We have further deleted a paragraph in the discussion where clonal heterogeneity and the advantages of single cell analyses in this context was discussed.
+
+<--- Page Split --->
+
+## Reviewer #2,  
+
+expertise in bioinformatics, mass cytometry analysis and AML (Remarks to the Author):  
+
+The authors used 36- dimensional mass cytometry data to identify genes sensitive to the chemotherapy response. These were further validated using RNAseq and mass spec. The reason of this study is well explained and further validated using DSRT.  
+
+18. The tSNE plot in Figure 1 can be colored with ID to see if clustering is not influenced by batch effect.  
+
+18.1 The issue of batch effect in mass cytometry data is an important question, and we are glad to elaborate on this in our data. Notably, to be able to correct for batch effects, we included an identical reference sample that was added to each of the 7 barcodepools (barcode 1-7) and used for normalization (quantile normalization). Thereby we were able to correct for the batch effects, and compare the different patients across barcodepools. A new figure has been added to supplementary Figure 16, (Supplementary figure 18e) to show the reference sample prior to and after normalization, the t-SNE plots illustrate how there was a batch effect prior to normalization and show how the batch effect is removed after normalization. CD4 and CD66b staining is shown in the normalized samples.  
+
+Supplementary figure 18e  
+
+![PLACEHOLDER_23_0]
+  
+
+19. Many clusters including MC1,2,3 are scattered. Isn't it because of batch effect? coloring based on patient ID will answer.  
+
+19.1 As suggested by the reviewer, we have made new t-SNE plots showing the different barcodepools (batches) and colored by patient ID. These figures have been added to the supplementary information as Supplementary figure 20. The data is not affected by batch effect in our opinion. However, due to the heterogeneity of the disease, there are some clusters that consist of few patients, like MC2, which is mainly consisting of P9 (batch 1) and P22 (batch 4). We have also added some text in the methods section to elaborate on potential batch effects (line 782) Supplementary figure 20a shows the 10 FlowSOM metaclusters. Figure b shows the 7 different barcodepools/batches. Figure c shows the 32 individuals and the 7 healthy donors.
+
+<--- Page Split --->
+![PLACEHOLDER_24_0]
+
+<center>Supplementary Figure 20 </center>  
+
+20. Why they are scattered? Is it due to the assignment to the existing clusters? How then they can be assigned to the same cluster? For instance, MC1 is scattered into to clusters? How these are defined as MC1?  
+
+20.1 FlowSOM consists of a clustering step that assigns cells to many clusters and a meta-clustering step that combines some of these clusters. The step of combining clusters can in some cases also result in meta-clusters that appear somewhat scattered on a t-SNE map. The visually scattered clusters indeed reflect phenotypic heterogeneity within the clusters identified by FlowSOM (e.g. variable surface marker expression levels). For instance, there are five known subsets of t-helper cells, and although specific markers to identify these subsets are not included in the antibody panel, the t-helper population will still have a slightly variable expression levels of lymphoid markers, which reflects different biological functions.  
+
+- Importantly, tSNE is a visualization tool, not a clustering tool. tSNE plots are used to visualize high-dimensional data in two dimensions. Its purpose is not to show clusters in data (https://arxiv.org/pdf/2110.02573.pdf). It would be surprising if there would be no correspondence at all between tSNE and clusters when they are both based on the same combination of surface markers, but we cannot expect complete overlap. - Clustering or gating on the tSNE-transformed data would lead to the least scattering on the tSNE, but it has been shown that this can result in less predictive clusters than clustering the high-dimensional data using unsupervised algorithms. We focused on the predictive power of our clustering rather than the visual appearance.
+
+<--- Page Split --->
+
+- FlowSOM was performed on all cells from all patients, without down sampling. The t-SNE plots show 20,000 cells per plot.  
+
+21. MC9 has been extensively used for the analysis. Why MC9 (the myeloid cluster) could become most informative cluster among other MCs? Any speculation on it?  
+
+21.1 MC9 was chosen from the LASSO regression analysis. This is the metacluster where we find the prognostic value of pERK1/2 and p-P38 at 24h. As shown in supplementary figure 8, we performed a new FlowSOM analysis on only MC9, identifying 10 sub-clusters (sub-C1-10). The immunophenotype of these sub-Cs for the 32 AML patients at diagnosis are shown in supplementary figure 8b. The largest sub-C, sub-C3 contained 61.07% of the cells in MC9, this cluster had a normal myeloid phenotype with C064, C038 and C033. Three clusters (sub-C 1, sub-C4 and sub-C2) had expression of C034 and were therefore classified as malignant cells (23.81%). The other sub-Cs had myeloid markers (C038, 033, C064) with aberrant lymphoid markers (C03, C04, C08, C020) and were therefore classified as aberrant clusters (15,12%). When we did a new FlowSOM analysis on only MC9 we discovered that the prognostic pERK1/2 signal did not seem to originate from these C034 positive cells, but from a smaller myeloid but aberrant cluster (sub-cluster 7) which contained 3.24% of the cells from the 32 AML patients (Supplementary figure 8d). This is discussed in the new section in our manuscript "Immunophenotypic characterization of metacluster (MC) 9."  
+
+22. Besides 24h-p-ERK1/2, p-p38 at 24, p-Rb at 24h were reported. The combination of several markers could lead a better performance. This has to be tested and discussed. In the same way, can the information obtained from RNAseq and Mass spec be used to in designing combination of markers?  
+
+22.1 Note that the Lasso Cox regression model does find a combination of markers, namely the combination of p-ERK1/2, p-p38, and p-Rb at 24h. We have chosen to report on these individually, because this makes it possible to show Kaplan-Meier curves split by median of each individual marker. But the overall model uses a linear combination of the makers.  
+
+It is true that the Lasso regression does not include any interaction terms between the different markers. While this is in principle possible, it increases the number of regression coefficients from around 600 to around 180000. For this to be reasonable, we would require a substantially larger sample size.  
+
+22.2 The RNAseq and Mass spec are bulk cell analyses and therefore they are difficult to use to find combinations of markers that can be applied in single cell analysis.  
+
+## Reviewer #3  
+
+expertise in mass cytometry and AML (Remarks to the Author):  
+
+In the present manuscript the authors assessed the correlation between single cell signaling and clinical response to induction chemotherapy in 32 patients with AML. Samples were collected at diagnosis and during the course of induction chemotherapy, and analyzed by mass cytometry, RNAseq and mass spectrometry proteomics in order to assess 1/ the chemotherapy- induced phospho- signaling modulations in the myeloid compartment and 2/ the prognostic significance of phospho- protein expression levels in myeloid cells 24h after treatment initiation. Overall, the manuscript is well written and the mass cytometry data are presented in a clear way. However
+
+<--- Page Split --->
+
+some issues detailed below prevent publication of this manuscript in its present form. Graphical abstract :  
+
+23. The legend is constructed as a standard legend; this legend should rather focus on the summary of the key points and the main conclusions of the article.  
+
+23.1 the graphical abstract has been reconstructed into figure 1.  
+
+24. The risk stratification by bone marrow tumor load is mentioned in the discussion section, but has not been displayed in the results section or in the supplementary data. The overall survival according to the ELN classification only appears in the graphical abstract as well, and should be integrated to the body of the article.  
+
+24.1 We thank the reviewer for this observation, the ELN risk stratification has now been described in the manuscript and is now displayed in the results section as Figure 1.  
+
+## Results section  
+
+25. Identification of an independent prognostic factor in AML should be confirmed using a Cox regression including the validated prognostic factors in order to exclude potential confounding factors. In the present work, this Cox regression should include 24h-pERK1/2, age, ELN, WBC count at diagnosis, as well as allogeneic stem cell transplantation, ideally as a time-dependent covariate. Of note, the poor survival in the ELN favorable group, especially taking into account the fact that most patients are young patients, underlies that there might be a bias in the present cohort and probably precludes proper Cox regression for a formal identification of a potential prognostic factor.  
+
+25.1 We have performed a new cox regression analysis that has been added to the results section of the revised manuscript (line 258), which included the following covariates for all patients: The 24h pERK1/2 value in MC9 (continuous variable), age (continuous variable), ELN 2017 risk (categorical variable: Favorable, Intermediate, Adverse), WBC count at diagnosis (continuous variable) and transplantation status as a time-dependent covariate. There were 14 patients who received allogeneic transplantation in our study. All patients were followed for 5 years or until they died. No patients dropped out of the study.  
+
+The result of this analysis was:  
+
+<table><tr><td>parameter</td><td>hazard_ratio</td><td>p</td></tr><tr><td>1 pERK24h</td><td>2.27</td><td>0.000581</td></tr><tr><td>2 Age</td><td>0.0123</td><td>0.571</td></tr><tr><td>3 WBC at diagnosis</td><td>-0.00690</td><td>0.290</td></tr><tr><td>4 ELN</td><td>0.0227</td><td>0.941</td></tr><tr><td>5 transplant</td><td>-0.252</td><td>0.639</td></tr></table>  
+
+Thus, pERK1/2 value at 24h was the only predictive marker for patient outcome (5 year survival, HR 2.27, p- value 0.000581).  
+
+25.2 To make it easier to evaluate the composition of the patient cohort, we have added a new summary table of the cohort to the supplementary information. (supplementary table 2) Notably, these patients were consecutively sampled over 2 years at Bergen university hospital and Oslo university hospital. The only inclusion criteria was AML, undergoing standard 3+7 induction therapy.
+
+<--- Page Split --->
+
+26. The MRD status is a cornerstone for prognostication of AML. An analysis with respect to the MDR would therefore be essential before concluding about the potential usefulness of measuring pERK1/2 to refine the therapeutic decision algorithm in AML.  
+
+26.1 MRD data was available for 14 of the patients in this study. Seven patients in low 24h-pERK1/2 group and seven in high pERK1/2 group. 5/7 patients in low group had negative MRD after cycle 2, the remaining two had no detectable leukemia associated immunophenotype (LAIP) at diagnosis (P10) and the other one had negative MRD but positive NPM1 MRD (P13). Among the seven patients in high group, four had negative MRD and three patients had positive MRD. The three patients with positive MRD were among the six patients with highest pERK1/2 values in MC9 at 24h after start of therapy. We found no significant differences in MRD status between patients in the high and low pERK1/2 group. The MRD data has now been added to the patients characteristics table, figure 3b and is described in the paragraph "Clinical parameters related to the p-ERK1/2 level in MC9" in the revised manuscript.  
+
+27. The authors claim in the abstract that they questioned whether the signaling response to therapy was more informative than analysis at time of diagnosis. In fig 2d, the basal level of pERK1/2 seems low in the low-24h-pERK1/2 group and high in the high-24h-pERK1/2 group. Did the authors assess the impact of baseline expression in the metacluster 9 on clinical outcome for a direct comparison? This information is important in order to appreciate the improvement of the quality of prediction using samples drawn during induction chemotherapy rather than baseline samples.  
+
+27.1 These analyses were done, and we found no significant correlations with differences in baseline expression and survival. This has been added to the results section of the revised manuscript. The Oh timepoint (in all metaclusters) is included in the LASSO cox regression and was not significant.  
+
+Additionally, when patients were stratified by Oh pERK value and divide into two groups (with 16 patients in each group) based on median pERK1/2 at Oh value, the survival between the two groups is not significant.  
+
+survival between the two groups is not significant.  
+
+Response letter figure 27.1  
+
+![PLACEHOLDER_27_0]
+  
+
+28. Why did the authors choose to include patients with MDS and B-ALL in this manuscript? At least the authors should retain only the 32 patients included in the study in the patient characteristics table.
+
+<--- Page Split --->
+
+28.1 We agree with the reviewer, these patients were not used in any of the presented analysis and have now been removed from the manuscript. We have chosen to keep the two patients P33 and P34 who received dose reduced \(3 + 7\) treatment. In a separate analysis (FlowSOM clustering and LASSO cox regression analysis) together with the other 32 patients, we still found a significant association with survival and pERK1/2 and pp38 at 24 hours. This is described in the methods section and strengthens the stability of the LASSO cox regression analysis.  
+
+29. In supplementary fig 3, the cluster of NK cells is located at 2 different places on the tSNE plot. Some cells that were annotated as NK cells clusterize in the region of myeloid cells, and might be leukemic blasts that express CD56. This would also explain the low CD45 mean expression of this cluster compared with CD45 expression in the clusters of CD4 and CD8 T cells. The authors should confirm by manual gating the exact nature of these cells.  
+
+29.1 To define the exact nature of these cells, we exported the NK cell metacluster (MC8) and did a new t-SNE analysis on only the NK cell cluster. These t-SNE plots show MC8 for all patients at the three different timepoints (all MC8 cells from each patient were concatenated at each timepoint) and for the 7 healthy donors. The t-SNE plots show the expression of all surface markers. From these plots there is an obvious difference between the patient samples and the healthy donors. One island on the tSNE plot is not present in the healthy donors, this island is positive for CD33, CD56, negative for CD7 and have a lower CD45 expression. We agree with the reviewer that these cells might be leukemic blasts that express CD56.  
+
+![PLACEHOLDER_28_0]
+  
+
+30. The authors should provide expression of the different markers projected on the tSNE plots in supplementary 3 in order to enable appreciation of the homogeneity of the different clusters. Minor: in supplementary fig 3, the lack of heatmap annotation in the lower panel makes the figure difficult to read.  
+
+30.1 A supplementary figure showing the expression of each surface marker on the tSNE plots have been added to the revised manuscript (Supplementary figure 19).
+
+<--- Page Split --->
+
+We have added the annotation on the lower panel of heatmaps in supplementary figure 3 (this figure is supplementary figure 2 in the revised supplementary information).  
+
+![PLACEHOLDER_29_0]
+
+<center>Supplementary figure 19 </center>  
+
+31. In Fig 3, the authors detail the manual gating strategy of leukemic blasts in order to confirm the results obtained with machine learning algorithms and assess the transferability of the results in a routine setting. However, the metacluster used to generate the initial survival curves according to pERK1/2 expression was MC9, which was considered as a myeloid cluster, by contrast with the MC1 and MC2 that were considered as the AML blast clusters (line 198). Could the authors clarify this discrepancy in the selection of the cell population they chose for manual confirmation of the results?  
+
+31.1 This is an important point, and we thank the reviewer for pointing this out. As further elaborated in the revised version of the manuscript, MC9 is a very heterogeneous cluster, and identification of similar/comparable cells in single AML samples by flow cytometry would not be feasible. Thus, the reason for this specific gating strategy was rather to investigate whether a gating for all immature/leukemic blasts cells would provide comparable information. In addition to assessing the transferability to routine flow diagnostics, another reason for investigating this population specifically was to increase the comparability between the CyTOF data and the validation datasets, consisting
+
+<--- Page Split --->
+
+of bulk proteomics and RNAseq data, since the samples used for proteomics and RNAseq were density centrifuged mononuclear cells and thus depleted of granulocytes.  
+
+32. In fig 6, results are presented for 16 and 9 drugs in the low 24h - p-ERK1/2 group and the high 24h - p-ERK1/2 group, respectively. The lack of consistence between panel b and c makes it difficult to compare the graphs. In addition, the drugs that are displayed in figure 6 should be listed in the materials and methods section ("MEK inhibitor", "DNA inhibitor" or "anti-oxidant" being too evasive). The barplots should be annotated so as the drug class appears as "Protein X inhibitor" rather than "Protein X".  
+
+32.2 We have revised figure 6 (called figure 7 in the revised manuscript) according to the reviewers' suggestions. We chose to visualize the 12 most efficient drugs in the entire cohort for both high and low 24h-pERK1/2 group (as shown in supplementary figure 16a). The same drug targets are visualized for both groups to make the two groups more comparable. We have specified that the different targets are inhibitors in the figure legend. We have added the specific drug names of the different drug target groups per patient in a supplementary excel table (Supplementary table 6) (sheet 2 "top 10 drugs per patient") together with the sDSS data of all patients (Sheet 1 "sDSS").  
+
+33. The aim of this figure is to correlate the intracellular signaling profiles detected by mass cytometry with ex vivo drug sensitivity (line 573). The way the results are presented does not enable a direct comparison between the two groups of patients.  
+
+33.1 We have rearranged the data to improve readability (figure 7 in the revised manuscript)  
+
+33.2 We have performed a new analysis where we do hierarchical clustering of the sDSS data comparing it to the pERK1/2 groups (Supplementary figure 15a) The selective drug sensitivity score (sDSS) for all patients were clustered by hierarchical clustering (Euclidean distance), we applied a sDSS cut-off at +/-5, drugs that did not have values above or below 5 were not included (Supplementary figure 15a). We also performed a grouped students t-test between the patients in high versus low Oh and 24h pERK1/2 group on the sDSS (cut-off +/-5). As the sDSS data were only performed at time of diagnosis, we chose to compare it with the pERK1/2 value at time of diagnosis. Two drugs were significantly different between the two groups, namely the HSP90 inhibitor Tanespimycin and the hypomethylating agent Azacitidine/Vidaza). Both had high sDSS in high Oh pERK1/2 group. The most significant drug that had a high sDSS in patients with high pERK1/2 at Oh, was Azacitidine (Vidaza)(p=0.0128). When we compared the low vs high 24h-pERK1/2 group, the most significant drug with high sDSS in high pERK1/2 group was the Akt inhibitor MK-2206 2HCl (p=0.026). Supplementary Fig.15b show the results of the students t-test for the Oh pERK1/2 groups, patients are sorted by the pERK1/2 value in MC9 at Oh. Supplementary Fig.15c show the correlation between sDSS for Azacitidine vs pERK1/2 in MC9 at Oh. (Simple linear regression analysis, p-value=0.0159). This figure have been added to the supplementary information and described in the Paragraph "Drug sensitivity data show sensitivity for HSP90, mTOR, BCL-2 and MEK inhibitors".  
+
+34. In supplementary figure 13, several drugs are designated as "MEK inhibitor", "BCL2 inhibitor",
+
+<--- Page Split --->
+
+etc; in figure 6, do the names of the therapeutic classes refer to the same drugs in the upper and the lower panel?  
+
+34.1 To improve interpretability of this data- intensive figure, we have annotated the figure using drug class (based on drug target) instead of using the specific drug name. This allows for an easier biological interpretation of the results. However, we appreciate that if could be interesting for some readers to compare the effect of specific drugs side- by- side, and we have included the drug names for the different drug targets for each patient in supplementary table 6, sheet 2.  
+
+35. Overall, the conclusions of the authors related to the results of fig 6 are not supported by the results, notably their conclusions regarding the RAS mutational status and the benefit from MEK inhibitors (line 608). The low number of patients included in this part of the work as well as the high heterogeneity of patients in terms of mutational status precludes any solid conclusion based on these results.  
+
+35.1 We apologize if this was poorly explained, the RAS mutation data comes from supplementary 14, where the other 2 patients with RAS mutations are shown, not from Figure 6. However, as these patients were not included in the main LASSO cox regression analysis we have chosen to remove this figure and the description of it in the revised manuscript. We have toned down our conclusion regarding the benefit from MEK inhibitors.  
+
+36. Minor: In supplementary figure 13, the scale should be identical for all patients. 36.1 The scale is now made identical for all patients. (Supplementary figure 16 in revised manuscript)  
+
+37.2D plot of phosphoprotein expression by group (in particular p-ERK1/2 in the high- vs the low- 24h-p-ERK1/2 group) is lacking in the supplementary data. This would enable to appreciate the quality of the staining, and would enable the readers to be more confident with the data currently presented as arcsinh transformed 90th percentile of p-ERK1/2.  
+
+37.1 A new figure has been added to the supplementary information of the revised manuscript (Supplementary figure 23). We exported MC9 for each patient and healthy donors. Then we did a new tSNE with all cells for each patient at the three timepoints and the bone marrow or peripheral blood from the healthy donor that were in the same barcode as that particular patient. These tSNE plots are shown in supplementary figure 23. Patients are stratified based on their 24h pERK1/2 value in MC9, from low to high. Patients in low 24h-pERK1/2 group are in Supplementary
+
+<--- Page Split --->
+![PLACEHOLDER_32_0]
+
+
+<--- Page Split --->
+
+38. In supplementary table 1, the 24h-pERK1/2 status would be relevant information to include, which would enable to appreciate the repartition of the confounding factors (age, ELN, WBC count, and allogeneic stem cell transplantation and MRD) according to pERK1/2 expression.  
+
+38.1 We thank the reviewer for this helpful suggestion. pERK1/2 status has now been included in the Patients characteristics table (supplementary table 1) 38.2 We have also done statistical analysis of the clinical data between the two groups. Age, ELN, WBC or allogeneic stem cell transplantation were not significantly different between the high and low 24h-pERK groups. This has been added and further described in the section "Clinical paramters related to the p-ERK1/2 level in MC9" in the revised manuscript. We also performed a new cox regression analysis where we included pERK1/2 in MC9, age, ELN, WBC at time of diagnosis and allogeneic transplantation as a time-dependent covariate. Only pERK1/2 was significant in predicting patient 5 year survival (HR 2.27, p-value 0.000581)(Line 258) in the revised manuscript.  
+
+## Conclusion  
+
+39. The authors claim that "early single cell signaling response to chemotherapy provided precise prognostic information independent of stratification by genetics". Taking into account the limitations of this work ie the small sample size (N=32), the absence of effect on survival of validated prediction tools in this cohort (ELN), as well as the absence of comparison with the prediction based on the MRD, the authors may moderate this conclusion.  
+
+39.1 We agree with the reviewer, and the discussion and conclusion of the revised manuscript has been re-written to moderate/tone down this conclusion (Line 690)  
+
+Minor comments and typos  Line 115: and overall poor overall survival  Thank you, this has been corrected  
+
+Line 618: the 18 patients that was analyzed  Thank you, this has been corrected  
+
+## Reviewer #4  
+
+expertise in proteomics/super-SILAC (Remarks to the Author):  
+
+The manuscript by Tislevoll et al. demonstrates an elegant new approach to predicting patient outcome based on molecular changes detected early in the treatment phase, rather than relying on genetic parameters only. The potential impact of such discoveries is vast, as it allows much more efficient treatment strategies to be determined and/or altered along the way if needed, thereby saving actual lives. Therefore I commend the authors for undertaking this work, which I expect will inspire many follow- up studies pursuing similar early treatment response markers across a range of diseases.  
+
+The manuscript is well written, the analyses conducted were thorough and nicely presented, and the combination of a range of technologies (and modalities) significantly strengthened the work as
+
+<--- Page Split --->
+
+well in my opinion. The authors demonstrate appropriate use of the various technologies, and I especially liked the barcoding and reference sample spike in approach for their CyTOF analyses. Seeing a combination of single cell proteomics (on more than 35 million individual cells!), bulk proteomics and RNAseq, and targeted sequencing, in a clinical setting and all integrated through machine learning, is not a mean feat, and the results presented speak for themselves.  
+
+We are grateful for these very encouraging remarks.  
+
+Thus, in my opinion, the work should be accepted for publication, and I have only a few minor comments laid out below:  
+
+40. On pg. 7, the authors declare the cells in MC1 and MC2 to be CD34+ blasts due to their high expression of CD34 and CD117. As these are also classical stem cell markers, could the authors clarify why they are deemed differentiated blasts rather than a more primitive cell type? 40.1 MC1 and MC2 were much more abundant in AML samples compared to healthy donors as shown in supplementary figure 3 (revised manuscript), and was only detectable in healthy bone marrow at very low levels (<0.5%) and in healthy peripheral blood in even lower levels. MC1 and MC2 had also aberrant marker expression like CD7 and CD56, we therefore classified these cells as CD34+ AML blasts. 40.2 We have annotated the CD34 positive metaclusters MC1 and MC2 for CD34+ cells blasts and not leukemic stem cell as we think that these terms may be controversial. There is an ongoing discussion about the definition of leukemic stem cells based on the immunophenotype (Khaldoyanidi SK et al. Crit Rev Oncol Hematol. 2022; Vetrie D, Helgason GV, Copland M. Nat Rev Cancer. 2020), and we have therefore chosen a broader term by calling these cells CD34+ blasts.  
+
+41. Fig2b - is this average expression across the single cells measured within cluster MC9? 41.1 The heatmap shows the Arcsinh transformed \(90^{\text{th}}\) percentile value of pERK1/2 in MC9.  
+
+42. Was any sub-segregation possible to determine the critical cell type in which response needs to be measured? In other words, given the single cell nature of their analyses, was there anything more specifically known about which cell types had high or low ERK? Or is it truly a homogeneous cell effect?  
+
+42. Was any sub-segregation possible to determine the critical cell type in which response needs to be measured? In other words, given the single cell nature of their analyses, was there anything more specifically known about which cell types had high or low ERK? Or is it truly a homogeneous cell effect? 42.1 We think this is a crucial question and have done additional analysis to address this. To identify the cells from where the pERK1/2 signal originated from we did a new FLOWSON with 10 metaclusters of only the cells in MC9 for all patients. A new LASSO cox regression analysis of these 10 sub-clusters identified that ERK1/2 and pp38 originated from two separate smaller clusters within the original MC9 (Sub-cluster 2 and Sub-cluster 7). These new findings have been described in a new paragraph in the revised manuscript "Immunophenotypical characterization of metacluster (MC) 9" and new supplementary figures have been added to the supplementary information (supplementary figure 7 and 8). When we investigated the immunophenotype per patient in Sub-C7 there is still a heterogeneous expression of surfacemarkers among the 32 AML patients (Response letter figure 42.1). We assume that the malignant clusters will never obtain the same homogeneity as we see in normal healthy cells (healthy cell clusters in supplementary figure 2). This is due to the heterogeneity of AML and one of the defining factor of malignant cells; they have aberrant surface marker expression. The surface markers might not be the best proxy for cell function, and they are expressed in a continuum over the course of the hematopoietic hierarchy. This continuum is very difficult to capture with clustering approaches and this
+
+<--- Page Split --->
+
+forces the cells into artificial clusters/metaclusters. Therefore, we chose an approach which underclusters the data and focused on the signaling.  
+
+![PLACEHOLDER_35_0]
+  
+
+42.2 To address the exact immunophenotype of cells with high pERK1/2 in MC9 we did a manual gating strategy where we gated the positive and negative pERK1/2 cells in MC9 for each patient. The results are shown in the new supplementary figure 7. Both the pERK1/2 positive cells and the negative had a heterogenous immunophenotype. AXL, CD90 and CD56 were significantly higher in pERK1/2 positive cells (grouped students t-test) and when we did a paired students t-test CD34 was also significantly higher in pERK1/2 positive cells. This has been described in the revised version of the manuscript (line 310).
+
+<--- Page Split --->
+![PLACEHOLDER_36_0]
+  
+
+43. Regarding the RNAseq vs MS (Super-SILAC) results -> were none of the RNA observations confirmed in proteomics? I.e. were those candidates not detected in MS or did they show different results? A bit more discussion on the value of bulk MS vs the single cell CyTOF would also help strengthen the reasoning for including both in the study, and help follow-up work to decide for one or the other vs both.  
+
+43.1 We have investigated if any of the significant genes in our RNAseq analysis could be detected in the proteomics. For the genes shown in figure 5d, only HSP90AA1 could be detected, the other ones were not detectable in our MS data. Among the AP-1 family genes shown in supplementary figure 13b, neither FOLS1 or ATF3 could be detected in the proteomics, neither could FOS, FOSB, ATF3, JUN or MAFF. FOLS-2 was found in the proteomics data but only in 7/15 patients. JUNB was only present in 6/15 patients. JUND was the only member of the AP-1 family that were present in most patients in our
+
+<--- Page Split --->
+
+proteomics data. (Response letter figure 43.2)  
+
+Response letter figure 43.2  
+
+![PLACEHOLDER_37_0]
+  
+
+43.3 In our unsupervised analysis between the 24h low and high pERK groups with an FDR cut- off at 0.05 (new supplementary figure 11), we found 76 significant genes in our RNA seq data. We did the same with our proteomics data, low vs high 24h- pERK1/2 group with an FDR cut off at 0.05 and found 193 significant proteins. IGF2BP2 (Insulin Like growth Factor 2) MRNA binding protein 2) was the only gene/protein that was significant in both the proteomics and RNAseq data analysis. It was high in high group in both the RNAseq data and the proteomics data. IGF2BP2 is an m6A reader gene and have been shown to have a negative prognostic value in AML.  
+
+44. I would like to request more detail on the in vitro drug screening experiments. E.g. what media was used, any growth factors or stromal cells, etc). Can they also comment on overall viability of the patient samples once put in vitro? As primary AML is notoriously difficult to culture in vitro while maintaining their hierarchical nature, it would be nice have this explored in a bit more detail. Their results were very encouraging, and could suggest potential alternative treatment strategies for those patients not responding to standard-of-care.  
+
+44.1 We apologies that the methodology of the DSRT experiments were not described in sufficient detail. A more detailed description of the methodological approach is added to the methods section of the revised manuscript line(853). MCM media was used, supplemented with 1%PS, no growth factors or stromal cells were used. The viability of the patient samples in vitro has not been assessed.
+
+<--- Page Split --->
+
+Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+The authors addressed my concerns in great detail. I am still troubled by the fact that it is not conclusively shown that the relevant cells in MC9 are malignant, but the authors' inference does seem reasonable.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+My questions are well answered.  
+
+Reviewer #3:  
+
+Remarks to the Author:  
+
+A very good quality review was carried out by the authors. All the issues raised have been addressed in this revised version and have led the authors to make changes that are definitely adapted to the weaknesses raised. The limitations of the work have been discussed; conclusions that are not supported by the data do not appear anymore. However, there is still a problem with question 25. In the results of the multivariate analysis, hazard ratios are negative for the WBC count as well as allogeneic stem cell transplantation. This part must be reviewed carefully by a statistician.  
+
+Reviewer #4:  
+
+Remarks to the Author:  
+
+Tislevoll et al have clearly taken into consideration the comments of the reviewer collective, and significantly enhanced the quality of their manuscript. The additional analyses conducted, clarifications provided and especially the in- depth analysis of MC9 and what cell- types might exist therein, have all contributed to raising the impact of this manuscript even further.  
+
+From my perspective, all my concerns have been addressed and I would recommend publication of the manuscript in its current form.  
+
+Reviewer #5:  
+
+Remarks to the Author:  
+
+Here are my concerns for the statistical analysis and, in particular, survival analysis.  
+
+The biggest concern is that the study would be severely underpowered with only 32 AML patients, among whom 20 deaths  
+
+were observed. With this sample size, it would be difficult to obtain an assertive statement that "Chemotherapy- induced changes in intracellular signaling during the first 24 hours of 233 treatment predicts long- term survival" as the obtained associations were likely to be spurious.  
+
+Second, all the p- values were without multiple comparison corrections and, as a result, the type I error (or more precisely the family wise error) could not be controlled.
+
+<--- Page Split --->
+
+Third, the machine learning results were not described clearly. How did you set up your training and validation samples? Did you do cross- validation? Normally, machine learning helps you generate hypotheses to test in future studies. I am not sure how machine learning strengthens your results here.
+
+<--- Page Split --->
+
+## Response letter to reviewers  
+
+## Reviewer #1 (Remarks to the Author):  
+
+1. The authors addressed my concerns in great detail. I am still troubled by the fact that it is not conclusively shown that the relevant cells in MC9 are malignant, but the authors' inference does seem reasonable.  
+
+1.1 We thank the reviewer for the positive feedback.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+2. My questions are well answered.  
+
+2.1 We are glad the reviewer found our revision to be satisfying.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+3. A very good quality review was carried out by the authors. All the issues raised have been addressed in this revised version and have led the authors to make changes that are definitely adapted to the weaknesses raised. The limitations of the work have been discussed; conclusions that are not supported by the data do not appear anymore. However, there is still a problem with question 25. In the results of the multivariate analysis, hazard ratios are negative for the WBC count as well as allogeneic stem cell transplantation. This part must be reviewed carefully by a statistician.  
+
+3.1 We thank the reviewer for this comment and sincerely apologize for our mistake regarding the calculations of the hazard ratio. The hazard ratios we have written in our manuscript were log hazard ratios. This explains the negative values. We have now added log to all hazard ratios in the revised manuscript.  
+
+## Reviewer #4 (Remarks to the Author):  
+
+4. Tislevoll et al have clearly taken into consideration the comments of the reviewer collective, and significantly enhanced the quality of their manuscript. The additional analyses conducted, clarifications provided and especially the in-depth analysis of MC9 and what cell-types might exist therein, have all contributed to raising the impact of this manuscript even further.  
+
+From my perspective, all my concerns have been addressed and I would recommend publication of the manuscript in its current form.  
+
+4.1 We thank the reviewer for these encouraging words.  
+
+## Reviewer #5 (Remarks to the Author): Expert in biostatistics  
+
+Here are my concerns for the statistical analysis and, in particular, survival analysis.  
+
+5. The biggest concern is that the study would be severely underpowered with only 32 AML
+
+<--- Page Split --->
+
+patients, among whom 20 deaths were observed. With this sample size, it would be difficult to obtain an assertive statement that "Chemotherapy- induced changes in intracellular signaling during the first 24 hours of 233 treatment predicts long- term survival" as the obtained associations were likely to be spurious.  
+
+5.1 We agree with the reviewer, the cohort size is unfortunately small. This is mainly because the patient material we present in this study is unique. Samples are collected shortly after start of treatment and processed using fix/lysis right after sample collection to preserve the phosphorylation status of intracellular signaling proteins. This requires immediate on- site sample processing by qualified personnel in appropriately equipped labs, which limits sample collection significantly. Additionally, AML is a relatively rare diagnosis, with approximately 150 new cases diagnosed per year in Norway (population: 5,4 million). Thus, to our knowledge, no other comparable datasets of AML patient material exist to date.  
+
+The small sample size was a major concern when deciding on an analytical approach to this dataset. Indeed, the selection of LASSO regression over many other machine learning algorithms (e.g., deep networks) was specifically to deal with the cohort size problem. LASSO has reported advantages to avoid overfitting, and thus it is an ideal method for smaller datasets. This method has also been used by others in similar studies with comparable sample sizes. A typical example is the seminal study published by Good et al in Nature Medicine 2018.  
+
+Of note, to test the stability of our findings, two additional AML patients (namely P33 and P34, treated with dose- reduced induction therapy) were added to the FlowSOM and LASSO cox regression analysis of the first 32 patients. The result of this second analysis was that p- ERK1/2 \((p = 0.0019\) , \(p - adj = 0.0038\) , Log- HR 1.25) and \(p - p38\) \((p = 0.0020\) , \(p - adj = 0.004\) , Log- HR 2.07) in MC 9 at 24 hours was significant at predicting two- year survival. When analyzing five- year survival with P33 and P34 included, significance was identified in p- ERK1/2 at 24h in myeloid blasts (MC9) \((p = 0.0011\) , \(p - adj = 0.0022\) , Log- HR 1.29) and p- Rb at 24 hours in the B- cells (MC5) \((p = 0.003\) , \(p - adj = 0.006\) , Log- HR 1.67) (Material and methods, line 836). We acknowledge that this is a small validation set, but these analyses indicate that the results are indeed stable in a slightly expanded patient cohort.  
+
+Nonetheless, we agree with the reviewer that the cohort size is small, and even though the results might be true for this particular small cohort of AML patients it may not apply to all AML patients. The results have to be validated in future studies of larger AML cohorts. We have made the following edits to the revised manuscript in order to make the readers aware of this limitation of the study:  
+
+i) "Chemotherapy-induced changes in intracellular signaling during the first 24 hours of treatment may predict long-term survival" (line 232)  
+
+ii) "One limitation of our study is the small cohort size, and follow-up studies with larger patient cohorts are required to further validate our findings." (line 694).
+
+<--- Page Split --->
+
+As a final note, we would like to point out that low statistical power increases the risk of type II error, not type I error; i.e. if the analysis did not return any statistically significant results, we would not be able to claim that there were no associations. However, our results are highly significant, and therefore we believe that the observed associations are not spurious, and should be tested with the available validation approaches.  
+
+6. Second, all the p-values were without multiple comparison corrections and, as a result, the type I error (or more precisely the family wise error) could not be controlled.  
+
+6.1 We have added the adjusted p-values for all analyses involving multiple testing in the revised manuscript.  
+
+7. Third, the machine learning results were not described clearly. How did you set up your training and validation samples? Did you do cross-validation?  
+
+7.1 The machine learning approach was validated using (nested) cross-validation. As described in the manuscript: "For survival analyses, we applied a Cox Lasso regression model with automatic feature selection and nested leave-one-out cross-validation to determine the regularization parameter." (Material and methods- line 820) and main text (line 237). Since we used leave-one-out cross-validation and only had 32 patients, we could use all possible one-patient subsets as test samples, all one-patient subsets as validation samples, and all 30-patient subsets as training data.  
+
+We agree that the description of the machine learning approach should be improved. To provide a better description of the nested leave-one-out cross validation, we added a supplementary figure (Supplementary Figure 5) to the revised manuscript. We have also added some additional text to the methods section (line 821)
+
+<--- Page Split --->
+![PLACEHOLDER_43_0]
+
+<center>Supplementary Figure 5 </center>  
+
+As the reviewer points out, we have used machine learning to generate hypotheses from a complex mass cytometry dataset. We used it to dissect the data, by linking single cells to predictive markers. Our aim was to present this as a framework that can be used in future larger studies, not only for AML but possibly also other forms of cancer.
+
+<--- Page Split --->
+
+8. Normally, machine learning helps you generate hypotheses to test in future studies. I am not sure how machine learning strengthens your results here.  
+
+8.1 We do not intend to claim that machine learning strengthens our results, but it was necessary for the discovery of phospho-ERK1/2 in a putative model for prediction of survival. As pointed out by the reviewer, CyTOF data are multidimensional and complex. Thus, the downstream analysis requires effective computational methods that can handle such data. In this case, we used lasso regression - as is commonly used in the mass cytometry literature.  
+
+Of note, the hypothesis generated by the machine learning algorithm was indeed validated using other methods in this study. First, manual analysis of the mass cytometry dataset confirmed the association of phospho- ERK1/2 with survival in the disease- specific cell population. Next, PB samples from a subset of the patients were also analyzed by RNAseq, proteomics, and drug sensitivity and resistance testing. These independent analyses also indicated activation of the ERK1/2 signaling pathway in the group of patients identified as "pERK1/2 high" in the initial analysis. In our opinion, the application of several independent experimental approaches to inspect the ERK1/2 signaling pathway at both the RNA-, protein- and functional levels strengthens our findings.
+
+<--- Page Split --->
+
+Reviewers' Comments:  
+
+Reviewer #3:  Remarks to the Author:  My last question has been well answered.  
+
+Reviewer #5:  None
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a3aec9dfee94996570423cfe6ec1e55b9660040642ae120e923878928dc3ac1a/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__a3aec9dfee94996570423cfe6ec1e55b9660040642ae120e923878928dc3ac1a/supplementary_0_Peer Review File__a3aec9dfee94996570423cfe6ec1e55b9660040642ae120e923878928dc3ac1a.mmd

@@ -0,0 +1,435 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Experimental confirmation of driving pressure boosting and smoothing for hybrid- drive inertial fusion at the 100- kJ laser facility  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+I believe that this paper is very interesting and it is an innovative and substantial contribution to the field of Inertial Confinement Fusion. Publication of this article would be very timely seen the recent breakthrough in laser fusion obtained at NIF.  
+
+In general I have only a few remarks which I would ask the authors to reply before publication.  
+
+1) in the conclusions the authors wriest that "Using DD laser with intensities \(\sim (1.8 - 2.0) \mathrm{PW / cm^2}\) , ... we stably obtained the boosted HD pressure \(P \approx 150 - 155\) Mbar about 3.5-3.6 times the radiation ablation pressure experimentally"  
+
+Well, this indeed the range of pressure which can be obtained in direct- drive or similar intensities, using e.g. Lindl's scaling. So in this case the contribution from the indirect drive part of the scheme does not seem to be really apparent. Con the authors comment on this?  
+
+2) While the comparison between the HD scheme and the "traditional" DD scheme is well described, I still have doubts on how the HD scheme compares to the Shock Ignition approach. SI is also based on a two step process with a first "compression" laser pulse and a final "ignition" laser spike.  
+
+The authors write that "the distance between the electron ablation front and critical surface for the traditional DD system is too short". Does this conclusion also apply to shock ignition?  
+
+3) Does this conclusion also apply to the proposed scheme of "foam buffered target" in which the pellet is surrounded by a low density material (foam) to create a long scale length plasma assumed to be able to smooth te non-uniformities thanks to the long stand-off distance available to thermal smoothing?  
+
+4) what is the measured "conversion efficiency for thermal X-rays" reported at the end of page 5 (section "Experimental target design")  
+
+5) the authors speak about a "hot electron energy fraction of the DD lasers in laser-plasma interaction in HD..." of about \(2\%\) and they say that this is "significantly lower than that in the traditional DD laser". However I found no measurement or estimation of the HE temperature. This is indeed an additional important parameters because if the HE are not too energetic, they will not be able to penetrate deeply into the fuel, and so their armful effect will be strongly mitigated.
+
+<--- Page Split --->
+
+6) Also, what is the fraction of HE produced during the ID phase?  
+
+7) The authors write that SRS amounts to about \(0.5\%\) . Since in these experimental conditions SRS is probably the main source of HE, how is this result compatible with a HE conversion efficiency of \(2\%\) ??  
+
+Reviewer #2 (Remarks to the Author):  
+
+This manuscript reports on simulation and experimental work on the hybrid- drive (HD) approach to inertial confinement fusion (ICF). This approach uses a spherical hohlraum and long- pulse laser drive to provide an 'indirect' X- ray drive to the ICF capsule, combined with a short pulse indirect laser drive applied sometime after the start of the long pulse. If tuned correctly, the relativistic electron heat wave generated by the short pulse laser slows to form a compression wave in the material ablated by the indirect drive. The net effect is a stable plateau in density between the indirect- drive ablation front and direct- drive critical surface, increasing the drive pressure on the ICF capsule and thermally smoothing perturbations. It is hoped that his hybrid approach will provide favorable scaling to ignition conditions and an effective path to high fusion yield from ICF implosions.  
+
+Inertial fusion experiments have been widely reported recently as a result of significant advances in a related approach using only a direct drive. This makes ICF experiments, and alternative paths to ignition, a timely contribution and of high interest to a wide audience. However, the work presented here is an incremental change to work that has already been reported; the HD approach has been widely described over several decades, and nearly identical simulations and experiments have been published recently (High Energy Density Physics 96, 100804). The only difference between this submission and the previously reported results is a slight increase in direct drive energy (4.0 KJ vs 3.6 KJ); the key results – observation of pressure boosting, peak radiation temperature, and laser backscatter - are unchanged.  
+
+Given the incremental nature of this work, I do not find it suitable for publishing in Nature Communications. However, the topic of hybrid- drive, and new experimental results for ICF, remains highly impactful and relevant to the Nature readership. I therefore recommend that the authors re- write the manuscript to avoid repeating previously published work, and to identify the novel and high- impact results of the present work. I also suggest that the authors pay attention to the clarity and conciseness of the new manuscript, since I found the current submission difficult to read and somewhat unclear in some places.
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+Dear Editor,  
+
+This manuscript reports on initial experiments at SG- III using a hybrid- drive (HD) scheme designed to take advantage of the drive smoothing provided by indirect- drive (ID), and the efficient absorption of laser energy provided by direct- drive (DD). The manuscript presents important proof- of- principle experiments for the HD scheme that can be used to validate modeling capabilities and assess the LPI threat. The manuscript is well organized and provides an appropriate level of detail and could be appropriate for publication with some changes. My specific comments follow:  
+
+1) Although the manuscript is well-written from the point-of-view of the results being presented in a logical and coherent manner, there are a large number of grammatical errors that make it difficult to read.  
+
+2) The abstract (and elsewhere) states that the HD shock "stops the asymmetric ID shock." I am not sure what this statement is supposed to mean, but certainly it is not meant to be taken literally.  
+
+3) As the HD scheme is proposed as an alternative to DD/ID, it would help if the manuscript was more explicit about the advantages/disadvantages of the HD scheme relative to those schemes. For example, the HD scheme provides increased ablation pressure relative to ID, but how does the pressure compare to DD alone?  
+
+4) In the caption of Fig. 1 LHEs should be LEHs.  
+
+5) Figure 2(c) shows an AI layer, but in the text this is referred to as a Mo layer (the figure caption says that it can be Mo or AI).  
+
+6) The manuscript states that the measured radiation temperatures agreed with predictions. It would be nice to include the predicted radiation temperature in Fig. 3(a).  
+
+7) What does the colormap represent in Fig. 4(a)?
+
+<--- Page Split --->
+
+8) In the first sentence on page 9 "reappearing" should be "reproducing".  
+
+9) In the 2nd paragraph on page 11 "75J for DD" should be "75J for HD".  
+
+10) The manuscript states in regards to the reduction in hot-electron fraction and backscatter fraction for the HD scheme: "Obviously, the thermal smoothing effect in HD significantly reduces the fractions compared with the traditional DD." This is not at all obvious to me. Traditionally I think of thermal smoothing as mitigating hydrodynamic instabilities, not laser plasma instabilities. The authors should explain how thermal smoothing is mitigating LPI in this case.  
+
+11) Despite the mitigation relative to DD, the backscatter and hot-electron production in the HD scheme is still quite high and is likely to get worse when scaling up to a high-gain design. Are these levels believed to be acceptable for the HD design or are there plans for mitigation?  
+
+12) What is the f-number of the drive beams, as this has a significant impact on SBS/SRS backscatter?
+
+<--- Page Split --->
+
+## Response Letter  
+
+Article ID: NCOMMS- 22- 52906  Title: Experimental confirmation of drive pressure boosting and smoothing for hybrid- drive inertial fusion at the 100- kJ laser facility  Authors: Ji Yan, Jiwei Li, X. T. He, et al.  
+
+We are grateful to the three referees for their insightful reviews and valuable comments on our manuscript. According to their comments, we have carefully revised the manuscript. Following are the specific responses to the referees' comments point by point.  
+
+### I. Responses to Referee #1  
+
+1. Comment: "in the conclusions the authors wriest that "Using DD laser with intensities \(\sim (1.8 - 2.0)\) PW/cm², ..., we stably obtained the boosted HD pressure \(P \approx 150 - 155\) Mbar about 3.5-3.6 times the radiation ablation pressure experimentally." Well, this indeed the range of pressure which can be obtained in direct-drive or similar intensities, using e.g. Lindl's scaling. So in this case the contribution from the indirect drive part of the scheme does not seem to be really apparent. Can the authors comment on this?"  
+
+## Response:  
+
+We understand the reviewer's concern. Yes, under the current laser drive condition of the direct- drive (DD) laser energies of \(3.6 - 4\mathrm{kJ}\) only, the hybrid- drive (HD) pressure of 150- 155 Mbar in our HD scheme is similar to that in the DD scheme. However, the physics is completely different. In our HD scheme, the important roles of the indirect drive (ID) lasers are to produce an ID corona plasma background and a large enough distance \(\Delta R_{ID}\) between the critical surface and radiation ablation front before the DD laser arriving, and also a radiation ablation pressure at the radiation ablation front in addition to pre- compression of the fuel. Afterwards, the DD laser with the intensities of \(I_{L} \sim (1.5 - 2.0)\) PW/cm² then is absorbed near the critical surface and converted into a supersonic- electron- heat- conduction wave (below, abbreviated the SEC wave). The SEC wave propagates within the large \(\Delta R_{ID}\) from the critical surface toward the radiation ablation front and slows down to a plasma compressive wave while smoothing. This compressive wave isothermally compresses low ID corona plasma density \(\rho_{ID}\) into high HD density \(\rho_{HD}\) , which can be fitted as the function of the ID laser energy \(\sim E_{DD}^{1 / 4}\) (see the manuscript), to change the radiation ablation pressure \(P_{ID} = \Gamma \rho_{ID}T_{r}\) at the radiation ablation front into a new smoothed HD pressure \(P_{HD} = \Gamma \rho_{HD}T_{r} \propto E_{DD}^{1 / 4}T_{r}\) , where \(E_{DD}\) is DD laser energy and \(T_{r}\) is radiation temperature. Therefore, we see that the first ID process plays a key role in our HD scheme, which results in that the HD pressure scales as both the DD laser energy \(E_{DD}^{1 / 4}\) and the radiation temperature, while in the DD scheme, the pressure only scales as \(I_{L}^{2 / 3}\) .  
+
+For the CH ablator, the HD pressure achieved in our HD scheme can be calculated as \(P_{HD}(\mathrm{Mbar}) \approx 62 E_{DD}^{1 / 4}T_{r}\) , while the DD pressure achieved in the DD scheme (without ID) is \(P_{DD} \approx 90 I_{L}^{2 / 3}\) . Under \(\sim\) current laser conditions in our experiments, the DD laser energy is \(E_{DD} = 3.6 - 4\mathrm{kJ}\) , \(I_{L} = 1.8\) PW/cm², and \(T_{r} = 200\mathrm{eV}\) , the pressure in our HD scheme is about \(P_{HD} \sim 155\) Mbar, which is indeed close to that in the DD scheme \(P_{DD} \sim 140\) Mbar. However, when we scale up to the ignition condition with the increased DD laser energy, such as \(E_{DD} = 825\mathrm{kJ}\) , the pressure in our HD scheme can be much boosted (due to the role of the ID) up to \(P_{HD} \sim 775\) Mbar, while that in the DD scheme still keeps to be as only 140 Mbar, since
+
+<--- Page Split --->
+
+the laser intensity is similar as \(I_{L} = 1.8 \mathrm{PW / cm^2}\) .  
+
+In the present revised manuscript, we also added a hemispherical ablator target performed in a recent experiment and under the same DD laser energy of \(3.6 - 4.0 \mathrm{kJ}\) and obtained the HD pressure \(P_{HD} \sim 170 - 180 \mathrm{Mbar}\) higher than \(\sim 150 - 155 \mathrm{Mbar}\) in the planar ablator target. These results completely verified the HD pressure boosting and smoothing effects- the heart of the HD scheme  
+
+2. Comment: "While the comparison between the HD scheme and the "traditional" DD scheme is well described, I still have doubts on how the HD scheme compares to the Shock Ignition approach. SI is also based on a two-step process with a first "compression" laser pulse and a final "ignition" laser spike. The authors write that "the distance between the electron ablation front and critical surface for the traditional DD system is too short". Does this conclusion also apply to shock ignition?"  
+
+## Response:  
+
+The two- step process of ID and DD in our HD scheme is inherently different from the "compression" and "ignition" in the DD scheme. This conclusion also applies to SI which is under the DD scheme. As explained in the manuscript, in HD, in the first step, the ID distance \(\Delta R_{ID}\) offered in advance by the ID laser is about \(250 - 300 \mu \mathrm{m}\) for radiation temperature \(T_{r} = 200 \mathrm{eV}\) , which is large enough for the SEC wave to slow down to a compressive wave thermally compressing the ID corona plasma density into high plasma density to make the HD pressure far higher than the radiation ablation pressure. However, in SI, no ID laser is used in advance, and thus of course there exists no large ID distance \(\Delta R_{ID}\) before the DD laser arrives at the critical surface. Although in the later stage of SI, the DD distance \(\Delta R_{DD}\) between the critical surface and electron ablation front offered by the pre- pulse DD laser ablating the ablator surface is getting to increase, about \(100 - 150 \mu \mathrm{m}\) for a model in Fig. 1 in Nucl. Fusion 54, 054004(2014), due to the mass ablation rate for electrons being smaller than for radiation, under the same laser intensity \(\Delta R_{DD}\) is shorter than \(\Delta R_{ID}\) , resulting in the SEC wave driven by the DD laser directly hitting the capsule without enough time slowing down to a compressive wave. Therefore, the electron ablation pressure \(P_{DD} \sim I_{L}^{2 / 3}\) can't be smoothed very well, which leads to an asymmetric and unstable implosion in the early stage of the implosion. In addition, in SI, the driving pressure (electron ablation pressure) \(P_{DD}\) over \(300 \mathrm{Mbar}\) is required for the hotspot ignition (Nucl. Fusion 54, 054004(2014)) in which the DD laser intensity of \(I_{L} \sim 5 \mathrm{PW / cm^2}\) is required. For such intensity, the velocity of the SEC wave slowing down to the compressive wave should be larger than that for \(I_{L} = 1.8 \mathrm{PW / cm^2}\) in HD, and therefore, the DD distance \(\Delta R_{DD}\) has too far larger than \(250 - 300 \mu \mathrm{m}\) .  
+
+3. Comment: "Does this conclusion also apply to the proposed scheme of "foam buffered target" in which the pellet is surrounded by a low-density material (foam) to create a long scale length plasma assumed to be able to smooth to non-uniformities thanks to the long stand-off distance available to thermal smoothing?"  
+
+## Response:  
+
+In HD, the distance \(\Delta R_{ID}\) offered in advance by the ID laser does at least two things, that is, the SEC wave will be slowed down to a compressive wave for boosting the HD pressure and thermally smoothed very well.  
+
+To our understanding, the foam (low density but above critical density) indeed removes the critical surface of the ablator in the pellet to the foam surface and is available for thermal smoothing well in the form \(\exp (- \ell \Delta R / R_{cr})\) if the foam thickness \(\Delta R\) is larger than the perturbation wavelength \(\lambda = 2\pi R_{CR} / \ell\)
+
+<--- Page Split --->
+
+from the DD laser imprinting and beam overlapping, \(\ell\) is the perturbative mode number, and \(R_{CR}\) is the radius of the critical surface.  
+
+However, the foam is difficult to offer a large distance, like \(\Delta R_{ID}\) in HD, since it will significantly reduce laser energy to the pellet.  
+
+4. Comment: "what is the measured "conversion efficiency for thermal X-rays" reported at the end of page 5 (section "Experimental target design")"  
+
+## Response:  
+
+Thanks for the reviewer's correction. It refers to the conversion efficiency \(\eta\) of the ID laser to thermal X- rays. The x- ray conversion efficiency is estimated based on the hohlraum power balance equation. The equation can be described as follow:  
+
+\[\eta (P_{laser} - P_{backscatter}) = [A_{wall}(1 - \alpha) + A_{LEH} + F_{CH}A_{CH}]\sigma T_r^4\]  
+
+where \(P_{laser}\) and \(P_{backscatter}\) are the incident laser power and the backscattered laser power due to LPI effect, respectively. \(A_{wall}\) , \(A_{LEH}\) and \(A_{CH}\) represent the surface areas of the hohlraum wall, laser entrance hole and CH sample, respectively. \(\alpha\) is the albedo of the Hohlraum and inferred by analytical model \((\alpha = 1 - 0.32T(100eV)_{r}^{- 0.7}\tau^{- 0.382})\) . \(F_{CH}\) represents the ratio of the absorbed flux over the incident flux of capsule and approximately as a constant and equal to \(0.7\sigma\) is the Stefan- Boltzmann constant. \(T_{r}\) is the radiation temperature which measured by calibrated flat response x- ray diode (FXRD). The uncertainty of radiation temperature using FXRD is \(3\%\) . In our experiments, the conversion efficiency is \(89\pm 8\%\) .  
+
+5. Comment: "the authors speak about a "hot electron energy fraction of the DD lasers in laser-plasma interaction in HD..." of about \(2\%\) and they say that this is "significantly lower than that in the traditional DD laser". However, I found no measurement or estimation of the HE temperature. This is indeed an additional important parameter because if the HE are not too energetic, they will not be able to penetrate deeply into the fuel, and so their armful effect will be strongly mitigated."  
+
+## Response:  
+
+Thanks for the reviewer's comments, we have added a diagram of the measured HE temperature in the present revised manuscript (see Fig. 5c). In our experiment, the HE temperature is about 21- 22 keV. The production of HE is mainly from two- plasmon decay (TPD) since the backscattering fraction for SRS is only \(\sim 0.4\%\) experimentally.  
+
+The simulation shows that for the modeling ignition target, the area density of the remainder of the ablating CH ablator is about \(0.06 \mathrm{g} / \mathrm{cm}^2\) at the time of the maximal implosion velocity in which \(0.02 \mathrm{g} / \mathrm{cm}^2\) is in the density plateau piled up by the isothermal compression. On the other hand, the HE's range can be expressed in the form \(\rho R(\mathrm{g} / \mathrm{cm}^2) \approx 0.59 \times 10^{- 5} E_e^{1.661}\) for \(E_e > 2 \mathrm{keV}\) in the CH material (Phys. Plasmas 18, 022703(2011)), where \(E_e\) is the HE energy in \(\mathrm{keV}\) . We see from the expression and numerical result that all hot electrons with energy less than \(250 \mathrm{keV}\) are completely trapped in the remaining CH, impossible to reach the fuel. Obviously, for the HE temperature of 21- 22 keV the number of the HEs with energy greater than \(250 \mathrm{keV}\) can be neglectable and they are impossible to preheat the fuel significantly.  
+
+6. Comment: "Also, what is the fraction of HE produced during the ID phase?" Response:
+
+<--- Page Split --->
+
+A calibrated filtered- Fluorescence (FF) hard x- ray spectrometer was employed to monitor the fraction of the hot- electrons. In HD shots, the measured hot- electrons include both ID phase and DD phase. But however, in ID only shots, the measured energy of hot- electrons is lower than 5 J (the detectable threshold of the spectrometer) and the fraction of HE produced during the ID phase is around \(5 \mathrm{~J} / 43 \mathrm{kJ} = 0.01\%\) , since the ID laser energy is \(43 \mathrm{~kJ}\) in our experiment.  
+
+In the HD experiment, the ID laser with an averaged intensity of about \(0.4 \mathrm{PW} / \mathrm{cm}^2\) and energy of 40- 50 kJ is coupled in the HD plasma with the DD laser with an intensity of \(1.5 - 2 \mathrm{PW} / \mathrm{cm}^2\) and energy of \(3.6 - 4.0 \mathrm{kJ}\) launching in a later pulse duration of the ID laser, and therefore, it is difficult to separate the HE energy from the ID laser and DD laser. We have measured HE produced by the ID laser energy of \(43 \mathrm{~kJ}\) without the DD laser, and its energy is about \(3 \mathrm{~J}\) , which means that the energy fraction of HE produced by the ID laser is \(3 \mathrm{~J} / 43 \mathrm{~kJ} = 0.0067\%\) .  
+
+7. Comment: "The authors write that SRS amounts to about \(0.5\%\) . Since in these experimental conditions, SRS is probably the main source of HE, how is this result compatible with a HE conversion efficiency of \(2\%\) ?"  
+
+## Response:  
+
+There should be some misunderstanding. As mentioned in our manuscript, in our HD scheme, as the laser intensity \(I_L \sim 1.8 \mathrm{PW} / \mathrm{cm}^2\) exceeds the threshold of two- plasma decay (TPD), TPD, instead of SRS, is the main source of HE. That is the reason why SRS amounts to only about \(0.5\%\) . The main component of HE (about \(2\%\) ) comes from TPD. In the HD experiment with the DD laser intensity of \(\sim 1.8 \mathrm{PW} / \mathrm{cm}^2\) , we measured the SRS fraction of about \(0.4\%\) on SG- III and compared it to the results on OMEGA under the DD scheme with the same intensity, as seen in Fig. 20 in the reference No. 23 and in Fig. 21 in the reference No. 24 in our manuscript, where the SRS fraction is about 2.5- 3 times the fraction of \(0.4\%\) in the HD experiment and SBS dominated. Considering that the radiation ablation offered a uniform ID corona plasma before the DD laser arrives, different from the DD laser- plasma interaction (LPI) in the DD scheme, the thermal smoothing effect on the ID corona plasma may result in the decrease of the SRS fraction in the HD experiment. So, the SRS fraction on SG- III may be reasonably compared with the results in OMEGA. Such a low SRS fraction in HD is unable to offer the HE energy fraction of \(\sim 2\%\) for the DD laser energy of \(3.6 - 4.0 \mathrm{~kJ}\) . Therefore, we think the HE fraction mainly comes from two- plasmon decay (TPD) since this intensity of \(I_L \sim 1.8 \mathrm{PW} / \mathrm{cm}^2\) exceeds the threshold of TPD.  
+
+At electron temperature of \(T_e = 2 - 3 \mathrm{keV}\) and the electron number density of \(n_e / n_c \sim 0.2 - 0.7\) , the maximal linear growth rate for TPD is \(\sim \gamma /\omega \sim 10^{- 2} / (n_e / n_c)\) , where \(n_c\) the critical number density, the laser frequency \(\omega_0 \approx 5.4 \times 10^{15} / s\) for the wavelength of \(0.35 \mu \mathrm{m}\) . Therefore, at a quarter of the critical density, we have the maximal linear growth rate of \(\gamma \sim 0.2 / \mathrm{fs}\) , and the occurrence of TPD is completely possible soon after the DD laser arrives. Further specialized investigation of course is necessary.  
+
+## II. Responses to Referee #2  
+
+1. Comment: "Inertial fusion experiments have been widely reported recently as a result of significant advances in a related approach using only a direct drive. This makes ICF experiments, and alternative paths to ignition, a timely contribution and of high interest to a wide audience. However, the work presented here is an incremental change to work that has already been reported; the HD approach has been widely
+
+<--- Page Split --->
+
+described over several decades, and nearly identical simulations and experiments have been published recently (High Energy Density Physics 96, 100804). The only difference between this submission and the previously reported results is a slight increase in direct drive energy (4.0 KJ vs 3.6KJ); the key results - observation of pressure boosting, peak radiation temperature, and laser backscatter - are unchanged."  
+
+## Response:  
+
+Thanks for the referee's comments. We now explained those as follows.  
+
+The hybrid- drive content is only a portion of the conference proceedings review paper issued in HEDP, where we simply presented with DD laser energy of \(3.6\mathrm{kJ}\) one early experiment data of the HD pressure of \(\sim 150\) Mbar without more direct experimental evidence and the detailed analyses to confirm the reliability of HD pressure boosting and smoothing, which are the key effects for our HD scheme.  
+
+In our previous manuscript, we analyzed new and published experimental data in detail to demonstrate and confirm the HD pressure boosting and smoothing effects.  
+
+Firstly, with the new DD laser energy of \(4.0\mathrm{kJ}\) and a Mo shield instead of Al, we obtained the new experimental result of the HD pressure of \(\sim 155\) Mbar, and with DD laser energy of \(3.6\mathrm{kJ}\) checked the result of \(\sim 150\) Mbar in the paper issued in HEDP. Thus, we further confirmed the HD pressure- boosting effect physically with more experimental data. In addition, we first showed HD pressure smoothing and observed the phenomenon of the HD shock merged with the ID shock (Fig. 3c), both are important for stable implosion. Second, we showed how under keeping the radiation temperature of \(200\mathrm{eV}\) unchanged the experimental target is designed by scaling down the size of the spherical hohlraum in the ignition target, which makes the experimental results can directly scale up to the ignition target experimentally. Finally, in the manuscript, we found that with a pre- offered ID corona plasma background of radiation temperature \(200\mathrm{eV}\) , tuning the distance between the critical surface and the radiation front and the slowing down length, the supersonic- electronic- heat wave converted by the DD laser intensity of \(1.8\mathrm{PW / cm}^2\) can provide a perfect bulldozer effect (a compressive wave) to generate significant HD pressure boosting and smoothing effect. we further discussed the time evolution process of indirect- driven and hybrid- driven shock velocities experimentally for the planar target and deduced the boosted and smoothed pressure in the density plateau by the simulations matched with experimental results in the quartz. This makes the manuscript a new paper essentially different from the paper in HEDP.  
+
+As for why the key results of peak radiation temperature and laser backscatter are unchanged. Because the peak radiation temperature of \(200\mathrm{eV}\) is the optimal temperature in the design of the ignition target, which can well provide the conditions of HD pressure boosting and smoothing, and also the right ID energy for the ignition target, in addition, according to the indirect- drive energy balance relationship, when the semicylindrical hohlraum used in the experiments is designed by scaling down the spherical hohlraum in the ignition target, the peak radiation temperature of \(T_{r} = 200\mathrm{eV}\) is the same as in the ignition target, which is beneficial to scaling up the experimental results to the ignition target. This is why peak radiation temperature \(T_{r} = 200\mathrm{eV}\) is unchanged in the review article in HEDP and the manuscript. In addition, the same radiation temperature heating the surface of the CH ablator leads to the same plasma environment for LPI, and therefore, for the DD laser intensity of \(\sim 1.8\mathrm{PW / cm}^2\) , the backscatter and the hot- electro energy should be close to those in similar experiments, which is plotted in Fig. 5c.  
+
+2. Comment: "Given the incremental nature of this work, I do not find it suitable for publishing in Nature Communications. However, the topic of hybrid-drive, and new experimental results for ICF, remains highly
+
+<--- Page Split --->
+
+impactful and relevant to the Nature readership. I therefore recommend that the authors re- write the manuscript to avoid repeating previously published work, and to identify the novel and high- impact results of the present work. I also suggest that the authors pay attention to the clarity and conciseness of the new manuscript, since I found the current submission difficult to read and somewhat unclear in some places."  
+
+## Response:  
+
+Although we think our previous manuscript is essentially different from the paper issued in HEDP, we, to respond to the reviewer's comment of "to avoid repeating previously published work, and to identify the novel and high- impact results of the present work", in rewriting the manuscript, added new experimental results with a hemispherical ablator target recently performed on SG- III. Under the same DD laser energies of \(\sim 3.6 \mathrm{kJ}\) and \(\sim 4.0 \mathrm{kJ}\) and radiation temperature of \(200 \mathrm{eV}\) , the hemispherical target due to the spherical convergence effect provided the peak HD pressure achieved 170- 180 Mbar, larger than 150- 155 Mbar of the planar target, as seen in Fig. 3c for the HD shock velocity and in Fig. 4c for the HD pressure of 180 Mbar. These results are the latest and greatest for the HD pressure experimentally to date and we discussed them in the text in detail.  
+
+We also tried to make some changes in the text to make the reading clear.  
+
+## III. Responses to Referee #3  
+
+1. Comment: "Although the manuscript is well-written from the point-of-view of the results being presented in a logical and coherent manner, there are a large number of grammatical errors that make it difficult to read."  
+
+## Response:  
+
+Sorry, we have made an effort to correct some grammatical errors in the present revised manuscript.  
+
+2. Comment: "The abstract (and elsewhere) states that the HD shock "stops the asymmetric ID shock." I am not sure what this statement is supposed to mean, but certainly it is not meant to be taken literally."  
+
+## Response:  
+
+We wrote it too simply in the abstract, more details are below:  
+
+A strong symmetric HD shock driven by the ideal HD pressure rapidly entering the imploding capsule collides in the opposite directions with the asymmetric relatively weak ID shock which is reflected from the center of the hotspot after pre- compressed the fuel and just arriving at the interface of the hotspot, and the ID shock reflected inward becomes weaker and is quickly caught up, swallowed and merged by the strong HD shock. Thus, the asymmetric ID shock in the early implosion stage is suppressed to prevent it from further asymmetric implosion.  
+
+We have made corresponding changes in the manuscript.  
+
+3. Comment: "As the HD scheme is proposed as an alternative to DD/ID, it would help if the manuscript was more explicit about the advantages/disadvantages of the HD scheme relative to those schemes. For example, the HD scheme provides increased ablation pressure relative to ID, but how does the pressure compare to DD alone?"  
+
+## Response:  
+
+In the DD scheme, its advantage is having a high conversion efficiency of laser energy to the capsule, however, due to the mass ablation rate for electrons smaller than that for radiation the electron- conduction
+
+<--- Page Split --->
+
+region between the critical surface and the electron ablation front is relatively narrow, which easily transfers the nonuniformities from laser imprinting and beam overlapping to the imploding capsule, especially in the early stage, resulting in hydrodynamic instabilities. In addition, at the electron ablation front, the electron ablation pressure driving implosion is \(P_{DD} = \Gamma \rho_{DD}T_e\) , where \(\Gamma\) is a pressure constant for an ideal gas, the electron corona density \(\rho_{DD}\) is approximately \(2\rho_{c}\) with \(\rho_{c}\) the critical density, and the electron temperature is \(T_{e} \propto I_{L}^{2 / 3}\) where \(I_{L}\) the DD laser intensity (Ref. 9 in the manuscript). For the CH ablator, the critical density is \(\rho_{c} \approx 0.03 \mathrm{~g / cc}\) , and therefore, \(P_{DD}(\mathrm{Mbar}) \approx 90 \mathrm{x} I_{L}^{2 / 3}\) for the laser wavelength of 0.35 micron, where \(I_{L}\) in units of \(\mathrm{PW} / \mathrm{cm}^{2}\) . It is seen from the above discussion that due to the expansion of the ablator surface by high- temperature ablating, the plasma density \(\rho_{DD}\) at the electron ablation front is low compared with the normal ablator density. Therefore, in order to boost the DD pressure, the only way is to increase the electron temperature \(T_{e}\) or the DD laser intensity \(I_{L}\) . As an example, In the DD scheme, if the pressure driving the implosion is required to be \(P_{DD} \approx 300 \mathrm{Mbar}\) (Batani et al., Nucl. Fusion 54, 2014), the laser intensity \(I_{L}\) must be over \(5 \mathrm{PW} / \mathrm{cm}^{2}\) , resulting in severe LPI. In addition, laser imprinting and beam overlapping would lead to asymmetric implosion and hydrodynamic instabilities.  
+
+As for HD, a coupling of ID and DD, we use the advantages of ID (large mass ablation rate for radiation) and DD (high efficiency) to improve their shortcomings involving hydrodynamic instabilities. As we explained in the manuscript that before the DD laser arrives, the ID laser has offered a large distance \(\Delta R_{ID}\) between the critical surface and the radiation ablation front, and then a supersonic- electron- heat wave converted by the DD laser with the intensity of \(I_{L} = (1 - 2) \mathrm{PW} / \mathrm{cm}^{2}\) propagates in \(\Delta R_{ID}\) and slows down to a compressive wave before reaching the radiation ablation front, while thermal smoothing. This compressive wave, like a "bulldoze, isothermally compresses low ID corona plasma density \(\rho_{ID}\) , between the compressive wave front and the radiation ablation front, into high HD plasma density \(\rho_{HD}\) (fitted to \(\sim E_{4D}^{1 / 4}\) , see the manuscript) far greater than \(\rho_{ID}\) , and therefore, by increasing the plasma density rather than radiation temperature \(T_{r}\) , the radiation ablation pressure is changed into a new smoothed HD pressure, much greater than the radiation ablation pressure, i.e., \(P_{HD} / P_{ID} = \rho_{HD} / \rho_{ID} \gg 1\) . Clearly, the HD pressure increases with the DD laser energy \(E_{4D}^{1 / 4}\) and radiation temperature \(T_{r}\) and is independent of the laser intensity. As an example, for the modeling ignition target with the radius \(5 \mathrm{mm}\) of the spherical hohlraum and radiation temperature \(T_{r} = 200 \mathrm{eV}\) , at the radiation ablation front the HD maximal pressure \(P_{HD}\) can reach as high as 775 Mbar for \(E_{DD} = 825 \mathrm{kJ}\) while the radiation ablation pressure is only \(P_{ID} \sim 43 \mathrm{Mbar}\) , and at the electron ablation front for the intensities of \(I_{L} = (1 - 2) \mathrm{PW} / \mathrm{cm}^{2}\) the DD pressure is \(P_{DD} \sim 90 - 158 \mathrm{Mbar}\) .  
+
+As shown in the experiment, with the target consisting of the semicylindrical hohlraum and the planar ablator, for the radiation temperature \(T_{r} = 200 \mathrm{eV}\) , only using the low DD laser energy of \(3.6 - 4 \mathrm{kJ}\) , we obtain the HD pressure of 150- 155 Mbar, about 3.5 times the radiation ablation pressure.  
+
+In the present revised manuscript, we added the new experimental results by using a new target consisting of the same semicylindrical hohlraum and a new hemispherical ablator. Using the same DD laser energy of 3.5- 4.0 kJ and the same radiation temperature of 200 eV, the new experimental results, due to the spherical convergent effect, achieve HD pressures of 170- 180 Mbar higher than the HD pressures of 152- 155 Mbar in the planar target. These results further verified the HD pressure boosting and smoothing effects- the heart of the HD scheme and are discussed in detail in the revised manuscript.  
+
+4. Comment: "In the caption of Fig. 1 LHEs should be LEHs."
+
+<--- Page Split --->
+
+## Response:  
+
+Thanks for the reviewer's correction, we have revised it in the present manuscript.  
+
+5. Comment: "Figure 2(c) shows an Al layer, but in the text this is referred to as a Mo layer (the figure caption says that it can be Mo or Al)."  
+
+## Response:  
+
+Sorry, we were careless. In the last manuscript, we did two rounds of experimental shots. In the first shot, with the DD laser energy of \(3.6\mathrm{kJ}\) , Al was used as the shielding layer as plotted in Figure 2(c), and in the second shot with the DD laser energy of \(4.0\mathrm{kJ}\) , the shielding layer was Mo. In the discussion, only Mo was used.  
+
+In the new experiments, with the DD laser energies of \(\sim 3.6\mathrm{kJ}\) and \(4.0\mathrm{kJ}\) the shielding layers for two shots all are Al, as shown in new Fig. 2(c).  
+
+6. Comment: "The manuscript states that the measured radiation temperatures agreed with predictions. It would be nice to include the predicted radiation temperature in Fig. 3(a)."  
+
+## Response:  
+
+Following the reviewer's suggestion, we have included the predicted radiation temperature in Fig. 3(a). After several experimental checks, the simulation results of temperature are consistent with the experimental results within the experimental error range of \(\pm 3\%\) .  
+
+7. Comment: "What does the colormap represent in Fig. 4(a)?"  
+
+## Response:  
+
+The simulated density versus the time and space.  
+
+8. Comment: "In the first sentence on page 9 "reappearing" should be "reproducing."  
+
+## Response:  
+
+Thanks for the reviewer's correction, we have revised it in the new manuscript.  
+
+9. Comment: "In the 2nd paragraph on page 11 "75J for DD" should be "75J for HD"."  
+
+## Response:  
+
+Sorry for the unclear statement. In the previous manuscript, the total laser energies of \(\sim 47\mathrm{kJ}\) including \(43\mathrm{kJ}\) for ID and \(3.6\mathrm{- }4.0\mathrm{kJ}\) for DD. Only 75J is for DD lasers, covering only a small portion. We have revised the texts.  
+
+10. Comment: "The manuscript states in regards to the reduction in hot-electron fraction and backscatter fraction for the HD scheme: "Obviously, the thermal smoothing effect in HD significantly reduces the fractions compared with the traditional DD." This is not at all obvious to me. Traditionally I think of thermal smoothing as mitigating hydrodynamic instabilities, not laser plasma instabilities. The authors should explain how thermal smoothing is mitigating LPI in this case."  
+
+## Response:  
+
+In the DD scheme, when the DD lasers are incident on the ablator surface, especially in the early incident stage, it is inevitable that nonuniformities, like small density-depletion hollows or density bulges, appear in
+
+<--- Page Split --->
+
+the rapidly expanding electron corona plasma. These nonuniformities are difficult to thermally smooth out due to the low mass ablation rate for electrons.  
+
+As an example, we discuss the self- focus and filamentation in these small hollows when the DD laser with the intensity \(I_{L} = 1.8 \mathrm{PW / cm^{2}}\) is incident on a quarter critical density surface with the plasma density of \(\rho \sim 0.0075 / \mathrm{cc}\) for the CH ablator, in which the "collision" doesn't dominate. In this case, the laser is soon focused into the hollows unstably due to the light refraction index increasing caused by electron number density depletion. For the DD laser intensity of \(I_{L} = 1.8 \mathrm{PW / cm^{2}}\) , electron temperature \(T_{e}\) can rise to \(\sim 4.5 \mathrm{keV}\) soon through inverse bremsstrahlung absorption while the ion temperature rises to \(T_{i} \sim 1.0 \mathrm{keV}\) by the time at least \(\sim 1 \mathrm{ns}\) (refer to ref. 9 in the manuscript). At the quarter critical surface, the thermal pressure of electrons is \(n_{e} T_{e} \sim 1.4 \times 10^{6} \mathrm{J / cc}\) while the laser radiation pressure is \(I_{L} / c \sim 0.6 \times 10^{5} \mathrm{J / cc}\) , where c light speed. Thus, if \(n_{e}\) is drained by \(10\%\) , due to the pressure balance the laser intensity must be increased to \(\sim 4.1 \mathrm{PW / cm^{2}}\) , which results in severe LPI.  
+
+In HD, as discussed in the manuscript, the ID laser provided the ID corona CH plasma (completely ionized) in local thermodynamic equilibrium with peak temperature \(T_{r} = T_{e} = T_{i} = 200 \mathrm{eV}\) within several nanoseconds before the arrival of the DD laser with the intensity of \(1.8 \mathrm{pw / cm^{2}}\) for the wavelength of 0.35 \(\mu \mathrm{m}\) , where \(T_{r}\) , \(T_{e}\) , \(T_{i}\) the temperatures for radiation (thermal X- rays), electron, and ion, respectively. Such LTE plasma offers a radiation- hydrodynamic sound velocity of \(C_{T} = 110 \mu \mathrm{m / ns}\) , the density nonuniformity around hollows or bulges can be thermally smoothed out, which, usually, have widths less than ten microns, within the time of tens of picoseconds. This is our explanation of why LPI in HD is smaller than that in the traditional DD scheme due to thermally smoothing.  
+
+We have briefly added these clarifications in the revised manuscript.  
+
+11. Comment: "Despite the mitigation relative to DD, the backscatter and hot-electron production in the HD scheme is still quite high and is likely to get worse when scaling up to a high-gain design. Are these levels believed to be acceptable for the HD design or are there plans for mitigation?"  
+
+## Response:  
+
+The fractions of the backscatter and the hot- electron production mainly depend on the parameters of the laser intensity \(I_{L}\) and the plasma environment of the density nonuniformities and density scale length \(L_{n}\) . In HD, the DD laser intensity of \(I_{L} = 1.8 \mathrm{PW / cm^{2}}\) , which serves for generating the supersonic- electronic- heat wave, and the ID corona plasma with radiation temperature of \(T_{r} = 200 \mathrm{eV}\) are the same as in the future ignition target, but there is some difference in the density scale length \(L_{n} \sim 300 \mu \mathrm{m}\) for the experiment target and \(\sim 400 \mu \mathrm{m}\) for the designed ignition target. Therefore, we think the fractions of the backscatter and hot electrons in the present experimental target and the future ignition target should not differ much. In addition, the backscattering fraction of \(\sim 5\%\) , as shown in the recent NIF experiments, is acceptable for the ignition target physically and in laser energy.  
+
+For the hot electrons, we would like to explain whether they are able to preheat the fuel. The hot- electron range in the CH plasma can be written theoretically in the form \(\rho R(\mathrm{g / cm^{2}}) \approx 0.59 \times 10^{- 5} E_{e}^{1.661}\) for \(E_{e} > 2 \mathrm{keV}\) (Phys. Plasmas 18, 022703(2011)), where \(E_{e}\) is the hot- electron energy in units of \(\mathrm{keV}\) . It is seen from the expression that even if the hot- electron energy is large enough, such as \(E_{e} = 100 \mathrm{keV}\) , the range is \(\rho R \approx 0.012 \mathrm{g / cm^{2}}\) . On the other hand, the simulations show that in the modeling ignition target, the area density of the remainder of the ablating CH ablator is about \(0.06 \mathrm{g / cm^{2}}\) at the time of the maximal implosion velocity in which \(0.02 \mathrm{g / cm^{2}}\) is from the density plateau piled up by the isothermal compression.
+
+<--- Page Split --->
+
+This indicates that the hot electrons are stopped in CH and prevented from preheating the fuel in the ignition target.  
+
+12. Comment: What is the f-number of the drive beams, as this has a significant impact on SBS/SRs backscatter?  
+
+## Response:  
+
+The f- number on the SG- III laser facility is 10. From the results of the HD experiment, the influence of this f- number on LPI seems to be small.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+I have read the revised version of the article and the answers to my comments from the authors. I am satisfied with their answers.  
+
+I could still have some minor doubts( e.g., the result obtained with the hemispherical taret is till a bit "marginal") but in general I believe that this contribution is important. In this sense I disagree with the comments of one of the other referees who consider this work "incremental". Indeed I think this is not incremental.  
+
+I have also looked to the other comments raised by the other referees and the authors' replies, d again in general I find the answers correct, even if not always fully convincing.  
+
+In conclusion I would say that the manuscript can be published as it is  
+
+Reviewer #2 (Remarks to the Author):  
+
+This is my report on a review of a second submission of this article following revisions by the authors. In my previous report I raised concerns that the results reported are an incremental change to those already published in a high energy density physics (HEDP) paper. I suggested that the authors "re- write the manuscript to avoid repeating previously published work, and to identify the novel and high- impact results of the present work."  
+
+In response, the authors provided a rebuttal that suggests that the new results provide experimental confirmation of the previous data (which I take to mean they have shown they are reproducible), and that provides a breakdown of three new insights that the new submission provides:  
+
+- Confirmed previous results at 3.6KJ indirect-drive laser energy and extended them with new experiments at 4.0KJ, providing more data to confirm pressure boosting with the HD scheme  
+
+- Demonstrated experimental changes which allow better scaling to ignition capable target designs  
+
+- Demonstrated the tuning of the DD laser to provide a "perfect" bulldozer effect and produce pressure boosting and smoothing  
+
+While the above statements are fairly clear, I don't see correspondingly clear statements in the updated manuscript, which to my mind significantly degrades the impact of the work.
+
+<--- Page Split --->
+
+The authors also responded to my comments by adding a series of new data using hemispherical ablator targets as a comparison with the planar targets in the previous draft. This is a welcome addition to the paper but does not directly address my concern  
+
+I still believe that the current submission lacks a proper discussion of the context of new data with respect to the previously published results. For example, the key result in the abstract of this paper is that "This article reports that... such a boosted and smoothed HD pressure is first verified experimentally" while the previous HEDP paper contains the statement "Thus, the experiments have verified the HD pressure boost compared to the ID pressure". These statements suggest that the same conclusions are being presented in two separate works, which according to the rebuttal (not the manuscript) is not the case.  
+
+Given the more detailed description in this paper compared to the previous, and the potential high impact of the hybrid- drive scheme, I am prepared to recommend this paper for publication. However, I would like to stress that the addition of a short paragraph discussing how these experiments add to the understanding of the HD scheme will result in a higher quality article where the impact is more clear.  
+
+Reviewer #3 (Remarks to the Author):  
+
+Dear Editor,  
+
+For the most part I am satisfied with the authors' response to the referees' comments. I think the question of whether similar results being previously published as part of a conference proceedings precludes publication in Nature Comms is a decision for the Editor. There are a couple of points in the authors' response that I am concerned about, but they are not critical to the main results:  
+
+1) The authors stated that electrons with energies less than 250 keV will be stopped in the CH ablator (and thus are not a preheat concern), but this is about an order of magnitude higher than the typical electron energies that are thought to be a preheat concern in direct-drive implosions.  
+
+2) I do not think that thermal smoothing in the HD scheme is the source of reduced hot-electron production relative to traditional DD. Even if there were significant density nonuniformities in the corona of a DD implosion, filamentation and self-focusing are effectively absolute instabilities, so an enhanced seed level would not have a significant impact on these instabilities (also the enhanced seed would have to be wavelength and phase matched to the beam-driven perturbation). That being said, it is not obvious to me why the hot-electron production is reduced in the HD scheme, but it could be as simple as the fact that the ID beams increase Te, which would result in increased thresholds for
+
+<--- Page Split --->
+
+filamentation and TPD. Regardless of the reason, LPI is not of primary importance to these results, and this is not the appropriate forum to be introducing an LPI mitigation mechanism that (to my knowledge) has not been previously discussed in the literature without adequate supporting evidence.
+
+<--- Page Split --->
+
+## Response Letter  
+
+Article ID: NCOMMS- 22- 52906  
+
+# Title: Experimental confirmation of drive pressure boosting and smoothing for hybrid-drive inertial fusion at the 100-kJ laser facility  
+
+Authors: Ji Yan, Jiwei Li, X. T. He, et al.  
+
+We are grateful to the referees for their further insightful reviews and valuable comments on our revised manuscript. According to their comments, we have carefully revised the manuscript again. Following are the specific responses to the referees' comments point by point. With these responses, we believe that we have overcome all referees' criticisms and fulfilled the criteria set by them for accepting our manuscript for publication in Nature Communications. The changes in the revised manuscript are listed in the following.  
+
+### I. Responses to Referee #1  
+
+Comment: "In conclusion I would say that the manuscript can be published as it is" Response: Thanks for the referee's recommendation of our manuscript for publication.  
+
+## II. Responses to Referee #2  
+
+Comment 1: the key result in the abstract of this paper is that "This article reports that... such a boosted and smoothed HD pressure is first verified experimentally" while the previous HEDP paper contains the statement "Thus, the experiments have verified the HD pressure boost compared to the ID pressure". These statements suggest that the same conclusions are being presented in two separate works, which according to the rebuttal (not the manuscript) is not the case.  
+
+Response: The previous result using a planar ablator target at DD laser energy of \(3.6\mathrm{kJ}\) to experimentally confirm HD pressure boosting for the first time is obviously important, but indeed applying the term "verify" is not strict enough because there is only one experimental result without more repeated or similar results to support it. In the present manuscript, new more sufficient data from the hemispherical and planar ablator targets at DD laser energy of \(3.6\mathrm{- 4kJ}\) are provided to fully confirm HD pressure boosting and smoothing, while among these data, the experiment on the planar target at DD laser energy of \(3.6\mathrm{kJ}\) is for "check and further confirm the previous result", which has been mentioned all in the abstract, the planar target experiment at DD laser of \(3.6\mathrm{kJ}\) , and the section of "Discussions and conclusions".  
+
+Comment 2: I would like to stress that the addition of a short paragraph discussing how these experiments add to the understanding of the HD scheme will result in a higher quality article
+
+<--- Page Split --->
+
+where the impact is more clear.  
+
+Response: In the present manuscript, we have already added explanations of how these experiments increase understanding of the HD scheme in the section "Discussions and Conclusions"  
+
+## III. Responses to Referee #3  
+
+Comment 1: The authors stated that electrons with energies less than \(250\mathrm{keV}\) will be stopped in the CH ablator (and thus are not a preheat concern), but this is about an order of magnitude higher than the typical electron energies that are thought to be a preheat concern in directdrive implosions.  
+
+Comment 1: The authors stated that electrons with energies less than \(250\mathrm{keV}\) will be stopped in the CH ablator (and thus are not a preheat concern), but this is about an order of magnitude higher than the typical electron energies that are thought to be a preheat concern in directdrive implosions.  Response: In the last response to comment 11 of reviewer #3, we used hot electron energy of \(100\mathrm{keV}\) , from which with the expression of \(\rho R(\mathrm{g / cm}^2)\) we estimated the area density (the penetrating range) to be \(0.012\mathrm{g / cm}^2\) less than \(0.06\mathrm{g / cm}^2\) for the remaining CH ablator at the time of maximal implosion velocity. As for energy \(250\mathrm{keV}\) mentioned in the response to the comment to reviewer #2, we mean that for hot electrons with energy of \(250\mathrm{keV}\) the area density is just equal to the remaining CH area density of \(0.06\mathrm{g / cm}^2\) , and therefore, only these electrons with energies greater than \(250\mathrm{keV}\) can penetrate the remaining CH preheating the fuel or to say that those electrons with energies less than \(250\mathrm{keV}\) are trapped in the remaining CH.  
+
+In the ID scheme, the experiment on hot electron preheating fuel in the ignition- scale target for NIC (PRL 108, 135006(2012)) has shown that with ID laser of \(1.3\mathrm{MJ}\) , the energy of energetic electron \(>100\mathrm{keV}\) deposited into the CH ablator is about \(570\pm 250\mathrm{J}\) , and \(5\pm 3\mathrm{J}\) (an upper bound) is absorbed in the DT ice, showing an acceptable increase of \(3.5\%\) for the adiabat of \(\alpha = 1.5\) . For the HD scheme, energetic electron preheating fuel needs further to be investigated.  
+
+Comment 2: I do not think that thermal smoothing in the HD scheme is the source of reduced hot- electron production relative to traditional DD. Even if there were significant density nonuniformities in the corona of a DD implosion, filamentation and self- focusing are effectively absolute instabilities, so an enhanced seed level would not have a significant impact on these instabilities (also the enhanced seed would have to be wavelength and phase matched to the beam- driven perturbation). That being said, it is not obvious to me why the hot- electron production is reduced in the HD scheme, but it could be as simple as the fact that the ID beams increase Te, which would result in increased thresholds for filamentation and TPD. Regardless of the reason, LPI is not of primary importance to these results, and this is not the appropriate forum to be introducing an LPI mitigation mechanism that (to my knowledge) has not been previously discussed in the literature without adequate supporting evidence.  
+
+Response: In the last response to the reviewer's comment, we mentioned that in the DD
+
+<--- Page Split --->
+
+scheme, the DD laser with the intensity of \(1.8 \mathrm{PW / cm}^2\) provides a nonuniform electron corona plasma, which results in filamentation and self- focusing increase to make the DD laser intensity over \(1.8 \mathrm{PW / cm}^2\) forming new laser speckles in filamentation and self- focusing, resulting in a new LPI source, here LPI refers to the laser plasma instability of SRS, SBS, TPD, etc., not only the filamentation and self- focusing itself. While for the HD scheme, the ID laser provides a more uniform ID corona plasma with local equilibrium temperature \(T_r = T_e = T_i = 200 \mathrm{eV}\) , which results in filamentation and self- focusing mitigation, and therefore, LPI reduction by thermal smoothing compared with the DD scheme, here LPI, of course, involves SRS, SBS, and TPD, etc. This phenomenon is indeed seen in the comparative study by numerical simulation in which SBS and SRS occur rapidly with the formation of filamentations when the DD laser with the intensity of \(1.8 \mathrm{PW / cm}^2\) is incident on a nonuniform plasma, but these are not clear in the uniform plasma.  
+
+We think in a novel HD scheme different from DD and ID schemes, how the thermodynamic equilibrium ID plasma affects LPI is still understudied not too much. In our report, we mainly investigate pressure smoothing and smoothing, which is key physics in the HD scheme, and there is no need for further discussion of the LPI mitigation mechanism. We thank the reviewer's comment: "LPI is not of primary importance to these results, and this is not the appropriate forum to be introducing an LPI mitigation mechanism that has not been previously discussed in the literature without adequate supporting evidence". At the end of the section "Fractions of stimulated backscattering and hot electron energy" in the present manuscript, we have revised the previous discussions related to the LPI mitigation mechanism to change into the statement "So far, why the LPI in the HD scheme is smaller than the LPI in the DD scheme that has no radiation ablation is still not very clear, which needs further exploration and discussion."  
+
+## IV. List of changes  
+
+## IV. List of changes  
+
+1. In response to comment 1 from reviewer #2, in addition to explaining the previous result in this Response Letter, in the present manuscript, we also stated it in red font in the Abstract, the section "Radiation temperature and ID and DD velocities", and the section "Discussions and Conclusions".  
+
+In response to comment 2 from reviewer #2, we added explanations of how these experiments increase understanding of the HD scheme in the section "Discussions and Conclusions" in red font.  
+
+2. In response to comment 2 from reviewer #3, in the present manuscript at the end of the section "Fraction of stimulated backscattering and hot-electron energy", we revised the previous discussions related to the LPI mitigation mechanism to change into the statement "So far, why the LPI in the HD scheme is smaller than the LPI in the DD scheme that has no radiation ablation is still not very clear, which needs further exploration and discussion" in red font.
+
+<--- Page Split --->
+
+REVIEWERS' COMMENTS  
+
+Reviewer #2 (Remarks to the Author):  
+
+The present manuscript has addressed my detailed questions, and I believe the changes have put the results of this work in context with previous publications. I believe that the current form of the manuscript is suitable for publication.  
+
+Reviewer #3 (Remarks to the Author):  
+
+I am satisfied with the Authors' response and recommend the manuscript for publication.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #2 (Remarks to the Author):  
+
+The present manuscript has addressed my detailed questions, and I believe the changes have put the results of this work in context with previous publications. I believe that the current form of the manuscript is suitable for publication.  
+
+Reviewer #3 (Remarks to the Author):  
+
+I am satisfied with the Authors' response and recommend the manuscript for publication.  
+
+## Response to reviewer' s comment  
+
+We are grateful for the reviewers' acknowledgement of our work. Thanks all the reviews for their contribution to improving the quality of the paper and promotion of our understanding of the work.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a3aec9dfee94996570423cfe6ec1e55b9660040642ae120e923878928dc3ac1a/supplementary_0_Peer Review File__a3aec9dfee94996570423cfe6ec1e55b9660040642ae120e923878928dc3ac1a_det.mmd

@@ -0,0 +1,604 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[106, 161, 863, 250]]<|/det|>
+Experimental confirmation of driving pressure boosting and smoothing for hybrid- drive inertial fusion at the 100- kJ laser facility  
+
+<|ref|>image<|/ref|><|det|>[[94, 732, 262, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[271, 732, 880, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 145, 394, 161]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 847, 256]]<|/det|>
+I believe that this paper is very interesting and it is an innovative and substantial contribution to the field of Inertial Confinement Fusion. Publication of this article would be very timely seen the recent breakthrough in laser fusion obtained at NIF.  
+
+<|ref|>text<|/ref|><|det|>[[115, 268, 806, 285]]<|/det|>
+In general I have only a few remarks which I would ask the authors to reply before publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 324, 880, 378]]<|/det|>
+1) in the conclusions the authors wriest that "Using DD laser with intensities \(\sim (1.8 - 2.0) \mathrm{PW / cm^2}\) , ... we stably obtained the boosted HD pressure \(P \approx 150 - 155\) Mbar about 3.5-3.6 times the radiation ablation pressure experimentally"  
+
+<|ref|>text<|/ref|><|det|>[[115, 389, 877, 444]]<|/det|>
+Well, this indeed the range of pressure which can be obtained in direct- drive or similar intensities, using e.g. Lindl's scaling. So in this case the contribution from the indirect drive part of the scheme does not seem to be really apparent. Con the authors comment on this?  
+
+<|ref|>text<|/ref|><|det|>[[115, 482, 870, 536]]<|/det|>
+2) While the comparison between the HD scheme and the "traditional" DD scheme is well described, I still have doubts on how the HD scheme compares to the Shock Ignition approach. SI is also based on a two step process with a first "compression" laser pulse and a final "ignition" laser spike.  
+
+<|ref|>text<|/ref|><|det|>[[115, 548, 848, 583]]<|/det|>
+The authors write that "the distance between the electron ablation front and critical surface for the traditional DD system is too short". Does this conclusion also apply to shock ignition?  
+
+<|ref|>text<|/ref|><|det|>[[115, 621, 877, 676]]<|/det|>
+3) Does this conclusion also apply to the proposed scheme of "foam buffered target" in which the pellet is surrounded by a low density material (foam) to create a long scale length plasma assumed to be able to smooth te non-uniformities thanks to the long stand-off distance available to thermal smoothing?  
+
+<|ref|>text<|/ref|><|det|>[[115, 715, 822, 751]]<|/det|>
+4) what is the measured "conversion efficiency for thermal X-rays" reported at the end of page 5 (section "Experimental target design")  
+
+<|ref|>text<|/ref|><|det|>[[115, 790, 880, 881]]<|/det|>
+5) the authors speak about a "hot electron energy fraction of the DD lasers in laser-plasma interaction in HD..." of about \(2\%\) and they say that this is "significantly lower than that in the traditional DD laser". However I found no measurement or estimation of the HE temperature. This is indeed an additional important parameters because if the HE are not too energetic, they will not be able to penetrate deeply into the fuel, and so their armful effect will be strongly mitigated.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 118, 590, 135]]<|/det|>
+6) Also, what is the fraction of HE produced during the ID phase?  
+
+<|ref|>text<|/ref|><|det|>[[115, 175, 864, 210]]<|/det|>
+7) The authors write that SRS amounts to about \(0.5\%\) . Since in these experimental conditions SRS is probably the main source of HE, how is this result compatible with a HE conversion efficiency of \(2\%\) ??  
+
+<|ref|>text<|/ref|><|det|>[[116, 279, 393, 295]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 335, 875, 500]]<|/det|>
+This manuscript reports on simulation and experimental work on the hybrid- drive (HD) approach to inertial confinement fusion (ICF). This approach uses a spherical hohlraum and long- pulse laser drive to provide an 'indirect' X- ray drive to the ICF capsule, combined with a short pulse indirect laser drive applied sometime after the start of the long pulse. If tuned correctly, the relativistic electron heat wave generated by the short pulse laser slows to form a compression wave in the material ablated by the indirect drive. The net effect is a stable plateau in density between the indirect- drive ablation front and direct- drive critical surface, increasing the drive pressure on the ICF capsule and thermally smoothing perturbations. It is hoped that his hybrid approach will provide favorable scaling to ignition conditions and an effective path to high fusion yield from ICF implosions.  
+
+<|ref|>text<|/ref|><|det|>[[114, 538, 880, 683]]<|/det|>
+Inertial fusion experiments have been widely reported recently as a result of significant advances in a related approach using only a direct drive. This makes ICF experiments, and alternative paths to ignition, a timely contribution and of high interest to a wide audience. However, the work presented here is an incremental change to work that has already been reported; the HD approach has been widely described over several decades, and nearly identical simulations and experiments have been published recently (High Energy Density Physics 96, 100804). The only difference between this submission and the previously reported results is a slight increase in direct drive energy (4.0 KJ vs 3.6 KJ); the key results – observation of pressure boosting, peak radiation temperature, and laser backscatter - are unchanged.  
+
+<|ref|>text<|/ref|><|det|>[[114, 722, 883, 850]]<|/det|>
+Given the incremental nature of this work, I do not find it suitable for publishing in Nature Communications. However, the topic of hybrid- drive, and new experimental results for ICF, remains highly impactful and relevant to the Nature readership. I therefore recommend that the authors re- write the manuscript to avoid repeating previously published work, and to identify the novel and high- impact results of the present work. I also suggest that the authors pay attention to the clarity and conciseness of the new manuscript, since I found the current submission difficult to read and somewhat unclear in some places.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 118, 393, 135]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 176, 204, 192]]<|/det|>
+Dear Editor,  
+
+<|ref|>text<|/ref|><|det|>[[115, 231, 857, 341]]<|/det|>
+This manuscript reports on initial experiments at SG- III using a hybrid- drive (HD) scheme designed to take advantage of the drive smoothing provided by indirect- drive (ID), and the efficient absorption of laser energy provided by direct- drive (DD). The manuscript presents important proof- of- principle experiments for the HD scheme that can be used to validate modeling capabilities and assess the LPI threat. The manuscript is well organized and provides an appropriate level of detail and could be appropriate for publication with some changes. My specific comments follow:  
+
+<|ref|>text<|/ref|><|det|>[[115, 379, 855, 433]]<|/det|>
+1) Although the manuscript is well-written from the point-of-view of the results being presented in a logical and coherent manner, there are a large number of grammatical errors that make it difficult to read.  
+
+<|ref|>text<|/ref|><|det|>[[115, 472, 875, 509]]<|/det|>
+2) The abstract (and elsewhere) states that the HD shock "stops the asymmetric ID shock." I am not sure what this statement is supposed to mean, but certainly it is not meant to be taken literally.  
+
+<|ref|>text<|/ref|><|det|>[[115, 547, 875, 619]]<|/det|>
+3) As the HD scheme is proposed as an alternative to DD/ID, it would help if the manuscript was more explicit about the advantages/disadvantages of the HD scheme relative to those schemes. For example, the HD scheme provides increased ablation pressure relative to ID, but how does the pressure compare to DD alone?  
+
+<|ref|>text<|/ref|><|det|>[[115, 658, 456, 676]]<|/det|>
+4) In the caption of Fig. 1 LHEs should be LEHs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 715, 865, 751]]<|/det|>
+5) Figure 2(c) shows an AI layer, but in the text this is referred to as a Mo layer (the figure caption says that it can be Mo or AI).  
+
+<|ref|>text<|/ref|><|det|>[[115, 790, 880, 826]]<|/det|>
+6) The manuscript states that the measured radiation temperatures agreed with predictions. It would be nice to include the predicted radiation temperature in Fig. 3(a).  
+
+<|ref|>text<|/ref|><|det|>[[115, 866, 480, 883]]<|/det|>
+7) What does the colormap represent in Fig. 4(a)?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 117, 653, 135]]<|/det|>
+8) In the first sentence on page 9 "reappearing" should be "reproducing".  
+
+<|ref|>text<|/ref|><|det|>[[113, 174, 643, 192]]<|/det|>
+9) In the 2nd paragraph on page 11 "75J for DD" should be "75J for HD".  
+
+<|ref|>text<|/ref|><|det|>[[115, 230, 862, 323]]<|/det|>
+10) The manuscript states in regards to the reduction in hot-electron fraction and backscatter fraction for the HD scheme: "Obviously, the thermal smoothing effect in HD significantly reduces the fractions compared with the traditional DD." This is not at all obvious to me. Traditionally I think of thermal smoothing as mitigating hydrodynamic instabilities, not laser plasma instabilities. The authors should explain how thermal smoothing is mitigating LPI in this case.  
+
+<|ref|>text<|/ref|><|det|>[[115, 361, 879, 415]]<|/det|>
+11) Despite the mitigation relative to DD, the backscatter and hot-electron production in the HD scheme is still quite high and is likely to get worse when scaling up to a high-gain design. Are these levels believed to be acceptable for the HD design or are there plans for mitigation?  
+
+<|ref|>text<|/ref|><|det|>[[115, 454, 857, 472]]<|/det|>
+12) What is the f-number of the drive beams, as this has a significant impact on SBS/SRS backscatter?
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[427, 75, 569, 92]]<|/det|>
+## Response Letter  
+
+<|ref|>text<|/ref|><|det|>[[101, 95, 860, 175]]<|/det|>
+Article ID: NCOMMS- 22- 52906  Title: Experimental confirmation of drive pressure boosting and smoothing for hybrid- drive inertial fusion at the 100- kJ laser facility  Authors: Ji Yan, Jiwei Li, X. T. He, et al.  
+
+<|ref|>text<|/ref|><|det|>[[101, 197, 896, 256]]<|/det|>
+We are grateful to the three referees for their insightful reviews and valuable comments on our manuscript. According to their comments, we have carefully revised the manuscript. Following are the specific responses to the referees' comments point by point.  
+
+<|ref|>sub_title<|/ref|><|det|>[[384, 295, 610, 312]]<|/det|>
+### I. Responses to Referee #1  
+
+<|ref|>text<|/ref|><|det|>[[101, 337, 898, 435]]<|/det|>
+1. Comment: "in the conclusions the authors wriest that "Using DD laser with intensities \(\sim (1.8 - 2.0)\) PW/cm², ..., we stably obtained the boosted HD pressure \(P \approx 150 - 155\) Mbar about 3.5-3.6 times the radiation ablation pressure experimentally." Well, this indeed the range of pressure which can be obtained in direct-drive or similar intensities, using e.g. Lindl's scaling. So in this case the contribution from the indirect drive part of the scheme does not seem to be really apparent. Can the authors comment on this?"  
+
+<|ref|>sub_title<|/ref|><|det|>[[101, 440, 185, 454]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[100, 458, 900, 780]]<|/det|>
+We understand the reviewer's concern. Yes, under the current laser drive condition of the direct- drive (DD) laser energies of \(3.6 - 4\mathrm{kJ}\) only, the hybrid- drive (HD) pressure of 150- 155 Mbar in our HD scheme is similar to that in the DD scheme. However, the physics is completely different. In our HD scheme, the important roles of the indirect drive (ID) lasers are to produce an ID corona plasma background and a large enough distance \(\Delta R_{ID}\) between the critical surface and radiation ablation front before the DD laser arriving, and also a radiation ablation pressure at the radiation ablation front in addition to pre- compression of the fuel. Afterwards, the DD laser with the intensities of \(I_{L} \sim (1.5 - 2.0)\) PW/cm² then is absorbed near the critical surface and converted into a supersonic- electron- heat- conduction wave (below, abbreviated the SEC wave). The SEC wave propagates within the large \(\Delta R_{ID}\) from the critical surface toward the radiation ablation front and slows down to a plasma compressive wave while smoothing. This compressive wave isothermally compresses low ID corona plasma density \(\rho_{ID}\) into high HD density \(\rho_{HD}\) , which can be fitted as the function of the ID laser energy \(\sim E_{DD}^{1 / 4}\) (see the manuscript), to change the radiation ablation pressure \(P_{ID} = \Gamma \rho_{ID}T_{r}\) at the radiation ablation front into a new smoothed HD pressure \(P_{HD} = \Gamma \rho_{HD}T_{r} \propto E_{DD}^{1 / 4}T_{r}\) , where \(E_{DD}\) is DD laser energy and \(T_{r}\) is radiation temperature. Therefore, we see that the first ID process plays a key role in our HD scheme, which results in that the HD pressure scales as both the DD laser energy \(E_{DD}^{1 / 4}\) and the radiation temperature, while in the DD scheme, the pressure only scales as \(I_{L}^{2 / 3}\) .  
+
+<|ref|>text<|/ref|><|det|>[[101, 781, 898, 920]]<|/det|>
+For the CH ablator, the HD pressure achieved in our HD scheme can be calculated as \(P_{HD}(\mathrm{Mbar}) \approx 62 E_{DD}^{1 / 4}T_{r}\) , while the DD pressure achieved in the DD scheme (without ID) is \(P_{DD} \approx 90 I_{L}^{2 / 3}\) . Under \(\sim\) current laser conditions in our experiments, the DD laser energy is \(E_{DD} = 3.6 - 4\mathrm{kJ}\) , \(I_{L} = 1.8\) PW/cm², and \(T_{r} = 200\mathrm{eV}\) , the pressure in our HD scheme is about \(P_{HD} \sim 155\) Mbar, which is indeed close to that in the DD scheme \(P_{DD} \sim 140\) Mbar. However, when we scale up to the ignition condition with the increased DD laser energy, such as \(E_{DD} = 825\mathrm{kJ}\) , the pressure in our HD scheme can be much boosted (due to the role of the ID) up to \(P_{HD} \sim 775\) Mbar, while that in the DD scheme still keeps to be as only 140 Mbar, since
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[102, 75, 476, 92]]<|/det|>
+the laser intensity is similar as \(I_{L} = 1.8 \mathrm{PW / cm^2}\) .  
+
+<|ref|>text<|/ref|><|det|>[[102, 95, 897, 174]]<|/det|>
+In the present revised manuscript, we also added a hemispherical ablator target performed in a recent experiment and under the same DD laser energy of \(3.6 - 4.0 \mathrm{kJ}\) and obtained the HD pressure \(P_{HD} \sim 170 - 180 \mathrm{Mbar}\) higher than \(\sim 150 - 155 \mathrm{Mbar}\) in the planar ablator target. These results completely verified the HD pressure boosting and smoothing effects- the heart of the HD scheme  
+
+<|ref|>text<|/ref|><|det|>[[101, 196, 898, 295]]<|/det|>
+2. Comment: "While the comparison between the HD scheme and the "traditional" DD scheme is well described, I still have doubts on how the HD scheme compares to the Shock Ignition approach. SI is also based on a two-step process with a first "compression" laser pulse and a final "ignition" laser spike. The authors write that "the distance between the electron ablation front and critical surface for the traditional DD system is too short". Does this conclusion also apply to shock ignition?"  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 299, 186, 314]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[100, 317, 898, 700]]<|/det|>
+The two- step process of ID and DD in our HD scheme is inherently different from the "compression" and "ignition" in the DD scheme. This conclusion also applies to SI which is under the DD scheme. As explained in the manuscript, in HD, in the first step, the ID distance \(\Delta R_{ID}\) offered in advance by the ID laser is about \(250 - 300 \mu \mathrm{m}\) for radiation temperature \(T_{r} = 200 \mathrm{eV}\) , which is large enough for the SEC wave to slow down to a compressive wave thermally compressing the ID corona plasma density into high plasma density to make the HD pressure far higher than the radiation ablation pressure. However, in SI, no ID laser is used in advance, and thus of course there exists no large ID distance \(\Delta R_{ID}\) before the DD laser arrives at the critical surface. Although in the later stage of SI, the DD distance \(\Delta R_{DD}\) between the critical surface and electron ablation front offered by the pre- pulse DD laser ablating the ablator surface is getting to increase, about \(100 - 150 \mu \mathrm{m}\) for a model in Fig. 1 in Nucl. Fusion 54, 054004(2014), due to the mass ablation rate for electrons being smaller than for radiation, under the same laser intensity \(\Delta R_{DD}\) is shorter than \(\Delta R_{ID}\) , resulting in the SEC wave driven by the DD laser directly hitting the capsule without enough time slowing down to a compressive wave. Therefore, the electron ablation pressure \(P_{DD} \sim I_{L}^{2 / 3}\) can't be smoothed very well, which leads to an asymmetric and unstable implosion in the early stage of the implosion. In addition, in SI, the driving pressure (electron ablation pressure) \(P_{DD}\) over \(300 \mathrm{Mbar}\) is required for the hotspot ignition (Nucl. Fusion 54, 054004(2014)) in which the DD laser intensity of \(I_{L} \sim 5 \mathrm{PW / cm^2}\) is required. For such intensity, the velocity of the SEC wave slowing down to the compressive wave should be larger than that for \(I_{L} = 1.8 \mathrm{PW / cm^2}\) in HD, and therefore, the DD distance \(\Delta R_{DD}\) has too far larger than \(250 - 300 \mu \mathrm{m}\) .  
+
+<|ref|>text<|/ref|><|det|>[[101, 721, 897, 780]]<|/det|>
+3. Comment: "Does this conclusion also apply to the proposed scheme of "foam buffered target" in which the pellet is surrounded by a low-density material (foam) to create a long scale length plasma assumed to be able to smooth to non-uniformities thanks to the long stand-off distance available to thermal smoothing?"  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 784, 186, 799]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[102, 802, 897, 859]]<|/det|>
+In HD, the distance \(\Delta R_{ID}\) offered in advance by the ID laser does at least two things, that is, the SEC wave will be slowed down to a compressive wave for boosting the HD pressure and thermally smoothed very well.  
+
+<|ref|>text<|/ref|><|det|>[[102, 862, 897, 920]]<|/det|>
+To our understanding, the foam (low density but above critical density) indeed removes the critical surface of the ablator in the pellet to the foam surface and is available for thermal smoothing well in the form \(\exp (- \ell \Delta R / R_{cr})\) if the foam thickness \(\Delta R\) is larger than the perturbation wavelength \(\lambda = 2\pi R_{CR} / \ell\)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 75, 896, 113]]<|/det|>
+from the DD laser imprinting and beam overlapping, \(\ell\) is the perturbative mode number, and \(R_{CR}\) is the radius of the critical surface.  
+
+<|ref|>text<|/ref|><|det|>[[101, 116, 896, 154]]<|/det|>
+However, the foam is difficult to offer a large distance, like \(\Delta R_{ID}\) in HD, since it will significantly reduce laser energy to the pellet.  
+
+<|ref|>text<|/ref|><|det|>[[101, 177, 896, 214]]<|/det|>
+4. Comment: "what is the measured "conversion efficiency for thermal X-rays" reported at the end of page 5 (section "Experimental target design")"  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 218, 185, 233]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 237, 896, 294]]<|/det|>
+Thanks for the reviewer's correction. It refers to the conversion efficiency \(\eta\) of the ID laser to thermal X- rays. The x- ray conversion efficiency is estimated based on the hohlraum power balance equation. The equation can be described as follow:  
+
+<|ref|>equation<|/ref|><|det|>[[277, 297, 717, 317]]<|/det|>
+\[\eta (P_{laser} - P_{backscatter}) = [A_{wall}(1 - \alpha) + A_{LEH} + F_{CH}A_{CH}]\sigma T_r^4\]  
+
+<|ref|>text<|/ref|><|det|>[[101, 318, 898, 457]]<|/det|>
+where \(P_{laser}\) and \(P_{backscatter}\) are the incident laser power and the backscattered laser power due to LPI effect, respectively. \(A_{wall}\) , \(A_{LEH}\) and \(A_{CH}\) represent the surface areas of the hohlraum wall, laser entrance hole and CH sample, respectively. \(\alpha\) is the albedo of the Hohlraum and inferred by analytical model \((\alpha = 1 - 0.32T(100eV)_{r}^{- 0.7}\tau^{- 0.382})\) . \(F_{CH}\) represents the ratio of the absorbed flux over the incident flux of capsule and approximately as a constant and equal to \(0.7\sigma\) is the Stefan- Boltzmann constant. \(T_{r}\) is the radiation temperature which measured by calibrated flat response x- ray diode (FXRD). The uncertainty of radiation temperature using FXRD is \(3\%\) . In our experiments, the conversion efficiency is \(89\pm 8\%\) .  
+
+<|ref|>text<|/ref|><|det|>[[101, 479, 897, 577]]<|/det|>
+5. Comment: "the authors speak about a "hot electron energy fraction of the DD lasers in laser-plasma interaction in HD..." of about \(2\%\) and they say that this is "significantly lower than that in the traditional DD laser". However, I found no measurement or estimation of the HE temperature. This is indeed an additional important parameter because if the HE are not too energetic, they will not be able to penetrate deeply into the fuel, and so their armful effect will be strongly mitigated."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 582, 185, 596]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 600, 897, 678]]<|/det|>
+Thanks for the reviewer's comments, we have added a diagram of the measured HE temperature in the present revised manuscript (see Fig. 5c). In our experiment, the HE temperature is about 21- 22 keV. The production of HE is mainly from two- plasmon decay (TPD) since the backscattering fraction for SRS is only \(\sim 0.4\%\) experimentally.  
+
+<|ref|>text<|/ref|><|det|>[[101, 681, 897, 839]]<|/det|>
+The simulation shows that for the modeling ignition target, the area density of the remainder of the ablating CH ablator is about \(0.06 \mathrm{g} / \mathrm{cm}^2\) at the time of the maximal implosion velocity in which \(0.02 \mathrm{g} / \mathrm{cm}^2\) is in the density plateau piled up by the isothermal compression. On the other hand, the HE's range can be expressed in the form \(\rho R(\mathrm{g} / \mathrm{cm}^2) \approx 0.59 \times 10^{- 5} E_e^{1.661}\) for \(E_e > 2 \mathrm{keV}\) in the CH material (Phys. Plasmas 18, 022703(2011)), where \(E_e\) is the HE energy in \(\mathrm{keV}\) . We see from the expression and numerical result that all hot electrons with energy less than \(250 \mathrm{keV}\) are completely trapped in the remaining CH, impossible to reach the fuel. Obviously, for the HE temperature of 21- 22 keV the number of the HEs with energy greater than \(250 \mathrm{keV}\) can be neglectable and they are impossible to preheat the fuel significantly.  
+
+<|ref|>text<|/ref|><|det|>[[101, 862, 700, 899]]<|/det|>
+6. Comment: "Also, what is the fraction of HE produced during the ID phase?" Response:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 75, 898, 175]]<|/det|>
+A calibrated filtered- Fluorescence (FF) hard x- ray spectrometer was employed to monitor the fraction of the hot- electrons. In HD shots, the measured hot- electrons include both ID phase and DD phase. But however, in ID only shots, the measured energy of hot- electrons is lower than 5 J (the detectable threshold of the spectrometer) and the fraction of HE produced during the ID phase is around \(5 \mathrm{~J} / 43 \mathrm{kJ} = 0.01\%\) , since the ID laser energy is \(43 \mathrm{~kJ}\) in our experiment.  
+
+<|ref|>text<|/ref|><|det|>[[101, 179, 898, 297]]<|/det|>
+In the HD experiment, the ID laser with an averaged intensity of about \(0.4 \mathrm{PW} / \mathrm{cm}^2\) and energy of 40- 50 kJ is coupled in the HD plasma with the DD laser with an intensity of \(1.5 - 2 \mathrm{PW} / \mathrm{cm}^2\) and energy of \(3.6 - 4.0 \mathrm{kJ}\) launching in a later pulse duration of the ID laser, and therefore, it is difficult to separate the HE energy from the ID laser and DD laser. We have measured HE produced by the ID laser energy of \(43 \mathrm{~kJ}\) without the DD laser, and its energy is about \(3 \mathrm{~J}\) , which means that the energy fraction of HE produced by the ID laser is \(3 \mathrm{~J} / 43 \mathrm{~kJ} = 0.0067\%\) .  
+
+<|ref|>text<|/ref|><|det|>[[101, 319, 898, 376]]<|/det|>
+7. Comment: "The authors write that SRS amounts to about \(0.5\%\) . Since in these experimental conditions, SRS is probably the main source of HE, how is this result compatible with a HE conversion efficiency of \(2\%\) ?"  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 381, 185, 397]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 400, 898, 680]]<|/det|>
+There should be some misunderstanding. As mentioned in our manuscript, in our HD scheme, as the laser intensity \(I_L \sim 1.8 \mathrm{PW} / \mathrm{cm}^2\) exceeds the threshold of two- plasma decay (TPD), TPD, instead of SRS, is the main source of HE. That is the reason why SRS amounts to only about \(0.5\%\) . The main component of HE (about \(2\%\) ) comes from TPD. In the HD experiment with the DD laser intensity of \(\sim 1.8 \mathrm{PW} / \mathrm{cm}^2\) , we measured the SRS fraction of about \(0.4\%\) on SG- III and compared it to the results on OMEGA under the DD scheme with the same intensity, as seen in Fig. 20 in the reference No. 23 and in Fig. 21 in the reference No. 24 in our manuscript, where the SRS fraction is about 2.5- 3 times the fraction of \(0.4\%\) in the HD experiment and SBS dominated. Considering that the radiation ablation offered a uniform ID corona plasma before the DD laser arrives, different from the DD laser- plasma interaction (LPI) in the DD scheme, the thermal smoothing effect on the ID corona plasma may result in the decrease of the SRS fraction in the HD experiment. So, the SRS fraction on SG- III may be reasonably compared with the results in OMEGA. Such a low SRS fraction in HD is unable to offer the HE energy fraction of \(\sim 2\%\) for the DD laser energy of \(3.6 - 4.0 \mathrm{~kJ}\) . Therefore, we think the HE fraction mainly comes from two- plasmon decay (TPD) since this intensity of \(I_L \sim 1.8 \mathrm{PW} / \mathrm{cm}^2\) exceeds the threshold of TPD.  
+
+<|ref|>text<|/ref|><|det|>[[101, 682, 898, 780]]<|/det|>
+At electron temperature of \(T_e = 2 - 3 \mathrm{keV}\) and the electron number density of \(n_e / n_c \sim 0.2 - 0.7\) , the maximal linear growth rate for TPD is \(\sim \gamma /\omega \sim 10^{- 2} / (n_e / n_c)\) , where \(n_c\) the critical number density, the laser frequency \(\omega_0 \approx 5.4 \times 10^{15} / s\) for the wavelength of \(0.35 \mu \mathrm{m}\) . Therefore, at a quarter of the critical density, we have the maximal linear growth rate of \(\gamma \sim 0.2 / \mathrm{fs}\) , and the occurrence of TPD is completely possible soon after the DD laser arrives. Further specialized investigation of course is necessary.  
+
+<|ref|>sub_title<|/ref|><|det|>[[381, 802, 616, 819]]<|/det|>
+## II. Responses to Referee #2  
+
+<|ref|>text<|/ref|><|det|>[[101, 843, 898, 921]]<|/det|>
+1. Comment: "Inertial fusion experiments have been widely reported recently as a result of significant advances in a related approach using only a direct drive. This makes ICF experiments, and alternative paths to ignition, a timely contribution and of high interest to a wide audience. However, the work presented here is an incremental change to work that has already been reported; the HD approach has been widely
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 75, 898, 154]]<|/det|>
+described over several decades, and nearly identical simulations and experiments have been published recently (High Energy Density Physics 96, 100804). The only difference between this submission and the previously reported results is a slight increase in direct drive energy (4.0 KJ vs 3.6KJ); the key results - observation of pressure boosting, peak radiation temperature, and laser backscatter - are unchanged."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 159, 185, 174]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[120, 177, 657, 194]]<|/det|>
+Thanks for the referee's comments. We now explained those as follows.  
+
+<|ref|>text<|/ref|><|det|>[[101, 197, 898, 275]]<|/det|>
+The hybrid- drive content is only a portion of the conference proceedings review paper issued in HEDP, where we simply presented with DD laser energy of \(3.6\mathrm{kJ}\) one early experiment data of the HD pressure of \(\sim 150\) Mbar without more direct experimental evidence and the detailed analyses to confirm the reliability of HD pressure boosting and smoothing, which are the key effects for our HD scheme.  
+
+<|ref|>text<|/ref|><|det|>[[101, 278, 897, 315]]<|/det|>
+In our previous manuscript, we analyzed new and published experimental data in detail to demonstrate and confirm the HD pressure boosting and smoothing effects.  
+
+<|ref|>text<|/ref|><|det|>[[100, 318, 898, 638]]<|/det|>
+Firstly, with the new DD laser energy of \(4.0\mathrm{kJ}\) and a Mo shield instead of Al, we obtained the new experimental result of the HD pressure of \(\sim 155\) Mbar, and with DD laser energy of \(3.6\mathrm{kJ}\) checked the result of \(\sim 150\) Mbar in the paper issued in HEDP. Thus, we further confirmed the HD pressure- boosting effect physically with more experimental data. In addition, we first showed HD pressure smoothing and observed the phenomenon of the HD shock merged with the ID shock (Fig. 3c), both are important for stable implosion. Second, we showed how under keeping the radiation temperature of \(200\mathrm{eV}\) unchanged the experimental target is designed by scaling down the size of the spherical hohlraum in the ignition target, which makes the experimental results can directly scale up to the ignition target experimentally. Finally, in the manuscript, we found that with a pre- offered ID corona plasma background of radiation temperature \(200\mathrm{eV}\) , tuning the distance between the critical surface and the radiation front and the slowing down length, the supersonic- electronic- heat wave converted by the DD laser intensity of \(1.8\mathrm{PW / cm}^2\) can provide a perfect bulldozer effect (a compressive wave) to generate significant HD pressure boosting and smoothing effect. we further discussed the time evolution process of indirect- driven and hybrid- driven shock velocities experimentally for the planar target and deduced the boosted and smoothed pressure in the density plateau by the simulations matched with experimental results in the quartz. This makes the manuscript a new paper essentially different from the paper in HEDP.  
+
+<|ref|>text<|/ref|><|det|>[[100, 641, 898, 860]]<|/det|>
+As for why the key results of peak radiation temperature and laser backscatter are unchanged. Because the peak radiation temperature of \(200\mathrm{eV}\) is the optimal temperature in the design of the ignition target, which can well provide the conditions of HD pressure boosting and smoothing, and also the right ID energy for the ignition target, in addition, according to the indirect- drive energy balance relationship, when the semicylindrical hohlraum used in the experiments is designed by scaling down the spherical hohlraum in the ignition target, the peak radiation temperature of \(T_{r} = 200\mathrm{eV}\) is the same as in the ignition target, which is beneficial to scaling up the experimental results to the ignition target. This is why peak radiation temperature \(T_{r} = 200\mathrm{eV}\) is unchanged in the review article in HEDP and the manuscript. In addition, the same radiation temperature heating the surface of the CH ablator leads to the same plasma environment for LPI, and therefore, for the DD laser intensity of \(\sim 1.8\mathrm{PW / cm}^2\) , the backscatter and the hot- electro energy should be close to those in similar experiments, which is plotted in Fig. 5c.  
+
+<|ref|>text<|/ref|><|det|>[[100, 883, 897, 920]]<|/det|>
+2. Comment: "Given the incremental nature of this work, I do not find it suitable for publishing in Nature Communications. However, the topic of hybrid-drive, and new experimental results for ICF, remains highly
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 75, 898, 154]]<|/det|>
+impactful and relevant to the Nature readership. I therefore recommend that the authors re- write the manuscript to avoid repeating previously published work, and to identify the novel and high- impact results of the present work. I also suggest that the authors pay attention to the clarity and conciseness of the new manuscript, since I found the current submission difficult to read and somewhat unclear in some places."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 158, 185, 173]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 177, 899, 355]]<|/det|>
+Although we think our previous manuscript is essentially different from the paper issued in HEDP, we, to respond to the reviewer's comment of "to avoid repeating previously published work, and to identify the novel and high- impact results of the present work", in rewriting the manuscript, added new experimental results with a hemispherical ablator target recently performed on SG- III. Under the same DD laser energies of \(\sim 3.6 \mathrm{kJ}\) and \(\sim 4.0 \mathrm{kJ}\) and radiation temperature of \(200 \mathrm{eV}\) , the hemispherical target due to the spherical convergence effect provided the peak HD pressure achieved 170- 180 Mbar, larger than 150- 155 Mbar of the planar target, as seen in Fig. 3c for the HD shock velocity and in Fig. 4c for the HD pressure of 180 Mbar. These results are the latest and greatest for the HD pressure experimentally to date and we discussed them in the text in detail.  
+
+<|ref|>text<|/ref|><|det|>[[120, 358, 668, 375]]<|/det|>
+We also tried to make some changes in the text to make the reading clear.  
+
+<|ref|>sub_title<|/ref|><|det|>[[377, 397, 620, 414]]<|/det|>
+## III. Responses to Referee #3  
+
+<|ref|>text<|/ref|><|det|>[[101, 438, 898, 496]]<|/det|>
+1. Comment: "Although the manuscript is well-written from the point-of-view of the results being presented in a logical and coherent manner, there are a large number of grammatical errors that make it difficult to read."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 500, 185, 515]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[115, 520, 864, 537]]<|/det|>
+Sorry, we have made an effort to correct some grammatical errors in the present revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[101, 560, 899, 597]]<|/det|>
+2. Comment: "The abstract (and elsewhere) states that the HD shock "stops the asymmetric ID shock." I am not sure what this statement is supposed to mean, but certainly it is not meant to be taken literally."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 602, 185, 617]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[120, 621, 585, 638]]<|/det|>
+We wrote it too simply in the abstract, more details are below:  
+
+<|ref|>text<|/ref|><|det|>[[101, 641, 898, 738]]<|/det|>
+A strong symmetric HD shock driven by the ideal HD pressure rapidly entering the imploding capsule collides in the opposite directions with the asymmetric relatively weak ID shock which is reflected from the center of the hotspot after pre- compressed the fuel and just arriving at the interface of the hotspot, and the ID shock reflected inward becomes weaker and is quickly caught up, swallowed and merged by the strong HD shock. Thus, the asymmetric ID shock in the early implosion stage is suppressed to prevent it from further asymmetric implosion.  
+
+<|ref|>text<|/ref|><|det|>[[120, 742, 545, 758]]<|/det|>
+We have made corresponding changes in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[101, 781, 898, 858]]<|/det|>
+3. Comment: "As the HD scheme is proposed as an alternative to DD/ID, it would help if the manuscript was more explicit about the advantages/disadvantages of the HD scheme relative to those schemes. For example, the HD scheme provides increased ablation pressure relative to ID, but how does the pressure compare to DD alone?"  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 864, 185, 878]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 883, 897, 920]]<|/det|>
+In the DD scheme, its advantage is having a high conversion efficiency of laser energy to the capsule, however, due to the mass ablation rate for electrons smaller than that for radiation the electron- conduction
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[100, 75, 898, 355]]<|/det|>
+region between the critical surface and the electron ablation front is relatively narrow, which easily transfers the nonuniformities from laser imprinting and beam overlapping to the imploding capsule, especially in the early stage, resulting in hydrodynamic instabilities. In addition, at the electron ablation front, the electron ablation pressure driving implosion is \(P_{DD} = \Gamma \rho_{DD}T_e\) , where \(\Gamma\) is a pressure constant for an ideal gas, the electron corona density \(\rho_{DD}\) is approximately \(2\rho_{c}\) with \(\rho_{c}\) the critical density, and the electron temperature is \(T_{e} \propto I_{L}^{2 / 3}\) where \(I_{L}\) the DD laser intensity (Ref. 9 in the manuscript). For the CH ablator, the critical density is \(\rho_{c} \approx 0.03 \mathrm{~g / cc}\) , and therefore, \(P_{DD}(\mathrm{Mbar}) \approx 90 \mathrm{x} I_{L}^{2 / 3}\) for the laser wavelength of 0.35 micron, where \(I_{L}\) in units of \(\mathrm{PW} / \mathrm{cm}^{2}\) . It is seen from the above discussion that due to the expansion of the ablator surface by high- temperature ablating, the plasma density \(\rho_{DD}\) at the electron ablation front is low compared with the normal ablator density. Therefore, in order to boost the DD pressure, the only way is to increase the electron temperature \(T_{e}\) or the DD laser intensity \(I_{L}\) . As an example, In the DD scheme, if the pressure driving the implosion is required to be \(P_{DD} \approx 300 \mathrm{Mbar}\) (Batani et al., Nucl. Fusion 54, 2014), the laser intensity \(I_{L}\) must be over \(5 \mathrm{PW} / \mathrm{cm}^{2}\) , resulting in severe LPI. In addition, laser imprinting and beam overlapping would lead to asymmetric implosion and hydrodynamic instabilities.  
+
+<|ref|>text<|/ref|><|det|>[[100, 358, 898, 698]]<|/det|>
+As for HD, a coupling of ID and DD, we use the advantages of ID (large mass ablation rate for radiation) and DD (high efficiency) to improve their shortcomings involving hydrodynamic instabilities. As we explained in the manuscript that before the DD laser arrives, the ID laser has offered a large distance \(\Delta R_{ID}\) between the critical surface and the radiation ablation front, and then a supersonic- electron- heat wave converted by the DD laser with the intensity of \(I_{L} = (1 - 2) \mathrm{PW} / \mathrm{cm}^{2}\) propagates in \(\Delta R_{ID}\) and slows down to a compressive wave before reaching the radiation ablation front, while thermal smoothing. This compressive wave, like a "bulldoze, isothermally compresses low ID corona plasma density \(\rho_{ID}\) , between the compressive wave front and the radiation ablation front, into high HD plasma density \(\rho_{HD}\) (fitted to \(\sim E_{4D}^{1 / 4}\) , see the manuscript) far greater than \(\rho_{ID}\) , and therefore, by increasing the plasma density rather than radiation temperature \(T_{r}\) , the radiation ablation pressure is changed into a new smoothed HD pressure, much greater than the radiation ablation pressure, i.e., \(P_{HD} / P_{ID} = \rho_{HD} / \rho_{ID} \gg 1\) . Clearly, the HD pressure increases with the DD laser energy \(E_{4D}^{1 / 4}\) and radiation temperature \(T_{r}\) and is independent of the laser intensity. As an example, for the modeling ignition target with the radius \(5 \mathrm{mm}\) of the spherical hohlraum and radiation temperature \(T_{r} = 200 \mathrm{eV}\) , at the radiation ablation front the HD maximal pressure \(P_{HD}\) can reach as high as 775 Mbar for \(E_{DD} = 825 \mathrm{kJ}\) while the radiation ablation pressure is only \(P_{ID} \sim 43 \mathrm{Mbar}\) , and at the electron ablation front for the intensities of \(I_{L} = (1 - 2) \mathrm{PW} / \mathrm{cm}^{2}\) the DD pressure is \(P_{DD} \sim 90 - 158 \mathrm{Mbar}\) .  
+
+<|ref|>text<|/ref|><|det|>[[101, 701, 897, 758]]<|/det|>
+As shown in the experiment, with the target consisting of the semicylindrical hohlraum and the planar ablator, for the radiation temperature \(T_{r} = 200 \mathrm{eV}\) , only using the low DD laser energy of \(3.6 - 4 \mathrm{kJ}\) , we obtain the HD pressure of 150- 155 Mbar, about 3.5 times the radiation ablation pressure.  
+
+<|ref|>text<|/ref|><|det|>[[101, 761, 898, 879]]<|/det|>
+In the present revised manuscript, we added the new experimental results by using a new target consisting of the same semicylindrical hohlraum and a new hemispherical ablator. Using the same DD laser energy of 3.5- 4.0 kJ and the same radiation temperature of 200 eV, the new experimental results, due to the spherical convergent effect, achieve HD pressures of 170- 180 Mbar higher than the HD pressures of 152- 155 Mbar in the planar target. These results further verified the HD pressure boosting and smoothing effects- the heart of the HD scheme and are discussed in detail in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[101, 903, 577, 920]]<|/det|>
+4. Comment: "In the caption of Fig. 1 LHEs should be LEHs."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[102, 77, 185, 92]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[118, 96, 730, 113]]<|/det|>
+Thanks for the reviewer's correction, we have revised it in the present manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[100, 138, 896, 177]]<|/det|>
+5. Comment: "Figure 2(c) shows an Al layer, but in the text this is referred to as a Mo layer (the figure caption says that it can be Mo or Al)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 181, 185, 196]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 200, 897, 276]]<|/det|>
+Sorry, we were careless. In the last manuscript, we did two rounds of experimental shots. In the first shot, with the DD laser energy of \(3.6\mathrm{kJ}\) , Al was used as the shielding layer as plotted in Figure 2(c), and in the second shot with the DD laser energy of \(4.0\mathrm{kJ}\) , the shielding layer was Mo. In the discussion, only Mo was used.  
+
+<|ref|>text<|/ref|><|det|>[[101, 279, 897, 318]]<|/det|>
+In the new experiments, with the DD laser energies of \(\sim 3.6\mathrm{kJ}\) and \(4.0\mathrm{kJ}\) the shielding layers for two shots all are Al, as shown in new Fig. 2(c).  
+
+<|ref|>text<|/ref|><|det|>[[101, 340, 897, 379]]<|/det|>
+6. Comment: "The manuscript states that the measured radiation temperatures agreed with predictions. It would be nice to include the predicted radiation temperature in Fig. 3(a)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 382, 185, 398]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 401, 896, 459]]<|/det|>
+Following the reviewer's suggestion, we have included the predicted radiation temperature in Fig. 3(a). After several experimental checks, the simulation results of temperature are consistent with the experimental results within the experimental error range of \(\pm 3\%\) .  
+
+<|ref|>text<|/ref|><|det|>[[101, 481, 585, 498]]<|/det|>
+7. Comment: "What does the colormap represent in Fig. 4(a)?"  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 503, 185, 518]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[120, 522, 485, 539]]<|/det|>
+The simulated density versus the time and space.  
+
+<|ref|>text<|/ref|><|det|>[[101, 562, 765, 579]]<|/det|>
+8. Comment: "In the first sentence on page 9 "reappearing" should be "reproducing."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 584, 185, 599]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[120, 603, 710, 620]]<|/det|>
+Thanks for the reviewer's correction, we have revised it in the new manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[101, 642, 775, 660]]<|/det|>
+9. Comment: "In the 2nd paragraph on page 11 "75J for DD" should be "75J for HD"."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 664, 185, 679]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 683, 897, 740]]<|/det|>
+Sorry for the unclear statement. In the previous manuscript, the total laser energies of \(\sim 47\mathrm{kJ}\) including \(43\mathrm{kJ}\) for ID and \(3.6\mathrm{- }4.0\mathrm{kJ}\) for DD. Only 75J is for DD lasers, covering only a small portion. We have revised the texts.  
+
+<|ref|>text<|/ref|><|det|>[[101, 764, 897, 861]]<|/det|>
+10. Comment: "The manuscript states in regards to the reduction in hot-electron fraction and backscatter fraction for the HD scheme: "Obviously, the thermal smoothing effect in HD significantly reduces the fractions compared with the traditional DD." This is not at all obvious to me. Traditionally I think of thermal smoothing as mitigating hydrodynamic instabilities, not laser plasma instabilities. The authors should explain how thermal smoothing is mitigating LPI in this case."  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 866, 185, 880]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 885, 897, 922]]<|/det|>
+In the DD scheme, when the DD lasers are incident on the ablator surface, especially in the early incident stage, it is inevitable that nonuniformities, like small density-depletion hollows or density bulges, appear in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[101, 75, 896, 112]]<|/det|>
+the rapidly expanding electron corona plasma. These nonuniformities are difficult to thermally smooth out due to the low mass ablation rate for electrons.  
+
+<|ref|>text<|/ref|><|det|>[[101, 115, 898, 316]]<|/det|>
+As an example, we discuss the self- focus and filamentation in these small hollows when the DD laser with the intensity \(I_{L} = 1.8 \mathrm{PW / cm^{2}}\) is incident on a quarter critical density surface with the plasma density of \(\rho \sim 0.0075 / \mathrm{cc}\) for the CH ablator, in which the "collision" doesn't dominate. In this case, the laser is soon focused into the hollows unstably due to the light refraction index increasing caused by electron number density depletion. For the DD laser intensity of \(I_{L} = 1.8 \mathrm{PW / cm^{2}}\) , electron temperature \(T_{e}\) can rise to \(\sim 4.5 \mathrm{keV}\) soon through inverse bremsstrahlung absorption while the ion temperature rises to \(T_{i} \sim 1.0 \mathrm{keV}\) by the time at least \(\sim 1 \mathrm{ns}\) (refer to ref. 9 in the manuscript). At the quarter critical surface, the thermal pressure of electrons is \(n_{e} T_{e} \sim 1.4 \times 10^{6} \mathrm{J / cc}\) while the laser radiation pressure is \(I_{L} / c \sim 0.6 \times 10^{5} \mathrm{J / cc}\) , where c light speed. Thus, if \(n_{e}\) is drained by \(10\%\) , due to the pressure balance the laser intensity must be increased to \(\sim 4.1 \mathrm{PW / cm^{2}}\) , which results in severe LPI.  
+
+<|ref|>text<|/ref|><|det|>[[101, 318, 899, 477]]<|/det|>
+In HD, as discussed in the manuscript, the ID laser provided the ID corona CH plasma (completely ionized) in local thermodynamic equilibrium with peak temperature \(T_{r} = T_{e} = T_{i} = 200 \mathrm{eV}\) within several nanoseconds before the arrival of the DD laser with the intensity of \(1.8 \mathrm{pw / cm^{2}}\) for the wavelength of 0.35 \(\mu \mathrm{m}\) , where \(T_{r}\) , \(T_{e}\) , \(T_{i}\) the temperatures for radiation (thermal X- rays), electron, and ion, respectively. Such LTE plasma offers a radiation- hydrodynamic sound velocity of \(C_{T} = 110 \mu \mathrm{m / ns}\) , the density nonuniformity around hollows or bulges can be thermally smoothed out, which, usually, have widths less than ten microns, within the time of tens of picoseconds. This is our explanation of why LPI in HD is smaller than that in the traditional DD scheme due to thermally smoothing.  
+
+<|ref|>text<|/ref|><|det|>[[120, 480, 630, 497]]<|/det|>
+We have briefly added these clarifications in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[101, 519, 897, 579]]<|/det|>
+11. Comment: "Despite the mitigation relative to DD, the backscatter and hot-electron production in the HD scheme is still quite high and is likely to get worse when scaling up to a high-gain design. Are these levels believed to be acceptable for the HD design or are there plans for mitigation?"  
+
+<|ref|>sub_title<|/ref|><|det|>[[102, 582, 185, 597]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[101, 600, 898, 780]]<|/det|>
+The fractions of the backscatter and the hot- electron production mainly depend on the parameters of the laser intensity \(I_{L}\) and the plasma environment of the density nonuniformities and density scale length \(L_{n}\) . In HD, the DD laser intensity of \(I_{L} = 1.8 \mathrm{PW / cm^{2}}\) , which serves for generating the supersonic- electronic- heat wave, and the ID corona plasma with radiation temperature of \(T_{r} = 200 \mathrm{eV}\) are the same as in the future ignition target, but there is some difference in the density scale length \(L_{n} \sim 300 \mu \mathrm{m}\) for the experiment target and \(\sim 400 \mu \mathrm{m}\) for the designed ignition target. Therefore, we think the fractions of the backscatter and hot electrons in the present experimental target and the future ignition target should not differ much. In addition, the backscattering fraction of \(\sim 5\%\) , as shown in the recent NIF experiments, is acceptable for the ignition target physically and in laser energy.  
+
+<|ref|>text<|/ref|><|det|>[[101, 782, 898, 920]]<|/det|>
+For the hot electrons, we would like to explain whether they are able to preheat the fuel. The hot- electron range in the CH plasma can be written theoretically in the form \(\rho R(\mathrm{g / cm^{2}}) \approx 0.59 \times 10^{- 5} E_{e}^{1.661}\) for \(E_{e} > 2 \mathrm{keV}\) (Phys. Plasmas 18, 022703(2011)), where \(E_{e}\) is the hot- electron energy in units of \(\mathrm{keV}\) . It is seen from the expression that even if the hot- electron energy is large enough, such as \(E_{e} = 100 \mathrm{keV}\) , the range is \(\rho R \approx 0.012 \mathrm{g / cm^{2}}\) . On the other hand, the simulations show that in the modeling ignition target, the area density of the remainder of the ablating CH ablator is about \(0.06 \mathrm{g / cm^{2}}\) at the time of the maximal implosion velocity in which \(0.02 \mathrm{g / cm^{2}}\) is from the density plateau piled up by the isothermal compression.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[100, 75, 896, 113]]<|/det|>
+This indicates that the hot electrons are stopped in CH and prevented from preheating the fuel in the ignition target.  
+
+<|ref|>text<|/ref|><|det|>[[100, 136, 895, 174]]<|/det|>
+12. Comment: What is the f-number of the drive beams, as this has a significant impact on SBS/SRs backscatter?  
+
+<|ref|>sub_title<|/ref|><|det|>[[101, 179, 185, 194]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[100, 198, 898, 234]]<|/det|>
+The f- number on the SG- III laser facility is 10. From the results of the HD experiment, the influence of this f- number on LPI seems to be small.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 145, 394, 161]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 861, 238]]<|/det|>
+I have read the revised version of the article and the answers to my comments from the authors. I am satisfied with their answers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 249, 868, 322]]<|/det|>
+I could still have some minor doubts( e.g., the result obtained with the hemispherical taret is till a bit "marginal") but in general I believe that this contribution is important. In this sense I disagree with the comments of one of the other referees who consider this work "incremental". Indeed I think this is not incremental.  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 868, 370]]<|/det|>
+I have also looked to the other comments raised by the other referees and the authors' replies, d again in general I find the answers correct, even if not always fully convincing.  
+
+<|ref|>text<|/ref|><|det|>[[115, 380, 626, 397]]<|/det|>
+In conclusion I would say that the manuscript can be published as it is  
+
+<|ref|>text<|/ref|><|det|>[[115, 465, 394, 482]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 521, 876, 614]]<|/det|>
+This is my report on a review of a second submission of this article following revisions by the authors. In my previous report I raised concerns that the results reported are an incremental change to those already published in a high energy density physics (HEDP) paper. I suggested that the authors "re- write the manuscript to avoid repeating previously published work, and to identify the novel and high- impact results of the present work."  
+
+<|ref|>text<|/ref|><|det|>[[115, 652, 860, 707]]<|/det|>
+In response, the authors provided a rebuttal that suggests that the new results provide experimental confirmation of the previous data (which I take to mean they have shown they are reproducible), and that provides a breakdown of three new insights that the new submission provides:  
+
+<|ref|>text<|/ref|><|det|>[[115, 716, 802, 753]]<|/det|>
+- Confirmed previous results at 3.6KJ indirect-drive laser energy and extended them with new experiments at 4.0KJ, providing more data to confirm pressure boosting with the HD scheme  
+
+<|ref|>text<|/ref|><|det|>[[115, 762, 840, 781]]<|/det|>
+- Demonstrated experimental changes which allow better scaling to ignition capable target designs  
+
+<|ref|>text<|/ref|><|det|>[[115, 791, 870, 827]]<|/det|>
+- Demonstrated the tuning of the DD laser to provide a "perfect" bulldozer effect and produce pressure boosting and smoothing  
+
+<|ref|>text<|/ref|><|det|>[[115, 837, 875, 874]]<|/det|>
+While the above statements are fairly clear, I don't see correspondingly clear statements in the updated manuscript, which to my mind significantly degrades the impact of the work.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 879, 143]]<|/det|>
+The authors also responded to my comments by adding a series of new data using hemispherical ablator targets as a comparison with the planar targets in the previous draft. This is a welcome addition to the paper but does not directly address my concern  
+
+<|ref|>text<|/ref|><|det|>[[115, 182, 867, 310]]<|/det|>
+I still believe that the current submission lacks a proper discussion of the context of new data with respect to the previously published results. For example, the key result in the abstract of this paper is that "This article reports that... such a boosted and smoothed HD pressure is first verified experimentally" while the previous HEDP paper contains the statement "Thus, the experiments have verified the HD pressure boost compared to the ID pressure". These statements suggest that the same conclusions are being presented in two separate works, which according to the rebuttal (not the manuscript) is not the case.  
+
+<|ref|>text<|/ref|><|det|>[[115, 349, 874, 421]]<|/det|>
+Given the more detailed description in this paper compared to the previous, and the potential high impact of the hybrid- drive scheme, I am prepared to recommend this paper for publication. However, I would like to stress that the addition of a short paragraph discussing how these experiments add to the understanding of the HD scheme will result in a higher quality article where the impact is more clear.  
+
+<|ref|>text<|/ref|><|det|>[[116, 490, 393, 505]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 546, 204, 562]]<|/det|>
+Dear Editor,  
+
+<|ref|>text<|/ref|><|det|>[[115, 574, 861, 645]]<|/det|>
+For the most part I am satisfied with the authors' response to the referees' comments. I think the question of whether similar results being previously published as part of a conference proceedings precludes publication in Nature Comms is a decision for the Editor. There are a couple of points in the authors' response that I am concerned about, but they are not critical to the main results:  
+
+<|ref|>text<|/ref|><|det|>[[115, 685, 859, 739]]<|/det|>
+1) The authors stated that electrons with energies less than 250 keV will be stopped in the CH ablator (and thus are not a preheat concern), but this is about an order of magnitude higher than the typical electron energies that are thought to be a preheat concern in direct-drive implosions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 778, 884, 905]]<|/det|>
+2) I do not think that thermal smoothing in the HD scheme is the source of reduced hot-electron production relative to traditional DD. Even if there were significant density nonuniformities in the corona of a DD implosion, filamentation and self-focusing are effectively absolute instabilities, so an enhanced seed level would not have a significant impact on these instabilities (also the enhanced seed would have to be wavelength and phase matched to the beam-driven perturbation). That being said, it is not obvious to me why the hot-electron production is reduced in the HD scheme, but it could be as simple as the fact that the ID beams increase Te, which would result in increased thresholds for
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 90, 877, 143]]<|/det|>
+filamentation and TPD. Regardless of the reason, LPI is not of primary importance to these results, and this is not the appropriate forum to be introducing an LPI mitigation mechanism that (to my knowledge) has not been previously discussed in the literature without adequate supporting evidence.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[432, 87, 574, 104]]<|/det|>
+## Response Letter  
+
+<|ref|>text<|/ref|><|det|>[[149, 106, 405, 122]]<|/det|>
+Article ID: NCOMMS- 22- 52906  
+
+<|ref|>title<|/ref|><|det|>[[149, 126, 830, 164]]<|/det|>
+# Title: Experimental confirmation of drive pressure boosting and smoothing for hybrid-drive inertial fusion at the 100-kJ laser facility  
+
+<|ref|>text<|/ref|><|det|>[[149, 166, 483, 183]]<|/det|>
+Authors: Ji Yan, Jiwei Li, X. T. He, et al.  
+
+<|ref|>text<|/ref|><|det|>[[148, 208, 851, 328]]<|/det|>
+We are grateful to the referees for their further insightful reviews and valuable comments on our revised manuscript. According to their comments, we have carefully revised the manuscript again. Following are the specific responses to the referees' comments point by point. With these responses, we believe that we have overcome all referees' criticisms and fulfilled the criteria set by them for accepting our manuscript for publication in Nature Communications. The changes in the revised manuscript are listed in the following.  
+
+<|ref|>sub_title<|/ref|><|det|>[[370, 368, 595, 385]]<|/det|>
+### I. Responses to Referee #1  
+
+<|ref|>text<|/ref|><|det|>[[148, 409, 795, 446]]<|/det|>
+Comment: "In conclusion I would say that the manuscript can be published as it is" Response: Thanks for the referee's recommendation of our manuscript for publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[370, 469, 605, 486]]<|/det|>
+## II. Responses to Referee #2  
+
+<|ref|>text<|/ref|><|det|>[[147, 525, 851, 645]]<|/det|>
+Comment 1: the key result in the abstract of this paper is that "This article reports that... such a boosted and smoothed HD pressure is first verified experimentally" while the previous HEDP paper contains the statement "Thus, the experiments have verified the HD pressure boost compared to the ID pressure". These statements suggest that the same conclusions are being presented in two separate works, which according to the rebuttal (not the manuscript) is not the case.  
+
+<|ref|>text<|/ref|><|det|>[[147, 649, 851, 829]]<|/det|>
+Response: The previous result using a planar ablator target at DD laser energy of \(3.6\mathrm{kJ}\) to experimentally confirm HD pressure boosting for the first time is obviously important, but indeed applying the term "verify" is not strict enough because there is only one experimental result without more repeated or similar results to support it. In the present manuscript, new more sufficient data from the hemispherical and planar ablator targets at DD laser energy of \(3.6\mathrm{- 4kJ}\) are provided to fully confirm HD pressure boosting and smoothing, while among these data, the experiment on the planar target at DD laser energy of \(3.6\mathrm{kJ}\) is for "check and further confirm the previous result", which has been mentioned all in the abstract, the planar target experiment at DD laser of \(3.6\mathrm{kJ}\) , and the section of "Discussions and conclusions".  
+
+<|ref|>text<|/ref|><|det|>[[148, 872, 850, 910]]<|/det|>
+Comment 2: I would like to stress that the addition of a short paragraph discussing how these experiments add to the understanding of the HD scheme will result in a higher quality article
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[149, 87, 381, 102]]<|/det|>
+where the impact is more clear.  
+
+<|ref|>text<|/ref|><|det|>[[148, 106, 850, 164]]<|/det|>
+Response: In the present manuscript, we have already added explanations of how these experiments increase understanding of the HD scheme in the section "Discussions and Conclusions"  
+
+<|ref|>sub_title<|/ref|><|det|>[[370, 204, 612, 221]]<|/det|>
+## III. Responses to Referee #3  
+
+<|ref|>text<|/ref|><|det|>[[147, 245, 850, 323]]<|/det|>
+Comment 1: The authors stated that electrons with energies less than \(250\mathrm{keV}\) will be stopped in the CH ablator (and thus are not a preheat concern), but this is about an order of magnitude higher than the typical electron energies that are thought to be a preheat concern in directdrive implosions.  
+
+<|ref|>text<|/ref|><|det|>[[147, 326, 851, 506]]<|/det|>
+Comment 1: The authors stated that electrons with energies less than \(250\mathrm{keV}\) will be stopped in the CH ablator (and thus are not a preheat concern), but this is about an order of magnitude higher than the typical electron energies that are thought to be a preheat concern in directdrive implosions.  Response: In the last response to comment 11 of reviewer #3, we used hot electron energy of \(100\mathrm{keV}\) , from which with the expression of \(\rho R(\mathrm{g / cm}^2)\) we estimated the area density (the penetrating range) to be \(0.012\mathrm{g / cm}^2\) less than \(0.06\mathrm{g / cm}^2\) for the remaining CH ablator at the time of maximal implosion velocity. As for energy \(250\mathrm{keV}\) mentioned in the response to the comment to reviewer #2, we mean that for hot electrons with energy of \(250\mathrm{keV}\) the area density is just equal to the remaining CH area density of \(0.06\mathrm{g / cm}^2\) , and therefore, only these electrons with energies greater than \(250\mathrm{keV}\) can penetrate the remaining CH preheating the fuel or to say that those electrons with energies less than \(250\mathrm{keV}\) are trapped in the remaining CH.  
+
+<|ref|>text<|/ref|><|det|>[[147, 529, 851, 628]]<|/det|>
+In the ID scheme, the experiment on hot electron preheating fuel in the ignition- scale target for NIC (PRL 108, 135006(2012)) has shown that with ID laser of \(1.3\mathrm{MJ}\) , the energy of energetic electron \(>100\mathrm{keV}\) deposited into the CH ablator is about \(570\pm 250\mathrm{J}\) , and \(5\pm 3\mathrm{J}\) (an upper bound) is absorbed in the DT ice, showing an acceptable increase of \(3.5\%\) for the adiabat of \(\alpha = 1.5\) . For the HD scheme, energetic electron preheating fuel needs further to be investigated.  
+
+<|ref|>text<|/ref|><|det|>[[147, 667, 851, 889]]<|/det|>
+Comment 2: I do not think that thermal smoothing in the HD scheme is the source of reduced hot- electron production relative to traditional DD. Even if there were significant density nonuniformities in the corona of a DD implosion, filamentation and self- focusing are effectively absolute instabilities, so an enhanced seed level would not have a significant impact on these instabilities (also the enhanced seed would have to be wavelength and phase matched to the beam- driven perturbation). That being said, it is not obvious to me why the hot- electron production is reduced in the HD scheme, but it could be as simple as the fact that the ID beams increase Te, which would result in increased thresholds for filamentation and TPD. Regardless of the reason, LPI is not of primary importance to these results, and this is not the appropriate forum to be introducing an LPI mitigation mechanism that (to my knowledge) has not been previously discussed in the literature without adequate supporting evidence.  
+
+<|ref|>text<|/ref|><|det|>[[145, 892, 848, 909]]<|/det|>
+Response: In the last response to the reviewer's comment, we mentioned that in the DD
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 85, 851, 328]]<|/det|>
+scheme, the DD laser with the intensity of \(1.8 \mathrm{PW / cm}^2\) provides a nonuniform electron corona plasma, which results in filamentation and self- focusing increase to make the DD laser intensity over \(1.8 \mathrm{PW / cm}^2\) forming new laser speckles in filamentation and self- focusing, resulting in a new LPI source, here LPI refers to the laser plasma instability of SRS, SBS, TPD, etc., not only the filamentation and self- focusing itself. While for the HD scheme, the ID laser provides a more uniform ID corona plasma with local equilibrium temperature \(T_r = T_e = T_i = 200 \mathrm{eV}\) , which results in filamentation and self- focusing mitigation, and therefore, LPI reduction by thermal smoothing compared with the DD scheme, here LPI, of course, involves SRS, SBS, and TPD, etc. This phenomenon is indeed seen in the comparative study by numerical simulation in which SBS and SRS occur rapidly with the formation of filamentations when the DD laser with the intensity of \(1.8 \mathrm{PW / cm}^2\) is incident on a nonuniform plasma, but these are not clear in the uniform plasma.  
+
+<|ref|>text<|/ref|><|det|>[[148, 350, 851, 553]]<|/det|>
+We think in a novel HD scheme different from DD and ID schemes, how the thermodynamic equilibrium ID plasma affects LPI is still understudied not too much. In our report, we mainly investigate pressure smoothing and smoothing, which is key physics in the HD scheme, and there is no need for further discussion of the LPI mitigation mechanism. We thank the reviewer's comment: "LPI is not of primary importance to these results, and this is not the appropriate forum to be introducing an LPI mitigation mechanism that has not been previously discussed in the literature without adequate supporting evidence". At the end of the section "Fractions of stimulated backscattering and hot electron energy" in the present manuscript, we have revised the previous discussions related to the LPI mitigation mechanism to change into the statement "So far, why the LPI in the HD scheme is smaller than the LPI in the DD scheme that has no radiation ablation is still not very clear, which needs further exploration and discussion."  
+
+<|ref|>sub_title<|/ref|><|det|>[[424, 592, 571, 608]]<|/det|>
+## IV. List of changes  
+
+<|ref|>sub_title<|/ref|><|det|>[[424, 647, 571, 663]]<|/det|>
+## IV. List of changes  
+
+<|ref|>text<|/ref|><|det|>[[148, 683, 850, 756]]<|/det|>
+1. In response to comment 1 from reviewer #2, in addition to explaining the previous result in this Response Letter, in the present manuscript, we also stated it in red font in the Abstract, the section "Radiation temperature and ID and DD velocities", and the section "Discussions and Conclusions".  
+
+<|ref|>text<|/ref|><|det|>[[148, 758, 850, 811]]<|/det|>
+In response to comment 2 from reviewer #2, we added explanations of how these experiments increase understanding of the HD scheme in the section "Discussions and Conclusions" in red font.  
+
+<|ref|>text<|/ref|><|det|>[[148, 814, 850, 904]]<|/det|>
+2. In response to comment 2 from reviewer #3, in the present manuscript at the end of the section "Fraction of stimulated backscattering and hot-electron energy", we revised the previous discussions related to the LPI mitigation mechanism to change into the statement "So far, why the LPI in the HD scheme is smaller than the LPI in the DD scheme that has no radiation ablation is still not very clear, which needs further exploration and discussion" in red font.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 300, 106]]<|/det|>
+REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 146, 393, 163]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 202, 857, 257]]<|/det|>
+The present manuscript has addressed my detailed questions, and I believe the changes have put the results of this work in context with previous publications. I believe that the current form of the manuscript is suitable for publication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 325, 393, 342]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[112, 381, 770, 399]]<|/det|>
+I am satisfied with the Authors' response and recommend the manuscript for publication.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[179, 107, 288, 116]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[179, 125, 333, 135]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[179, 143, 820, 172]]<|/det|>
+The present manuscript has addressed my detailed questions, and I believe the changes have put the results of this work in context with previous publications. I believe that the current form of the manuscript is suitable for publication.  
+
+<|ref|>text<|/ref|><|det|>[[179, 199, 333, 208]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[179, 217, 541, 226]]<|/det|>
+I am satisfied with the Authors' response and recommend the manuscript for publication.  
+
+<|ref|>sub_title<|/ref|><|det|>[[179, 272, 370, 283]]<|/det|>
+## Response to reviewer' s comment  
+
+<|ref|>text<|/ref|><|det|>[[179, 291, 820, 320]]<|/det|>
+We are grateful for the reviewers' acknowledgement of our work. Thanks all the reviews for their contribution to improving the quality of the paper and promotion of our understanding of the work.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a40eea9572c21a6973f0adc9033763b52d2f19965adff59ac8a944801a1f5cf5/images_list.json

@@ -0,0 +1,25 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure: \\(\\mathrm{Cu(N)_3(C)_2}\\) (left)",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "\\(\\mathrm{PF}_5\\) , \\(D_{3\\mathrm{h}}\\) symmetry(right)",
+    "footnote": [],
+    "bbox": [
+      [
+        542,
+        216,
+        740,
+        375
+      ]
+    ],
+    "page_idx": 5
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a40eea9572c21a6973f0adc9033763b52d2f19965adff59ac8a944801a1f5cf5/supplementary_0_Peer Review File__a40eea9572c21a6973f0adc9033763b52d2f19965adff59ac8a944801a1f5cf5.mmd

@@ -0,0 +1,306 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Heterospin Frustration in a Metal-Fullerene-Bonded Semiconductor Antiferromagnet  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+In my opinion, the work is interesting and well founded. The results are remarkable and constitute an important contribution to the field. Furthermore, the conclusions are duly supported by the results.  
+
+However, I have a couple of objections. The first one is not very important, but the second one I think it is mandatory that it be corrected:  
+
+1) In the Introduction section I miss some important references of calculations made on Computational studies about encapsulated lithium cation in C60. Without wishing to be exhaustive, some of them of relevance:  
+
+V. Bernshtein, I. Oref, Phys. Rev. A, 2000, 62, 03320. 
+H. U. Rehman, N. A. McKee, M. L. McKee, J. Comput.Chem., 2016, 37, 194. 
+I. Gonzalez-Veloso, J. Rodriguez-Otero, E. M. Cabaleiro-Lago, PCCP; 2019, 21, 16665.  
+
+2) The choice of the method for obtaining the absorption spectrum (TD-DFT method at the B3LYP/6-311G(d,p) level) is not the best possible one. It is widely established that to obtain an acceptable reproduction of the absorption spectrum, the use of a long-range corrected functional, such as LC- \(\omega\) PBE or CAM-B3LYP, is highly recommended.  
+
+Therefore, it would be mandatory that the authors repeat the quantum mechanical calculations with a more appropriate functional. Perhaps the results are not very different, but I think that the method recommended by the literature should always be tried.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The manuscript by Shen and coworkers details the structure and physical properties of a new material assembled from Li@C60 and Cu complexes. The synthetic and structural results are impressive but the authors make claims about the magnetic properties of this materials that are not supported by the data presented in the manuscript. For this reason, I cannot recommend publication in the current form. If the authors can bulk up the magnetism with more measurement and/or analysis section to support their claim of spin liquid, or re- write the paper to shift the focus, then I could reconsider my recommendation.  
+
+Here are a few things to consider.  
+
+1. The fit of 1/X data to extract the Weiss temperature is not valid as there is no truly linear region. The authors should subtract a temperature independent contribution. Take a look at this paper for an in-depth discussion: https://www.sciencedirect.com/science/article/pii/S0304885316324581 
+2. Spin liquid are notoriously difficult to probe. The authors should look into other methods to show there is still quantum fluctuations at low T, including muon spin relaxation and ac susceptibility. 
+3. Figure 2 is useless. Figure is also difficult to visualize the structure. I find the SI Figures are actually better than those in manuscript. 
+4. There is no discussion of the synthesis in the manuscript. 
+5. Page 3, Line 58: The authors claim that it is easy to deduce that Li@C60 can be doped with alkali metals to reach a superconducting state. I am not aware of any reports of superconductivity in Li@C60. I assume the authors meant something else. The text should be modified. 
+6. The authors talk about bandgap on line 169. It's not clear to me that these materials are actually band semiconductors. The authors should be careful with this. 
+7. I'm still not convinced the structure is geometrically frustrated. It relies on a number of hypothesis. Is the structure distorting at low T, which would remove the frustration? Is the radical localized at certain T?
+
+<--- Page Split --->
+
+This work reports the synthesis and characterization of an impressive 1D- coordination polymer, constructed upon the reaction of the dinuclear copper(II) complex [Cu2L(py)4] (L= 1,2,4,5- tetrakis(methanesulfonamido)benzene) with the lithium encapsulated fullerene salt (Li+@C60)(NTf2- ) (NTf2- = bis(trifluoromethane)sulfonamide anion), forming compound [[Cu4 (Li+@C60)L(py)4](NTf2)(hexane)]n (1). Compound 1 describes a 1D- coordination polymer (chain) in which each (Li+@C60) "molecule" coordinates to four Cu(II) centres belonging to two [Cu2L(py)4], while the remaining Cu(II) centres serve as nodes to a neighboring (Li+@C60) "molecule", thus resulting in the chain- like motif. Electrical conductivity studies reveal that 1 displays long- range electrical conductivity, while magnetic studies demonstrate that 1 may be treated as a spin- frustrated system.  
+
+The quality of the ms. is good, although at some points the authors should revise the use of the English language. All data presented fully agree with the analysis presented by the authors, while in addition the theoretical studies performed support their claims.  
+
+My personal view is that this work is quite novel and exciting, since it is the first time that such exotic coordination polymers exhibit promising electrical and magnetic properties. I believe the results reported in this work will be of great significance to the fields of inorganic/coordination chemistry, magnetochemistry, physics and materials, with potential applications in spintronic devices. Therefore, I am happy to suggest acceptance of the ms. in Nature Communications, since it will attract the wider readership of scientists working in the above- mentioned fields.  
+
+The points the authors should consider are the following ones:  
+
+1) Regarding the purity of the bulk samples included in the work (metallic precursor and compound 1) no data are presented besides the single-crystal structure. Therefore, for each compound: a) elemental C,H,N analysis should be provided, b) p-XRD diagrams should be provided along with comparison with the theoretical patterns, c) EDS measurements should be provided regarding the metallic content of compound 1.  
+
+2) In 1 the coordination environment of the Cu centres is described as distorted trigonal bipyramidal. The distortion should be quantified in terms of the deviation parameter.  
+
+3) No deviation is given in the magnetic exchange parameters and the goodness value of the fit is missing.  
+
+4) I can understand the existence of five exchange parameters for the fitting of the magnetic susceptibility, since there are indeed five different pathways. However, so many parameters often lead to overparameterization or to conclusions with no physical/real meaning. I would suggest the authors to perform the analysis with four J exchange parameters (i.e. J1-radical = J2-radical) and report/compare the two different models (in the SI).  
+
+Reviewer #4 (Remarks to the Author):  
+
+This manuscript reports a novel coordination polymer constructed by using Li+@C60 as acceptor and a specific Cu complex as donor. In the obtained framework, each Li+@C60 coordinates with four Cu2+ ions forming infinite 1D ladder- like patterns along the crystallographic b- axis. This is accompanied by the strong charge transfer from the Cu species to the fullerene core. As a result, the four Cu2+ ions (S = 1/2) and Li+@C- 60 (S = 1/2) interact with each other, showing magnetic frustration in a triangular- like lattice.  
+
+The paper is well- presented, the experiments and theoretical analysis are very solid. Although some fullerene- based 1D polymers have been reported previously, the assembly of Li+@C60 with a Cu complex is a new finding. Moreover, the obtained coordination polymer is conductive and features strong spin frustration. I would recommend the acceptance of this paper after addressing some minor issues.  
+
+1. Is the electrical conductivity of the crystal anisotropic? For example, along the ladder-chain direction (Figure 1d) versus other directions.  
+
+2. The Z value in the crystal data of 1 (Datablock: exp.860) should be corrected. Although the 1D ladder-like structure of this crystal is clear, I would still suggest the authors to do more refinement on
+
+<--- Page Split --->
+
+the anion NTf2- moieties and the disordered hexane molecules.  
+
+3. Some typos:  
+
+a. In page 2, line 28, the close brace is missing;  
+b. Please unify/check the writing of ions/compounds. For example, NTf2 versus NTf2-;  
+'[Li+@C60](SbCl6)' in Page 3, line 45.
+
+<--- Page Split --->
+
+## Response to reviewers' comments  
+
+We thank the four reviewers' great efforts in reviewing our manuscript. All modifications in the revised manuscript are highlighted with a yellow background. A point- by- point response to the reviewers' comments are shown below (reviewers' comments in blue, authors response in black):  
+
+Reviewer #1 (Remarks to the Author):  
+
+In my opinion, the work is interesting and well founded. The results are remarkable and constitute an important contribution to the field. Furthermore, the conclusions are duly supported by the results.  
+
+However, I have a couple of objections. The first one is not very important, but the second one I think it is mandatory that it be corrected:  
+
+Authors response: Thank you so much for reviewing our manuscript and positive comments.  
+
+1) In the Introduction section I miss some important references of calculations made on Computational studies about encapsulated lithium cation in C60. Without wishing to be exhaustive, some of them of relevance:  
+
+V. Bernshtein, I. Oref, Phys. Rev. A, 2000, 62, 03320. 
+H. U. Rehman, N. A. McKee, M. L. McKee, J. Comput.Chem., 2016, 37, 194. 
+I. Gonzalez-Veloso, J. Rodriguez-Otero, E. M. Cabaleiro-Lago, PCCP; 2019, 21, 16665.  
+
+Authors response: After carefully checking these references, we added these references in the revised manuscript (in ref 4, 6, 11) because they helped Nature Communications readers to better understand the \(\mathrm{C_{60}}\) electronic structures.  
+
+2) The choice of the method for obtaining the absorption spectrum (TD-DFT method at the B3LYP/6-311G(d,p) level) is not the best possible one. It is widely established that to obtain an acceptable reproduction of the absorption spectrum, the use of a long-range corrected functional, such as LC-αPBE or CAM-B3LYP, is highly recommended.  
+
+Therefore, it would be mandatory that the authors repeat the quantum mechanical calculations with a more appropriate functional. Perhaps the results are not very different, but I think that the method recommended by the literature should always be tried.  
+
+Authors response: In the revised manuscript, we tried the TD-DFT calculation by using CAM- B3LYP/de2tzvp and B3LYP/def2tzvp for \(\mathrm{Cu_2(L)(py)_4}\) . The electron transitions from HOMO to LUMO were observed at 920, 765, 760, 860 nm by using experimental, B3LYP/6- 311G(d,p), B3LYP/def2tzvp and CAM- B3LYP/def2tvp, respectively. The results are summarized in the table and figures. From the results, CAM- B3LYP/de2tzvp method showed better results.
+
+<--- Page Split --->
+
+
+<table><tr><td>Basis sets</td><td>band (1)</td><td>band (2)</td></tr><tr><td>Experimental</td><td>920 nm</td><td>444 nm</td></tr><tr><td>B3LYP/6-311G(d,p)</td><td>765 nm</td><td>412</td></tr><tr><td>B3LYP/def2tzvp</td><td>760 nm</td><td>340 nm</td></tr><tr><td>CAM-B3LYP/de2tzvp</td><td>860 nm</td><td>470 nm, 507 nm</td></tr></table>
+
+![](images/Figure_unknown_1.jpg)
+
+Reviewer #2 (Remarks to the Author):
+
+The manuscript by Shen and coworkers details the structure and physical properties of a new material assembled from Li@C60 and Cu complexes. The synthetic and structural results are impressive but the authors make claims about the magnetic properties of this materials that are not supported by the data presented in the manuscript. For this reason, I cannot recommend publication in the current form. If the authors can bulk up the magnetism with more measurement and/or analysis section to support their claim of spin liquid, or re-write the paper to shift the focus,then I could reconsider my recommendation.
+
+Authors response: Thank you so much for reviewing our manuscript and comments.
+
+Here are a few things to consider.
+
+1. The fit of 1/X data to extract the Weiss temperature is not valid as there is no truly linear region. The authors should subtract a temperature independent contribution.
+
+Take a look at this paper for an in-depth
+
+discussion: https://www.sciencedirect.com/science/article/pii/S0304885316324581
+
+Authors response: Thank you so much for your reference. Based on this reference, we subtracted the temperature independence paramagnetism ( \(χ_{0}\) ), the \(χ_{0}\) value was estimated to be \(8.8\times 10^{-4}\mathrm {\sim cm}^{3}\mathrm {\sim mol}^{-1}\) in 1 T field by extrapolating \((χ-χ_{0})\) vs \(T^{-1}\) curve to \(T^{-1}\rightarrow 0\) . In addition, we examined the effects on \(χ-χ_{0}\) by changing \(χ_{0}\) values from 0 to 5 \(\times 10^{-3}\mathrm {\sim cm}^{3}\mathrm {\sim mol}^{-1}\) . The Weiss temperatures ( \(θ_{cw}\) ) were varied from -305 to 25 K. In our case, the \(θ_{cw}\) value was estimated to be -190 K when \(χ_{0}=8.8\times 10^{-4}\mathrm {\sim cm}^{3}\mathrm {\sim mol}^{-1}\) .
+
+<--- Page Split --->
+![PLACEHOLDER_6_0]
+  
+
+2. Spin liquid are notoriously difficult to probe. The authors should look into other methods to show there is still quantum fluctuations at low T, including muon spin relaxation and ac susceptibility.  
+
+Authors response: To probe spin liquid, muon spin relaxation is powerful, however, we don't have such collaborators who can measure in the UK. Alternatively, we think heat capacity is another powerful tool to unearth hidden magnetic phases and has been widely used for the spin liquid system. Therefore, we performed low- temperature heat capacity and ac susceptibility measurements. Both the results indicated there is no anomaly in 2- 20 K, suggesting no magnetic ordering in this temperature range. Usually, heat capacity is given by \(\gamma T\) (density of state) \(+ \beta T^{3}\) (lattice). The \(\gamma\) is finite in metals, but it is 0 in insulators. Interestingly, the \(\gamma\) value of compound 1 is estimated to be \((12 \pm 2) \mathrm{mJ K}^{- 2} \mathrm{mol}^{- 1}\) in an semiconducting ground state \((T < 7 \mathrm{K})\) and is quite similar to that of quantum spin liquids candidates in organic and inorganic crystals as reported by the following two papers: 1) Kitagawa, K., Takayama, T., Matsumoto, Y. et al. A spin- orbital- entangled quantum liquid on a honeycomb lattice. Nature 554, 341- 345 (2018). 2) Yamashita, S., Nakazawa, Y., Oguni, M. et al. Thermodynamic properties of a spin- 1/2 spin- liquid state in a \(\kappa\) - type organic salt. Nature Phys 4, 459- 462 (2008). These results indicated that compound 1 is a possible candidate of QSL. We added these results to the revised manuscript to give more evidence of spin liquid of 1.  
+
+We added these results to the revised manuscript to give more evidence of spin liquid of 1.
+
+<--- Page Split --->
+![PLACEHOLDER_7_0]
+  
+
+3. Figure 2 is useless. Figure is also difficult to visualize the structure. I find the SI Figures are actually better than those in manuscript.  
+
+Authors response: Thank you so much for your suggestions. We moved Figure 2 to the Supporting Information and Figure S1 moved to the manuscript. In addition, we rearranged the figures in the revised manuscript.  
+
+4. There is no discussion of the synthesis in the manuscript.  
+
+Authors response: We added short discussion about the synthetic procedure in the revised manuscript.  
+
+5. Page 3, Line 58: The authors claim that it is easy to deduce that \(\mathrm{Li@C60}\) can be doped with alkali metals to reach a superconducting state. I am not aware of any reports of superconductivity in \(\mathrm{Li@C60}\) . I assume the authors meant something else. The text should be modified.  
+
+Authors response: We are sorry for this confusing expression. We agree that there is no superconductivity reported so far about \(\mathrm{Li@C60}\) . In this text, we would like to express the potential presence of superconductivity in \(\mathrm{Li@C60}\) by comparison of A3C60 superconductors. We have revised this text as 'It is possible that \(\mathrm{Li^{+}@C_{60}}\) can be doped by alkali metals to produce \(\mathrm{A_3(Li@C_{60})}\) ( \(\mathrm{A} = \mathrm{K}^{+}\) , \(\mathrm{Rb}^{+}\) and \(\mathrm{Cs}^{+}\) ) species in imitation of \(\mathrm{M_3C_{60}}\) superconductors with three electrons accommodated in the triply degenerated LUMO.'  
+
+6. The authors talk about bandgap on line 169. It's not clear to me that these materials are actually band semiconductors. The authors should be careful with this. Authors response: We revised this sentence in the revised manuscript.  
+
+7. I'm still not convinced the structure is geometrically frustrated. It relies on a number of hypothesis. Is the structure distorting at low T, which would remove the frustration? Is the radical localized at certain T?  
+
+Authors response: Thanks for your good question. In the revised manuscript, we determined the low- temperature single- crystal structures by synchrotron radiation measurements CCDC NO: 2121505- 2121509. From the structure and PXRD analysis, first, we did not observe the structural changes in 25- 300 K; second, \(\mathrm{Li^{+}}\) ion inside the \(\mathrm{C_{60}}\) cage was found to be split into half and localized at a certain position from 100 K to 25 K. The position of \(\mathrm{Li^{+}}\) ion remained unchanged below 100 K. Above 100
+
+<--- Page Split --->
+
+K, we did not detect the \(\mathrm{Li^{+}}\) position due to the fast motion of the \(\mathrm{Li^{+}}\) ion. As the localized \(\mathrm{Li^{+}}\) ion would strongly attract the negative radical (due to \(\mathrm{Li^{+}}\) - C bonds), the radicals should localize on the six- carbon ring (pink carbon atoms in Figure 2).  
+
+![PLACEHOLDER_8_0]
+  
+
+Reviewer #3 (Remarks to the Author):  
+
+This work reports the synthesis and characterization of an impressive 1D- coordination polymer, constructed upon the reaction of the dinuclear copper(II) complex [Cu2L(py)4] (L= 1,2,4,5- tetrakis(methanesulfonamido)benzene) with the lithium encapsulated fullerene salt ( \(\mathrm{Li + @C60}\) ) (NTf2- ) (NTf2- = bis(trifluoromethane)sulfonamide anion), forming compound {[Cu4 ( \(\mathrm{Li + @C60}\) )L(py)4](NTf2)(hexane)}n (1). Compound 1 describes a 1D- coordination polymer (chain) in which each ( \(\mathrm{Li + @C60}\) ) “molecule” coordinates to four Cu(II) centres belonging to two [Cu2L(py)4], while the remaining Cu(II) centres serve as nodes to a neighboring ( \(\mathrm{Li + @C60}\) ) “molecule”, thus resulting in the chain- like motif. Electrical conductivity studies reveal that 1 displays long- range electrical conductivity, while magnetic studies demonstrate that 1 may be treated as a spin- frustrated system.  
+
+The quality of the ms. is good, although at some points the authors should revise the use of the English language. All data presented fully agree with the analysis presented by the authors, while in addition the theoretical studies performed support their claims.  
+
+My personal view is that this work is quite novel and exciting, since it is the first time that such exotic coordination polymers exhibit promising electrical and magnetic properties. I believe the results reported in this work will be of great significance to the fields of inorganic/coordination chemistry, magnetochemistry, physics and materials, with potential applications in spintronic devices. Therefore, I am happy to suggest acceptance of the ms. in Nature Communications, since it will attract the wider readership of scientists working in the above- mentioned fields.  
+
+Authors response: Thank you so much for your positive comments.  
+
+The points the authors should consider are the following ones:
+
+<--- Page Split --->
+
+1) Regarding the purity of the bulk samples included in the work (metallic precursor and compound 1) no data are presented besides the single-crystal structure. Therefore, for each compound: a) elemental C,H,N analysis should be provided, b) p-XRD diagrams should be provided along with comparison with the theoretical patterns, c) EDS measurements should be provided regarding the metallic content of compound 1. Authors response: The elemental (CHN) analysis was performed in the Research and Analytical Center for Giant Molecules at Tohoku University; the results are tabulated as below:  
+
+<table><tr><td>Compounds</td><td>C (%) Cal. / Exp.</td><td>H (%) Cal. / Exp.</td><td>N (%) Cal. / Exp.</td></tr><tr><td>Cu2(L)(py)4</td><td>40.49 / 40.46</td><td>3.85 / 3.86</td><td>12.59 / 12.47</td></tr><tr><td>{[Cu4(Li+@C60)L(py)4](NTf2)(hexane)}n</td><td>55.76 / 55.34</td><td>2.29 / 1.96</td><td>5.97 / 5.71</td></tr></table>  
+
+The PXRD patterns for \(\mathrm{Cu_2(L)(py)_4}\) (upper) and compound 1 (down).  
+
+![PLACEHOLDER_9_0]
+  
+
+2) In 1 the coordination environment of the Cu centres is described as distorted trigonal bipyramidal. The distortion should be quantified in terms of the deviation
+
+<--- Page Split --->
+
+parameter.  
+
+Authors response: Thank you so much for your comments. Unlike the \(\mathrm{PF}_5\) and \(\mathrm{PCl}_5\) , they have a D3h symmetry, \(\mathrm{Cu(N)_3(C)_2}\) has a distorted trigonal bipyramidal geometry because of the spherical surface of \(\mathrm{C}_{60}\) and the two carbon atoms are bonded.  
+
+Therefore, it shows very low symmetry and lacks of axis. Compared to trigonal bipyramidal geometry with a \(D_{3\mathrm{h}}\) symmetry, the deviation parameters are defined as (angle- \(90^{\circ}) / 90^{\circ}\) and \((180^{\circ} - \mathrm{angle}) / 180^{\circ}\) and calculated to be \(15.6\%\) , \(21.1\%\) and \(22.2\%\) .  
+
+![PLACEHOLDER_10_0]
+
+<center>Figure: \(\mathrm{Cu(N)_3(C)_2}\) (left) </center>  
+
+![PLACEHOLDER_10_1]
+
+<center>\(\mathrm{PF}_5\) , \(D_{3\mathrm{h}}\) symmetry(right) </center>  
+
+\(\angle \mathrm{N5Cu1N2} = 104.1^{\circ}\) \(\angle \mathrm{N5Cu1N1} = 104.1^{\circ}\) \(\angle \mathrm{N5Cu1C1} = 109.0^{\circ}\) \(\angle \mathrm{N5Cu1C1} = 141.9^{\circ}\)  
+
+So, deviation parameters \(\mathrm{P}\) are:  
+
+\(\mathrm{P1} = (104.1 - 90) / 90 * 100\% = 15.6\%\) \(\mathrm{P2} = (109 - 90) / 90 * 100\% = 21.1\%\) \(\mathrm{P3} = (180 - 140) / 180 * 100\% = 22.2\%\)  
+
+3) No deviation is given in the magnetic exchange parameters and the goodness value of the fit is missing.  
+
+Authors response: The effective exchange coupling parameters of \(J_{\mathrm{Cu - Cu}} = - 170\pm 5\mathrm{K}\) and \(J_{1 - C_{60}} = J_{2 - C_{60}} = - 185\pm 3\mathrm{K}\) with \(g_{\mathrm{Cu}} = 2.09(0)\) were obtained after the best fit \((g_{\mathrm{C_{60}}}\) was fixed to 2.0). The \(R\) - value is \(1.9\times 10^{- 3}\)  
+
+4) I can understand the existence of five exchange parameters for the fitting of the magnetic susceptibility, since there are indeed five different pathways. However, so many parameters often lead to overparameterization or to conclusions with no physical/real meaning. I would suggest the authors to perform the analysis with four J exchange parameters (i.e. J1-radical = J2-radical) and report/compare the two different models (in the SI).  
+
+Authors response: We reduced the parameters by only considering the exchange coupling \(J_{\mathrm{Cu - Cu}}\) and \(J_{\mathrm{Cu - C_{60}}}\) . To obtain the \(J\) values, we used the following spin Hamiltonian \(\widehat{H}\) using equations (1) by considering two kinds of exchange coupling \(J_{\mathrm{Cu - Cu}}\) and \(J_{\mathrm{Cu - C_{60}}}\) . \(\begin{array}{r}\widehat{H} = - 2J_{\mathrm{Cu - Cu}}(\hat{S}_{\mathrm{Cu1}}\cdot \hat{S}_{\mathrm{Cu2}} + \hat{S}_{\mathrm{Cu3}}\cdot \hat{S}_{\mathrm{Cu4}}) - 2J_{\mathrm{Cu - C_{60}}}(\hat{S}_{\mathrm{Cu1}}\cdot \hat{S}_{\mathrm{C_{60}}} + \hat{S}_{\mathrm{Cu2}}\cdot \hat{S}_{\mathrm{C_{60}}} + \hat{S}_{\mathrm{Cu3}}\cdot \hat{S}_{\mathrm{C_{60}}} + \hat{S}_{\mathrm{Cu4}}\cdot \hat{S}_{\mathrm{C_{60}}}) \end{array}\) \(+\mu_{B}\bar{B}\cdot g_{\mathrm{Cu}}\cdot \hat{S}_{\mathrm{Cu}} + \mu_{B}g_{\mathrm{C_{60}}}\bar{B}\hat{S}_{\mathrm{C_{60}}}\) (1)
+
+<--- Page Split --->
+
+To avoid overparameterization, we assumed the interactions of \(J_{14}\) and \(J_{23}\) are ignored compared to \(J_{12}\) , \(J_{34}\) and \(J_{\mathrm{Cu - C_{60}}}\) due to their long metal- metal distances and the interactions between \(\mathrm{C_{60}^{- }}\) and \(\mathrm{Cu1}\) , \(\mathrm{Cu2}\) are equivalent. The effective exchange coupling parameters of \(J_{\mathrm{Cu - Cu}} = - 170\pm 5\) K and \(J_{1 - \mathrm{C_{60}^{- }}} = J_{2 - \mathrm{C_{60}^{- }}} = - 185\pm 3\) K with \(g_{\mathrm{Cu}} = 2.09(0)\) were obtained after the best fit \((g_{\mathrm{C_{60}^{- }}})\) was fixed to 2.0).  
+
+Reviewer #4 (Remarks to the Author):  
+
+This manuscript reports a novel coordination polymer constructed by using \(\mathrm{Li + @C60}\) as acceptor and a specific Cu complex as donor. In the obtained framework, each \(\mathrm{Li + @C60}\) coordinates with four \(\mathrm{Cu2 + }\) ions forming infinite 1D ladder- like patterns along the crystallographic b- axis. This is accompanied by the strong charge transfer from the Cu species to the fullerene core. As a result, the four \(\mathrm{Cu2 + }\) ions ( \(\mathrm{S} = \frac{1}{2}\) ) and \(\mathrm{Li + @C_{60}^{- } - 60}\) ( \(\mathrm{S} = \frac{1}{2}\) ) interact with each other, showing magnetic frustration in a triangular- like lattice.  
+
+The paper is well- presented, the experiments and theoretical analysis are very solid. Although some fullerene- based 1D polymers have been reported previously, the assembly of \(\mathrm{Li + @C60}\) with a Cu complex is a new finding. Moreover, the obtained coordination polymer is conductive and features strong spin frustration. I would recommend the acceptance of this paper after addressing some minor issues.  
+
+1. Is the electrical conductivity of the crystal anisotropic? For example, along the ladder-chain direction (Figure 1d) versus other directions.  
+
+Authors response: Thank you so much for your positive comments. Unfortunately, the single- crystals morphology is not so nice and the crystal size is relatively small. We cannot determine the \(a\) , \(b\) , \(c\) axes by using such a single crystal. However, we can expect the conductive pathway from the crystal structure. One conductive pathway could be the 1D ladder- like chain as electrons move from the donor \(\mathrm{Cu_2(L)(py)_2}\) to the acceptor \(\mathrm{Li^{+}@C_{60}}\) cage (through bond); the other pathway is the electrons move between \(\mathrm{Li^{+}@C_{60}}\) cages along the \(a\) - axis (pi- pi stacking, through space).  
+
+2. The Z value in the crystal data of 1 (Datablock: exp_860) should be corrected. Although the 1D ladder-like structure of this crystal is clear, I would still suggest the authors to do more refinement on the anion NTf2- moieties and the disordered hexane molecules.  
+
+Authors response: We resolved the crystal structures and reduced and explained the alert A level in the cif file.  
+
+## 3. Some typos:  
+
+a. In page 2, line 28, the close brace is missing; Authors response: We added it in the revised manuscript.  
+
+b. Please unify/check the writing of ions/compounds. For example, NTf2 versus
+
+<--- Page Split --->
+
+NTf2- ; '[Li+@C60](SbCl6)' in Page 3, line 45. Authors response: We corrected them in the revised manuscript.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+All my objections were fully addressed and therefore I think the paper is suitable for publication. Only a small mistake: in the legend of figure S11 the experimental data are not shown in orange but in black  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors made a valiant effort to answer my comments. Considering how challenging it is to characterize spin liquids and how careful the authors are in the revised manuscript, I recommend accepting this work for publication.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In the revised ms., the authors answered with clarity to all issues initially raised. Therefore, I am more than happy to suggest acceptance of the manuscript at its current form.  
+
+Reviewer #4 (Remarks to the Author):  
+
+I think the authors have fully addressed the concerns from all four reviewers. The texts and experimental data are very compelling. In my opinion it is now ready to be published without further change.
+
+<--- Page Split --->
+
+## Response to reviewers' comments (2nd round)  
+
+We thank the four reviewers again for reviewing our revised manuscript. A point- by- point response to the reviewers' comments are shown below (reviewers' comments in blue, authors response in black):  
+
+Reviewer #1 (Remarks to the Author):  
+
+All my objections were fully addressed and therefore I think the paper is suitable for publication.  
+
+Only a small mistake: in the legend of figure S11 the experimental data are not shown in orange but in black.  
+
+Authors response: Thank you so much for pointing it out. We have corrected them in the supporting information.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors made a valiant effort to answer my comments. Considering how challenging it is to characterize spin liquids and how careful the authors are in the revised manuscript, I recommend accepting this work for publication. Authors response: Thank you so much for your comments. We tried our best to probe the possible presence of a quantum spin liquid state in compound 1.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In the revised ms., the authors answered with clarity to all issues initially raised. Therefore, I am more than happy to suggest acceptance of the manuscript at its current form.  
+
+Authors response: Thank you so much for your acceptance.  
+
+Reviewer #4 (Remarks to the Author):  
+
+I think the authors have fully addressed the concerns from all four reviewers. The texts and experimental data are very compelling. In my opinion it is now ready to be published without further change. Authors response: Thank you so much for your recommendation.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a40eea9572c21a6973f0adc9033763b52d2f19965adff59ac8a944801a1f5cf5/supplementary_0_Peer Review File__a40eea9572c21a6973f0adc9033763b52d2f19965adff59ac8a944801a1f5cf5_det.mmd

@@ -0,0 +1,409 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[67, 110, 361, 138]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[135, 155, 861, 210]]<|/det|>
+# Heterospin Frustration in a Metal-Fullerene-Bonded Semiconductor Antiferromagnet  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 782]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 911, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 83, 312, 98]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[120, 112, 402, 128]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 142, 868, 172]]<|/det|>
+In my opinion, the work is interesting and well founded. The results are remarkable and constitute an important contribution to the field. Furthermore, the conclusions are duly supported by the results.  
+
+<|ref|>text<|/ref|><|det|>[[118, 186, 870, 216]]<|/det|>
+However, I have a couple of objections. The first one is not very important, but the second one I think it is mandatory that it be corrected:  
+
+<|ref|>text<|/ref|><|det|>[[118, 230, 855, 270]]<|/det|>
+1) In the Introduction section I miss some important references of calculations made on Computational studies about encapsulated lithium cation in C60. Without wishing to be exhaustive, some of them of relevance:  
+
+<|ref|>text<|/ref|><|det|>[[118, 272, 775, 318]]<|/det|>
+V. Bernshtein, I. Oref, Phys. Rev. A, 2000, 62, 03320. 
+H. U. Rehman, N. A. McKee, M. L. McKee, J. Comput.Chem., 2016, 37, 194. 
+I. Gonzalez-Veloso, J. Rodriguez-Otero, E. M. Cabaleiro-Lago, PCCP; 2019, 21, 16665.  
+
+<|ref|>text<|/ref|><|det|>[[118, 332, 866, 392]]<|/det|>
+2) The choice of the method for obtaining the absorption spectrum (TD-DFT method at the B3LYP/6-311G(d,p) level) is not the best possible one. It is widely established that to obtain an acceptable reproduction of the absorption spectrum, the use of a long-range corrected functional, such as LC- \(\omega\) PBE or CAM-B3LYP, is highly recommended.  
+
+<|ref|>text<|/ref|><|det|>[[118, 393, 870, 437]]<|/det|>
+Therefore, it would be mandatory that the authors repeat the quantum mechanical calculations with a more appropriate functional. Perhaps the results are not very different, but I think that the method recommended by the literature should always be tried.  
+
+<|ref|>text<|/ref|><|det|>[[120, 481, 402, 496]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 510, 878, 614]]<|/det|>
+The manuscript by Shen and coworkers details the structure and physical properties of a new material assembled from Li@C60 and Cu complexes. The synthetic and structural results are impressive but the authors make claims about the magnetic properties of this materials that are not supported by the data presented in the manuscript. For this reason, I cannot recommend publication in the current form. If the authors can bulk up the magnetism with more measurement and/or analysis section to support their claim of spin liquid, or re- write the paper to shift the focus, then I could reconsider my recommendation.  
+
+<|ref|>text<|/ref|><|det|>[[119, 629, 370, 644]]<|/det|>
+Here are a few things to consider.  
+
+<|ref|>text<|/ref|><|det|>[[115, 657, 876, 894]]<|/det|>
+1. The fit of 1/X data to extract the Weiss temperature is not valid as there is no truly linear region. The authors should subtract a temperature independent contribution. Take a look at this paper for an in-depth discussion: https://www.sciencedirect.com/science/article/pii/S0304885316324581 
+2. Spin liquid are notoriously difficult to probe. The authors should look into other methods to show there is still quantum fluctuations at low T, including muon spin relaxation and ac susceptibility. 
+3. Figure 2 is useless. Figure is also difficult to visualize the structure. I find the SI Figures are actually better than those in manuscript. 
+4. There is no discussion of the synthesis in the manuscript. 
+5. Page 3, Line 58: The authors claim that it is easy to deduce that Li@C60 can be doped with alkali metals to reach a superconducting state. I am not aware of any reports of superconductivity in Li@C60. I assume the authors meant something else. The text should be modified. 
+6. The authors talk about bandgap on line 169. It's not clear to me that these materials are actually band semiconductors. The authors should be careful with this. 
+7. I'm still not convinced the structure is geometrically frustrated. It relies on a number of hypothesis. Is the structure distorting at low T, which would remove the frustration? Is the radical localized at certain T?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 115, 872, 262]]<|/det|>
+This work reports the synthesis and characterization of an impressive 1D- coordination polymer, constructed upon the reaction of the dinuclear copper(II) complex [Cu2L(py)4] (L= 1,2,4,5- tetrakis(methanesulfonamido)benzene) with the lithium encapsulated fullerene salt (Li+@C60)(NTf2- ) (NTf2- = bis(trifluoromethane)sulfonamide anion), forming compound [[Cu4 (Li+@C60)L(py)4](NTf2)(hexane)]n (1). Compound 1 describes a 1D- coordination polymer (chain) in which each (Li+@C60) "molecule" coordinates to four Cu(II) centres belonging to two [Cu2L(py)4], while the remaining Cu(II) centres serve as nodes to a neighboring (Li+@C60) "molecule", thus resulting in the chain- like motif. Electrical conductivity studies reveal that 1 displays long- range electrical conductivity, while magnetic studies demonstrate that 1 may be treated as a spin- frustrated system.  
+
+<|ref|>text<|/ref|><|det|>[[118, 260, 868, 305]]<|/det|>
+The quality of the ms. is good, although at some points the authors should revise the use of the English language. All data presented fully agree with the analysis presented by the authors, while in addition the theoretical studies performed support their claims.  
+
+<|ref|>text<|/ref|><|det|>[[118, 305, 872, 393]]<|/det|>
+My personal view is that this work is quite novel and exciting, since it is the first time that such exotic coordination polymers exhibit promising electrical and magnetic properties. I believe the results reported in this work will be of great significance to the fields of inorganic/coordination chemistry, magnetochemistry, physics and materials, with potential applications in spintronic devices. Therefore, I am happy to suggest acceptance of the ms. in Nature Communications, since it will attract the wider readership of scientists working in the above- mentioned fields.  
+
+<|ref|>text<|/ref|><|det|>[[118, 393, 583, 407]]<|/det|>
+The points the authors should consider are the following ones:  
+
+<|ref|>text<|/ref|><|det|>[[118, 407, 870, 480]]<|/det|>
+1) Regarding the purity of the bulk samples included in the work (metallic precursor and compound 1) no data are presented besides the single-crystal structure. Therefore, for each compound: a) elemental C,H,N analysis should be provided, b) p-XRD diagrams should be provided along with comparison with the theoretical patterns, c) EDS measurements should be provided regarding the metallic content of compound 1.  
+
+<|ref|>text<|/ref|><|det|>[[118, 480, 861, 510]]<|/det|>
+2) In 1 the coordination environment of the Cu centres is described as distorted trigonal bipyramidal. The distortion should be quantified in terms of the deviation parameter.  
+
+<|ref|>text<|/ref|><|det|>[[118, 510, 848, 540]]<|/det|>
+3) No deviation is given in the magnetic exchange parameters and the goodness value of the fit is missing.  
+
+<|ref|>text<|/ref|><|det|>[[118, 541, 847, 615]]<|/det|>
+4) I can understand the existence of five exchange parameters for the fitting of the magnetic susceptibility, since there are indeed five different pathways. However, so many parameters often lead to overparameterization or to conclusions with no physical/real meaning. I would suggest the authors to perform the analysis with four J exchange parameters (i.e. J1-radical = J2-radical) and report/compare the two different models (in the SI).  
+
+<|ref|>text<|/ref|><|det|>[[118, 657, 401, 672]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 687, 875, 776]]<|/det|>
+This manuscript reports a novel coordination polymer constructed by using Li+@C60 as acceptor and a specific Cu complex as donor. In the obtained framework, each Li+@C60 coordinates with four Cu2+ ions forming infinite 1D ladder- like patterns along the crystallographic b- axis. This is accompanied by the strong charge transfer from the Cu species to the fullerene core. As a result, the four Cu2+ ions (S = 1/2) and Li+@C- 60 (S = 1/2) interact with each other, showing magnetic frustration in a triangular- like lattice.  
+
+<|ref|>text<|/ref|><|det|>[[118, 777, 872, 850]]<|/det|>
+The paper is well- presented, the experiments and theoretical analysis are very solid. Although some fullerene- based 1D polymers have been reported previously, the assembly of Li+@C60 with a Cu complex is a new finding. Moreover, the obtained coordination polymer is conductive and features strong spin frustration. I would recommend the acceptance of this paper after addressing some minor issues.  
+
+<|ref|>text<|/ref|><|det|>[[118, 851, 812, 880]]<|/det|>
+1. Is the electrical conductivity of the crystal anisotropic? For example, along the ladder-chain direction (Figure 1d) versus other directions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 881, 866, 910]]<|/det|>
+2. The Z value in the crystal data of 1 (Datablock: exp.860) should be corrected. Although the 1D ladder-like structure of this crystal is clear, I would still suggest the authors to do more refinement on
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 592, 98]]<|/det|>
+the anion NTf2- moieties and the disordered hexane molecules.  
+
+<|ref|>text<|/ref|><|det|>[[118, 99, 232, 113]]<|/det|>
+3. Some typos:  
+
+<|ref|>text<|/ref|><|det|>[[118, 114, 761, 160]]<|/det|>
+a. In page 2, line 28, the close brace is missing;  
+b. Please unify/check the writing of ions/compounds. For example, NTf2 versus NTf2-;  
+'[Li+@C60](SbCl6)' in Page 3, line 45.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[353, 86, 643, 103]]<|/det|>
+## Response to reviewers' comments  
+
+<|ref|>text<|/ref|><|det|>[[148, 122, 848, 196]]<|/det|>
+We thank the four reviewers' great efforts in reviewing our manuscript. All modifications in the revised manuscript are highlighted with a yellow background. A point- by- point response to the reviewers' comments are shown below (reviewers' comments in blue, authors response in black):  
+
+<|ref|>text<|/ref|><|det|>[[150, 215, 459, 232]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 251, 844, 305]]<|/det|>
+In my opinion, the work is interesting and well founded. The results are remarkable and constitute an important contribution to the field. Furthermore, the conclusions are duly supported by the results.  
+
+<|ref|>text<|/ref|><|det|>[[148, 307, 822, 343]]<|/det|>
+However, I have a couple of objections. The first one is not very important, but the second one I think it is mandatory that it be corrected:  
+
+<|ref|>text<|/ref|><|det|>[[148, 344, 808, 380]]<|/det|>
+Authors response: Thank you so much for reviewing our manuscript and positive comments.  
+
+<|ref|>text<|/ref|><|det|>[[147, 400, 840, 454]]<|/det|>
+1) In the Introduction section I miss some important references of calculations made on Computational studies about encapsulated lithium cation in C60. Without wishing to be exhaustive, some of them of relevance:  
+
+<|ref|>text<|/ref|><|det|>[[147, 455, 800, 528]]<|/det|>
+V. Bernshtein, I. Oref, Phys. Rev. A, 2000, 62, 03320. 
+H. U. Rehman, N. A. McKee, M. L. McKee, J. Comput.Chem., 2016, 37, 194. 
+I. Gonzalez-Veloso, J. Rodriguez-Otero, E. M. Cabaleiro-Lago, PCCP; 2019, 21, 16665.  
+
+<|ref|>text<|/ref|><|det|>[[148, 530, 800, 584]]<|/det|>
+Authors response: After carefully checking these references, we added these references in the revised manuscript (in ref 4, 6, 11) because they helped Nature Communications readers to better understand the \(\mathrm{C_{60}}\) electronic structures.  
+
+<|ref|>text<|/ref|><|det|>[[147, 603, 838, 696]]<|/det|>
+2) The choice of the method for obtaining the absorption spectrum (TD-DFT method at the B3LYP/6-311G(d,p) level) is not the best possible one. It is widely established that to obtain an acceptable reproduction of the absorption spectrum, the use of a long-range corrected functional, such as LC-αPBE or CAM-B3LYP, is highly recommended.  
+
+<|ref|>text<|/ref|><|det|>[[148, 697, 842, 769]]<|/det|>
+Therefore, it would be mandatory that the authors repeat the quantum mechanical calculations with a more appropriate functional. Perhaps the results are not very different, but I think that the method recommended by the literature should always be tried.  
+
+<|ref|>text<|/ref|><|det|>[[147, 770, 835, 881]]<|/det|>
+Authors response: In the revised manuscript, we tried the TD-DFT calculation by using CAM- B3LYP/de2tzvp and B3LYP/def2tzvp for \(\mathrm{Cu_2(L)(py)_4}\) . The electron transitions from HOMO to LUMO were observed at 920, 765, 760, 860 nm by using experimental, B3LYP/6- 311G(d,p), B3LYP/def2tzvp and CAM- B3LYP/def2tvp, respectively. The results are summarized in the table and figures. From the results, CAM- B3LYP/de2tzvp method showed better results.
+
+<--- Page Split --->
+<|ref|>table<|/ref|><|det|>[[148, 82, 848, 180]]<|/det|>
+
+<table><tr><td>Basis sets</td><td>band (1)</td><td>band (2)</td></tr><tr><td>Experimental</td><td>920 nm</td><td>444 nm</td></tr><tr><td>B3LYP/6-311G(d,p)</td><td>765 nm</td><td>412</td></tr><tr><td>B3LYP/def2tzvp</td><td>760 nm</td><td>340 nm</td></tr><tr><td>CAM-B3LYP/de2tzvp</td><td>860 nm</td><td>470 nm, 507 nm</td></tr></table>
+
+<|ref|>image<|/ref|><|det|>[[312, 210, 690, 434]]<|/det|>
+
+<|ref|>text<|/ref|><|det|>[[148, 462, 460, 474]]<|/det|>
+Reviewer #2 (Remarks to the Author):
+
+<|ref|>text<|/ref|><|det|>[[147, 499, 848, 641]]<|/det|>
+The manuscript by Shen and coworkers details the structure and physical properties of a new material assembled from Li@C60 and Cu complexes. The synthetic and structural results are impressive but the authors make claims about the magnetic properties of this materials that are not supported by the data presented in the manuscript. For this reason, I cannot recommend publication in the current form. If the authors can bulk up the magnetism with more measurement and/or analysis section to support their claim of spin liquid, or re-write the paper to shift the focus,then I could reconsider my recommendation.
+
+<|ref|>text<|/ref|><|det|>[[148, 648, 828, 661]]<|/det|>
+Authors response: Thank you so much for reviewing our manuscript and comments.
+
+<|ref|>text<|/ref|><|det|>[[148, 685, 421, 697]]<|/det|>
+Here are a few things to consider.
+
+<|ref|>text<|/ref|><|det|>[[148, 721, 833, 754]]<|/det|>
+1. The fit of 1/X data to extract the Weiss temperature is not valid as there is no truly linear region. The authors should subtract a temperature independent contribution.
+
+<|ref|>text<|/ref|><|det|>[[148, 759, 476, 772]]<|/det|>
+Take a look at this paper for an in-depth
+
+<|ref|>text<|/ref|><|det|>[[148, 777, 820, 790]]<|/det|>
+discussion: https://www.sciencedirect.com/science/article/pii/S0304885316324581
+
+<|ref|>text<|/ref|><|det|>[[147, 796, 850, 902]]<|/det|>
+Authors response: Thank you so much for your reference. Based on this reference, we subtracted the temperature independence paramagnetism ( \(χ_{0}\) ), the \(χ_{0}\) value was estimated to be \(8.8\times 10^{-4}\mathrm {\sim cm}^{3}\mathrm {\sim mol}^{-1}\) in 1 T field by extrapolating \((χ-χ_{0})\) vs \(T^{-1}\) curve to \(T^{-1}\rightarrow 0\) . In addition, we examined the effects on \(χ-χ_{0}\) by changing \(χ_{0}\) values from 0 to 5 \(\times 10^{-3}\mathrm {\sim cm}^{3}\mathrm {\sim mol}^{-1}\) . The Weiss temperatures ( \(θ_{cw}\) ) were varied from -305 to 25 K. In our case, the \(θ_{cw}\) value was estimated to be -190 K when \(χ_{0}=8.8\times 10^{-4}\mathrm {\sim cm}^{3}\mathrm {\sim mol}^{-1}\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[258, 81, 740, 380]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[147, 400, 819, 456]]<|/det|>
+2. Spin liquid are notoriously difficult to probe. The authors should look into other methods to show there is still quantum fluctuations at low T, including muon spin relaxation and ac susceptibility.  
+
+<|ref|>text<|/ref|><|det|>[[147, 456, 850, 760]]<|/det|>
+Authors response: To probe spin liquid, muon spin relaxation is powerful, however, we don't have such collaborators who can measure in the UK. Alternatively, we think heat capacity is another powerful tool to unearth hidden magnetic phases and has been widely used for the spin liquid system. Therefore, we performed low- temperature heat capacity and ac susceptibility measurements. Both the results indicated there is no anomaly in 2- 20 K, suggesting no magnetic ordering in this temperature range. Usually, heat capacity is given by \(\gamma T\) (density of state) \(+ \beta T^{3}\) (lattice). The \(\gamma\) is finite in metals, but it is 0 in insulators. Interestingly, the \(\gamma\) value of compound 1 is estimated to be \((12 \pm 2) \mathrm{mJ K}^{- 2} \mathrm{mol}^{- 1}\) in an semiconducting ground state \((T < 7 \mathrm{K})\) and is quite similar to that of quantum spin liquids candidates in organic and inorganic crystals as reported by the following two papers: 1) Kitagawa, K., Takayama, T., Matsumoto, Y. et al. A spin- orbital- entangled quantum liquid on a honeycomb lattice. Nature 554, 341- 345 (2018). 2) Yamashita, S., Nakazawa, Y., Oguni, M. et al. Thermodynamic properties of a spin- 1/2 spin- liquid state in a \(\kappa\) - type organic salt. Nature Phys 4, 459- 462 (2008). These results indicated that compound 1 is a possible candidate of QSL. We added these results to the revised manuscript to give more evidence of spin liquid of 1.  
+
+<|ref|>text<|/ref|><|det|>[[147, 751, 840, 788]]<|/det|>
+We added these results to the revised manuscript to give more evidence of spin liquid of 1.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[151, 88, 848, 228]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[147, 252, 822, 288]]<|/det|>
+3. Figure 2 is useless. Figure is also difficult to visualize the structure. I find the SI Figures are actually better than those in manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[148, 289, 828, 344]]<|/det|>
+Authors response: Thank you so much for your suggestions. We moved Figure 2 to the Supporting Information and Figure S1 moved to the manuscript. In addition, we rearranged the figures in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[147, 363, 630, 380]]<|/det|>
+4. There is no discussion of the synthesis in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[148, 382, 810, 417]]<|/det|>
+Authors response: We added short discussion about the synthetic procedure in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[147, 437, 833, 510]]<|/det|>
+5. Page 3, Line 58: The authors claim that it is easy to deduce that \(\mathrm{Li@C60}\) can be doped with alkali metals to reach a superconducting state. I am not aware of any reports of superconductivity in \(\mathrm{Li@C60}\) . I assume the authors meant something else. The text should be modified.  
+
+<|ref|>text<|/ref|><|det|>[[147, 512, 839, 640]]<|/det|>
+Authors response: We are sorry for this confusing expression. We agree that there is no superconductivity reported so far about \(\mathrm{Li@C60}\) . In this text, we would like to express the potential presence of superconductivity in \(\mathrm{Li@C60}\) by comparison of A3C60 superconductors. We have revised this text as 'It is possible that \(\mathrm{Li^{+}@C_{60}}\) can be doped by alkali metals to produce \(\mathrm{A_3(Li@C_{60})}\) ( \(\mathrm{A} = \mathrm{K}^{+}\) , \(\mathrm{Rb}^{+}\) and \(\mathrm{Cs}^{+}\) ) species in imitation of \(\mathrm{M_3C_{60}}\) superconductors with three electrons accommodated in the triply degenerated LUMO.'  
+
+<|ref|>text<|/ref|><|det|>[[147, 659, 843, 714]]<|/det|>
+6. The authors talk about bandgap on line 169. It's not clear to me that these materials are actually band semiconductors. The authors should be careful with this. Authors response: We revised this sentence in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[147, 733, 820, 787]]<|/det|>
+7. I'm still not convinced the structure is geometrically frustrated. It relies on a number of hypothesis. Is the structure distorting at low T, which would remove the frustration? Is the radical localized at certain T?  
+
+<|ref|>text<|/ref|><|det|>[[147, 790, 847, 900]]<|/det|>
+Authors response: Thanks for your good question. In the revised manuscript, we determined the low- temperature single- crystal structures by synchrotron radiation measurements CCDC NO: 2121505- 2121509. From the structure and PXRD analysis, first, we did not observe the structural changes in 25- 300 K; second, \(\mathrm{Li^{+}}\) ion inside the \(\mathrm{C_{60}}\) cage was found to be split into half and localized at a certain position from 100 K to 25 K. The position of \(\mathrm{Li^{+}}\) ion remained unchanged below 100 K. Above 100
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 85, 842, 140]]<|/det|>
+K, we did not detect the \(\mathrm{Li^{+}}\) position due to the fast motion of the \(\mathrm{Li^{+}}\) ion. As the localized \(\mathrm{Li^{+}}\) ion would strongly attract the negative radical (due to \(\mathrm{Li^{+}}\) - C bonds), the radicals should localize on the six- carbon ring (pink carbon atoms in Figure 2).  
+
+<|ref|>image<|/ref|><|det|>[[320, 144, 673, 358]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[148, 382, 459, 399]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[147, 419, 848, 639]]<|/det|>
+This work reports the synthesis and characterization of an impressive 1D- coordination polymer, constructed upon the reaction of the dinuclear copper(II) complex [Cu2L(py)4] (L= 1,2,4,5- tetrakis(methanesulfonamido)benzene) with the lithium encapsulated fullerene salt ( \(\mathrm{Li + @C60}\) ) (NTf2- ) (NTf2- = bis(trifluoromethane)sulfonamide anion), forming compound {[Cu4 ( \(\mathrm{Li + @C60}\) )L(py)4](NTf2)(hexane)}n (1). Compound 1 describes a 1D- coordination polymer (chain) in which each ( \(\mathrm{Li + @C60}\) ) “molecule” coordinates to four Cu(II) centres belonging to two [Cu2L(py)4], while the remaining Cu(II) centres serve as nodes to a neighboring ( \(\mathrm{Li + @C60}\) ) “molecule”, thus resulting in the chain- like motif. Electrical conductivity studies reveal that 1 displays long- range electrical conductivity, while magnetic studies demonstrate that 1 may be treated as a spin- frustrated system.  
+
+<|ref|>text<|/ref|><|det|>[[148, 641, 845, 713]]<|/det|>
+The quality of the ms. is good, although at some points the authors should revise the use of the English language. All data presented fully agree with the analysis presented by the authors, while in addition the theoretical studies performed support their claims.  
+
+<|ref|>text<|/ref|><|det|>[[148, 715, 843, 844]]<|/det|>
+My personal view is that this work is quite novel and exciting, since it is the first time that such exotic coordination polymers exhibit promising electrical and magnetic properties. I believe the results reported in this work will be of great significance to the fields of inorganic/coordination chemistry, magnetochemistry, physics and materials, with potential applications in spintronic devices. Therefore, I am happy to suggest acceptance of the ms. in Nature Communications, since it will attract the wider readership of scientists working in the above- mentioned fields.  
+
+<|ref|>text<|/ref|><|det|>[[149, 845, 695, 862]]<|/det|>
+Authors response: Thank you so much for your positive comments.  
+
+<|ref|>text<|/ref|><|det|>[[148, 882, 652, 899]]<|/det|>
+The points the authors should consider are the following ones:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 85, 847, 233]]<|/det|>
+1) Regarding the purity of the bulk samples included in the work (metallic precursor and compound 1) no data are presented besides the single-crystal structure. Therefore, for each compound: a) elemental C,H,N analysis should be provided, b) p-XRD diagrams should be provided along with comparison with the theoretical patterns, c) EDS measurements should be provided regarding the metallic content of compound 1. Authors response: The elemental (CHN) analysis was performed in the Research and Analytical Center for Giant Molecules at Tohoku University; the results are tabulated as below:  
+
+<|ref|>table<|/ref|><|det|>[[162, 250, 833, 329]]<|/det|>
+
+<table><tr><td>Compounds</td><td>C (%) Cal. / Exp.</td><td>H (%) Cal. / Exp.</td><td>N (%) Cal. / Exp.</td></tr><tr><td>Cu2(L)(py)4</td><td>40.49 / 40.46</td><td>3.85 / 3.86</td><td>12.59 / 12.47</td></tr><tr><td>{[Cu4(Li+@C60)L(py)4](NTf2)(hexane)}n</td><td>55.76 / 55.34</td><td>2.29 / 1.96</td><td>5.97 / 5.71</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[148, 347, 711, 366]]<|/det|>
+The PXRD patterns for \(\mathrm{Cu_2(L)(py)_4}\) (upper) and compound 1 (down).  
+
+<|ref|>image<|/ref|><|det|>[[264, 375, 737, 840]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[147, 865, 812, 902]]<|/det|>
+2) In 1 the coordination environment of the Cu centres is described as distorted trigonal bipyramidal. The distortion should be quantified in terms of the deviation
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 87, 234, 101]]<|/det|>
+parameter.  
+
+<|ref|>text<|/ref|><|det|>[[148, 103, 842, 156]]<|/det|>
+Authors response: Thank you so much for your comments. Unlike the \(\mathrm{PF}_5\) and \(\mathrm{PCl}_5\) , they have a D3h symmetry, \(\mathrm{Cu(N)_3(C)_2}\) has a distorted trigonal bipyramidal geometry because of the spherical surface of \(\mathrm{C}_{60}\) and the two carbon atoms are bonded.  
+
+<|ref|>text<|/ref|><|det|>[[148, 158, 837, 212]]<|/det|>
+Therefore, it shows very low symmetry and lacks of axis. Compared to trigonal bipyramidal geometry with a \(D_{3\mathrm{h}}\) symmetry, the deviation parameters are defined as (angle- \(90^{\circ}) / 90^{\circ}\) and \((180^{\circ} - \mathrm{angle}) / 180^{\circ}\) and calculated to be \(15.6\%\) , \(21.1\%\) and \(22.2\%\) .  
+
+<|ref|>image<|/ref|><|det|>[[260, 217, 480, 375]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[247, 381, 454, 399]]<|/det|>
+<center>Figure: \(\mathrm{Cu(N)_3(C)_2}\) (left) </center>  
+
+<|ref|>image<|/ref|><|det|>[[542, 216, 740, 375]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[539, 381, 748, 399]]<|/det|>
+<center>\(\mathrm{PF}_5\) , \(D_{3\mathrm{h}}\) symmetry(right) </center>  
+
+<|ref|>text<|/ref|><|det|>[[148, 417, 555, 453]]<|/det|>
+\(\angle \mathrm{N5Cu1N2} = 104.1^{\circ}\) \(\angle \mathrm{N5Cu1N1} = 104.1^{\circ}\) \(\angle \mathrm{N5Cu1C1} = 109.0^{\circ}\) \(\angle \mathrm{N5Cu1C1} = 141.9^{\circ}\)  
+
+<|ref|>text<|/ref|><|det|>[[148, 455, 400, 471]]<|/det|>
+So, deviation parameters \(\mathrm{P}\) are:  
+
+<|ref|>text<|/ref|><|det|>[[148, 473, 454, 528]]<|/det|>
+\(\mathrm{P1} = (104.1 - 90) / 90 * 100\% = 15.6\%\) \(\mathrm{P2} = (109 - 90) / 90 * 100\% = 21.1\%\) \(\mathrm{P3} = (180 - 140) / 180 * 100\% = 22.2\%\)  
+
+<|ref|>text<|/ref|><|det|>[[147, 547, 842, 582]]<|/det|>
+3) No deviation is given in the magnetic exchange parameters and the goodness value of the fit is missing.  
+
+<|ref|>text<|/ref|><|det|>[[148, 585, 843, 639]]<|/det|>
+Authors response: The effective exchange coupling parameters of \(J_{\mathrm{Cu - Cu}} = - 170\pm 5\mathrm{K}\) and \(J_{1 - C_{60}} = J_{2 - C_{60}} = - 185\pm 3\mathrm{K}\) with \(g_{\mathrm{Cu}} = 2.09(0)\) were obtained after the best fit \((g_{\mathrm{C_{60}}}\) was fixed to 2.0). The \(R\) - value is \(1.9\times 10^{- 3}\)  
+
+<|ref|>text<|/ref|><|det|>[[147, 658, 843, 768]]<|/det|>
+4) I can understand the existence of five exchange parameters for the fitting of the magnetic susceptibility, since there are indeed five different pathways. However, so many parameters often lead to overparameterization or to conclusions with no physical/real meaning. I would suggest the authors to perform the analysis with four J exchange parameters (i.e. J1-radical = J2-radical) and report/compare the two different models (in the SI).  
+
+<|ref|>text<|/ref|><|det|>[[147, 770, 837, 900]]<|/det|>
+Authors response: We reduced the parameters by only considering the exchange coupling \(J_{\mathrm{Cu - Cu}}\) and \(J_{\mathrm{Cu - C_{60}}}\) . To obtain the \(J\) values, we used the following spin Hamiltonian \(\widehat{H}\) using equations (1) by considering two kinds of exchange coupling \(J_{\mathrm{Cu - Cu}}\) and \(J_{\mathrm{Cu - C_{60}}}\) . \(\begin{array}{r}\widehat{H} = - 2J_{\mathrm{Cu - Cu}}(\hat{S}_{\mathrm{Cu1}}\cdot \hat{S}_{\mathrm{Cu2}} + \hat{S}_{\mathrm{Cu3}}\cdot \hat{S}_{\mathrm{Cu4}}) - 2J_{\mathrm{Cu - C_{60}}}(\hat{S}_{\mathrm{Cu1}}\cdot \hat{S}_{\mathrm{C_{60}}} + \hat{S}_{\mathrm{Cu2}}\cdot \hat{S}_{\mathrm{C_{60}}} + \hat{S}_{\mathrm{Cu3}}\cdot \hat{S}_{\mathrm{C_{60}}} + \hat{S}_{\mathrm{Cu4}}\cdot \hat{S}_{\mathrm{C_{60}}}) \end{array}\) \(+\mu_{B}\bar{B}\cdot g_{\mathrm{Cu}}\cdot \hat{S}_{\mathrm{Cu}} + \mu_{B}g_{\mathrm{C_{60}}}\bar{B}\hat{S}_{\mathrm{C_{60}}}\) (1)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[148, 84, 843, 177]]<|/det|>
+To avoid overparameterization, we assumed the interactions of \(J_{14}\) and \(J_{23}\) are ignored compared to \(J_{12}\) , \(J_{34}\) and \(J_{\mathrm{Cu - C_{60}}}\) due to their long metal- metal distances and the interactions between \(\mathrm{C_{60}^{- }}\) and \(\mathrm{Cu1}\) , \(\mathrm{Cu2}\) are equivalent. The effective exchange coupling parameters of \(J_{\mathrm{Cu - Cu}} = - 170\pm 5\) K and \(J_{1 - \mathrm{C_{60}^{- }}} = J_{2 - \mathrm{C_{60}^{- }}} = - 185\pm 3\) K with \(g_{\mathrm{Cu}} = 2.09(0)\) were obtained after the best fit \((g_{\mathrm{C_{60}^{- }}})\) was fixed to 2.0).  
+
+<|ref|>text<|/ref|><|det|>[[149, 196, 459, 212]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 233, 844, 360]]<|/det|>
+This manuscript reports a novel coordination polymer constructed by using \(\mathrm{Li + @C60}\) as acceptor and a specific Cu complex as donor. In the obtained framework, each \(\mathrm{Li + @C60}\) coordinates with four \(\mathrm{Cu2 + }\) ions forming infinite 1D ladder- like patterns along the crystallographic b- axis. This is accompanied by the strong charge transfer from the Cu species to the fullerene core. As a result, the four \(\mathrm{Cu2 + }\) ions ( \(\mathrm{S} = \frac{1}{2}\) ) and \(\mathrm{Li + @C_{60}^{- } - 60}\) ( \(\mathrm{S} = \frac{1}{2}\) ) interact with each other, showing magnetic frustration in a triangular- like lattice.  
+
+<|ref|>text<|/ref|><|det|>[[148, 363, 828, 454]]<|/det|>
+The paper is well- presented, the experiments and theoretical analysis are very solid. Although some fullerene- based 1D polymers have been reported previously, the assembly of \(\mathrm{Li + @C60}\) with a Cu complex is a new finding. Moreover, the obtained coordination polymer is conductive and features strong spin frustration. I would recommend the acceptance of this paper after addressing some minor issues.  
+
+<|ref|>text<|/ref|><|det|>[[148, 474, 806, 510]]<|/det|>
+1. Is the electrical conductivity of the crystal anisotropic? For example, along the ladder-chain direction (Figure 1d) versus other directions.  
+
+<|ref|>text<|/ref|><|det|>[[148, 512, 848, 641]]<|/det|>
+Authors response: Thank you so much for your positive comments. Unfortunately, the single- crystals morphology is not so nice and the crystal size is relatively small. We cannot determine the \(a\) , \(b\) , \(c\) axes by using such a single crystal. However, we can expect the conductive pathway from the crystal structure. One conductive pathway could be the 1D ladder- like chain as electrons move from the donor \(\mathrm{Cu_2(L)(py)_2}\) to the acceptor \(\mathrm{Li^{+}@C_{60}}\) cage (through bond); the other pathway is the electrons move between \(\mathrm{Li^{+}@C_{60}}\) cages along the \(a\) - axis (pi- pi stacking, through space).  
+
+<|ref|>text<|/ref|><|det|>[[148, 660, 844, 732]]<|/det|>
+2. The Z value in the crystal data of 1 (Datablock: exp_860) should be corrected. Although the 1D ladder-like structure of this crystal is clear, I would still suggest the authors to do more refinement on the anion NTf2- moieties and the disordered hexane molecules.  
+
+<|ref|>text<|/ref|><|det|>[[148, 735, 825, 769]]<|/det|>
+Authors response: We resolved the crystal structures and reduced and explained the alert A level in the cif file.  
+
+<|ref|>sub_title<|/ref|><|det|>[[149, 791, 270, 806]]<|/det|>
+## 3. Some typos:  
+
+<|ref|>text<|/ref|><|det|>[[148, 809, 610, 844]]<|/det|>
+a. In page 2, line 28, the close brace is missing; Authors response: We added it in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[147, 882, 796, 899]]<|/det|>
+b. Please unify/check the writing of ions/compounds. For example, NTf2 versus
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 85, 666, 121]]<|/det|>
+NTf2- ; '[Li+@C60](SbCl6)' in Page 3, line 45. Authors response: We corrected them in the revised manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 83, 312, 98]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[120, 112, 402, 128]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 142, 878, 186]]<|/det|>
+All my objections were fully addressed and therefore I think the paper is suitable for publication. Only a small mistake: in the legend of figure S11 the experimental data are not shown in orange but in black  
+
+<|ref|>text<|/ref|><|det|>[[120, 216, 402, 231]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 245, 842, 290]]<|/det|>
+The authors made a valiant effort to answer my comments. Considering how challenging it is to characterize spin liquids and how careful the authors are in the revised manuscript, I recommend accepting this work for publication.  
+
+<|ref|>text<|/ref|><|det|>[[120, 320, 402, 335]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 349, 872, 379]]<|/det|>
+In the revised ms., the authors answered with clarity to all issues initially raised. Therefore, I am more than happy to suggest acceptance of the manuscript at its current form.  
+
+<|ref|>text<|/ref|><|det|>[[120, 408, 402, 423]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 437, 857, 482]]<|/det|>
+I think the authors have fully addressed the concerns from all four reviewers. The texts and experimental data are very compelling. In my opinion it is now ready to be published without further change.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[302, 86, 694, 103]]<|/det|>
+## Response to reviewers' comments (2nd round)  
+
+<|ref|>text<|/ref|><|det|>[[148, 122, 837, 177]]<|/det|>
+We thank the four reviewers again for reviewing our revised manuscript. A point- by- point response to the reviewers' comments are shown below (reviewers' comments in blue, authors response in black):  
+
+<|ref|>text<|/ref|><|det|>[[149, 197, 459, 213]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 234, 839, 270]]<|/det|>
+All my objections were fully addressed and therefore I think the paper is suitable for publication.  
+
+<|ref|>text<|/ref|><|det|>[[148, 272, 794, 307]]<|/det|>
+Only a small mistake: in the legend of figure S11 the experimental data are not shown in orange but in black.  
+
+<|ref|>text<|/ref|><|det|>[[148, 308, 844, 344]]<|/det|>
+Authors response: Thank you so much for pointing it out. We have corrected them in the supporting information.  
+
+<|ref|>text<|/ref|><|det|>[[149, 364, 459, 380]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 400, 819, 492]]<|/det|>
+The authors made a valiant effort to answer my comments. Considering how challenging it is to characterize spin liquids and how careful the authors are in the revised manuscript, I recommend accepting this work for publication. Authors response: Thank you so much for your comments. We tried our best to probe the possible presence of a quantum spin liquid state in compound 1.  
+
+<|ref|>text<|/ref|><|det|>[[149, 512, 459, 528]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 548, 804, 602]]<|/det|>
+In the revised ms., the authors answered with clarity to all issues initially raised. Therefore, I am more than happy to suggest acceptance of the manuscript at its current form.  
+
+<|ref|>text<|/ref|><|det|>[[148, 604, 635, 621]]<|/det|>
+Authors response: Thank you so much for your acceptance.  
+
+<|ref|>text<|/ref|><|det|>[[149, 641, 459, 657]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[148, 677, 839, 751]]<|/det|>
+I think the authors have fully addressed the concerns from all four reviewers. The texts and experimental data are very compelling. In my opinion it is now ready to be published without further change. Authors response: Thank you so much for your recommendation.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a429eaadff2c5827b3e3038598b02ea58c59236cae97e1e8231cc6460632f9b2/images_list.json

@@ -0,0 +1,340 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 | Temperature-dependence third-order nonlinear Hall effect measurement results for the \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) . a. The longitudinal conductivity \\(\\sigma\\) as a function of temperature. b. The dependence of \\(V_{xxx}^{3\\omega}\\) on the cubic of \\(V_{xx}^{0\\omega}\\) is linear within the",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        280,
+        850,
+        556
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Fig. S17 | Angle-dependence third-order nonlinear Hall effect measurement results for the \\(\\mathbf{Fe}_{5}\\mathbf{GeTe}_{2}\\) . a. Optical micrograph of the 12-electrode circle device. b. Angle-dependent third-harmonic Hall voltage \\(\\left(\\frac{V_{xx}^{3\\omega}}{(V_{xx}^{3\\omega})^{3}}\\right)\\) measured at 300 K. c. Angle-dependent third-harmonic Hall voltage \\(\\left(\\frac{V_{xx}^{3\\omega}}{(V_{xx}^{3\\omega})^{3}}\\right)\\) measured at 200 K.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        567,
+        850,
+        732
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Fig. S8 | Ferromagnetic properties of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) . a. Optical micrograph of the sample used for RMCD measurements, with the thickness indicated. b. Field-cooled RMCD measurements of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) over the temperature range from \\(1.6\\mathrm{K}\\) to \\(250\\mathrm{K}\\) . c. Field-",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        422,
+        848,
+        818
+      ]
+    ],
+    "page_idx": 13
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Fig. R1 | Angle-dependence third-order nonlinear Hall effect under out-of-plane magnetic field of \\(\\mathrm{Fe_5GeTe_2}\\) . a. Optical micrograph of the sample used for RMCD measurements, with the thickness indicated. b. Schematic diagram illustrating the spatial orientation of the device and the directional configuration of in-plane \\((\\mu_0\\mathrm{H}_1)\\) and out-of-plane \\((\\mu_0\\mathrm{H}_1)\\) magnetic fields. c. \\(V_{xxy}^{3\\omega}\\) as a function of out-of-plane magnetic field at \\(350\\mathrm{K}\\) . d. \\(V_{xxy}^{3\\omega}\\) as a function of out-of-plane magnetic field.",
+    "footnote": [],
+    "bbox": [
+      [
+        198,
+        415,
+        797,
+        675
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Fig. R2 | Angle-dependence third-order nonlinear Hall effect under in-plane magnetic field of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) . a. \\(V_{xxy}^{3\\omega}\\) as a function of in-plane magnetic field at \\(350\\mathrm{K}\\) . b. \\(V_{xxy}^{3\\omega}\\) as a function of in-plane magnetic field at 300 K. c. \\(V_{xxy}^{3\\omega}\\) as a function of in-plane magnetic field at \\(200\\mathrm{K}\\) . d. \\(V_{xxy}^{3\\omega}\\) as a function of in-plane magnetic field at 100 K.",
+    "footnote": [],
+    "bbox": [
+      [
+        225,
+        153,
+        775,
+        478
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Fig. R3 | Magnetic field intensity-dependence third-order nonlinear Hall effect under out-of-plane magnetic field of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) . a. Optical micrograph of the \\(84~\\mathrm{nm}\\) hall device. b. \\(V_{xxy}^{3\\omega}\\) as a function of \\(V_{xx}^{w}\\) at different in-plane magnetic field under 100 K. c. \\(V_{xxy}^{3\\omega}\\) as a function of out-of-plane magnetic field at 100 K.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        530,
+        855,
+        692
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Fig. R4 | STEM image corresponding fast Fourier translate (FFT) pattern, the orange circular annotation in the figure denotes the {003} family of crystallographic planes.",
+    "footnote": [],
+    "bbox": [
+      [
+        340,
+        314,
+        689,
+        559
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Fig. 1 | Crystal structure and crystal growth of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) . a. Crystal structure of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) viewed along the [110]. b. The crystal structure of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) from the top view. The fragment describes the red-purple shaded area as the threefold helical symmetry axes (31, 32). green sphere: Fe atoms, orange sphere: Ge atoms, blue sphere: Te atoms. c. Schematic diagram of the growth of high-quality \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) single crystals. d. STEM image of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) . e. STEM image corresponding fast Fourier translate (FFT) pattern.",
+    "footnote": [],
+    "bbox": [
+      [
+        181,
+        83,
+        860,
+        550
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Fig. S5 | a-c. The XPS region scan of the Fe<sub>5</sub>Ge<sub>Te<sub>2</sub></sub> single crystal is focused on the Fe 2p, Ge 3d, and Te 3d elements.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        311,
+        850,
+        469
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Fig. S7b | The polar plots of Raman intensity versus rotation angle for the \\(E_{y}\\) Raman active modes.",
+    "footnote": [],
+    "bbox": [
+      [
+        378,
+        333,
+        620,
+        500
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Fig. R3 | Magnetic field intensity-dependence third-order nonlinear Hall effect under out-of-plane magnetic field of \\(\\mathrm{Fe}_{3}\\mathrm{GeTe}_{2}\\) . a. Optical micrograph of the \\(84~\\mathrm{nm}\\) hall device. b. \\(V_{xxy}^{3\\omega}\\) as a function of \\(V_{xxy}^{\\omega}\\) at different in-plane magnetic field under 100 K. c. \\(V_{xxy}^{3\\omega}\\) as a function of out-of-plane magnetic field at \\(100~\\mathrm{K}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        85,
+        856,
+        250
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Fig. S9 | Sample thickness of the \\(\\mathrm{Fe}_{5} \\mathrm{GeTe}_{2}\\) #Device3. a. Optical image of \\(\\mathrm{Fe}_{5} \\mathrm{GeTe}_{2}\\) device b. AFM images of the \\(\\mathrm{Fe}_{5} \\mathrm{GeTe}_{2}\\) device. c. Line profile of the corresponding position in (a).",
+    "footnote": [],
+    "bbox": [
+      [
+        188,
+        660,
+        886,
+        820
+      ]
+    ],
+    "page_idx": 28
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4d.jpg",
+    "caption": "Fig. 4d, e | Blue dots for the high temperature regime, red dots for the low temperature regime.",
+    "footnote": [],
+    "bbox": [
+      [
+        180,
+        444,
+        865,
+        651
+      ]
+    ],
+    "page_idx": 29
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4f.jpg",
+    "caption": "Fig. 4f | Comparison of the third-order conductivity \\((\\chi_{xxy}^{3\\omega})\\) and third-order current conversion ratio \\((|E_{xxy}^{3\\omega}| / (E_{xx}^{\\omega})^{3})\\) of \\(\\mathrm{Fe_5GeTe_2}\\) with other materials.",
+    "footnote": [],
+    "bbox": [
+      [
+        310,
+        400,
+        720,
+        640
+      ]
+    ],
+    "page_idx": 30
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Fig. S8 | Ferromagnetic properties of Fe\\(GeTe_2\\). a. Optical micrograph of the sample used for RMCD measurements, with the thickness indicated. b. Field-cooled RMCD measurements of Fe\\(GeTe_2\\) over the temperature range from 1.6 K to 250 K. c. Field-cooled RMCD measurements of Fe\\(GeTe_2\\) across the temperature range from 250 K to 300 K. d. Optical micrograph of the field-cooled Anomalous Hall resistance measurements device. e. Temperature-dependent anomalous Hall resistance of Fe\\(GeTe_2\\) measured over the temperature range of 100 K to 250 K. f. Temperature-dependent anomalous Hall resistance of Fe\\(GeTe_2\\) measured over the temperature range",
+    "footnote": [],
+    "bbox": [
+      [
+        150,
+        90,
+        848,
+        485
+      ]
+    ],
+    "page_idx": 33
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_11.jpg",
+    "caption": "Fig. S13 | Phase information in lock-in measurements. Black dots depict the nonlinear longitudinal voltage as a function of the applied current, while red dots correspond to the phase information obtained during measurement of nonlinear longitudinal voltage.",
+    "footnote": [],
+    "bbox": [
+      [
+        288,
+        445,
+        744,
+        686
+      ]
+    ],
+    "page_idx": 35
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Fig. 4 | Temperature-dependence third-order nonlinear Hall effect measurement results for the \\(\\mathrm{Fe}_{5} \\mathrm{GeTe}_{2}\\) . a. The longitudinal conductivity \\(\\sigma\\) as a function of temperature. b. The dependence of \\(V_{xxx}^{3\\omega}\\) on the cubic of \\(V_{xx}^{3\\omega}\\) is linear within the temperature range of 30 to \\(300 \\mathrm{~K}\\) . c. The \\(|E_{xxy}^{3\\omega}| / (E_{xx}^{3\\omega})^{3}\\) as a function of temperature.",
+    "footnote": [],
+    "bbox": [
+      [
+        146,
+        508,
+        849,
+        787
+      ]
+    ],
+    "page_idx": 38
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4f.jpg",
+    "caption": "Fig. 4f | Comparison of the third-order conductivity \\((\\chi_{1}^{\\mathrm{3}\\omega})\\) and third-order current conversion ratio \\((|E_{1}^{\\mathrm{3}\\omega}| / (E_{1}^{\\omega})^{3})\\) of \\(\\mathrm{Fe}_5\\mathrm{GeTe}_2\\) with other materials.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 47
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_12.jpg",
+    "caption": "Fig. R2 | Magnetic field intensity-dependence third-order nonlinear Hall effect under out-of-plane magnetic field of \\(\\mathrm{Fe}_{5}\\mathrm{GeTe}_{2}\\) . a. Optical micrograph of the \\(84 \\mathrm{nm}\\) hall device. b. \\(V_{1}^{3\\omega}\\) as a function of \\(V_{1}^{\\omega}\\) at different in-plane magnetic field under 100 K. c. \\(V_{1}^{3\\omega}\\) as a function of out-of-plane magnetic field at 100 K.",
+    "footnote": [],
+    "bbox": [
+      [
+        180,
+        590,
+        849,
+        744
+      ]
+    ],
+    "page_idx": 51
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_13.jpg",
+    "caption": "Fig. R3 | Angle-dependence third-order nonlinear Hall effect under out-of-plane magnetic field of \\(\\mathrm{Fe_5GeTe_2}\\) . a. Optical micrograph of the sample used for magnetic-field-tunable third-order nonlinear Hall measurements, with the thickness indicated. b. Schematic diagram illustrating the spatial orientation of the device and the directional configuration of in-plane \\((\\mu_0\\mathrm{H}_1)\\) and out-of-plane \\((\\mu_0\\mathrm{H}_1)\\) magnetic fields. c. \\(V_1^{3\\omega}\\) as a function of out-of-plane magnetic field at \\(350\\mathrm{K}\\) . d. \\(V_1^{3\\omega}\\) as a function of out-of-plane magnetic field at 300 K. e. \\(V_1^{3\\omega}\\) as a function of out-of-plane magnetic field at \\(200\\mathrm{K}\\) . f. \\(V_1^{3\\omega}\\) as a function of out-of-plane magnetic field at 100 K.",
+    "footnote": [],
+    "bbox": [
+      [
+        181,
+        85,
+        816,
+        360
+      ]
+    ],
+    "page_idx": 52
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_14.jpg",
+    "caption": "Fig. R4 | Angle-dependence third-order nonlinear Hall effect under in-plane magnetic field of \\(\\mathbf{Fe}_5\\mathbf{GeTe}_2\\) . a. \\(V_{1}^{3\\omega}\\) as a function of in-plane magnetic field at \\(350\\mathrm{K}\\) . b. \\(V_{xxy}^{3\\omega}\\) as a function of in-plane magnetic field at \\(300\\mathrm{K}\\) . c. \\(V_{1}^{3\\omega}\\) as a function of in-plane magnetic field at \\(200\\mathrm{K}\\) . d. \\(V_{1}^{3\\omega}\\) as a function of in-plane magnetic field at \\(100\\mathrm{K}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        234,
+        92,
+        768,
+        410
+      ]
+    ],
+    "page_idx": 55
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_15.jpg",
+    "caption": "Fig. R5 | \\(V_{xx}^{3\\omega}\\) as a function of out-of-plane magnetic field at 300 K.",
+    "footnote": [],
+    "bbox": [
+      [
+        293,
+        396,
+        705,
+        643
+      ]
+    ],
+    "page_idx": 56
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_16.jpg",
+    "caption": "Fig. R6 | Angle-dependence third-order nonlinear Hall effect under out-of-plane magnetic field of \\(\\mathrm{Fe_5GeTe_2}\\) . a. Optical micrograph of the sample used for magnetic-field-tunable third-order nonlinear Hall measurements, with the thickness indicated. b. Schematic diagram illustrating the spatial orientation of the device and the directional configuration of in-plane \\((\\mu_0\\mathrm{H}_{\\mathrm{H}})\\) and out-of-plane \\((\\mu_0\\mathrm{H}_{\\mathrm{L}})\\) magnetic fields. c. \\(V_{3}^{1\\omega}\\) as a function of out-of-plane magnetic field at \\(350\\mathrm{K}\\) . d. \\(V_{3}^{1\\omega}\\) as a function of out-of-plane magnetic field at 300 K. e. \\(V_{3}^{1\\omega}\\) as a function of out-of-plane magnetic field at \\(200\\mathrm{K}\\) . f. \\(V_{3}^{1\\omega}\\) as a function of out-of-plane magnetic field at 100 K.",
+    "footnote": [],
+    "bbox": [
+      [
+        147,
+        357,
+        852,
+        666
+      ]
+    ],
+    "page_idx": 60
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a42b23600537fd23c88e5370b65f268fab4924233bc8334a78821c082fc84d54/images_list.json

@@ -0,0 +1,10 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Phylogeographic reconstruction with Phytools",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a42b23600537fd23c88e5370b65f268fab4924233bc8334a78821c082fc84d54/supplementary_0_Peer Review File__a42b23600537fd23c88e5370b65f268fab4924233bc8334a78821c082fc84d54.mmd

@@ -0,0 +1,326 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Genomic perspective on the bacillus causing paratyphoid B fever  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors present compelling analyses documenting the genetic diversity, population structure, and historical distribution from a well- curated dataset of S. Paratyphi B d- tartrate(- ) strains. These results provide novel insight into the global distribution of SPB(- ), which continues to be an important public health concern. Furthermore, informed by their genomic analyses, the authors developed a pipeline for rapid and accurate typing of SPB(- ) strains for facilitating public health surveillance efforts. The data analysis, interpretation and conclusions are sound. There are some speculations about the historical events that could have led to the observed distribution in SPB, but these points are brought up in the discussions section of the paper.  
+
+Overall, this manuscript is based on compelling, sound, and robust evidence. I have only minor suggestions for the authors to consider.  
+
+Line 116: what is synthetic cream? Non- dairy cream?  
+
+Lines 200- 201: How were the higher- resolution genotypes named? Some have locations associated with their names? Since attaching geographic locations to names has some important public health implications, more details about \(\%\) of isolates from a given geographic location to assign clade name would be helpful to include, especially in the cases where the groups have multiple geo locations (ex. NorthAfrica_Europe).  
+
+Line 319: I might have missed this, but how was the presence of sopE confirmed/determined? With blastn? Or was this denoted from the Phaster annotation? I think this is important to describe because the authors present support for using sopE as a marker for SPB- strains (line 455)  
+
+Lines 334- 337: Were the sequences of sopE at the three genomic insertion sites different? Was sopE at insertion site 1 distinguishable from sopE at insertion site 2 or 3? The analyses suggest that they may be carried on different prophages.  
+
+Line 430: add 'in recent years' – it is noted at the beginning of the section that the surveillance dataset is used for this conclusion. Also, the three reasons listed are the 'primary' reasons  
+
+Line 490: Why 10 lineages? Earlier in the paper (Fig. 1) the authors note the presence of 11 lineages including 3 with singletons, of which lineage 11 is one  
+
+Reviewer #2 (Remarks to the Author):  
+
+Thank you for the opportunity to review this manuscript which describes the population diversity and evolution of Salmonella enterica serotype Paratyphi B or SPB. This organism of one of four causes of
+
+<--- Page Split --->
+
+enteric fever, a global public health challenge for many centuries.  
+
+This is an original, well- planned and well- presented study by experts in microbiology of enteropathogens.  
+
+The authors focused largely on D- tartrate non- fermenting strains of SPB and examined the population structure and temporal evolution of 568 genomes, the majority of which were generated for this study. The study identified 11 lineages (with the predominance of L10) and 38 SNVs unique to each SPB genotype and proposed and implemented a hierarchical SNV- based genotyping scheme that can split SPB populations into phylogeographically informative genotypes. The core genome of 4,044 genes was identified and proportions of accessory genes belonging to prophages, plasmids and transposases were quantified.  
+
+They also assembled and investigated a set of 336 genomes from four major public health laboratories in North America and Europe, which were uploaded between 2015 and 2023. The evolutionary timescale analysis and Bayesian analysis of population structure was applied to the data.  
+
+The paper builds on a rich history of discovery of salmonellas associated with clinical typhoid infections, fills a current gap in our understanding of the genomic epidemiology of this pathogen and should be of interest to readers of the Journal. The methods employed in this study are cutting- edge in microbial genomics and are adequate for the research aims and hypotheses.  
+
+In order to further improve the paper, the authors may consider addressing the following suggestions and questions:  
+
+- How informative could the suggested SPB genotyping system be for the public health investigation of outbreaks? What does the surveillance dataset tell us in that respect?  
+
+- Similar population genomics studies of Salmonella Paratyphi A emphasised the role gene acquisitions or losses play as key molecular events in the evolution of new lineages (e.g., Jacob JJ et al. Genomic analysis unveils genome degradation events and gene flux in the emergence and persistence of S. Paratyphi A lineages. PLoS Pathogens 2023;19(4): e1010650). Have the authors observed any parallels in their studies of SPB?  
+
+- Clarify whether any SPB isolates from Connor's collection (Connor TR et al. mBio 2016; 7(4): e00527-16 or Reference 27) were included in the study dataset of 568 genomes?  
+
+- Please provide percentages for cases where quinolone resistance conferring mutations were documented (lines 294-298).  
+
+- The important observation of azithromycin resistance presented in Figure 5 deserves to be mentioned in the text with the mechanism of resistance explained. The term 'azithromycin' (i.e. an individual antibiotic name) in Figure 5 can be replaced with 'macrolides' (antibiotic class name) to be consistent with other terms in the legend.  
+
+- Table 2 – Suggested heading for the second column – “Years of MRCA (95% HPD)”.  
+
+- Some minor editorial suggestions:  
+
+o Rephrase the statement in the Abstract “We show that this pathogen existed in the 13th century” to “Our comparative genomics findings suggest that this pathogen has existed since the 13th century”.  
+
+o Change “exploitation’ to ‘exploration’ (line 391), ‘their colonial and immigration histories’ (line 431) to ‘migration and travel patterns over recent centuries’  
+
+o Change to ‘Universities’ (line 544).
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+This is an interesting and very detailed study of the lineage of Salmonella that is the causative agent of Paratyphi B. It is well written, and I enjoyed reading it. It is certainly worthy of publication and will be of broad interest.  
+
+## Novelty  
+
+I think the paper has broad appeal, both within the field and in related fields. The work itself is original, and also builds well on the existing literature. As well as bringing together and drawing a line under findings from other studies (including those by the authors and others), the paper includes some novel/important findings. Firstly, the detailed historical collection shows strong evidence of phylogeographic clustering, which is not something that has been reported before, and is of epidemiological value and scientific value. Secondly, the pattern of resistance is also interesting, including evidence for increasing resistance in SPB isolates recently. Thirdly, The work around the prophages within SPB is also novel, and the pangenome analysis is also more comprehensive than previous work, due to a combination of the larger historic dataset and the use of long read technologies. Lastly, the observation around S. Onarimon, in the supplementary is also of interest to those who are enthusiasts of the Salmonella, and it was an unexpected nugget that was also new and interesting.  
+
+## Support for claims  
+
+Overall, the paper has a large amount of detail, and represents a very in- depth genomic study of this interesting pathogen.  
+
+The paper brings together historical work very well along with a historical and contemporary genomic dataset to provide the highest resolution exploration to date of SPB. The approaches used are appropriate and well matched to the aims, and the methods, both laboratory and bioinformatics are explained in sufficient detail to enable reproducibility. The use of Enterobase is welcome and enhances the reproducibility of the work. The release of the probes for Mykrobe is also welcome as this is open- source software.  
+
+Overall, the data analysis is sensible, and the methods are appropriate and meet the expected standards in the field.  
+
+Other questions/comments for the authors:  
+
+Although I like this paper a lot and think it should be published, there are a number of questions/clarifications that would be helpful, these are;  
+
+The paper reports the correlation with geography and the evidence of geographic clustering; however, I couldn't see a formal analysis for the phylogeography. Given that the authors had made use of BEAST, I wondered if they had made use of any of the phylogeographic models to examine predicted origins and phylogeographic spread more formally.
+
+<--- Page Split --->
+
+So  
+
+1. Were analyses performed to substantiate the suggestion of export from Europe?  
+2. The detail around the BEAST analysis is in the supplementary, and these results are interesting, although broadly agreeing with what had been found before - but the BEAST date/context information isn't that well referenced in the main text - so could the authors consider how to better signpost to the detail in the supplementary information regarding BEAST.  
+
+The AMR analysis is interesting, but it wasn't clear to me from the paper in what context the resistance genes were found. So;  
+
+3. Were the genes found on chromosomal mobile elements, carried with/on phage or were they plasmid carried?  
+
+4. From the figure in the supplementary, it looks like there weren't many plasmids in the samples analysed, though it would be good to have this described in a little more detail. The paper mentions a phylogenetic comparison between plasmid genes and core genome, but it isn't clear to me what this means, or what it shows - can the authors explain what is meant and consider if figure 9 in the supplementary is appropriate to communicate this.  
+
+With respect to the phylogenetics and cgMLST analyses;  
+
+5. How many (if any) HC400_1620 samples in Enterobase did not contain the STM 336 SNV? It isn't wholly clear from the text if HC400_1620 is just SPB-strains, or if it includes both SPB+ and SPB-strains.  
+7. Did the Onarimon samples cluster in a single lineage? were they grouped with other more contemporary samples in a single lineage? it isn't clear from the paper/results where these sit with respect to other lineages within SPB-samples.  
+
+Lastly;  
+
+8. In the validation work around Mykrobe, was the probe tested against a wider database to identify if the probe could miss-detect SPB? what validation was done on this - as the MS only lists testing against the study datasets.  
+
+9. In the discussion there are assertions around long term carriage in elderly patients - it wasn't clear to me what the evidence was for this. In those contemporaneous samples from older patients, was there evidence of carriage (e.g., diversity) or long-term chronic infection in their genomes?  
+10. Were the 'carriage' isolates resistant to antibiotics (which one might have expected - as one would have expected some antibiotic exposure in older patients at some point, which you might have expected to impact carriage of SPB)?
+
+<--- Page Split --->
+
+Authors: We would like to thank the reviewers for their many insightful comments. We address these comments, point by point, below.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors present compelling analyses documenting the genetic diversity, population structure, and historical distribution from a well- curated dataset of S. Paratyphi B d- tartrate(- ) strains. These results provide novel insight into the global distribution of SPB(- ), which continues to be an important public health concern. Furthermore, informed by their genomic analyses, the authors developed a pipeline for rapid and accurate typing of SPB(- ) strains for facilitating public health surveillance efforts. The data analysis, interpretation and conclusions are sound. There are some speculations about the historical events that could have led to the observed distribution in SPB, but these points are brought up in the discussions section of the paper.  
+
+Overall, this manuscript is based on compelling, sound, and robust evidence. I have only minor suggestions for the authors to consider.  
+
+Line 116: what is synthetic cream? Non- dairy cream?  
+
+Authors: It is an emulsion of vegetable oils or fat with water, with or without the addition of other substances, some of which may be of dairy origin. It was used in the UK during World War II when the sale of natural cream was forbidden. This information has been added to the Supplementary Note "Epidemiology of paratyphoid B fever during the first half of the \(20^{\text{th}}\) century".  
+
+Lines 200- 201: How were the higher- resolution genotypes named? Some have locations associated with their names? Since attaching geographic locations to names has some important public health implications, more details about \(1\%\) of isolates from a given geographic location to assign clade name would be helpful to include, especially in the cases where the groups have multiple geo locations (ex. NorthAfrica_Europe).  
+
+Authors: We now indicate in the results section: "Strong geographic patterns with differences from country to continent level were observed for 17 genotypes, whereas two genotypes were more widespread, isolated from two continents (Supplementary Data 1). These two genotypes were genotype 7.2 found in Europe (26.3%, 5/19) and Asia (73.7%, 14/19, mostly East Asia), and genotype 10.3.8.5 found in Europe (71.4%, 20/28) and North Africa (25%, 7/28). One genotype, 7.3.2 — also found in Europe (41.2%, 7/17) and North Africa (52.9%, 9/17) — was associated with a particular PT, BAOR (see Comparison of phylogenomics data with other typing schemes) (Supplementary Data 1). This geographic or PT information was added to the genotype nomenclature as an alias, to make it more informative." and in the discussion section: "New genotypes with geographic information (in cases of new emerging genotypes in a defined area) or new updates on alias names (in cases of the establishment of known genotypes in new areas) should be added to the scheme in the future."  
+
+Line 319: I might have missed this, but how was the presence of sope confirmed/determined? With blastn? Or was this denoted from the Phaster annotation? I
+
+<--- Page Split --->
+
+think this is important to describe because the authors present support for using sopE as a marker for SPB- strains (line 455)  
+
+Authors: To clarify this point, we have added the following sentences to the M&M section entitled "Prophage content analysis and sopE copy- number variation": "By combining the annotation of the 14 SPB complete genomes and the prophage delineation described above, we were able to locate the sopE gene in SEN34- like and P88- like prophages. We used the blastn algorithm and the sopE reference sequence (GenBank accession no. L78932) from S. enterica serotype Dublin<sup>38</sup> to confirm our identification of the prophage- borne sopE gene."  
+
+Lines 334- 337: Were the sequences of sopE at the three genomic insertion sites different? Was sopE at insertion site 1 distinguishable from sopE at insertion site 2 or 3? The analyses suggest that they may be carried on different prophages.  
+
+Authors: The sopE gene nucleotide sequence is highly conserved across prophages. We now indicate the following in results section: "In the 14 complete genomes, the nucleotide sequence of the sopE gene was 100% identical, regardless of the prophages in which it is inserted. One exception was the sopE gene inserted at site #3 of the B624 genome, which differed by eight (five being non- synonymous) out of 723 nucleotides from the sopE consensus sequence of SPB."  
+
+Line 430: add 'in recent years' - it is noted at the beginning of the section that the surveillance dataset is used for this conclusion. Also, the three reasons listed are the 'primary' reasons  
+
+Authors: This has been done.  
+
+Line 490: Why 10 lineages? Earlier in the paper (Fig. 1) the authors note the presence of 11 lineages including 3 with singletons, of which lineage 11 is one  
+
+Authors: We were actually referring to the 10 lineages or phylogroups described by Connor et al for the global population SPB and not the lineages we have identified for SPB PG1. We have clarified this in the following sentence: "Firstly, SPB isolates can be assigned to the 10 known PGs described by Connor and coworkers<sup>27</sup> with the EnteroBase cgMLST scheme, with the HC400_1620 cluster considered a signature of SPB PG1.".  
+
+Reviewer #2 (Remarks to the Author):  
+
+Thank you for the opportunity to review this manuscript which describes the population diversity and evolution of Salmonella enterica serotype Paratyphi B or SPB. This organism of one of four causes of enteric fever, a global public health challenge for many centuries. This is an original, well- planned and well- presented study by experts in microbiology of enteropathogens.  
+
+The authors focused largely on D- tartrate non- fermenting strains of SPB and examined the population structure and temporal evolution of 568 genomes, the majority of which were generated for this study. The study identified 11 lineages (with the predominance of L10) and 38 SNVs unique to each SPB genotype and proposed and implemented a hierarchical SNV- based genotyping scheme that can split SPB populations into phylogeographically informative genotypes. The core genome of 4,044 genes was identified and proportions of accessory genes belonging to prophages, plasmids and transposases were quantified. They also assembled and investigated a set of 336 genomes from four major public health
+
+<--- Page Split --->
+
+laboratories in North America and Europe, which were uploaded between 2015 and 2023. The evolutionary timescale analysis and Bayesian analysis of population structure was applied to the data.  
+
+The paper builds on a rich history of discovery of salmonellas associated with clinical typhoid infections, fills a current gap in our understanding of the genomic epidemiology of this pathogen and should be of interest to readers of the Journal. The methods employed in this study are cutting- edge in microbial genomics and are adequate for the research aims and hypotheses.  
+
+In order to further improve the paper, the authors may consider addressing the following suggestions and questions:  
+
+- How informative could the suggested SPB genotyping system be for the public health investigation of outbreaks? What does the surveillance dataset tell us in that respect? Authors: We now expand on this aspect in the revised MS (discussion section). The text now reads: "We anticipate that the use of this scheme and its universal nomenclature will improve the laboratory surveillance of PTB. It is now easier to track the different SPB PG1 populations at global scale. We were able to identify the different genotypes detected recently in travellers from or migrants to four high-income countries in North America and Europe (surveillance dataset). However, more global studies involving countries experiencing PTB across the globe and performed regularly over time — as for the PT surveys — would be helpful to monitor the diversity, spread and evolution (of AMR in particular) of this pathogen. New genotypes with geographic information (in cases of new emerging genotypes in a defined area) or new updates on alias names (in cases of the establishment of known genotypes in new areas) should be added to the scheme in the future. The use of a common nomenclature would also be helpful during outbreak investigations. It is now straightforward to define the bacterial types of SPB PG1 and to share this information during transborder outbreak investigations. This should also facilitate the identification of transmission chains associated with particular geographic regions, particularly in areas in which PTB is not endemic. For example, genomic surveillance in the UK identified an imported SPB outbreak in travellers coinciding with a mass gathering in Iraq in 2021 (ref.29). The isolates from these patients clustered in one of the two clades labelled "travel to Iraq". According to our genotyping scheme, this clade corresponds to genotype 10.3.2 MiddleEast1. The oldest isolates belonging to this genotype were collected in Iran in 1965 and Iraq in 1975, suggesting that this strain has been endemic in the region for many decades. The second clade labelled "travel to Iraq" identified by UKHSA corresponded to genotype 10.3.8.3 MiddleEast3. In endemic regions, this genotyping scheme might also be useful for determining the genotype of a potential outbreak strain if several genotypes are known to circulate regionally."  
+
+- Similar population genomics studies of Salmonella Paratyphi A emphasised the role gene acquisitions or losses play as key molecular events in the evolution of new lineages (e.g., Jacob JJ et al. Genomic analysis unveils genome degradation events and gene flux in the emergence and persistence of S. Paratyphi A lineages. PLoS Pathogens 2023;19(4): e1010650). Have the authors observed any parallels in their studies of SPB? Authors: This interesting aspect of genome degradation events and gene flux leading to the emergence of SPB PG1, a human-adapted pathogen causing paratyphoid fever, was previously studied by Connor et al (our reference #27) in their comparative genomics
+
+<--- Page Split --->
+
+analysis of the different PGs (1- 10) of SPB. We did not address this question as our study focused exclusively on the PG1 population. Regarding the evolution of PG1, our comprehensive pangenome analysis failed to identify genomic signals leading to increases in virulence or transmissibility during recent microevolution (as previously reported for other enteric fever pathogens, such as S. enterica serovars Typhi (Roumagnac et al. Science 2016) and Paratyphi A (Zhou et al. PNAS 2014)). Regarding virulence, we documented up to three copies of the sopE gene in some lineages, but we have no other data linking this microbial genomics finding to an increase in virulence in vivo or in human populations.  
+
+- Clarify whether any SPB isolates from Connor's collection (Connor TR et al. mBio 2016; 7(4): e00527-16 or Reference 27) were included in the study dataset of 568 genomes? Authors: The section entitled "S. enterica Paratyphi B sequence data collections" of the M&M (lines 579-583) previously included the following: "We first studied a diversity dataset of 568 SPB genomes, 446 of which were generated specifically for this study, 109 had already been published<sup>27,28,29,31,48,49,50</sup>, and the other 13 were unpublished but deposited in EnteroBase (https://enterobase.warwick.ac.uk/species/index/senterica) or GenBank (https://www.ncbi.nlm.nih.gov/genbank/) (Supplementary Data 1)."  
+
+In Supplementary Data 1 (the list of 568 genomes), the column (BH) entitled "Source" can be used to identify the 28 strains from Connor et al.  
+
+As we used 109 published genomes from seven different papers, we did not want to overload the text by providing the numbers of strains used per published paper. Instead, we provide all the necessary details in Supplementary Table 1. However, if the editor would prefer all this information to be provided in the main text, we are willing to comply with this request.  
+
+- Please provide percentages for cases where quinolone resistance conferring mutations were documented (lines 294-298). Authors: This has been done.  
+
+- The important observation of azithromycin resistance presented in Figure 5 deserves to be mentioned in the text with the mechanism of resistance explained.  
+
+Authors: This has been done: "One isolate (83282), acquired in South America in 2014, contained the mph(A) gene encoding a macrolide 2' phosphotransferase, a common determinant of azithromycin resistance (however, the antimicrobial susceptibility pattern of this isolate was unavailable)."  
+
+The term 'azithromycin' (i.e. an individual antibiotic name) in Figure 5 can be replaced with 'macrolides' (antibiotic class name) to be consistent with other terms in the legend. Authors: This has been done.  
+
+- Table 2 – Suggested heading for the second column – "Years of MRCA (95% HPD)". Authors: This has been changed to "Time (year) of the MRCA".  
+
+- Some minor editorial suggestions:
+
+<--- Page Split --->
+
+o Rephrase the statement in the Abstract "We show that this pathogen existed in the 13th century" to "Our comparative genomics findings suggest that this pathogen has existed since the 13th century".  
+
+Authors: This has been done.  
+
+o Change "exploitation' to 'exploration' (line 391), 'their colonial and immigration histories' (line 431) to 'migration and travel patterns over recent centuries' Authors: This has been done.  
+
+o Change to 'Universities' (line 544). Authors: This has been done.  
+
+Reviewer #3 (Remarks to the Author):  
+
+This is an interesting and very detailed study of the lineage of Salmonella that is the causative agent of Paratyphi B. It is well written, and I enjoyed reading it. It is certainly worthy of publication and will be of broad interest.  
+
+## Novelty  
+
+I think the paper has broad appeal, both within the field and in related fields. The work itself is original, and also builds well on the existing literature. As well as bringing together and drawing a line under findings from other studies (including those by the authors and others), the paper includes some novel/important findings. Firstly, the detailed historical collection shows strong evidence of phylogeographic clustering, which is not something that has been reported before, and is of epidemiological value and scientific value. Secondly, the pattern of resistance is also interesting, including evidence for increasing resistance in SPB isolates recently. Thirdly, The work around the prophages within SPB is also novel, and the pangenome analysis is also more comprehensive than previous work, due to a combination of the larger historic dataset and the use of long read technologies. Lastly, the observation around S. Onarimon, in the supplementary is also of interest to those who are enthusiasts of the Salmonella, and it was an unexpected nugget that was also new and interesting.  
+
+## Support for claims  
+
+Overall, the paper has a large amount of detail, and represents a very in- depth genomic study of this interesting pathogen.  
+
+The paper brings together historical work very well along with a historical and contemporary genomic dataset to provide the highest resolution exploration to date of SPB. The approaches used are appropriate and well matched to the aims, and the methods, both laboratory and bioinformatics are explained in sufficient detail to enable reproducibility. The use of Enterobase is welcome and enhances the reproducibility of the work. The release of the probes for Mykrobe is also welcome as this is open- source software.  
+
+Overall, the data analysis is sensible, and the methods are appropriate and meet the
+
+<--- Page Split --->
+
+expected standards in the field.  
+
+Other questions/comments for the authors:  
+
+Although I like this paper a lot and think it should be published, there are a number of questions/clarifications that would be helpful, these are;  
+
+The paper reports the correlation with geography and the evidence of geographic clustering; however, I couldn't see a formal analysis for the phylogeography. Given that the authors had made use of BEAST, I wondered if they had made use of any of the phylogeographic models to examine predicted origins and phylogeographic spread more formally. So  
+
+1. Were analyses performed to substantiate the suggestion of export from Europe? Authors: We initially performed two phylogeographic reconstructions with phytools (https://peerj.com/articles/16505/) and PastML (https://academic.oup.com/mbe/article/36/9/2069/5498561) to document the exportation of SPB PG1 from Europe at different points.  
+
+![PLACEHOLDER_10_0]
+
+
+<--- Page Split --->
+![PLACEHOLDER_11_0]
+
+<center>Phylogeographic reconstruction with Phytools </center>  
+
+However, due to a deeper sampling for European historical isolates, we were not comfortable presenting these phylogeographic reconstructions, which might be perceived as a formal proof. Instead, we simply suggest that export from Europe may have occurred, based on historical data and the tree topology.
+
+<--- Page Split --->
+
+2. The detail around the BEAST analysis is in the supplementary, and these results are interesting, although broadly agreeing with what had been found before - but the BEAST date/context information isn't that well referenced in the main text - so could the authors consider how to better signpost to the detail in the supplementary information regarding BEAST.  
+
+Authors: We have now moved all the supplementary data dealing with the Beast analysis to the main text (discussion section). It now reads "Based on our dataset of 568 SPB PG1 genomes, we estimated the age of this pathogen at \~750 years (1274 CE; 95% CI, 915 - 1583), which is very close to the previous median date of origin estimated by Connor and coworkers27 (1188 CE; 95% CI, 469 BC - 1799 CE), who used only 25 SPB PG1 genomes (i.e., those with a known year of isolation). SPB is older than SPA, which is estimated to have originated 450 - 700 years ago25. SPA was discovered two years after SPB (in the USA in 1898)1,2,32 but is currently the most frequent agent of paratyphoid fever28. Due to lineage extinction, in particular, times to the MRCA are often underestimated and the inclusion of ancient DNA in the analysis would increase precision and make it possible to establish dates of origin further in the past25. The dating of a representative collection of modern isolates of SPC estimated the origin of SPC to 456 - 664 years ago. When a draft SPC genome from an 800-year-old Norwegian skeleton was added to the analysis, the time to the MRCA increased to 1162 - 1526 years26. Unfortunately, no ancient DNA is currently available for SPB strains."  
+
+The AMR analysis is interesting, but it wasn't clear to me from the paper in what context the resistance genes were found. So;  
+
+3. Were the genes found on chromosomal mobile elements, carried with/on phage or were they plasmid carried?  
+
+Authors: Most of the AMR detected was due to single point mutations of chromosomal genes (gyrA and gyrB). The five remaining AMR isolates (those with ESBL, carbapenemase or mphA genes in particular) had AMR plasmids. This is now indicated in the main text: "Between 1898 and 2000, only one isolate (0.3%, 1/345) had antibiotic resistance genes (ARGs). This human isolate (B73-1117), collected in France in 1973, displayed resistance to ampicillin (blaTEM-1D), streptomycin (strAB, aadA1, and aadA2b), sulfonamides (sul1), chloramphenicol (cmIA1), and tetracycline (tetA). Between 2001 and 2021, 23.1% (52/223) of isolates had ARGs. One isolate acquired in Turkey in 2001 (01-7995) produced a CTX-M-3 extended-spectrum beta-lactamase30, whereas another isolate (P7704) acquired in South America in 2019 produced an OXA-48 carbapenemase31. One isolate (83282), acquired in South America in 2014, contained the mph(A) gene encoding a macrolide 2' phosphotransferase, a common determinant of azithromycin resistance (however, the antimicrobial susceptibility pattern of this isolate was unavailable). All these rare isolates contained AMR plasmids (Supplementary Data 1); however, the mechanisms of resistance (21.5%, 48/223) most prevalent during the 2001-2021 period involved mutations of the quinolone resistance-determining regions of the chromosomal gyrA and gyrB DNA gyrase genes leading to resistance to nalidixic acid and/or decreased susceptibility or resistance to ciprofloxacin (minimum inhibitory concentration [MIC] between 0.125 and 0.5 mg/L).".  
+
+4. From the figure in the supplementary, it looks like there weren't many plasmids in the samples analysed, though it would be good to have this described in a little more detail. The
+
+<--- Page Split --->
+
+paper mentions a phylogenetic comparison between plasmid genes and core genome, but it isn't clear to me what this means, or what it shows - can the authors explain what is meant and consider if figure 9 in the supplementary is appropriate to communicate this.  
+
+Authors: Plasmid- borne AMR was indeed very rare, with only five isolates displaying such resistance. We now provide more detail about the plasmid versus chromosome locations of the AMR genes in the main text (please see above our answer to point 3). The 242 plasmid genes (Supplementary Data 4 and Supplementary Fig. 9) were identified in the pangenome analysis, which was based principally on complete genomes. However, due to the large proportion of short- read sequences in our study, we are not comfortable with the idea of linking the AMR determinants to these 242 plasmid genes. We therefore prefer not to link Supplementary Data 4 and Fig. 9 to the section dedicated to AMR. However, we now indicate in the main text (results section) that these plasmid genes were found in plasmids with and without AMR genes. It reads: "Of the 1,506 accessory genes present in \(< 95\%\) of the genomes, 696 (46.2%), 242 (16.1%), and 28 (1.9%) were found to belong to prophages, plasmids (with or without AMR genes), and transposases, respectively (Supplementary Data 4)."  
+
+With respect to the phylogenetics and cgMLST analyses;  
+
+5. How many (if any) HC400_1620 samples in Enterobase did not contain the STM 336 SNV? It isn't wholly clear from the text if HC400_1620 is just SPB- strains, or if it includes both SPB+ and SPB- strains.  
+
+Authors: All the HC400_1620 genomes contained the STM 3356 d-Tar- specific SNV.  
+
+In the main text (lines 175- 181) of our first submission we stated: "We ensured that this diversity dataset comprised exclusively of genomes (i) with the correct in silico serotype, (ii) containing the specific SNV associated with the loss of d- Tar fermentation in SPB- (ref.22), (iii) and belonging to the invasive lineage, PG1 (ref.27). This was achieved in a straightforward manner using the HC400_1620 cluster of the EnteroBase Salmonella core- genome MLST scheme (https://enterobase.warwick.ac.uk/species/index/senteria) as a proxy (Supplementary Note "Validation of the SPB' PG1 diversity dataset", Supplementary Fig. 1)."  
+
+In the Supplementary Note "Validation of the SPB' PG1 diversity dataset" of our first submission we stated (lines 82- 87): "All 446 genomes from SPB isolates and strains contributed by various reference laboratories across the world for this study, and the 109 previously published genomes belonged to HC400_1620 and contained the d- Tar specific SNV (Supplementary Data 1)." and "Furthermore, only the HC400_1620 genomes contained the specific SNV within STM 3356 described in SPB' strains15."  
+
+In the Supplementary Data 1 of our first submission, the column (R) entitled "STM 3356 d- Tar- specific SNV" showed that all 568 PG1 genomes contain this SNV.  
+
+7. Did the Onarimon samples cluster in a single lineage? were they grouped with other more contemporary samples in a single lineage? it isn't clear from the paper/results where these sit with respect to other lineages within SPB- samples.  
+
+Authors: As this serovar was extremely rare, with only one isolate (the serotype reference
+
+<--- Page Split --->
+
+strain) in our collection (which includes more than 300,000 clinical isolates and all serovar and variant reference strains of Salmonella), it is discussed in the Supplementary Note entitled "Validation of the SPB' PG1 diversity dataset" (lines 82- 87): "During this search, we unexpectedly found within HC400_1620, a reference strain (116K) of an extremely rare serotype, Onarimon (antigenic formula: 1,9,12:b:1,2), deposited independently by one of the participating laboratories (Institut Pasteur). Serotype prediction based on genomic sequence confirmed this serotype and the d Tar- specific SNV was present. There were only 28 serotype Onarimon strains reported in 1965 (among the 547,386 strains from diverse sources across the world)17 and this serotype has been reported to cause paratyphoid fevers18. As in Salmonella spp., serotype antigens can be subject to horizontal gene transfer and homologous recombination19, we therefore considered 116K to be an O antigen- variant of SPB- and we included it in the study.  
+
+We have now modified one sentence and added another to the main text which now reads "Lineage L5 was more frequent in Asia (particularly East Asia), whereas L2 and L9 were more frequently identified in Europe. The only strain (116K) of serotype Onarimon (an O:9 antigen variant of SPB) — isolated in Japan in 1935 — belonged to lineage L5.".  
+
+## Lastly;  
+
+8. In the validation work around Mykrobe, was the probe tested against a wider database to identify if the probe could miss-detect SPB? what validation was done on this - as the MS only lists testing against the study datasets.  
+
+Authors: We have now tested our genomic tool on a genomic dataset of 102 non- SPB PG1 Salmonella reference genomes from the SARA, SARB and ATCC collections (new Supplementary Data 10). Interestingly, two rare genotypes (genotypes 7.0 and 7.3) were commonly called (in most Salmonella serotypes). The data are presented in detail in the Supplementary Note entitled "Development of a new SNV-based genotyping tool for SPB- PG1". In the main text (discussion section), we have added a sentence stressing the importance of applying this scheme only to genuine SPB PG1 genomes. It reads "It is, however, important to apply this scheme only to confirmed SPB' PG1 genomes — identified by the cgMLST HC400_1620 cluster or using MLST7 plus the specific d-Tar SNV — as Mykrobe may otherwise assign the rare genotype 7.0 and to a lesser extent genotype 7.3 to most non-PG1 Salmonella genomes (for non-PG1 genomes the tool will, however, always yield "Unknown" instead of "Salmonella_Paratyphi B" under the column "species" of the output table).".  
+
+9. In the discussion there are assertions around long term carriage in elderly patients - it wasn't clear to me what the evidence was for this. In those contemporaneous samples from older patients, was there evidence of carriage (e.g., diversity) or long-term chronic infection in their genomes?  
+
+Authors: This assertion on long-term carriage in elderly patients in Europe is based on the fact that old SPB PG1 genotypes (those that were epidemic in Europe several decades ago, such as genotype 9.1 in France) are currently found only in elderly patients with no recent history of travel to countries in which PTB is endemic and without enteric fever. Instead, their isolates (some obtained regularly and consistently over time from the same patient) originate from stools (rather than blood samples retrieved during acute bloodstream
+
+<--- Page Split --->
+
+infections). We previously wrote "Interestingly, old genotypes are still being isolated in Europe. For example, \(4\%\) of the surveillance isolates from the UK belonged to genotypes 2.1 and 5, and \(9.5\%\) of those in France belonged to the 9.1_ France genotype. The eight French cases, for which isolates were not recovered from blood samples, were patients between 81 and 98 years of age, suggesting that they may be long- term carriers infected several decades ago.". This has been changed to: "Interestingly, old genotypes are still being isolated in Europe. For example, \(4\%\) of the surveillance isolates from the UK belonged to genotypes 2.1 \((n = 7)\) and 5 \((n = 1)\) , and \(9.5\%\) \((n = 8)\) of those in France belonged to the 9.1_ France genotype. The eight French cases, for which isolates were not recovered from blood samples, were patients between 81 and 98 years of age with no recent history of travel to countries in which PTB is endemic, suggesting that they may be long- term carriers infected several decades ago.".  
+
+10. Were the 'carriage' isolates resistant to antibiotics (which one might have expected - as one would have expected some antibiotic exposure in older patients at some point, which you might have expected to impact carriage of SPB)?  
+
+Authors: This would make a lot of sense. However, due to the design of our microbiological study, with the inclusion of very limited metadata, it was impossible to determine whether all our isolates originated from long-term carriers or other types of patient. We cannot, therefore, present a formal comparison of AMR data between long-term carriers and other patients. Three of the eight French potential long-term carriers of genotype 9.1 identified in the surveillance dataset (please see point 9) had gyrA mutations (three different types). We now indicate in the main text (discussion section): "Long-term SPB: PG1 carriers may have particularly high levels of exposure; for example, three of the eight (37.5%) genotype 9.1 isolates recently obtained from elderly French patients (see above) had gyrA mutations (of three different types)."
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author): The authors have satisfactorily addressed all comments. No further suggestions.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have satisfactorily addressed all comments. No further suggestions.  
+
+Authors: We would like to thank again the three reviewers for their many insightful comments.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a42b23600537fd23c88e5370b65f268fab4924233bc8334a78821c082fc84d54/supplementary_0_Peer Review File__a42b23600537fd23c88e5370b65f268fab4924233bc8334a78821c082fc84d54_det.mmd

@@ -0,0 +1,446 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[70, 110, 362, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 154, 881, 209]]<|/det|>
+Genomic perspective on the bacillus causing paratyphoid B fever  
+
+<|ref|>image<|/ref|><|det|>[[56, 732, 240, 782]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 911, 785]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 290, 107]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 127, 392, 143]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 163, 872, 308]]<|/det|>
+The authors present compelling analyses documenting the genetic diversity, population structure, and historical distribution from a well- curated dataset of S. Paratyphi B d- tartrate(- ) strains. These results provide novel insight into the global distribution of SPB(- ), which continues to be an important public health concern. Furthermore, informed by their genomic analyses, the authors developed a pipeline for rapid and accurate typing of SPB(- ) strains for facilitating public health surveillance efforts. The data analysis, interpretation and conclusions are sound. There are some speculations about the historical events that could have led to the observed distribution in SPB, but these points are brought up in the discussions section of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[116, 327, 806, 362]]<|/det|>
+Overall, this manuscript is based on compelling, sound, and robust evidence. I have only minor suggestions for the authors to consider.  
+
+<|ref|>text<|/ref|><|det|>[[116, 382, 503, 399]]<|/det|>
+Line 116: what is synthetic cream? Non- dairy cream?  
+
+<|ref|>text<|/ref|><|det|>[[114, 418, 874, 510]]<|/det|>
+Lines 200- 201: How were the higher- resolution genotypes named? Some have locations associated with their names? Since attaching geographic locations to names has some important public health implications, more details about \(\%\) of isolates from a given geographic location to assign clade name would be helpful to include, especially in the cases where the groups have multiple geo locations (ex. NorthAfrica_Europe).  
+
+<|ref|>text<|/ref|><|det|>[[114, 529, 866, 584]]<|/det|>
+Line 319: I might have missed this, but how was the presence of sopE confirmed/determined? With blastn? Or was this denoted from the Phaster annotation? I think this is important to describe because the authors present support for using sopE as a marker for SPB- strains (line 455)  
+
+<|ref|>text<|/ref|><|det|>[[114, 602, 874, 656]]<|/det|>
+Lines 334- 337: Were the sequences of sopE at the three genomic insertion sites different? Was sopE at insertion site 1 distinguishable from sopE at insertion site 2 or 3? The analyses suggest that they may be carried on different prophages.  
+
+<|ref|>text<|/ref|><|det|>[[114, 675, 879, 711]]<|/det|>
+Line 430: add 'in recent years' – it is noted at the beginning of the section that the surveillance dataset is used for this conclusion. Also, the three reasons listed are the 'primary' reasons  
+
+<|ref|>text<|/ref|><|det|>[[114, 729, 844, 765]]<|/det|>
+Line 490: Why 10 lineages? Earlier in the paper (Fig. 1) the authors note the presence of 11 lineages including 3 with singletons, of which lineage 11 is one  
+
+<|ref|>text<|/ref|><|det|>[[116, 821, 392, 838]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 858, 857, 893]]<|/det|>
+Thank you for the opportunity to review this manuscript which describes the population diversity and evolution of Salmonella enterica serotype Paratyphi B or SPB. This organism of one of four causes of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 590, 107]]<|/det|>
+enteric fever, a global public health challenge for many centuries.  
+
+<|ref|>text<|/ref|><|det|>[[115, 109, 752, 144]]<|/det|>
+This is an original, well- planned and well- presented study by experts in microbiology of enteropathogens.  
+
+<|ref|>text<|/ref|><|det|>[[115, 145, 870, 272]]<|/det|>
+The authors focused largely on D- tartrate non- fermenting strains of SPB and examined the population structure and temporal evolution of 568 genomes, the majority of which were generated for this study. The study identified 11 lineages (with the predominance of L10) and 38 SNVs unique to each SPB genotype and proposed and implemented a hierarchical SNV- based genotyping scheme that can split SPB populations into phylogeographically informative genotypes. The core genome of 4,044 genes was identified and proportions of accessory genes belonging to prophages, plasmids and transposases were quantified.  
+
+<|ref|>text<|/ref|><|det|>[[115, 273, 880, 327]]<|/det|>
+They also assembled and investigated a set of 336 genomes from four major public health laboratories in North America and Europe, which were uploaded between 2015 and 2023. The evolutionary timescale analysis and Bayesian analysis of population structure was applied to the data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 328, 874, 400]]<|/det|>
+The paper builds on a rich history of discovery of salmonellas associated with clinical typhoid infections, fills a current gap in our understanding of the genomic epidemiology of this pathogen and should be of interest to readers of the Journal. The methods employed in this study are cutting- edge in microbial genomics and are adequate for the research aims and hypotheses.  
+
+<|ref|>text<|/ref|><|det|>[[115, 401, 860, 436]]<|/det|>
+In order to further improve the paper, the authors may consider addressing the following suggestions and questions:  
+
+<|ref|>text<|/ref|><|det|>[[115, 437, 872, 472]]<|/det|>
+- How informative could the suggested SPB genotyping system be for the public health investigation of outbreaks? What does the surveillance dataset tell us in that respect?  
+
+<|ref|>text<|/ref|><|det|>[[115, 473, 875, 565]]<|/det|>
+- Similar population genomics studies of Salmonella Paratyphi A emphasised the role gene acquisitions or losses play as key molecular events in the evolution of new lineages (e.g., Jacob JJ et al. Genomic analysis unveils genome degradation events and gene flux in the emergence and persistence of S. Paratyphi A lineages. PLoS Pathogens 2023;19(4): e1010650). Have the authors observed any parallels in their studies of SPB?  
+
+<|ref|>text<|/ref|><|det|>[[115, 565, 880, 601]]<|/det|>
+- Clarify whether any SPB isolates from Connor's collection (Connor TR et al. mBio 2016; 7(4): e00527-16 or Reference 27) were included in the study dataset of 568 genomes?  
+
+<|ref|>text<|/ref|><|det|>[[115, 602, 803, 637]]<|/det|>
+- Please provide percentages for cases where quinolone resistance conferring mutations were documented (lines 294-298).  
+
+<|ref|>text<|/ref|><|det|>[[115, 639, 875, 711]]<|/det|>
+- The important observation of azithromycin resistance presented in Figure 5 deserves to be mentioned in the text with the mechanism of resistance explained. The term 'azithromycin' (i.e. an individual antibiotic name) in Figure 5 can be replaced with 'macrolides' (antibiotic class name) to be consistent with other terms in the legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 712, 727, 729]]<|/det|>
+- Table 2 – Suggested heading for the second column – “Years of MRCA (95% HPD)”.  
+
+<|ref|>text<|/ref|><|det|>[[115, 731, 378, 746]]<|/det|>
+- Some minor editorial suggestions:  
+
+<|ref|>text<|/ref|><|det|>[[115, 748, 870, 800]]<|/det|>
+o Rephrase the statement in the Abstract “We show that this pathogen existed in the 13th century” to “Our comparative genomics findings suggest that this pathogen has existed since the 13th century”.  
+
+<|ref|>text<|/ref|><|det|>[[115, 792, 872, 828]]<|/det|>
+o Change “exploitation’ to ‘exploration’ (line 391), ‘their colonial and immigration histories’ (line 431) to ‘migration and travel patterns over recent centuries’  
+
+<|ref|>text<|/ref|><|det|>[[115, 828, 380, 843]]<|/det|>
+o Change to ‘Universities’ (line 544).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 91, 392, 107]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 126, 871, 179]]<|/det|>
+This is an interesting and very detailed study of the lineage of Salmonella that is the causative agent of Paratyphi B. It is well written, and I enjoyed reading it. It is certainly worthy of publication and will be of broad interest.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 200, 174, 216]]<|/det|>
+## Novelty  
+
+<|ref|>text<|/ref|><|det|>[[114, 235, 875, 437]]<|/det|>
+I think the paper has broad appeal, both within the field and in related fields. The work itself is original, and also builds well on the existing literature. As well as bringing together and drawing a line under findings from other studies (including those by the authors and others), the paper includes some novel/important findings. Firstly, the detailed historical collection shows strong evidence of phylogeographic clustering, which is not something that has been reported before, and is of epidemiological value and scientific value. Secondly, the pattern of resistance is also interesting, including evidence for increasing resistance in SPB isolates recently. Thirdly, The work around the prophages within SPB is also novel, and the pangenome analysis is also more comprehensive than previous work, due to a combination of the larger historic dataset and the use of long read technologies. Lastly, the observation around S. Onarimon, in the supplementary is also of interest to those who are enthusiasts of the Salmonella, and it was an unexpected nugget that was also new and interesting.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 456, 250, 472]]<|/det|>
+## Support for claims  
+
+<|ref|>text<|/ref|><|det|>[[116, 492, 850, 528]]<|/det|>
+Overall, the paper has a large amount of detail, and represents a very in- depth genomic study of this interesting pathogen.  
+
+<|ref|>text<|/ref|><|det|>[[114, 546, 866, 655]]<|/det|>
+The paper brings together historical work very well along with a historical and contemporary genomic dataset to provide the highest resolution exploration to date of SPB. The approaches used are appropriate and well matched to the aims, and the methods, both laboratory and bioinformatics are explained in sufficient detail to enable reproducibility. The use of Enterobase is welcome and enhances the reproducibility of the work. The release of the probes for Mykrobe is also welcome as this is open- source software.  
+
+<|ref|>text<|/ref|><|det|>[[114, 674, 875, 710]]<|/det|>
+Overall, the data analysis is sensible, and the methods are appropriate and meet the expected standards in the field.  
+
+<|ref|>text<|/ref|><|det|>[[116, 729, 437, 746]]<|/det|>
+Other questions/comments for the authors:  
+
+<|ref|>text<|/ref|><|det|>[[116, 765, 744, 801]]<|/det|>
+Although I like this paper a lot and think it should be published, there are a number of questions/clarifications that would be helpful, these are;  
+
+<|ref|>text<|/ref|><|det|>[[116, 820, 871, 893]]<|/det|>
+The paper reports the correlation with geography and the evidence of geographic clustering; however, I couldn't see a formal analysis for the phylogeography. Given that the authors had made use of BEAST, I wondered if they had made use of any of the phylogeographic models to examine predicted origins and phylogeographic spread more formally.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 92, 135, 104]]<|/det|>
+So  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 872, 199]]<|/det|>
+1. Were analyses performed to substantiate the suggestion of export from Europe?  
+2. The detail around the BEAST analysis is in the supplementary, and these results are interesting, although broadly agreeing with what had been found before - but the BEAST date/context information isn't that well referenced in the main text - so could the authors consider how to better signpost to the detail in the supplementary information regarding BEAST.  
+
+<|ref|>text<|/ref|><|det|>[[115, 217, 870, 253]]<|/det|>
+The AMR analysis is interesting, but it wasn't clear to me from the paper in what context the resistance genes were found. So;  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 875, 290]]<|/det|>
+3. Were the genes found on chromosomal mobile elements, carried with/on phage or were they plasmid carried?  
+
+<|ref|>text<|/ref|><|det|>[[115, 290, 860, 383]]<|/det|>
+4. From the figure in the supplementary, it looks like there weren't many plasmids in the samples analysed, though it would be good to have this described in a little more detail. The paper mentions a phylogenetic comparison between plasmid genes and core genome, but it isn't clear to me what this means, or what it shows - can the authors explain what is meant and consider if figure 9 in the supplementary is appropriate to communicate this.  
+
+<|ref|>text<|/ref|><|det|>[[115, 419, 525, 437]]<|/det|>
+With respect to the phylogenetics and cgMLST analyses;  
+
+<|ref|>text<|/ref|><|det|>[[115, 455, 872, 547]]<|/det|>
+5. How many (if any) HC400_1620 samples in Enterobase did not contain the STM 336 SNV? It isn't wholly clear from the text if HC400_1620 is just SPB-strains, or if it includes both SPB+ and SPB-strains.  
+7. Did the Onarimon samples cluster in a single lineage? were they grouped with other more contemporary samples in a single lineage? it isn't clear from the paper/results where these sit with respect to other lineages within SPB-samples.  
+
+<|ref|>text<|/ref|><|det|>[[115, 566, 163, 582]]<|/det|>
+Lastly;  
+
+<|ref|>text<|/ref|><|det|>[[115, 601, 869, 656]]<|/det|>
+8. In the validation work around Mykrobe, was the probe tested against a wider database to identify if the probe could miss-detect SPB? what validation was done on this - as the MS only lists testing against the study datasets.  
+
+<|ref|>text<|/ref|><|det|>[[115, 692, 875, 803]]<|/det|>
+9. In the discussion there are assertions around long term carriage in elderly patients - it wasn't clear to me what the evidence was for this. In those contemporaneous samples from older patients, was there evidence of carriage (e.g., diversity) or long-term chronic infection in their genomes?  
+10. Were the 'carriage' isolates resistant to antibiotics (which one might have expected - as one would have expected some antibiotic exposure in older patients at some point, which you might have expected to impact carriage of SPB)?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 84, 833, 118]]<|/det|>
+Authors: We would like to thank the reviewers for their many insightful comments. We address these comments, point by point, below.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 136, 313, 151]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[117, 154, 428, 170]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 187, 877, 344]]<|/det|>
+The authors present compelling analyses documenting the genetic diversity, population structure, and historical distribution from a well- curated dataset of S. Paratyphi B d- tartrate(- ) strains. These results provide novel insight into the global distribution of SPB(- ), which continues to be an important public health concern. Furthermore, informed by their genomic analyses, the authors developed a pipeline for rapid and accurate typing of SPB(- ) strains for facilitating public health surveillance efforts. The data analysis, interpretation and conclusions are sound. There are some speculations about the historical events that could have led to the observed distribution in SPB, but these points are brought up in the discussions section of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 361, 840, 396]]<|/det|>
+Overall, this manuscript is based on compelling, sound, and robust evidence. I have only minor suggestions for the authors to consider.  
+
+<|ref|>text<|/ref|><|det|>[[115, 413, 552, 430]]<|/det|>
+Line 116: what is synthetic cream? Non- dairy cream?  
+
+<|ref|>text<|/ref|><|det|>[[115, 431, 873, 518]]<|/det|>
+Authors: It is an emulsion of vegetable oils or fat with water, with or without the addition of other substances, some of which may be of dairy origin. It was used in the UK during World War II when the sale of natural cream was forbidden. This information has been added to the Supplementary Note "Epidemiology of paratyphoid B fever during the first half of the \(20^{\text{th}}\) century".  
+
+<|ref|>text<|/ref|><|det|>[[115, 535, 875, 621]]<|/det|>
+Lines 200- 201: How were the higher- resolution genotypes named? Some have locations associated with their names? Since attaching geographic locations to names has some important public health implications, more details about \(1\%\) of isolates from a given geographic location to assign clade name would be helpful to include, especially in the cases where the groups have multiple geo locations (ex. NorthAfrica_Europe).  
+
+<|ref|>text<|/ref|><|det|>[[115, 622, 875, 848]]<|/det|>
+Authors: We now indicate in the results section: "Strong geographic patterns with differences from country to continent level were observed for 17 genotypes, whereas two genotypes were more widespread, isolated from two continents (Supplementary Data 1). These two genotypes were genotype 7.2 found in Europe (26.3%, 5/19) and Asia (73.7%, 14/19, mostly East Asia), and genotype 10.3.8.5 found in Europe (71.4%, 20/28) and North Africa (25%, 7/28). One genotype, 7.3.2 — also found in Europe (41.2%, 7/17) and North Africa (52.9%, 9/17) — was associated with a particular PT, BAOR (see Comparison of phylogenomics data with other typing schemes) (Supplementary Data 1). This geographic or PT information was added to the genotype nomenclature as an alias, to make it more informative." and in the discussion section: "New genotypes with geographic information (in cases of new emerging genotypes in a defined area) or new updates on alias names (in cases of the establishment of known genotypes in new areas) should be added to the scheme in the future."  
+
+<|ref|>text<|/ref|><|det|>[[115, 866, 852, 900]]<|/det|>
+Line 319: I might have missed this, but how was the presence of sope confirmed/determined? With blastn? Or was this denoted from the Phaster annotation? I
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 84, 860, 118]]<|/det|>
+think this is important to describe because the authors present support for using sopE as a marker for SPB- strains (line 455)  
+
+<|ref|>text<|/ref|><|det|>[[115, 119, 872, 223]]<|/det|>
+Authors: To clarify this point, we have added the following sentences to the M&M section entitled "Prophage content analysis and sopE copy- number variation": "By combining the annotation of the 14 SPB complete genomes and the prophage delineation described above, we were able to locate the sopE gene in SEN34- like and P88- like prophages. We used the blastn algorithm and the sopE reference sequence (GenBank accession no. L78932) from S. enterica serotype Dublin<sup>38</sup> to confirm our identification of the prophage- borne sopE gene."  
+
+<|ref|>text<|/ref|><|det|>[[115, 240, 866, 291]]<|/det|>
+Lines 334- 337: Were the sequences of sopE at the three genomic insertion sites different? Was sopE at insertion site 1 distinguishable from sopE at insertion site 2 or 3? The analyses suggest that they may be carried on different prophages.  
+
+<|ref|>text<|/ref|><|det|>[[115, 293, 870, 396]]<|/det|>
+Authors: The sopE gene nucleotide sequence is highly conserved across prophages. We now indicate the following in results section: "In the 14 complete genomes, the nucleotide sequence of the sopE gene was 100% identical, regardless of the prophages in which it is inserted. One exception was the sopE gene inserted at site #3 of the B624 genome, which differed by eight (five being non- synonymous) out of 723 nucleotides from the sopE consensus sequence of SPB."  
+
+<|ref|>text<|/ref|><|det|>[[115, 414, 810, 465]]<|/det|>
+Line 430: add 'in recent years' - it is noted at the beginning of the section that the surveillance dataset is used for this conclusion. Also, the three reasons listed are the 'primary' reasons  
+
+<|ref|>text<|/ref|><|det|>[[117, 467, 355, 483]]<|/det|>
+Authors: This has been done.  
+
+<|ref|>text<|/ref|><|det|>[[115, 501, 864, 535]]<|/det|>
+Line 490: Why 10 lineages? Earlier in the paper (Fig. 1) the authors note the presence of 11 lineages including 3 with singletons, of which lineage 11 is one  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 868, 621]]<|/det|>
+Authors: We were actually referring to the 10 lineages or phylogroups described by Connor et al for the global population SPB and not the lineages we have identified for SPB PG1. We have clarified this in the following sentence: "Firstly, SPB isolates can be assigned to the 10 known PGs described by Connor and coworkers<sup>27</sup> with the EnteroBase cgMLST scheme, with the HC400_1620 cluster considered a signature of SPB PG1.".  
+
+<|ref|>text<|/ref|><|det|>[[117, 640, 428, 656]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 675, 872, 760]]<|/det|>
+Thank you for the opportunity to review this manuscript which describes the population diversity and evolution of Salmonella enterica serotype Paratyphi B or SPB. This organism of one of four causes of enteric fever, a global public health challenge for many centuries. This is an original, well- planned and well- presented study by experts in microbiology of enteropathogens.  
+
+<|ref|>text<|/ref|><|det|>[[115, 762, 861, 899]]<|/det|>
+The authors focused largely on D- tartrate non- fermenting strains of SPB and examined the population structure and temporal evolution of 568 genomes, the majority of which were generated for this study. The study identified 11 lineages (with the predominance of L10) and 38 SNVs unique to each SPB genotype and proposed and implemented a hierarchical SNV- based genotyping scheme that can split SPB populations into phylogeographically informative genotypes. The core genome of 4,044 genes was identified and proportions of accessory genes belonging to prophages, plasmids and transposases were quantified. They also assembled and investigated a set of 336 genomes from four major public health
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 857, 135]]<|/det|>
+laboratories in North America and Europe, which were uploaded between 2015 and 2023. The evolutionary timescale analysis and Bayesian analysis of population structure was applied to the data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 135, 877, 222]]<|/det|>
+The paper builds on a rich history of discovery of salmonellas associated with clinical typhoid infections, fills a current gap in our understanding of the genomic epidemiology of this pathogen and should be of interest to readers of the Journal. The methods employed in this study are cutting- edge in microbial genomics and are adequate for the research aims and hypotheses.  
+
+<|ref|>text<|/ref|><|det|>[[115, 223, 852, 256]]<|/det|>
+In order to further improve the paper, the authors may consider addressing the following suggestions and questions:  
+
+<|ref|>text<|/ref|><|det|>[[113, 272, 881, 745]]<|/det|>
+- How informative could the suggested SPB genotyping system be for the public health investigation of outbreaks? What does the surveillance dataset tell us in that respect? Authors: We now expand on this aspect in the revised MS (discussion section). The text now reads: "We anticipate that the use of this scheme and its universal nomenclature will improve the laboratory surveillance of PTB. It is now easier to track the different SPB PG1 populations at global scale. We were able to identify the different genotypes detected recently in travellers from or migrants to four high-income countries in North America and Europe (surveillance dataset). However, more global studies involving countries experiencing PTB across the globe and performed regularly over time — as for the PT surveys — would be helpful to monitor the diversity, spread and evolution (of AMR in particular) of this pathogen. New genotypes with geographic information (in cases of new emerging genotypes in a defined area) or new updates on alias names (in cases of the establishment of known genotypes in new areas) should be added to the scheme in the future. The use of a common nomenclature would also be helpful during outbreak investigations. It is now straightforward to define the bacterial types of SPB PG1 and to share this information during transborder outbreak investigations. This should also facilitate the identification of transmission chains associated with particular geographic regions, particularly in areas in which PTB is not endemic. For example, genomic surveillance in the UK identified an imported SPB outbreak in travellers coinciding with a mass gathering in Iraq in 2021 (ref.29). The isolates from these patients clustered in one of the two clades labelled "travel to Iraq". According to our genotyping scheme, this clade corresponds to genotype 10.3.2 MiddleEast1. The oldest isolates belonging to this genotype were collected in Iran in 1965 and Iraq in 1975, suggesting that this strain has been endemic in the region for many decades. The second clade labelled "travel to Iraq" identified by UKHSA corresponded to genotype 10.3.8.3 MiddleEast3. In endemic regions, this genotyping scheme might also be useful for determining the genotype of a potential outbreak strain if several genotypes are known to circulate regionally."  
+
+<|ref|>text<|/ref|><|det|>[[115, 761, 866, 900]]<|/det|>
+- Similar population genomics studies of Salmonella Paratyphi A emphasised the role gene acquisitions or losses play as key molecular events in the evolution of new lineages (e.g., Jacob JJ et al. Genomic analysis unveils genome degradation events and gene flux in the emergence and persistence of S. Paratyphi A lineages. PLoS Pathogens 2023;19(4): e1010650). Have the authors observed any parallels in their studies of SPB? Authors: This interesting aspect of genome degradation events and gene flux leading to the emergence of SPB PG1, a human-adapted pathogen causing paratyphoid fever, was previously studied by Connor et al (our reference #27) in their comparative genomics
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 875, 223]]<|/det|>
+analysis of the different PGs (1- 10) of SPB. We did not address this question as our study focused exclusively on the PG1 population. Regarding the evolution of PG1, our comprehensive pangenome analysis failed to identify genomic signals leading to increases in virulence or transmissibility during recent microevolution (as previously reported for other enteric fever pathogens, such as S. enterica serovars Typhi (Roumagnac et al. Science 2016) and Paratyphi A (Zhou et al. PNAS 2014)). Regarding virulence, we documented up to three copies of the sopE gene in some lineages, but we have no other data linking this microbial genomics finding to an increase in virulence in vivo or in human populations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 880, 380]]<|/det|>
+- Clarify whether any SPB isolates from Connor's collection (Connor TR et al. mBio 2016; 7(4): e00527-16 or Reference 27) were included in the study dataset of 568 genomes? Authors: The section entitled "S. enterica Paratyphi B sequence data collections" of the M&M (lines 579-583) previously included the following: "We first studied a diversity dataset of 568 SPB genomes, 446 of which were generated specifically for this study, 109 had already been published<sup>27,28,29,31,48,49,50</sup>, and the other 13 were unpublished but deposited in EnteroBase (https://enterobase.warwick.ac.uk/species/index/senterica) or GenBank (https://www.ncbi.nlm.nih.gov/genbank/) (Supplementary Data 1)."  
+
+<|ref|>text<|/ref|><|det|>[[115, 396, 877, 431]]<|/det|>
+In Supplementary Data 1 (the list of 568 genomes), the column (BH) entitled "Source" can be used to identify the 28 strains from Connor et al.  
+
+<|ref|>text<|/ref|><|det|>[[115, 448, 875, 536]]<|/det|>
+As we used 109 published genomes from seven different papers, we did not want to overload the text by providing the numbers of strains used per published paper. Instead, we provide all the necessary details in Supplementary Table 1. However, if the editor would prefer all this information to be provided in the main text, we are willing to comply with this request.  
+
+<|ref|>text<|/ref|><|det|>[[115, 553, 846, 605]]<|/det|>
+- Please provide percentages for cases where quinolone resistance conferring mutations were documented (lines 294-298). Authors: This has been done.  
+
+<|ref|>text<|/ref|><|det|>[[115, 622, 875, 673]]<|/det|>
+- The important observation of azithromycin resistance presented in Figure 5 deserves to be mentioned in the text with the mechanism of resistance explained.  
+
+<|ref|>text<|/ref|><|det|>[[115, 655, 870, 727]]<|/det|>
+Authors: This has been done: "One isolate (83282), acquired in South America in 2014, contained the mph(A) gene encoding a macrolide 2' phosphotransferase, a common determinant of azithromycin resistance (however, the antimicrobial susceptibility pattern of this isolate was unavailable)."  
+
+<|ref|>text<|/ref|><|det|>[[115, 744, 866, 797]]<|/det|>
+The term 'azithromycin' (i.e. an individual antibiotic name) in Figure 5 can be replaced with 'macrolides' (antibiotic class name) to be consistent with other terms in the legend. Authors: This has been done.  
+
+<|ref|>text<|/ref|><|det|>[[115, 813, 808, 848]]<|/det|>
+- Table 2 – Suggested heading for the second column – "Years of MRCA (95% HPD)". Authors: This has been changed to "Time (year) of the MRCA".  
+
+<|ref|>text<|/ref|><|det|>[[117, 866, 412, 883]]<|/det|>
+- Some minor editorial suggestions:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 880, 152]]<|/det|>
+o Rephrase the statement in the Abstract "We show that this pathogen existed in the 13th century" to "Our comparative genomics findings suggest that this pathogen has existed since the 13th century".  
+
+<|ref|>text<|/ref|><|det|>[[117, 154, 355, 170]]<|/det|>
+Authors: This has been done.  
+
+<|ref|>text<|/ref|><|det|>[[115, 187, 866, 240]]<|/det|>
+o Change "exploitation' to 'exploration' (line 391), 'their colonial and immigration histories' (line 431) to 'migration and travel patterns over recent centuries' Authors: This has been done.  
+
+<|ref|>text<|/ref|><|det|>[[115, 257, 415, 292]]<|/det|>
+o Change to 'Universities' (line 544). Authors: This has been done.  
+
+<|ref|>text<|/ref|><|det|>[[117, 309, 428, 326]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 344, 826, 396]]<|/det|>
+This is an interesting and very detailed study of the lineage of Salmonella that is the causative agent of Paratyphi B. It is well written, and I enjoyed reading it. It is certainly worthy of publication and will be of broad interest.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 414, 182, 430]]<|/det|>
+## Novelty  
+
+<|ref|>text<|/ref|><|det|>[[115, 447, 881, 658]]<|/det|>
+I think the paper has broad appeal, both within the field and in related fields. The work itself is original, and also builds well on the existing literature. As well as bringing together and drawing a line under findings from other studies (including those by the authors and others), the paper includes some novel/important findings. Firstly, the detailed historical collection shows strong evidence of phylogeographic clustering, which is not something that has been reported before, and is of epidemiological value and scientific value. Secondly, the pattern of resistance is also interesting, including evidence for increasing resistance in SPB isolates recently. Thirdly, The work around the prophages within SPB is also novel, and the pangenome analysis is also more comprehensive than previous work, due to a combination of the larger historic dataset and the use of long read technologies. Lastly, the observation around S. Onarimon, in the supplementary is also of interest to those who are enthusiasts of the Salmonella, and it was an unexpected nugget that was also new and interesting.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 675, 268, 691]]<|/det|>
+## Support for claims  
+
+<|ref|>text<|/ref|><|det|>[[117, 709, 836, 744]]<|/det|>
+Overall, the paper has a large amount of detail, and represents a very in- depth genomic study of this interesting pathogen.  
+
+<|ref|>text<|/ref|><|det|>[[115, 761, 878, 866]]<|/det|>
+The paper brings together historical work very well along with a historical and contemporary genomic dataset to provide the highest resolution exploration to date of SPB. The approaches used are appropriate and well matched to the aims, and the methods, both laboratory and bioinformatics are explained in sufficient detail to enable reproducibility. The use of Enterobase is welcome and enhances the reproducibility of the work. The release of the probes for Mykrobe is also welcome as this is open- source software.  
+
+<|ref|>text<|/ref|><|det|>[[115, 883, 809, 900]]<|/det|>
+Overall, the data analysis is sensible, and the methods are appropriate and meet the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 84, 375, 100]]<|/det|>
+expected standards in the field.  
+
+<|ref|>text<|/ref|><|det|>[[117, 118, 478, 135]]<|/det|>
+Other questions/comments for the authors:  
+
+<|ref|>text<|/ref|><|det|>[[117, 153, 822, 188]]<|/det|>
+Although I like this paper a lot and think it should be published, there are a number of questions/clarifications that would be helpful, these are;  
+
+<|ref|>text<|/ref|><|det|>[[115, 205, 880, 293]]<|/det|>
+The paper reports the correlation with geography and the evidence of geographic clustering; however, I couldn't see a formal analysis for the phylogeography. Given that the authors had made use of BEAST, I wondered if they had made use of any of the phylogeographic models to examine predicted origins and phylogeographic spread more formally. So  
+
+<|ref|>text<|/ref|><|det|>[[115, 309, 872, 398]]<|/det|>
+1. Were analyses performed to substantiate the suggestion of export from Europe? Authors: We initially performed two phylogeographic reconstructions with phytools (https://peerj.com/articles/16505/) and PastML (https://academic.oup.com/mbe/article/36/9/2069/5498561) to document the exportation of SPB PG1 from Europe at different points.  
+
+<|ref|>image<|/ref|><|det|>[[115, 414, 872, 670]]<|/det|>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[120, 85, 680, 805]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 808, 433, 822]]<|/det|>
+<center>Phylogeographic reconstruction with Phytools </center>  
+
+<|ref|>text<|/ref|><|det|>[[115, 839, 880, 909]]<|/det|>
+However, due to a deeper sampling for European historical isolates, we were not comfortable presenting these phylogeographic reconstructions, which might be perceived as a formal proof. Instead, we simply suggest that export from Europe may have occurred, based on historical data and the tree topology.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 858, 186]]<|/det|>
+2. The detail around the BEAST analysis is in the supplementary, and these results are interesting, although broadly agreeing with what had been found before - but the BEAST date/context information isn't that well referenced in the main text - so could the authors consider how to better signpost to the detail in the supplementary information regarding BEAST.  
+
+<|ref|>text<|/ref|><|det|>[[115, 188, 875, 432]]<|/det|>
+Authors: We have now moved all the supplementary data dealing with the Beast analysis to the main text (discussion section). It now reads "Based on our dataset of 568 SPB PG1 genomes, we estimated the age of this pathogen at \~750 years (1274 CE; 95% CI, 915 - 1583), which is very close to the previous median date of origin estimated by Connor and coworkers27 (1188 CE; 95% CI, 469 BC - 1799 CE), who used only 25 SPB PG1 genomes (i.e., those with a known year of isolation). SPB is older than SPA, which is estimated to have originated 450 - 700 years ago25. SPA was discovered two years after SPB (in the USA in 1898)1,2,32 but is currently the most frequent agent of paratyphoid fever28. Due to lineage extinction, in particular, times to the MRCA are often underestimated and the inclusion of ancient DNA in the analysis would increase precision and make it possible to establish dates of origin further in the past25. The dating of a representative collection of modern isolates of SPC estimated the origin of SPC to 456 - 664 years ago. When a draft SPC genome from an 800-year-old Norwegian skeleton was added to the analysis, the time to the MRCA increased to 1162 - 1526 years26. Unfortunately, no ancient DNA is currently available for SPB strains."  
+
+<|ref|>text<|/ref|><|det|>[[115, 448, 875, 483]]<|/det|>
+The AMR analysis is interesting, but it wasn't clear to me from the paper in what context the resistance genes were found. So;  
+
+<|ref|>text<|/ref|><|det|>[[115, 500, 870, 535]]<|/det|>
+3. Were the genes found on chromosomal mobile elements, carried with/on phage or were they plasmid carried?  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 880, 850]]<|/det|>
+Authors: Most of the AMR detected was due to single point mutations of chromosomal genes (gyrA and gyrB). The five remaining AMR isolates (those with ESBL, carbapenemase or mphA genes in particular) had AMR plasmids. This is now indicated in the main text: "Between 1898 and 2000, only one isolate (0.3%, 1/345) had antibiotic resistance genes (ARGs). This human isolate (B73-1117), collected in France in 1973, displayed resistance to ampicillin (blaTEM-1D), streptomycin (strAB, aadA1, and aadA2b), sulfonamides (sul1), chloramphenicol (cmIA1), and tetracycline (tetA). Between 2001 and 2021, 23.1% (52/223) of isolates had ARGs. One isolate acquired in Turkey in 2001 (01-7995) produced a CTX-M-3 extended-spectrum beta-lactamase30, whereas another isolate (P7704) acquired in South America in 2019 produced an OXA-48 carbapenemase31. One isolate (83282), acquired in South America in 2014, contained the mph(A) gene encoding a macrolide 2' phosphotransferase, a common determinant of azithromycin resistance (however, the antimicrobial susceptibility pattern of this isolate was unavailable). All these rare isolates contained AMR plasmids (Supplementary Data 1); however, the mechanisms of resistance (21.5%, 48/223) most prevalent during the 2001-2021 period involved mutations of the quinolone resistance-determining regions of the chromosomal gyrA and gyrB DNA gyrase genes leading to resistance to nalidixic acid and/or decreased susceptibility or resistance to ciprofloxacin (minimum inhibitory concentration [MIC] between 0.125 and 0.5 mg/L).".  
+
+<|ref|>text<|/ref|><|det|>[[115, 866, 874, 902]]<|/det|>
+4. From the figure in the supplementary, it looks like there weren't many plasmids in the samples analysed, though it would be good to have this described in a little more detail. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 84, 875, 135]]<|/det|>
+paper mentions a phylogenetic comparison between plasmid genes and core genome, but it isn't clear to me what this means, or what it shows - can the authors explain what is meant and consider if figure 9 in the supplementary is appropriate to communicate this.  
+
+<|ref|>text<|/ref|><|det|>[[115, 136, 875, 361]]<|/det|>
+Authors: Plasmid- borne AMR was indeed very rare, with only five isolates displaying such resistance. We now provide more detail about the plasmid versus chromosome locations of the AMR genes in the main text (please see above our answer to point 3). The 242 plasmid genes (Supplementary Data 4 and Supplementary Fig. 9) were identified in the pangenome analysis, which was based principally on complete genomes. However, due to the large proportion of short- read sequences in our study, we are not comfortable with the idea of linking the AMR determinants to these 242 plasmid genes. We therefore prefer not to link Supplementary Data 4 and Fig. 9 to the section dedicated to AMR. However, we now indicate in the main text (results section) that these plasmid genes were found in plasmids with and without AMR genes. It reads: "Of the 1,506 accessory genes present in \(< 95\%\) of the genomes, 696 (46.2%), 242 (16.1%), and 28 (1.9%) were found to belong to prophages, plasmids (with or without AMR genes), and transposases, respectively (Supplementary Data 4)."  
+
+<|ref|>text<|/ref|><|det|>[[117, 379, 578, 396]]<|/det|>
+With respect to the phylogenetics and cgMLST analyses;  
+
+<|ref|>text<|/ref|><|det|>[[115, 413, 870, 465]]<|/det|>
+5. How many (if any) HC400_1620 samples in Enterobase did not contain the STM 336 SNV? It isn't wholly clear from the text if HC400_1620 is just SPB- strains, or if it includes both SPB+ and SPB- strains.  
+
+<|ref|>text<|/ref|><|det|>[[115, 466, 802, 483]]<|/det|>
+Authors: All the HC400_1620 genomes contained the STM 3356 d-Tar- specific SNV.  
+
+<|ref|>text<|/ref|><|det|>[[115, 501, 875, 640]]<|/det|>
+In the main text (lines 175- 181) of our first submission we stated: "We ensured that this diversity dataset comprised exclusively of genomes (i) with the correct in silico serotype, (ii) containing the specific SNV associated with the loss of d- Tar fermentation in SPB- (ref.22), (iii) and belonging to the invasive lineage, PG1 (ref.27). This was achieved in a straightforward manner using the HC400_1620 cluster of the EnteroBase Salmonella core- genome MLST scheme (https://enterobase.warwick.ac.uk/species/index/senteria) as a proxy (Supplementary Note "Validation of the SPB' PG1 diversity dataset", Supplementary Fig. 1)."  
+
+<|ref|>text<|/ref|><|det|>[[115, 657, 876, 761]]<|/det|>
+In the Supplementary Note "Validation of the SPB' PG1 diversity dataset" of our first submission we stated (lines 82- 87): "All 446 genomes from SPB isolates and strains contributed by various reference laboratories across the world for this study, and the 109 previously published genomes belonged to HC400_1620 and contained the d- Tar specific SNV (Supplementary Data 1)." and "Furthermore, only the HC400_1620 genomes contained the specific SNV within STM 3356 described in SPB' strains15."  
+
+<|ref|>text<|/ref|><|det|>[[115, 779, 857, 813]]<|/det|>
+In the Supplementary Data 1 of our first submission, the column (R) entitled "STM 3356 d- Tar- specific SNV" showed that all 568 PG1 genomes contain this SNV.  
+
+<|ref|>text<|/ref|><|det|>[[115, 831, 875, 882]]<|/det|>
+7. Did the Onarimon samples cluster in a single lineage? were they grouped with other more contemporary samples in a single lineage? it isn't clear from the paper/results where these sit with respect to other lineages within SPB- samples.  
+
+<|ref|>text<|/ref|><|det|>[[115, 884, 856, 900]]<|/det|>
+Authors: As this serovar was extremely rare, with only one isolate (the serotype reference
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 880, 292]]<|/det|>
+strain) in our collection (which includes more than 300,000 clinical isolates and all serovar and variant reference strains of Salmonella), it is discussed in the Supplementary Note entitled "Validation of the SPB' PG1 diversity dataset" (lines 82- 87): "During this search, we unexpectedly found within HC400_1620, a reference strain (116K) of an extremely rare serotype, Onarimon (antigenic formula: 1,9,12:b:1,2), deposited independently by one of the participating laboratories (Institut Pasteur). Serotype prediction based on genomic sequence confirmed this serotype and the d Tar- specific SNV was present. There were only 28 serotype Onarimon strains reported in 1965 (among the 547,386 strains from diverse sources across the world)17 and this serotype has been reported to cause paratyphoid fevers18. As in Salmonella spp., serotype antigens can be subject to horizontal gene transfer and homologous recombination19, we therefore considered 116K to be an O antigen- variant of SPB- and we included it in the study.  
+
+<|ref|>text<|/ref|><|det|>[[116, 309, 875, 379]]<|/det|>
+We have now modified one sentence and added another to the main text which now reads "Lineage L5 was more frequent in Asia (particularly East Asia), whereas L2 and L9 were more frequently identified in Europe. The only strain (116K) of serotype Onarimon (an O:9 antigen variant of SPB) — isolated in Japan in 1935 — belonged to lineage L5.".  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 398, 170, 413]]<|/det|>
+## Lastly;  
+
+<|ref|>text<|/ref|><|det|>[[116, 431, 874, 483]]<|/det|>
+8. In the validation work around Mykrobe, was the probe tested against a wider database to identify if the probe could miss-detect SPB? what validation was done on this - as the MS only lists testing against the study datasets.  
+
+<|ref|>text<|/ref|><|det|>[[115, 484, 866, 708]]<|/det|>
+Authors: We have now tested our genomic tool on a genomic dataset of 102 non- SPB PG1 Salmonella reference genomes from the SARA, SARB and ATCC collections (new Supplementary Data 10). Interestingly, two rare genotypes (genotypes 7.0 and 7.3) were commonly called (in most Salmonella serotypes). The data are presented in detail in the Supplementary Note entitled "Development of a new SNV-based genotyping tool for SPB- PG1". In the main text (discussion section), we have added a sentence stressing the importance of applying this scheme only to genuine SPB PG1 genomes. It reads "It is, however, important to apply this scheme only to confirmed SPB' PG1 genomes — identified by the cgMLST HC400_1620 cluster or using MLST7 plus the specific d-Tar SNV — as Mykrobe may otherwise assign the rare genotype 7.0 and to a lesser extent genotype 7.3 to most non-PG1 Salmonella genomes (for non-PG1 genomes the tool will, however, always yield "Unknown" instead of "Salmonella_Paratyphi B" under the column "species" of the output table).".  
+
+<|ref|>text<|/ref|><|det|>[[115, 727, 872, 795]]<|/det|>
+9. In the discussion there are assertions around long term carriage in elderly patients - it wasn't clear to me what the evidence was for this. In those contemporaneous samples from older patients, was there evidence of carriage (e.g., diversity) or long-term chronic infection in their genomes?  
+
+<|ref|>text<|/ref|><|det|>[[115, 797, 861, 900]]<|/det|>
+Authors: This assertion on long-term carriage in elderly patients in Europe is based on the fact that old SPB PG1 genotypes (those that were epidemic in Europe several decades ago, such as genotype 9.1 in France) are currently found only in elderly patients with no recent history of travel to countries in which PTB is endemic and without enteric fever. Instead, their isolates (some obtained regularly and consistently over time from the same patient) originate from stools (rather than blood samples retrieved during acute bloodstream
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 878, 293]]<|/det|>
+infections). We previously wrote "Interestingly, old genotypes are still being isolated in Europe. For example, \(4\%\) of the surveillance isolates from the UK belonged to genotypes 2.1 and 5, and \(9.5\%\) of those in France belonged to the 9.1_ France genotype. The eight French cases, for which isolates were not recovered from blood samples, were patients between 81 and 98 years of age, suggesting that they may be long- term carriers infected several decades ago.". This has been changed to: "Interestingly, old genotypes are still being isolated in Europe. For example, \(4\%\) of the surveillance isolates from the UK belonged to genotypes 2.1 \((n = 7)\) and 5 \((n = 1)\) , and \(9.5\%\) \((n = 8)\) of those in France belonged to the 9.1_ France genotype. The eight French cases, for which isolates were not recovered from blood samples, were patients between 81 and 98 years of age with no recent history of travel to countries in which PTB is endemic, suggesting that they may be long- term carriers infected several decades ago.".  
+
+<|ref|>text<|/ref|><|det|>[[115, 309, 868, 362]]<|/det|>
+10. Were the 'carriage' isolates resistant to antibiotics (which one might have expected - as one would have expected some antibiotic exposure in older patients at some point, which you might have expected to impact carriage of SPB)?  
+
+<|ref|>text<|/ref|><|det|>[[115, 362, 872, 536]]<|/det|>
+Authors: This would make a lot of sense. However, due to the design of our microbiological study, with the inclusion of very limited metadata, it was impossible to determine whether all our isolates originated from long-term carriers or other types of patient. We cannot, therefore, present a formal comparison of AMR data between long-term carriers and other patients. Three of the eight French potential long-term carriers of genotype 9.1 identified in the surveillance dataset (please see point 9) had gyrA mutations (three different types). We now indicate in the main text (discussion section): "Long-term SPB: PG1 carriers may have particularly high levels of exposure; for example, three of the eight (37.5%) genotype 9.1 isolates recently obtained from elderly French patients (see above) had gyrA mutations (of three different types)."
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 303, 106]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 127, 700, 162]]<|/det|>
+Reviewer #1 (Remarks to the Author): The authors have satisfactorily addressed all comments. No further suggestions.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 101, 328, 117]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[117, 135, 437, 152]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 170, 788, 187]]<|/det|>
+The authors have satisfactorily addressed all comments. No further suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 205, 800, 238]]<|/det|>
+Authors: We would like to thank again the three reviewers for their many insightful comments.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a4501f2f5fc141c371fc4a9c6c00e00eda96fbbe28e2750079e2577f0d22b3f4/images_list.json

@@ -0,0 +1 @@
+[]

peer_reviews/supplementary_0_Peer Review File__a4501f2f5fc141c371fc4a9c6c00e00eda96fbbe28e2750079e2577f0d22b3f4/supplementary_0_Peer Review File__a4501f2f5fc141c371fc4a9c6c00e00eda96fbbe28e2750079e2577f0d22b3f4.mmd

@@ -0,0 +1,340 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Self- mixing in microtubule- kinesin active fluid from nonuniform to uniform distribution of activity  
+
+![PLACEHOLDER_0_0]
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+In this paper, the authors perform experiments on mixing in an active fluid composed of microtubules and kinesin clusters powered by ATP. Using light activated ATP, they are able to turn on activity over a finite region of space, thus creating an interface between an active fluid and a passive fluid. As the light- activated ATP disperses across the interface due to both diffusion and active mixing, the region where active flows take place expands spatially, thereby further enhancing transport and mixing. The authors characterize the rate at which the ATP gets mixed between the two regions, and find that the mixing is diffusive at low levels of activity, but superdiffusive at high levels of activity. They compare their experimental results with two types of models: the solution of a simple 1D diffusion equation, and a more complex PDE model for an active nematic coupled to fluid flow.  
+
+Overall, the experiments are impressive and well executed, and the paper is well written. The main result, namely the transition from diffusive to superdiffusive spreading with increasing of activity, is interesting, although perhaps not entirely surprising. However, the discussion and analysis as well as the comparison with models are sometimes a bit weak, and it is not entirely clear what fundamental new physics is learned as a result of this study. In particular, the transition to superdiffusive transport is not really understood (and in fact it is not captured by the models). As a result, I cannot recommend publication of the manuscript in its present form in Nature Communications.  
+
+Here are some more detailed comments and suggestions that the authors should addressing to improve the quality of the paper:  
+
+1. It's not completely clear how to interpret the velocity of equation (4): ATP should directly affect the velocity of kinesin motors, but how it translates into the velocity the interface is not trivial: indeed, the motors produce local microtubule motions, which then interact through the fluid the causing interfacial spreading. Furthermore, the fact that the interface in the 1D model spreads diffusively is not very surprising, since the model is based on the diffusion equation and the velocity of equation (4) is directly tied to the ATP concentration (in fact, when [ATP]<<k in the early stages of mixing, the velocity is directly proportional to [ATP]). As a result, I'm not sure exactly what is learned from the model.  
+
+2. It is likely that the transition to superdiffusion is tied to the emergence of strong large-scale motions in the fluid, which result in advective (ballistic) transport on top of the diffusive spreading. This is supported by the observation on page 4 that in order to increase the velocities in the fluid the authors increasing the thickness of the sample. This should have a direct effect on the correlation length scales of the turbulent motions in the active fluid (this point is very briefly addressed in the discussion section). It would be very interesting to characterize these length scales and determine how they affect the mixing process.  
+
+3. The fact the checkerboard patterns result in faster mixing than a single interface is again not very surprising, since gradients are introduced on short length scales.  
+
+4. The discussion of rheology in the discussion section is quite interesting, in particular the fact that the microtubule network really behaves as crosslinked gel until the kinesin motors are activated. However, it is not completely obvious why this would contribute to superdiffusive transport as hypothesized by the authors: in fact, I would expect transport to
+
+<--- Page Split --->
+
+be weaker due to crosslinking.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The manuscript reports the results of a combined experimental and theoretical/numerical study of fluid mixing by active fluids. Ultraviolet light- activated caged ATP and fluorescent dyes are used to activate regions of a microtubule- kinesin active fluid. Both a system with an interface between an activated and a passive microtubule- kinesin mixture, and a checkerboard activation pattern are investigated. As activity is turned on, the interfaces broaden and move into the initially inactivated regions. t low activation levels, mixing is found to be governed by an active diffusion- like processes at the active- inactive interface, while at higher activity levels, superdiffusion- like processes dominate. Samples activated in a checkerboard pattern reach homogeneity faster than those with a single dividing interface. A model of active nematohydrodynamics coupled to ATP transport is employed to describe the coupled mixing process numerically. The results are in qualitative agreement, but differ on the quantitative level significantly.  
+
+Mixing of fluids on the microscale is difficult to achieve, because Reynolds number is low and mixing is  
+
+dominated by molecular diffusion. Active fluids have the potential to generate a pronounced speed- up of the mixing process. The current study focuses on systems with spatiotemporally varying activities and the role of interfaces in the mixing process. This is a very interesting aspect, which has not received much attention so far.  
+
+The following questions and comments should be addressed:  
+
+(1) Why would a diffusive MSD of the interface position, which would naively be expected for an underlying diffusive process, give any hint about a super-diffusive process? 
+(2) How does the flow speed, which is determined by Michaelis-Menten kinetics, enter into model given by eqs. (1-3)? 
+(3) What changes at the molecular scale at the velocity threshold of 5 micron/sec? 
+(4) Why does mixing time increase linearly with grid size (Fig. 7b), and extrapolate to zero mixing time at a grid size of about \(0.9 \text{mm}\) ? 
+(5) The theoretical model predicts a scaling of interface progression coefficient, which is consistent with the results from experiments. However, the coefficient magnitudes differed by a factor of 5 between the model and the experiments.
+
+<--- Page Split --->
+
+Furthermore, the simulation does not show the transition from an active diffusion- like to a superdiffusion- like processes observed experimentally.  
+
+The authors propose as a possible explanation that the rheology of microtubule network is not taken  
+
+into account in the numerical model. Is this a plausible explanation, in particular for the factor 5?  
+
+(6) The authors emphasize that the mixing efficacy of the nonuniform active fluid systems depends on the distribution of activity, with systems consisting of more small active areas  
+
+evolve to a homogeneous state faster than systems with the same total active area distributed as  
+
+one piece. They conclude that the activity-uniform active fluid has the highest mixing performance.  
+
+However, this is exactly the behavior I would expect for a passive systems: As interfaces propagate diffusively, shorter distances can be covered more quickly. Thus, what is the significance  
+
+of the active process in this general conclusion?  
+
+(7) Arguments about maximizing system entropy are difficult because this a non-equilibrium active system.  
+
+(8) The following references might also be relevant:  
+
+-- A. Doostmohammadi et al., Nat. Commun. 8, 15326 (2017);  
+
+-- G.A. Vliegenthart et al., Sci. Adv. 6, eaaw9975 (2020);  
+
+-- H. Reinken et al., Commun. Phys. 3, 76 (2020);  
+
+-- R. Alert et al., Nature Physics 16, 682 (2020);  
+
+-- K. Qi et al., Commun. Phys. 5, 49 (2022).  
+
+## Reviewer #3 (Remarks to the Author):  
+
+The authors have performed mixing experiments on microtubule- kinesin based active nematics by taking advantage of an interesting molecule called "caged ATP". This ATP can be introduced uniformly into the system in its inactive state and then activated remotely via a UV light to give spatial patterning of activity in the active nematic system.  
+
+They observed diffusion- like and super- diffusive regimes dependent on the speed of the microtubule flows and compare results with some simulations of the system with a variety of quantitative experiments. Beautiful videos demonstrate the concepts very well. The data are fit to a fairly simple transport model that seems to describe the behaviors well. An additional and interesting experiment towards the end of the paper tests checker- board patterns of activation to look at how fast the system reaches homogeneity as a function of active/inactive interfacial length. This section of the paper could probably be a separate publication if explored in more detail.  
+
+Overall I think that the paper covers an exciting topic and one that should be focused on by the active nematic community. Self- mixing in active nematics represents an important direction in the evolution of this field. I like the hypothesis driven approach and the data is presented in a logical and interesting fashion. A large amount of data is shown with a focus largely on statistical measures such as 1D 'diffusion' constants (i.e. P in this case).  
+
+I found a few issues in the paper that should be corrected pre- publication but I think that the paper is appropriate for Nature Communications in scope and impact.
+
+<--- Page Split --->
+
+1. The title of the paper can be improved for clarity - is it missing a comma or colon? I would use "activity" not "activities"  
+
+2. I found that the description of the results at the start of that section to be a bit lacking in necessary detail. The authors need to be clearer about the nature of the "initially inactivated fluid" in the first results paragraph. Is it mixed Mts and kinesin but not formed into bundles (random filaments, or small bundles not yet aligned)? Presumably the solution has not yet been in the active nematic state? Are the kinesin clusters bound to MTs yet but the Mts are isotropic? It appears from the video that this is the case and I can guess, but clarification must be added. It seems that initiation of activity is not a reversible process - I.e. once bundled the system will never go back to the initial state (e.g. when ATP runs out). This can also be clarified.  
+
+3. In the first results paragraph we also need to know where the dyes are located to interpret the videos and Figures easily - is it on the MT? The kinesin or perhaps in the water. This needs to be added in the first results paragraph. How about the tracer particles? Are they coupled to the MTs? How are they confined to the active layer? I realize that there is plenty of detail in the methods section at the end of the paper but the most important points for figure interpretation should be in the main text otherwise it's too difficult to read the paper - especially for the uninitiated general reader.  
+
+4. The authors don't address the concept of chaotic mixing and advective flows at all in the paper and I found that to be an omission - how does that concept relate to the measures for quantifying mixing in the submitted work? A short discussion might help put this work in context with that recent work cited as ref 24 where chaotic mixing was described for the same system. Some other groups are also considering advective flows in these systems. How would this be related to the present work? I also think that the idea that mixing is driven primarily by defects should be addressed. Can this idea be related to the present work? This was first introduced in theoretical works by Marchetti and shown experimentally in ref 24. Since the submitted paper is not the first to talk about self-mixing in active nematics, a discussion should be added to see how these different papers fit together and can be connected to the submitted work.  
+
+5. The results presented in the paper focus on large length-scales (i.e. much larger that the active length scale. Can the authors discuss their results where the length scales approach the active length scale or even go below it? Transport measures may be different on small scales and should be ballistic-like. Can local flows o the scale of defects be tracked to get more detail on these scales?  
+
+6. It appears that Fig 3d might fit to Michaelis-Menten kinetics, tt looks like the trend is approximately there. Is that trend expected for this system? Comment on the shape of the curve and prior work.  
+
+7. In all the Figures the captions need to make it much clearer which panels are calculations and which are experimental data. The reader should not have search around for this information.  
+
+Line 241 - You can't use "turbulences" - this should phrased better - do you mean vortices? Areas of "active turbulence". Please clarify.
+
+<--- Page Split --->
+
+Fig S3 - the plural of spectrum is spectra.
+
+<--- Page Split --->
+
+## Reviewer 1:  
+
+In this paper, the authors perform experiments on mixing in an active fluid composed of microtubules and kinesin clusters powered by ATP. Using light activated ATP, they are able to turn on activity over a finite region of space, thus creating an interface between an active fluid and a passive fluid. As the light- activated ATP disperses across the interface due to both diffusion and active mixing, the region where active flows take place expands spatially, thereby further enhancing transport and mixing. The authors characterize the rate at which the ATP gets mixed between the two regions, and find that the mixing is diffusive at low levels of activity, but superdiffusive at high levels of activity. They compare their experimental results with two types of models: the solution of a simple 1D diffusion equation, and a more complex PDE model for an active nematic coupled to fluid flow.  
+
+Overall, the experiments are impressive and well executed, and the paper is well written. The main result, namely the transition from diffusive to superdiffusive spreading with increasing of activity, is interesting, although perhaps not entirely surprising. However, the discussion and analysis as well as the comparison with models are sometimes a bit weak, and it is not entirely clear what fundamental new physics is learned as a result of this study. In particular, the transition to superdiffusive transport is not really understood (and in fact it is not captured by the models). As a result, I cannot recommend publication of the manuscript in its present form in Nature Communications.  
+
+We thank the reviewer for their assessment. The essence of our work was to investigate the mixing dynamics of active fluid systems with a dynamic activity gradient. Recently, there has been a trend within active fluid research to study systems with activity gradients. For example, Zhang et al. Nature Materials 20, 875 (2021) and Ross et al. Nature 572, 224 (2019) investigated the role of activity gradients in the self- organization of active fluid and Shankar and Marchetti PRX 9, 041047 (2019) predicted the defect dynamics in an activity gradient. However, most existing work about activity gradients is focused on systems with steady state, time- independent activity gradients. Our work is novel in that it explores an activity gradient that changes spontaneously with time. Also, this paper hints that melting dynamics of the microtubule network play a role in the behavior of a dynamic activity gradient, and that such dynamics have been overlooked in existing active fluid models. As such, we believe our work can promote development of active fluid modeling to include the network melting mechanism and thus more accurately describe complex active fluid systems with dynamic distributions of activity.  
+
+Here are some more detailed comments and suggestions that the authors should addressing to improve the quality of the paper:  
+
+1. It's not completely clear how to interpret the velocity of equation (4): ATP should directly affect the velocity of kinesin motors, but how it translates into the velocity the interface is not trivial: indeed, the motors produce local microtubule motions, which then interact through the fluid the causing interfacial spreading.  
+
+We thank the reviewer for the comment. The velocity in Eq. 4 represents mean speed of active fluid flow driven by extensile microtubules whose motions are driven by kinesin motor proteins consuming local ATP and stepping along microtubules (Fig. 1a). The progression of the interface is a complex phenomenon that involves the coupling of ATP dispersion and active fluid flows. We agree with the reviewer that how dispersion of ATP relates to progression of the interface is not trivial. To further explore the issue, we derived an analytical expression for interface progression coefficient, \(P_{I}\) , (Eq. S7 in Supplementary Discussion 3) that showed that \(P_{I}\) depended on the initial ATP concentration \(C_{0}\) via an inverse complementary error function. Then we explored how \(P_{I}\) varied with \(C_{0}\) numerically, which revealed that
+
+<--- Page Split --->
+
+a higher initial ATP concentration led to a faster- progressing interface (Fig. 3d). Surprisingly, we further found that the interface progression coefficients \(P_{i}\) are well fit to the Michaelis- Menten equation (the second curve from the top in Supplementary Fig. 4a). This result demonstrated that how the speed of active fluid depended on ATP concentration directly led to how the interface progression depended on ATP concentration. For readers who are interested in the connection between these two ATP dependences, we added two supplementary discussions (Supplementary Discussions 2 & 3) to explore the connection numerically as well as analytically.  
+
+Furthermore, the fact that the interface in the 1D model spreads diffusively is not very surprising, since the model is based on the diffusion equation and the velocity of equation (4) is directly tied to the ATP concentration (in fact, when [ATP] \(\leq \leq \mathrm{k}\) in the early stages of mixing, the velocity is directly proportional to [ATP]). As a result, I'm not sure exactly what is learned from the model.  
+
+We thank the reviewer for asking about the significance of our 1D Fick's law- based active fluid model. The purpose of introducing this model was to demonstrate that the mixing of active and inactive fluids we observed (Fig. 2) could be described with a minimal model considering only diffusion of ATP and Michaelis- Menten kinetics (Fig. 3), without the need for a more complex hydrodynamic model such as the one we developed later in the manuscript (Fig. 6). We found that the simple 1D Fick's law- based active fluid model did agree with experimental observations (Fig. 3d). This consistency between the Fick's law- based model and our experimental results suggests that Michaelis- Menten kinetics successfully approximate the conversion of ATP concentration into local active fluid speed. For readers who are interested in learning about why we introduced the Fick's law- based model, we revised the manuscript to clarify the motivation of developing the model and its associated intellectual merits in the main text (lines 100- 104 and 129- 133) as well as in Supplementary Discussion 2.  
+
+In response to the reviewer's comment that the connection between diffusion- like interface progression \((\gamma = 1)\) and diffusive ATP \((a = 1)\) is an expected, trivial consequence of adopting the Michaelis- Menten equation (Eq. 4), we performed additional studies (Supplementary Discussion 2) that showed that the connection was not necessarily built upon using the Michaelis- Menten equation, because even when the relation between ATP concentration and flow speed became square root, quadratic, cubic, or higher power- law dependent, the resulting interface progression exponent remained \(\gamma = 1\) (Supplementary Fig. 3d inset). These additional studies show that connection between \(\gamma = 1\) and \(a = 1\) was not sensitive to which flow speed- ATP relation we chose (although the interface progression coefficient \(P_{i}\) was highly sensitive to the relation [Supplementary Fig. 3d]). For readers who are interested in learning about the role of flow speed- ATP relation in our model, we added Supplementary Discussion 2 to demonstrate how the modeling results would be influenced with 10 other relations. We thank the reviewer for questioning the intellectual merit of our Fick's law- based model which drove us to enrich the exploration of our model and strengthen the manuscript.  
+
+2. It is likely that the transition to superdiffusion is tied to the emergence of strong large-scale motions in the fluid, which result in advective (ballistic) transport on top of the diffusive spreading. This is supported by the observation on page 4 that in order to increase the velocities in the fluid the authors increasing the thickness of the sample. This should have a direct effect on the correlation length scales of the turbulent motions in the active fluid (this point is very briefly addressed in the discussion section). It would be very interesting to characterize these length scales and determine how they affect the mixing process.  
+
+We greatly thank the reviewer for the insightful comment. The reviewer was correct that increasing the sample container height would affect the correlation length scales of the turbulent motion. We added a supplementary discussion (Supplementary Discussion 4) in which we quantify how the correlation length
+
+<--- Page Split --->
+
+increased with sample container height (Supplementary Fig. 5d). To explore how the increased correlation length would affect the mixing of active- inactive fluid systems, we adopted a dimensionless quantity, the Péclet number \((P e)\) , defined as \(P e\equiv \bar{v}_{a b}l_{c} / D\) , where \(\bar{v}_{a b}\) represents the mean flow speed of active fluid, \(D\) is the diffusion coefficient of ATP, and \(l_{c}\) is the correlation length of flow velocity. The physical interpretation of this quantity is the ratio of convective transport rate to diffusive transport rate. When the Péclet number is greater than of order 1, the active transport is dominated by convection, whereas when the Péclet number is smaller than of order 1, the active transport is dominated by diffusion. To explore how the behaviors of active transport affect mixing in an active- inactive fluid system, we first analyzed the interface progression exponent \(\gamma\) as a function of Péclet number (Fig. 4b) and found that as \(P e\lesssim 3\) , \(\gamma \approx 1\) which corresponded to the diffusion- like mixing captured in our Fick's law- based model (Fig. 3). Interestingly, as the Péclet number is increased to \(P e\gtrsim 3\) , we observed \(\gamma\) became greater than 1, which corresponded to the transition in interface progression from diffusion- like to superdiffusion- like. As such, the reviewer was correct; the transition to superdiffusion was tied to the gradual emerging convection mechanism that governed the mixing process of active- inactive fluid system. We have revised the manuscript to clarify the underlying mechanism causing the transition (lines 146- 161 with accompanying Fig. 4b). We greatly appreciate the reviewer's constructive feedback that significantly improved the manuscript.  
+
+3. The fact the checkerboard patterns result in faster mixing than a single interface is again not very surprising, since gradients are introduced on short length scales.  
+
+We agree with the reviewer that the results of the checkerboard pattern experiments were expected and thus the intellectual contribution was minimal. The reason we introduced these experiments was to explore how the distribution of activity would affect the mixing process of the active- inactive fluid system, as previously we had only explored one configuration of activity distribution (one side active and one side inactive).  
+
+To enrich this part of the work, we further explored the checkerboard system in our established simulation platform (Fig. 8), revealing two interesting results: (1) At a sufficiently small grid size, the simulated mixing times of active and inactive systems were indistinguishable. We believe that this is because the active fluid needs time to "warm up". In experiments, the system had a warm- up time caused by network melting (Supplementary Discussion 6). Although a network melting mechanism was not included in the model, the simulated active fluid flow still took dimensionless time to rise because the onset of the flows was triggered by the initial activity- driven instability in extensile \(Q\) field which took finite dimensionless time to develop \((\sim 1\) dimensionless time in this case; Supplementary Video 6), and during the warm- up time, molecular diffusion is the main driving force of mixing. When the grid size was very small, the mixing could be completed by diffusion before the fluid activity could speed up the mixing. Thus, the simulation based on the checkerboard pattern suggested a limitation of active fluid mixing: The mixing effect of active fluid can only take place when the mixture is sufficiently nonuniform. (2) The checkerboard- pattern simulation also revealed that the mixing process of active fluid is less susceptible to changes in grid size than in inactive fluid systems (Fig. 8b). Altering the dimensionless grid size from 2 to 22 changed the mixing time of the inactive fluid system by a factor of 40 but only changed the mixing time of the active fluid system by a factor of 3. This result suggests that an active fluid system is less susceptible to the initial distribution of the mixture than an inactive fluid system.  
+
+These two new results (added to Results lines 261- 268, Discussion lines 307- 319, Methods lines 516- 529) demonstrate limits and advantages of introducing active fluid to microfluidic mixing systems with nonuniform activity. We thank the reviewer for the comment that drove us to enrich the manuscript. We believe that the checkerboard work is now more relevant and could inspire further exploration of the active mixing process with other complex distributions of activity.
+
+<--- Page Split --->
+
+4. The discussion of rheology in the discussion section is quite interesting, in particular the fact that the microtubule network really behaves as crosslinked gel until the kinesin motors are activated. However, it is not completely obvious why this would contribute to superdiffusive transport as hypothesized by the authors: in fact, I would expect transport to be weaker due to crosslinking.  
+
+We thank the reviewer for asking this question. After following the reviewer's second comment, we found that the transition to superdiffusion- like interface progression ( \(\gamma >1\) ) was tied to the underlying transition of active transport from being diffusion- like ( \(Pe \leq 3\) ) to superdiffusion- like ( \(Pe \geq 3\) ), and thus the network rheology might not have played a significant role in this transition. Thus, we removed that hypothesis from the manuscript. Nevertheless, we explored our data more deeply and found that the network rheology could indeed slow the progression of active- inactive interface (Supplementary Fig. 7e), as the reviewer expected, and this slower progression was not captured in our active- fluid hydrodynamic model because the model did not include the network melting mechanism (Supplementary Fig. 8c). These results demonstrate that the network melting mechanism plays a role in the mixing process of activity- nonuniform active fluid systems, and this mechanism has thus far been overlooked in the modeling of microtubule- kinesin active fluid systems. In the revised manuscript, we addressed the significance of the network rheology in mixing dynamics (lines 284- 297 and Supplementary Discussion 6) and suggested potential future experiments and simulations to further investigate the network melting dynamics (lines 297- 302). We believe that our current manuscript will encourage the active fluid community to investigate this topic further and inspire further exploration in self- organization of activity- nonuniform active fluid systems.
+
+<--- Page Split --->
+
+## Reviewer 2:  
+
+The manuscript reports the results of a combined experimental and theoretical/numerical study of fluid mixing by active fluids. Ultraviolet light- activated caged ATP and fluorescent dyes are used to activate regions of a microtubule- kinesin active fluid. Both a system with an interface between an activated and a passive microtubule- kinesin mixture, and a checkerboard activation pattern are investigated. As activity is turned on, the interfaces broaden and move into the initially inactivated regions. t low activation levels, mixing is found to be governed by an active diffusion- like processes at the active- inactive interface, while at higher activity levels, superdiffusion- like processes dominate. Samples activated in a checkerboard pattern reach homogeneity faster than those with a single dividing interface. A model of active nematohydrodynamics coupled to ATP transport is employed to describe the coupled mixing process numerically. The results are in qualitative agreement, but differ on the quantitative level significantly.  
+
+Mixing of fluids on the microscale is difficult to achieve, because Reynolds number is low and mixing is dominated by molecular diffusion. Active fluids have the potential to generate a pronounced speed- up of the mixing process. The current study focuses on systems with spatiotemporally varying activities and the role of interfaces in the mixing process. This is a very interesting aspect, which has not received much attention so far.  
+
+We thank the reviewer for their comments on the intellectual merit of our work. Please find our point- by- point responses to the reviewer's feedback below.  
+
+The following questions and comments should be addressed:  
+
+(1) Why would a diffusive MSD of the interface position, which would naively be expected for an underlying diffusive process, give any hint about a super-diffusive process?  
+
+We thank the reviewer for their comment. We have revised the paper to clarify this point. In the previous version of the paper, we used a proof by contradiction: proving a point by demonstrating that its opposite is a contradiction. We sought to prove that \(\gamma = 1\) is the consequence of ATP diffusion \((a = 1)\) by first stating that \(\gamma = 1\) because \(a > 1\) (superdiffusion). Then we used the Fick's law model to show that this statement was incorrect; thus, the invalidity of the statement indicated that \(\gamma = 1\) from \(a = 1\) (or \(a \leq 1\) , more strictly).  
+
+We agree with the reviewer that this proof was too complicated and led to unnecessary confusion, and we removed the proof from the manuscript. As an alternative, we revised the Fick's law model section in the manuscript to directly show that modeling ATP dispersion with the diffusion equation leads to diffusion- like interface progression (lines 100- 133). To provide deeper insight into the connection between ATP dispersion and interface progression, we presented an algebraic proof that squared interface displacement is proportional to time: \(\Delta x^{2} \propto t\) (Eq. S6) and provided an analytical perspective on how diffusive ATP transport led to diffusion- like interface progression (Supplementary Discussion 3). We believe that the manuscript is now clearer and more understandable.  
+
+(2) How does the flow speed, which is determined by Michaelis-Menten kinetics, enter into model given by eqs. (1-3)?  
+
+We thank the reviewer for asking. The Michaelis- Menten kinetics equation is separate from Eqs. 1- 3. The reason we modeled the ATP- dependent flow speed with Michaelis- Menten kinetics is because our previous studies of the flow speed of active fluid for various ATP concentrations (Figs. 3A- C in Bate et al. Soft Matter 15, 5006 [2019]) showed that the flow speed of active fluid followed Michaelis- Menten kinetics for ATP concentrations greater than \(100 \mu \mathrm{M}\) . Additionally, we believe this choice of model is reasonable
+
+<--- Page Split --->
+
+because the active fluid is driven by the motion of microtubules, which are driven by kinesin motors whose stepping rates are known to follow the Michaelis- Menten kinetics (Schnitzer et al. Nature 388, 386 [1997]), and thus these kinetics could be scaled up to active fluid scale.  
+
+To clarify how our choice of flow speed- ATP relation affected the modeling results in terms of predicted interface progression exponents \((\gamma)\) and coefficients \((P_{l})\) , in the revised manuscript we explored ten other plausible relations to connect ATP and flow speed and found that the Michaelis- Menten equation led to the best match between the predictions and the experimental results (Supplementary Discussion 2, Supplementary Fig. 3d). We are now more confident that the Michaelis- Menten equation is an appropriate coarse- grained approach to model local flow speed in terms of ATP concentrations. We thank the reviewer for pointing out the weakness in our manuscript which drove us to make the manuscript more robust.  
+
+(3) What changes at the molecular scale at the velocity threshold of 5 micron/sec?  
+
+We thank the reviewer for asking this important question. In the previous version of manuscript, we stated that the active- inactive interface progression occurred at an active fluid flow speed of \(5 \mu \mathrm{m / s}\) . In the revised version of the manuscript, we adopted a new perspective on this issue: we found that the transition is in fact related to Péclet number rather than active fluid speed. Péclet number is defined as \(Pe \equiv \bar{v}_{ab} l_{c} / D\) , where \(\bar{v}_{ab}\) represents the mean flow speed of active fluid, \(D\) is the diffusion coefficient of ATP, and \(l_{c}\) is the correlation length of flow velocity. The physical interpretation of this quantity is the ratio of convective transport rate to diffusive transport rate. When the Péclet number is smaller than of order 1, the active transport is dominated by diffusion, whereas when the Péclet number is greater than of order 1, the active transport is dominated by convection. To explore how the behaviors of active transport affect mixing in an active- inactive fluid system, we first analyzed the interface progression exponent \(\gamma\) as a function of Péclet number (Fig. 4b) and found that as \(Pe \leq 3\) , \(\gamma \approx 1\) , which corresponds to the diffusion- like mixing captured in our Fick's law- based model (Fig. 3). Interestingly, as the Péclet number increased to \(Pe \geq 3\) , \(\gamma\) became greater than 1, which corresponded to the transition in interface progression from being diffusion- like to being superdiffusion- like. Our further analysis revealed that the transition from diffusion- like to superdiffusion- like progression of the active- inactive interface was not directly relevant to the \(5 \mu \mathrm{m / s}\) flow speed. Instead, it was due to the transition of active transport from diffusion- dominated to superdiffusion- dominated, as characterized by a dimensionless Péclet number increasing from of order 1 to of order greater than 1. We revised the manuscript to clarify the mechanism underlying the diffusion- superdiffusion transition of the active- inactive interface (lines 146- 161, Fig. 4b). We thank the reviewer for their question, which drove us to dive more deeply into our data, pursue a fuller understanding of interface progression, and make the manuscript more robust.  
+
+(4) Why does mixing time increase linearly with grid size (Fig. 7b), and extrapolate to zero mixing time at a grid size of about \(0.9 \mathrm{mm}\) ?  
+
+We thank the reviewer for the observations on our experimental checkerboard data (Fig. 7b). However, we think it was not clear that the mixing time varied linearly with grid size. We believe that as the grid size approached zero, the mixing time would vary more slowly and approach zero (see the figure below). To learn more about grid size- dependence of mixing time, we performed a corresponding checkerboard active fluid simulation (Fig. 8) and found that as the grid sizes decreased, the mixing time in active fluid system became indistinguishable with the mixing time in inactive fluid system where the mixing was driven only by molecular diffusion (Fig. 8b). We believe that this was because our active fluid system needed time to "warm up". In experiments, the system had a warm- up time caused by network melting (Supplementary Discussion 6). Although a network melting mechanism was not included in the model, the simulated active fluid flow still took dimensionless time to rise because the onset of the flows was triggered by the initial
+
+<--- Page Split --->
+
+activity- driven instability in extensile \(Q\) field which took finite dimensionless time to develop ( \(\sim 1\) dimensionless time in this case; Supplementary Video 6), and during the warm- up time, molecular diffusion is the main driving force of mixing. Thus, the simulation showed that as the grid sizes approached zero, the mixing dynamics would also change from active flow- dominated to molecular diffusion- dominated, so the mixing time was not expected to vary linearly with all grid sizes. We thank the reviewer for the comment on our checkerboard data which motivated us to explore the checkerboard system more deeply. For readers who are interested in learning about checkerboard mixing dynamics in the limit of small grid sizes, we added our simulation checkerboard work in Results (lines 261- 268), Discussion (lines 307- 319), and Methods (lines 516- 529) along with accompanying figure (Fig. 8). We believe that our checkerboard work is now richer and can provide readers with deeper understanding of the mixing dynamics of nonuniform active fluid systems.  
+
+![PLACEHOLDER_12_0]
+  
+
+(5) The theoretical model predicts a scaling of interface progression coefficient, which is consistent with the results from experiments. However, the coefficient magnitudes differed by a factor of 5 between the model and the experiments. Furthermore, the simulation does not show the transition from an active diffusion-like to a superdiffusion-like processes observed experimentally. The authors propose as a possible explanation that the rheology of microtubule network is not taken into account in the numerical model. Is this a plausible explanation, in particular for the factor 5?  
+
+We thank the reviewer for raising the concern about the five- factor discrepancy of interface progression coefficient \(P_{I}\) between the model and experiment. In response to this comment, we examined our data more closely and concluded that the discrepancy mainly resulted from using an incorrect ATP diffusion coefficient. In the previous version of this manuscript, the model used the diffusion coefficient of ATP in water, because the base of the microtubule- kinesin active fluid is water. However, Gagnon et al. showed that the microtubule network has a viscosity greater than water (PRL 125, 178003, [2020]), which implies that the actual diffusion coefficient of ATP should be lower than its value in water. To better estimate the diffusion coefficient of ATP in our microtubule network, we performed extensive additional studies to explore the diffusion coefficient in inactive microtubule- kinesin fluid. We were not able to track ATP directly, so as an alternative we tracked fluorescein, which we could visualize with fluorescent microscopy. We found that the diffusion coefficient of fluorescein in the inactive microtubule network was one- fifth the value reported for aqueous solution (Supplementary Discussion 1), which suggested that the diffusion coefficient of ATP in our active fluid is also 5 times lower than the value reported in pure water. After making this correction, we found that our model was consistent with experimental observation (with \(\sim 10\%\) relative difference; Fig. 3d). We revised the Fick’s law model section in the manuscript (lines 100- 133) and added a description of how we estimated the diffusion coefficient of ATP in our active fluid system
+
+<--- Page Split --->
+
+(Supplementary Discussion 1). We thank the reviewer for questioning about the weakness in our model which drove us to strengthen our model and make our finding more robust.  
+
+(6) The authors emphasize that the mixing efficacy of the nonuniform active fluid systems depends on the distribution of activity, with systems consisting of more small active areas evolve to a homogeneous state faster than systems with the same total active area distributed as one piece. They conclude that the activity-uniform active fluid has the highest mixing performance. However, this is exactly the behavior I would expect for a passive systems: As interfaces propagate diffusively, shorter distances can be covered more quickly. Thus, what is the significance of the active process in this general conclusion?  
+
+We thank the reviewer for questioning the significance of fluid activity in the checkerboard work. We performed the checkerboard experiments because throughout our investigation on active- inactive fluid system, we only focused on one configuration of activity distribution: One side activated and the other side inactive. Thus, we were curious about how the mixing process would have been different if the activity had been distributed in a different arrangement. In response to the reviewer's comment, we further explored the checkerboard mixing system via our active fluid simulation (Fig. 8) and compared the mixing of the active and inactive checkerboard- pattern systems, to delve more deeply into this topic and provide more relevant findings in the manuscript. Surprisingly, our simulation showed that the mixing time of the active fluid system depended less on the grid size than the mixing time of an inactive fluid system (Fig. 8b). Increasing the dimensionless grid size from 2 to 22 increased the mixing time of inactive fluid system by a factor of 40, whereas in the active fluid system, the mixing time was only increased by a factor of 3. Such drastic difference showed that the mixing efficacy of active fluid is less sensitive to changes in the initial condition of the mixture. We added these new checkerboard simulation results to Results (lines 261- 268), Discussion (lines 307- 319), and Methods (lines 516- 529) along with accompanied figure (Fig. 8). We thank the reviewer for questioning about the intellectual merits of our checkerboard work which drove us to delve into the checkerboard mixing systems and enriched the checkerboard work in our manuscript.  
+
+(7) Arguments about maximizing system entropy are difficult because this a non-equilibrium active system.  
+
+We thank the reviewer for the correction. We removed arguments about entropy.  
+
+(8) The following references might also be relevant:  
+
+-- A. Doostmohammadi et al., Nat. Commun. 8, 15326 (2017);  
+
+-- G.A. Vliegenthart et al., Sci. Adv. 6, eaaw9975 (2020);  
+
+-- H. Reinken et al., Commun. Phys. 3, 76 (2020);  
+
+-- R. Alert et al., Nature Physics 16, 682 (2020);  
+
+-- K. Qi et al., Commun. Phys. 5, 49 (2022).  
+
+We thank the reviewer for bringing our attention to these important works. We have cited these works in relevant places in our manuscript.
+
+<--- Page Split --->
+
+## Reviewer 3:  
+
+The authors have performed mixing experiments on microtubule- kinesin based active nematics by taking advantage of an interesting molecule called "caged ATP". This ATP can be introduced uniformly into the system in its inactive state and then activated remotely via a UV light to give spatial patterning of activity in the active nematic system. They observed diffusion- like and super- diffusive regimes dependent on the speed of the microtubule flows and compare results with some simulations of the system with a variety of quantitative experiments. Beautiful videos demonstrate the concepts very well. The data are fit to a fairly simple transport model that seems to describe the behaviors well. An additional and interesting experiment towards the end of the paper tests checker- board patterns of activation to look at how fast the system reaches homogeneity as a function of active/inactive interfacial length. This section of the paper could probably be a separate publication if explored in more detail.  
+
+Overall I think that the paper covers an exciting topic and one that should be focused on by the active nematic community. Self- mixing in active nematics represents an important direction in the evolution of this field. I like the hypothesis driven approach and the data is presented in a logical and interesting fashion. A large amount of data is shown with a focus largely on statistical measures such as 1D 'diffusion' constants (i.e. P in this case).  
+
+I found a few issues in the paper that should be corrected pre- publication but I think that the paper is appropriate for Nature Communications in scope and impact.  
+
+We thank the reviewer for the support.  
+
+1. The title of the paper can be improved for clarity - is it missing a comma or colon? I would use "activity" not "activities"  
+
+We thank for the reviewer's correction. We have changed the title to "Self-mixing in microtubule- kinesin active fluid from nonuniform to uniform distribution of activity."  
+
+2. I found that the description of the results at the start of that section to be a bit lacking in necessary detail. The authors need to be clearer about the nature of the "initially inactivated fluid" in the first results paragraph. Is it mixed Mts and kinesin but not formed into bundles (random filaments, or small bundles not yet aligned)? Presumably the solution has not yet been in the active nematic state? Are the kinesin clusters bound to MTs yet but the Mts are isotropic? It appears from the video that this is the case and I can guess, but clarification must be added. It seems that initiation of activity is not a reversible process - I.e. once bundled the system will never go back to the initial state (e.g. when ATP runs out). This can also be clarified.  
+
+We thank the reviewer for highlighting the need to clarify the initial, inactive state of our active fluid system. Our inactive fluid contained microtubules that spontaneously formed bundles by depletion; these bundles were further crosslinked by kinesin motor dimers, forming an elastic gel network. We prepared the inactive gel sample in a test tube where microtubules were expected to orient isotropically, but when we loaded the mixture into the flow cell, shear flow was induced and drove the microtubules to align along the flow cell, so microtubules initially had a preferred alignment along the long edge of the flow cell (Supplementary Fig. 1). After the fluid was activated by ultraviolet light, the microtubule network underwent an irreversible process of becoming a 3D self- rearranging isotropic active gel consisting of extensile microtubule bundles that buckled and annealed repeatedly until the ATP ran out. We have clarified the initial state of microtubule network before UV activation along with various details about the network dynamics such as bundle formation and the irreversibility of the network structure in the main text (lines 55- 77) and added a supplementary figure to show the microtubule alignment right after loading (Supplementary Fig. 1). We
+
+<--- Page Split --->
+
+noticed that Najma et al. recently posted an arXiv reporting the similar initial network structure (Fig. 1c in Najma et al. arXiv: 2112.11364 [2022])) so we also cited the arXiv for readers who wanted to know more details about the initial state of microtubule- kinesin active fluid. We believe that now the readers have access to sufficient information about the initial state of our inactive fluid. We thank the reviewer for the request that increased the clarity of our manuscript.  
+
+3. In the first results paragraph we also need to know where the dyes are located to interpret the videos and Figures easily - is it on the MT? The kinesin or perhaps in the water. This needs to be added in the first results paragraph. How about the tracer particles? Are they coupled to the MTs? How are they confined to the active layer? I realize that there is plenty of detail in the methods section at the end of the paper but the most important points for figure interpretation should be in the main text otherwise it's too difficult to read the paper - especially for the uninitiated general reader.  
+
+We thank the reviewer for addressing the issue of important methods details in the main text. We added explicit description in the main text that the microtubules were labeled with Alexa 647 (Line 69- 70) and that tracers were freely suspended in solvent (lines 67- 69). Also, we clarified that our system was a 3D microtubule- based active fluid system consisting of self- rearranging isotropic active gel, so there were no active layers like there would be in a 2D active nematic system (lines 64- 67). We thank the reviewer for the suggestion, which increased the readability of the manuscript.  
+
+4. The authors don't address the concept of chaotic mixing and advective flows at all in the paper and I found that to be an omission - how does that concept relate to the measures for quantifying mixing in the submitted work? A short discussion might help put this work in context with that recent work cited as ref 24 where chaotic mixing was described for the same system. Some other groups are also considering advective flows in these systems. How would this be related to the present work?  
+
+We thank the reviewer for the suggestion on discussing the chaotic mixing and advective flows in our manuscript. We realize that our method of quantifying mixing by using the interface progression exponent, \(\gamma\) , and mixing time, \(t_0\) , is not a complete characterization of the mixing dynamics of the active fluid system. Tan et al. characterized the chaotic mixing by introducing topological entropy and Lyapunov exponents. We considered characterizing the mixing dynamics in our system the same way, but soon we realized that it was not practical because our system is different from the one used by Tan et al. Their system was a 2D active nematic system in which the embedded tracers could remain in one focal plane and thus could be tracked for a long period of time, whereas our experimental system was a 3D isotropic active gel where tracers frequently moved out of the focal plane, which prevented us from continuously tracking them. If the tracers could be imaged and tracked in 3D, it would be possible to measure Lyapunov exponents and topological entropies. We are working on developing 3D imaging and tracking; but we have not fully developed the technique at this time. Nevertheless, we agree with the reviewer that measuring Lyapunov exponents and topological entropies would have provided deeper insight into the mixing dynamics of our system from the perspective of system chaotic degree. In the revised Discussion section, we address the omission of chaotic characterization and suggest potential future work characterizing the chaotic degree in the nonuniform active fluid system to gain a deeper understanding of its mixing dynamics (lines 322- 325).  
+
+In response to reviewer's suggestion on exploring advective flows in our active- inactive fluid system, we adopted a dimensionless quantity, Péclet number, defined as \(Pe \equiv \bar{v}_{ab} l_c / D\) , where \(\bar{v}_{ab}\) represents the mean flow speed of active fluid, \(D\) is the diffusion coefficient of ATP, and \(l_c\) is the correlation length of flow velocity. The physical interpretation of this quantity is the ratio of convective transport rate to diffusive transport rate. When the Péclet number is greater than of order 1, the active transport is dominated by convection, whereas when the Péclet number is smaller than of order 1, the active transport is dominated
+
+<--- Page Split --->
+
+by diffusion. To explore how the behaviors of active transport affect the mixing of an active- inactive fluid system, we first analyzed the interface progression exponent \(\gamma\) as a function of Péclet number (Fig. 4b) and found that as \(Pe \lesssim 3\) , \(\gamma \approx 1\) , which corresponded to the diffusion- like mixing captured in our Fick's law- based model (Fig. 3). Interestingly, as the Péclet number is increased to \(Pe \gtrsim 3\) , we observed \(\gamma\) became greater than 1, which corresponded to the transition in interface progression to being superdiffusion- like as the convection mechanisms started to emerge. Overall, introducing the concept of advection and quantifying it with the dimensionless Péclet number allowed us to understand the progression of active- inactive interface from a more fundamental perspective of material transport. We have revised the manuscript to include the concept of advection to interpret the observed transition of interface progression throughout the manuscript (e.g., lines 146- 161). We believe our manuscript is now more self- explanatory. We thank the reviewer for the suggestion that significantly improved the manuscript.  
+
+I also think that the idea that mixing is driven primarily by defects should be addressed. Can this idea be related to the present work? This was first introduced in theoretical works by Marchetti and shown experimentally in ref 24. Since the submitted paper is not the first to talk about self- mixing in active nematics, a discussion should be added to see how these different papers fit together and can be connected to the submitted work.  
+
+We thank the reviewer for pointing out a confusion in our manuscript. The work of Tan et al. and Marchetti (such as PRX 9, 041047 [2019]) mainly focuses on 2D active nematic systems with high nematic order in which defect dynamics play a dominant role in mixing. In contrast, our experimental system is a 3D isotropic active gel with nematic order parameter close to zero, in which mixing is mainly driven by extensile microtubule bundles. Because of this fundamental difference in mixing dynamics, we think that our work cannot be directly compared with theirs. To avoid readers misinterpreting our work as the results from a 2D active nematic system, we revised the first paragraph of Results section to clarify that our system is a 3D isotropic active gel.  
+
+With these being said, we think that it would be more elucidative to compare our work with previous research on 3D isotropic active gels. As such, we compared our works with experimental research by Sanchez et al. (Nature 491, 431 [2012]) and Henkin et al. (Philosophical Transactions. Series A, Mathematical, Physical, and Engineering Sciences 372, 20140142 [2014]) and modeling research by Varghese et al. (PRL 125, 268003 [2020]) and Saintillan and Shelly (Physics of Fluids 20, 123304 [2008]). We believe these comparisons better demonstrate how self- organization and mixing dynamics of 3D isotropic microtubule- kinesin active fluids with nonuniform activity are different from those with uniform activity.  
+
+5. The results presented in the paper focus on large length-scales (i.e. much larger that the active length scale. Can the authors discuss their results where the length scales approach the active length scale or even go below it? Transport measures may be different on small scales and should be ballistic-like. Can local flows o the scale of defects be tracked to get more detail on these scales?  
+
+We thank the reviewer for suggesting that we look into the smaller- scale kinematics. Indeed, our results for interface progression transitioning from diffusion- like to superdiffusion- like behaviors would be better elucidated if we could have measured the mean squared displacement (MSD) of tracers across the interface and extract the diffusion exponents (a), like previous studies by Sanchez et al. Nature 491, 431 (2012). We attempted to perform such experiments and analyses; however, we soon learned that it was difficult to realize because, unlike Sanchez et al.'s work where the active fluid had steady uniform activity, our system had a dynamic active- inactive interface whose position and width changed with time. Such a time- varying interface prevented us from collecting MSD of a tracer at a fixed activity level of the interface because the
+
+<--- Page Split --->
+
+tracer initially at the diffusion zone (low ATP concentration) of the interface may later be in the superdiffusion zone (high ATP concentration) as the interfaces passed by so it would be difficult to distinguish between the diffusive and superdiffusive data, not to mention to analyze the corresponding diffusion exponents \((a)\) . Thus, we think that investigating the tracer motions at small length scales of the active- inactive interface in our system was impractical, at least with the tools and methods we currently have.  
+
+However, we agree with the reviewer that such a small length- scale study would provide a deeper insight into the active transport of active fluid at the active- inactive interface from the perspective of microscopic kinematics. Perhaps such a characterization can be better realized in an active fluid system whose active- inactive interface does not change with time, which would make it possible to measure tracer MSD at a fixed activity level of the interface and reveal how the tracer behaviors change across the interface. Such an experimental system could be established by adopting active fluid systems that are only activated upon light exposure and become inactive when the light is turned off (such as the systems developed by Ross et al. Nature 572, 224 [2019] and by Zhang et al. Nature Materials 20, 875 [2021]) because one can use such active fluid systems to create a steady, time- independent activity gradient by applying a fixed light intensity gradient.  
+
+We thank the reviewer for the insightful suggestions. We have revised the manuscript to clarify the limitations of our studies and suggested experiments for potential future work (lines 326- 338) for readers who are interested in learning about the microscopic kinematics of fluid flows at active- inactive interface.  
+
+6. It appears that Fig 3d might fit to Michaelis-Menten kinetics, tt looks like the trend is approximately there. Is that trend expected for this system? Comment on the shape of the curve and prior work.  
+
+We thank the reviewer for sharing the keen observation. When we originally prepared the manuscript, we did not think the interface progression coefficient would follow the Michaelis- Menten trend because we thought it was irrelevant, but after we followed the reviewer's suggestion to fit \(P_{I}\) vs. \(C_{0}\) to the Michaelis- Menten equation (Supplementary Fig. 4a), we found out that we were wrong because \(P_{I}\) vs. \(C_{0}\) was well fit to the equation with goodness of fit \(R^{2} \geq 0.99\) (Supplementary Fig. 4b). In fact, we tried ten other flow speed- ATP relations and found that in each relation, \(P_{I}\) vs. \(C_{0}\) followed the corresponding ATP dependence (fit curves in Supplementary Fig. 4a). This is an unexpected result; we never thought that two such unrelated ATP dependences— \(P_{I}(C_{0})\) and \(\bar{\nu} (C)\) —were connected in our model. To explore whether there is an underlying algebra that connect \(P_{I}\) and \(C_{0}\) via a Michaelis- Menten equation, we derived an analytical expression for the interface progression coefficient as a function of initial ATP concentration \(P_{I}(C_{0})\) (Eq. S7 in Supplementary Discussion 3), which reproduced the numerical results (see magenta curve and red dots in Fig. 3d). This expression shows that \(P_{I}\) was connected to \(C_{0}\) via an inverse complementary error function. Coincidentally, this functional form is extremely well approximated by a Michaelis- Menten- type equation \((R^{2} \geq 0.99)\) . We have summarized this finding in Supplementary Discussions 2 and 3 for readers who are interested in how the choice of flow speed- ATP relation \(\bar{\nu} (C)\) affects the resulting interface progression coefficient \(P_{I}(C_{0})\) . We thank the reviewer for sharing with us this interesting perspective in our interface progression coefficient that made our manuscript more inspiring.  
+
+7. In all the Figures the captions need to make it much clearer which panels are calculations and which are experimental data. The reader should not have search around for this information.  
+
+We thank the reviewer for pointing out this lack of clarity in our figure captions. To address this concern, we explicitly stated whether each figure shows experimental or simulation results in the beginning of each
+
+<--- Page Split --->
+
+figure caption. In Fig. 3, where simulations and experiments were both presented for comparison, we specified in each panel whether the data shown was simulation or experimental results.  
+
+Line 241 - You can't use "turbulences" - this should phrased better - do you mean vortices? Areas of "active turbulence". Please clarify.  
+
+We thank the reviewer for highlighting the confusion in our manuscript. We have changed the phrase "active turbulence" to "chaotic, turbulence- like mixing flows" to improve the readability of our manuscript. We thank the reviewer for the suggestion, which has made our manuscript more understandable and clearer to a wider range of readers.  
+
+Fig S3 - the plural of spectrum is spectra.  
+
+We thank the reviewer for their correction. We have corrected the spelling mistakes in the caption of Supplementary Fig. 11.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+## Reviewer #2 (Remarks to the Author):  
+
+In their rebuttal letter, the authors have responded in detail to all points raised in my previous report. They have modified and extended their manuscript accordingly. Thus, I support the publication of the manuscript in its present form.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+I am satisfied with the changes, they have greatly improved the manuscript and I recommend this work be accepted.
+
+<--- Page Split --->
+
+## Reviewer 2:  
+
+In their rebuttal letter, the authors have responded in detail to all points raised in my previous report. They have modified and extended their manuscript accordingly. Thus, I support the publication of the manuscript in its present form.  
+
+We thank the reviewer for the support.  
+
+## Reviewer 3:  
+
+I am satisfied with the changes, they have greatly improved the manuscript and I recommend this work be accepted.  
+
+We thank the reviewer for the recommendation.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a4501f2f5fc141c371fc4a9c6c00e00eda96fbbe28e2750079e2577f0d22b3f4/supplementary_0_Peer Review File__a4501f2f5fc141c371fc4a9c6c00e00eda96fbbe28e2750079e2577f0d22b3f4_det.mmd

@@ -0,0 +1,461 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[104, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[177, 154, 818, 209]]<|/det|>
+Self- mixing in microtubule- kinesin active fluid from nonuniform to uniform distribution of activity  
+
+<|ref|>image<|/ref|><|det|>[[93, 732, 262, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[270, 732, 880, 784]]<|/det|>
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[120, 83, 332, 99]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[120, 115, 450, 132]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 148, 875, 310]]<|/det|>
+In this paper, the authors perform experiments on mixing in an active fluid composed of microtubules and kinesin clusters powered by ATP. Using light activated ATP, they are able to turn on activity over a finite region of space, thus creating an interface between an active fluid and a passive fluid. As the light- activated ATP disperses across the interface due to both diffusion and active mixing, the region where active flows take place expands spatially, thereby further enhancing transport and mixing. The authors characterize the rate at which the ATP gets mixed between the two regions, and find that the mixing is diffusive at low levels of activity, but superdiffusive at high levels of activity. They compare their experimental results with two types of models: the solution of a simple 1D diffusion equation, and a more complex PDE model for an active nematic coupled to fluid flow.  
+
+<|ref|>text<|/ref|><|det|>[[118, 326, 878, 456]]<|/det|>
+Overall, the experiments are impressive and well executed, and the paper is well written. The main result, namely the transition from diffusive to superdiffusive spreading with increasing of activity, is interesting, although perhaps not entirely surprising. However, the discussion and analysis as well as the comparison with models are sometimes a bit weak, and it is not entirely clear what fundamental new physics is learned as a result of this study. In particular, the transition to superdiffusive transport is not really understood (and in fact it is not captured by the models). As a result, I cannot recommend publication of the manuscript in its present form in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[118, 472, 878, 505]]<|/det|>
+Here are some more detailed comments and suggestions that the authors should addressing to improve the quality of the paper:  
+
+<|ref|>text<|/ref|><|det|>[[118, 521, 880, 651]]<|/det|>
+1. It's not completely clear how to interpret the velocity of equation (4): ATP should directly affect the velocity of kinesin motors, but how it translates into the velocity the interface is not trivial: indeed, the motors produce local microtubule motions, which then interact through the fluid the causing interfacial spreading. Furthermore, the fact that the interface in the 1D model spreads diffusively is not very surprising, since the model is based on the diffusion equation and the velocity of equation (4) is directly tied to the ATP concentration (in fact, when [ATP]<<k in the early stages of mixing, the velocity is directly proportional to [ATP]). As a result, I'm not sure exactly what is learned from the model.  
+
+<|ref|>text<|/ref|><|det|>[[118, 667, 878, 780]]<|/det|>
+2. It is likely that the transition to superdiffusion is tied to the emergence of strong large-scale motions in the fluid, which result in advective (ballistic) transport on top of the diffusive spreading. This is supported by the observation on page 4 that in order to increase the velocities in the fluid the authors increasing the thickness of the sample. This should have a direct effect on the correlation length scales of the turbulent motions in the active fluid (this point is very briefly addressed in the discussion section). It would be very interesting to characterize these length scales and determine how they affect the mixing process.  
+
+<|ref|>text<|/ref|><|det|>[[118, 797, 864, 829]]<|/det|>
+3. The fact the checkerboard patterns result in faster mixing than a single interface is again not very surprising, since gradients are introduced on short length scales.  
+
+<|ref|>text<|/ref|><|det|>[[118, 845, 861, 911]]<|/det|>
+4. The discussion of rheology in the discussion section is quite interesting, in particular the fact that the microtubule network really behaves as crosslinked gel until the kinesin motors are activated. However, it is not completely obvious why this would contribute to superdiffusive transport as hypothesized by the authors: in fact, I would expect transport to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 368, 100]]<|/det|>
+be weaker due to crosslinking.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 131, 450, 148]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 163, 880, 475]]<|/det|>
+The manuscript reports the results of a combined experimental and theoretical/numerical study of fluid mixing by active fluids. Ultraviolet light- activated caged ATP and fluorescent dyes are used to activate regions of a microtubule- kinesin active fluid. Both a system with an interface between an activated and a passive microtubule- kinesin mixture, and a checkerboard activation pattern are investigated. As activity is turned on, the interfaces broaden and move into the initially inactivated regions. t low activation levels, mixing is found to be governed by an active diffusion- like processes at the active- inactive interface, while at higher activity levels, superdiffusion- like processes dominate. Samples activated in a checkerboard pattern reach homogeneity faster than those with a single dividing interface. A model of active nematohydrodynamics coupled to ATP transport is employed to describe the coupled mixing process numerically. The results are in qualitative agreement, but differ on the quantitative level significantly.  
+
+<|ref|>text<|/ref|><|det|>[[117, 488, 867, 535]]<|/det|>
+Mixing of fluids on the microscale is difficult to achieve, because Reynolds number is low and mixing is  
+
+<|ref|>text<|/ref|><|det|>[[117, 526, 867, 636]]<|/det|>
+dominated by molecular diffusion. Active fluids have the potential to generate a pronounced speed- up of the mixing process. The current study focuses on systems with spatiotemporally varying activities and the role of interfaces in the mixing process. This is a very interesting aspect, which has not received much attention so far.  
+
+<|ref|>text<|/ref|><|det|>[[118, 650, 622, 667]]<|/det|>
+The following questions and comments should be addressed:  
+
+<|ref|>text<|/ref|><|det|>[[115, 666, 875, 911]]<|/det|>
+(1) Why would a diffusive MSD of the interface position, which would naively be expected for an underlying diffusive process, give any hint about a super-diffusive process? 
+(2) How does the flow speed, which is determined by Michaelis-Menten kinetics, enter into model given by eqs. (1-3)? 
+(3) What changes at the molecular scale at the velocity threshold of 5 micron/sec? 
+(4) Why does mixing time increase linearly with grid size (Fig. 7b), and extrapolate to zero mixing time at a grid size of about \(0.9 \text{mm}\) ? 
+(5) The theoretical model predicts a scaling of interface progression coefficient, which is consistent with the results from experiments. However, the coefficient magnitudes differed by a factor of 5 between the model and the experiments.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 83, 865, 115]]<|/det|>
+Furthermore, the simulation does not show the transition from an active diffusion- like to a superdiffusion- like processes observed experimentally.  
+
+<|ref|>text<|/ref|><|det|>[[115, 115, 870, 148]]<|/det|>
+The authors propose as a possible explanation that the rheology of microtubule network is not taken  
+
+<|ref|>text<|/ref|><|det|>[[115, 149, 875, 180]]<|/det|>
+into account in the numerical model. Is this a plausible explanation, in particular for the factor 5?  
+
+<|ref|>text<|/ref|><|det|>[[115, 181, 875, 230]]<|/det|>
+(6) The authors emphasize that the mixing efficacy of the nonuniform active fluid systems depends on the distribution of activity, with systems consisting of more small active areas  
+
+<|ref|>text<|/ref|><|det|>[[115, 231, 800, 262]]<|/det|>
+evolve to a homogeneous state faster than systems with the same total active area distributed as  
+
+<|ref|>text<|/ref|><|det|>[[115, 263, 810, 294]]<|/det|>
+one piece. They conclude that the activity-uniform active fluid has the highest mixing performance.  
+
+<|ref|>text<|/ref|><|det|>[[115, 295, 847, 343]]<|/det|>
+However, this is exactly the behavior I would expect for a passive systems: As interfaces propagate diffusively, shorter distances can be covered more quickly. Thus, what is the significance  
+
+<|ref|>text<|/ref|><|det|>[[115, 344, 513, 360]]<|/det|>
+of the active process in this general conclusion?  
+
+<|ref|>text<|/ref|><|det|>[[115, 361, 870, 392]]<|/det|>
+(7) Arguments about maximizing system entropy are difficult because this a non-equilibrium active system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 393, 541, 408]]<|/det|>
+(8) The following references might also be relevant:  
+
+<|ref|>text<|/ref|><|det|>[[115, 409, 627, 425]]<|/det|>
+-- A. Doostmohammadi et al., Nat. Commun. 8, 15326 (2017);  
+
+<|ref|>text<|/ref|><|det|>[[115, 426, 580, 441]]<|/det|>
+-- G.A. Vliegenthart et al., Sci. Adv. 6, eaaw9975 (2020);  
+
+<|ref|>text<|/ref|><|det|>[[115, 443, 525, 457]]<|/det|>
+-- H. Reinken et al., Commun. Phys. 3, 76 (2020);  
+
+<|ref|>text<|/ref|><|det|>[[115, 459, 520, 473]]<|/det|>
+-- R. Alert et al., Nature Physics 16, 682 (2020);  
+
+<|ref|>text<|/ref|><|det|>[[115, 475, 480, 490]]<|/det|>
+-- K. Qi et al., Commun. Phys. 5, 49 (2022).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 537, 450, 553]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 570, 875, 636]]<|/det|>
+The authors have performed mixing experiments on microtubule- kinesin based active nematics by taking advantage of an interesting molecule called "caged ATP". This ATP can be introduced uniformly into the system in its inactive state and then activated remotely via a UV light to give spatial patterning of activity in the active nematic system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 636, 874, 765]]<|/det|>
+They observed diffusion- like and super- diffusive regimes dependent on the speed of the microtubule flows and compare results with some simulations of the system with a variety of quantitative experiments. Beautiful videos demonstrate the concepts very well. The data are fit to a fairly simple transport model that seems to describe the behaviors well. An additional and interesting experiment towards the end of the paper tests checker- board patterns of activation to look at how fast the system reaches homogeneity as a function of active/inactive interfacial length. This section of the paper could probably be a separate publication if explored in more detail.  
+
+<|ref|>text<|/ref|><|det|>[[117, 780, 866, 862]]<|/det|>
+Overall I think that the paper covers an exciting topic and one that should be focused on by the active nematic community. Self- mixing in active nematics represents an important direction in the evolution of this field. I like the hypothesis driven approach and the data is presented in a logical and interesting fashion. A large amount of data is shown with a focus largely on statistical measures such as 1D 'diffusion' constants (i.e. P in this case).  
+
+<|ref|>text<|/ref|><|det|>[[115, 878, 870, 911]]<|/det|>
+I found a few issues in the paper that should be corrected pre- publication but I think that the paper is appropriate for Nature Communications in scope and impact.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 870, 133]]<|/det|>
+1. The title of the paper can be improved for clarity - is it missing a comma or colon? I would use "activity" not "activities"  
+
+<|ref|>text<|/ref|><|det|>[[117, 148, 870, 295]]<|/det|>
+2. I found that the description of the results at the start of that section to be a bit lacking in necessary detail. The authors need to be clearer about the nature of the "initially inactivated fluid" in the first results paragraph. Is it mixed Mts and kinesin but not formed into bundles (random filaments, or small bundles not yet aligned)? Presumably the solution has not yet been in the active nematic state? Are the kinesin clusters bound to MTs yet but the Mts are isotropic? It appears from the video that this is the case and I can guess, but clarification must be added. It seems that initiation of activity is not a reversible process - I.e. once bundled the system will never go back to the initial state (e.g. when ATP runs out). This can also be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[117, 311, 874, 424]]<|/det|>
+3. In the first results paragraph we also need to know where the dyes are located to interpret the videos and Figures easily - is it on the MT? The kinesin or perhaps in the water. This needs to be added in the first results paragraph. How about the tracer particles? Are they coupled to the MTs? How are they confined to the active layer? I realize that there is plenty of detail in the methods section at the end of the paper but the most important points for figure interpretation should be in the main text otherwise it's too difficult to read the paper - especially for the uninitiated general reader.  
+
+<|ref|>text<|/ref|><|det|>[[117, 440, 876, 617]]<|/det|>
+4. The authors don't address the concept of chaotic mixing and advective flows at all in the paper and I found that to be an omission - how does that concept relate to the measures for quantifying mixing in the submitted work? A short discussion might help put this work in context with that recent work cited as ref 24 where chaotic mixing was described for the same system. Some other groups are also considering advective flows in these systems. How would this be related to the present work? I also think that the idea that mixing is driven primarily by defects should be addressed. Can this idea be related to the present work? This was first introduced in theoretical works by Marchetti and shown experimentally in ref 24. Since the submitted paper is not the first to talk about self-mixing in active nematics, a discussion should be added to see how these different papers fit together and can be connected to the submitted work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 634, 864, 715]]<|/det|>
+5. The results presented in the paper focus on large length-scales (i.e. much larger that the active length scale. Can the authors discuss their results where the length scales approach the active length scale or even go below it? Transport measures may be different on small scales and should be ballistic-like. Can local flows o the scale of defects be tracked to get more detail on these scales?  
+
+<|ref|>text<|/ref|><|det|>[[118, 732, 857, 780]]<|/det|>
+6. It appears that Fig 3d might fit to Michaelis-Menten kinetics, tt looks like the trend is approximately there. Is that trend expected for this system? Comment on the shape of the curve and prior work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 797, 871, 845]]<|/det|>
+7. In all the Figures the captions need to make it much clearer which panels are calculations and which are experimental data. The reader should not have search around for this information.  
+
+<|ref|>text<|/ref|><|det|>[[115, 862, 866, 894]]<|/det|>
+Line 241 - You can't use "turbulences" - this should phrased better - do you mean vortices? Areas of "active turbulence". Please clarify.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 456, 117]]<|/det|>
+Fig S3 - the plural of spectrum is spectra.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 208, 106]]<|/det|>
+## Reviewer 1:  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 882, 272]]<|/det|>
+In this paper, the authors perform experiments on mixing in an active fluid composed of microtubules and kinesin clusters powered by ATP. Using light activated ATP, they are able to turn on activity over a finite region of space, thus creating an interface between an active fluid and a passive fluid. As the light- activated ATP disperses across the interface due to both diffusion and active mixing, the region where active flows take place expands spatially, thereby further enhancing transport and mixing. The authors characterize the rate at which the ATP gets mixed between the two regions, and find that the mixing is diffusive at low levels of activity, but superdiffusive at high levels of activity. They compare their experimental results with two types of models: the solution of a simple 1D diffusion equation, and a more complex PDE model for an active nematic coupled to fluid flow.  
+
+<|ref|>text<|/ref|><|det|>[[115, 282, 882, 402]]<|/det|>
+Overall, the experiments are impressive and well executed, and the paper is well written. The main result, namely the transition from diffusive to superdiffusive spreading with increasing of activity, is interesting, although perhaps not entirely surprising. However, the discussion and analysis as well as the comparison with models are sometimes a bit weak, and it is not entirely clear what fundamental new physics is learned as a result of this study. In particular, the transition to superdiffusive transport is not really understood (and in fact it is not captured by the models). As a result, I cannot recommend publication of the manuscript in its present form in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[115, 412, 882, 619]]<|/det|>
+We thank the reviewer for their assessment. The essence of our work was to investigate the mixing dynamics of active fluid systems with a dynamic activity gradient. Recently, there has been a trend within active fluid research to study systems with activity gradients. For example, Zhang et al. Nature Materials 20, 875 (2021) and Ross et al. Nature 572, 224 (2019) investigated the role of activity gradients in the self- organization of active fluid and Shankar and Marchetti PRX 9, 041047 (2019) predicted the defect dynamics in an activity gradient. However, most existing work about activity gradients is focused on systems with steady state, time- independent activity gradients. Our work is novel in that it explores an activity gradient that changes spontaneously with time. Also, this paper hints that melting dynamics of the microtubule network play a role in the behavior of a dynamic activity gradient, and that such dynamics have been overlooked in existing active fluid models. As such, we believe our work can promote development of active fluid modeling to include the network melting mechanism and thus more accurately describe complex active fluid systems with dynamic distributions of activity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 629, 880, 664]]<|/det|>
+Here are some more detailed comments and suggestions that the authors should addressing to improve the quality of the paper:  
+
+<|ref|>text<|/ref|><|det|>[[115, 673, 882, 742]]<|/det|>
+1. It's not completely clear how to interpret the velocity of equation (4): ATP should directly affect the velocity of kinesin motors, but how it translates into the velocity the interface is not trivial: indeed, the motors produce local microtubule motions, which then interact through the fluid the causing interfacial spreading.  
+
+<|ref|>text<|/ref|><|det|>[[115, 752, 882, 891]]<|/det|>
+We thank the reviewer for the comment. The velocity in Eq. 4 represents mean speed of active fluid flow driven by extensile microtubules whose motions are driven by kinesin motor proteins consuming local ATP and stepping along microtubules (Fig. 1a). The progression of the interface is a complex phenomenon that involves the coupling of ATP dispersion and active fluid flows. We agree with the reviewer that how dispersion of ATP relates to progression of the interface is not trivial. To further explore the issue, we derived an analytical expression for interface progression coefficient, \(P_{I}\) , (Eq. S7 in Supplementary Discussion 3) that showed that \(P_{I}\) depended on the initial ATP concentration \(C_{0}\) via an inverse complementary error function. Then we explored how \(P_{I}\) varied with \(C_{0}\) numerically, which revealed that
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 883, 210]]<|/det|>
+a higher initial ATP concentration led to a faster- progressing interface (Fig. 3d). Surprisingly, we further found that the interface progression coefficients \(P_{i}\) are well fit to the Michaelis- Menten equation (the second curve from the top in Supplementary Fig. 4a). This result demonstrated that how the speed of active fluid depended on ATP concentration directly led to how the interface progression depended on ATP concentration. For readers who are interested in the connection between these two ATP dependences, we added two supplementary discussions (Supplementary Discussions 2 & 3) to explore the connection numerically as well as analytically.  
+
+<|ref|>text<|/ref|><|det|>[[115, 220, 882, 289]]<|/det|>
+Furthermore, the fact that the interface in the 1D model spreads diffusively is not very surprising, since the model is based on the diffusion equation and the velocity of equation (4) is directly tied to the ATP concentration (in fact, when [ATP] \(\leq \leq \mathrm{k}\) in the early stages of mixing, the velocity is directly proportional to [ATP]). As a result, I'm not sure exactly what is learned from the model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 299, 883, 489]]<|/det|>
+We thank the reviewer for asking about the significance of our 1D Fick's law- based active fluid model. The purpose of introducing this model was to demonstrate that the mixing of active and inactive fluids we observed (Fig. 2) could be described with a minimal model considering only diffusion of ATP and Michaelis- Menten kinetics (Fig. 3), without the need for a more complex hydrodynamic model such as the one we developed later in the manuscript (Fig. 6). We found that the simple 1D Fick's law- based active fluid model did agree with experimental observations (Fig. 3d). This consistency between the Fick's law- based model and our experimental results suggests that Michaelis- Menten kinetics successfully approximate the conversion of ATP concentration into local active fluid speed. For readers who are interested in learning about why we introduced the Fick's law- based model, we revised the manuscript to clarify the motivation of developing the model and its associated intellectual merits in the main text (lines 100- 104 and 129- 133) as well as in Supplementary Discussion 2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 499, 883, 725]]<|/det|>
+In response to the reviewer's comment that the connection between diffusion- like interface progression \((\gamma = 1)\) and diffusive ATP \((a = 1)\) is an expected, trivial consequence of adopting the Michaelis- Menten equation (Eq. 4), we performed additional studies (Supplementary Discussion 2) that showed that the connection was not necessarily built upon using the Michaelis- Menten equation, because even when the relation between ATP concentration and flow speed became square root, quadratic, cubic, or higher power- law dependent, the resulting interface progression exponent remained \(\gamma = 1\) (Supplementary Fig. 3d inset). These additional studies show that connection between \(\gamma = 1\) and \(a = 1\) was not sensitive to which flow speed- ATP relation we chose (although the interface progression coefficient \(P_{i}\) was highly sensitive to the relation [Supplementary Fig. 3d]). For readers who are interested in learning about the role of flow speed- ATP relation in our model, we added Supplementary Discussion 2 to demonstrate how the modeling results would be influenced with 10 other relations. We thank the reviewer for questioning the intellectual merit of our Fick's law- based model which drove us to enrich the exploration of our model and strengthen the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 734, 882, 838]]<|/det|>
+2. It is likely that the transition to superdiffusion is tied to the emergence of strong large-scale motions in the fluid, which result in advective (ballistic) transport on top of the diffusive spreading. This is supported by the observation on page 4 that in order to increase the velocities in the fluid the authors increasing the thickness of the sample. This should have a direct effect on the correlation length scales of the turbulent motions in the active fluid (this point is very briefly addressed in the discussion section). It would be very interesting to characterize these length scales and determine how they affect the mixing process.  
+
+<|ref|>text<|/ref|><|det|>[[115, 848, 882, 900]]<|/det|>
+We greatly thank the reviewer for the insightful comment. The reviewer was correct that increasing the sample container height would affect the correlation length scales of the turbulent motion. We added a supplementary discussion (Supplementary Discussion 4) in which we quantify how the correlation length
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 366]]<|/det|>
+increased with sample container height (Supplementary Fig. 5d). To explore how the increased correlation length would affect the mixing of active- inactive fluid systems, we adopted a dimensionless quantity, the Péclet number \((P e)\) , defined as \(P e\equiv \bar{v}_{a b}l_{c} / D\) , where \(\bar{v}_{a b}\) represents the mean flow speed of active fluid, \(D\) is the diffusion coefficient of ATP, and \(l_{c}\) is the correlation length of flow velocity. The physical interpretation of this quantity is the ratio of convective transport rate to diffusive transport rate. When the Péclet number is greater than of order 1, the active transport is dominated by convection, whereas when the Péclet number is smaller than of order 1, the active transport is dominated by diffusion. To explore how the behaviors of active transport affect mixing in an active- inactive fluid system, we first analyzed the interface progression exponent \(\gamma\) as a function of Péclet number (Fig. 4b) and found that as \(P e\lesssim 3\) , \(\gamma \approx 1\) which corresponded to the diffusion- like mixing captured in our Fick's law- based model (Fig. 3). Interestingly, as the Péclet number is increased to \(P e\gtrsim 3\) , we observed \(\gamma\) became greater than 1, which corresponded to the transition in interface progression from diffusion- like to superdiffusion- like. As such, the reviewer was correct; the transition to superdiffusion was tied to the gradual emerging convection mechanism that governed the mixing process of active- inactive fluid system. We have revised the manuscript to clarify the underlying mechanism causing the transition (lines 146- 161 with accompanying Fig. 4b). We greatly appreciate the reviewer's constructive feedback that significantly improved the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 376, 881, 411]]<|/det|>
+3. The fact the checkerboard patterns result in faster mixing than a single interface is again not very surprising, since gradients are introduced on short length scales.  
+
+<|ref|>text<|/ref|><|det|>[[115, 420, 882, 490]]<|/det|>
+We agree with the reviewer that the results of the checkerboard pattern experiments were expected and thus the intellectual contribution was minimal. The reason we introduced these experiments was to explore how the distribution of activity would affect the mixing process of the active- inactive fluid system, as previously we had only explored one configuration of activity distribution (one side active and one side inactive).  
+
+<|ref|>text<|/ref|><|det|>[[114, 498, 883, 793]]<|/det|>
+To enrich this part of the work, we further explored the checkerboard system in our established simulation platform (Fig. 8), revealing two interesting results: (1) At a sufficiently small grid size, the simulated mixing times of active and inactive systems were indistinguishable. We believe that this is because the active fluid needs time to "warm up". In experiments, the system had a warm- up time caused by network melting (Supplementary Discussion 6). Although a network melting mechanism was not included in the model, the simulated active fluid flow still took dimensionless time to rise because the onset of the flows was triggered by the initial activity- driven instability in extensile \(Q\) field which took finite dimensionless time to develop \((\sim 1\) dimensionless time in this case; Supplementary Video 6), and during the warm- up time, molecular diffusion is the main driving force of mixing. When the grid size was very small, the mixing could be completed by diffusion before the fluid activity could speed up the mixing. Thus, the simulation based on the checkerboard pattern suggested a limitation of active fluid mixing: The mixing effect of active fluid can only take place when the mixture is sufficiently nonuniform. (2) The checkerboard- pattern simulation also revealed that the mixing process of active fluid is less susceptible to changes in grid size than in inactive fluid systems (Fig. 8b). Altering the dimensionless grid size from 2 to 22 changed the mixing time of the inactive fluid system by a factor of 40 but only changed the mixing time of the active fluid system by a factor of 3. This result suggests that an active fluid system is less susceptible to the initial distribution of the mixture than an inactive fluid system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 802, 882, 890]]<|/det|>
+These two new results (added to Results lines 261- 268, Discussion lines 307- 319, Methods lines 516- 529) demonstrate limits and advantages of introducing active fluid to microfluidic mixing systems with nonuniform activity. We thank the reviewer for the comment that drove us to enrich the manuscript. We believe that the checkerboard work is now more relevant and could inspire further exploration of the active mixing process with other complex distributions of activity.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 882, 158]]<|/det|>
+4. The discussion of rheology in the discussion section is quite interesting, in particular the fact that the microtubule network really behaves as crosslinked gel until the kinesin motors are activated. However, it is not completely obvious why this would contribute to superdiffusive transport as hypothesized by the authors: in fact, I would expect transport to be weaker due to crosslinking.  
+
+<|ref|>text<|/ref|><|det|>[[114, 167, 883, 428]]<|/det|>
+We thank the reviewer for asking this question. After following the reviewer's second comment, we found that the transition to superdiffusion- like interface progression ( \(\gamma >1\) ) was tied to the underlying transition of active transport from being diffusion- like ( \(Pe \leq 3\) ) to superdiffusion- like ( \(Pe \geq 3\) ), and thus the network rheology might not have played a significant role in this transition. Thus, we removed that hypothesis from the manuscript. Nevertheless, we explored our data more deeply and found that the network rheology could indeed slow the progression of active- inactive interface (Supplementary Fig. 7e), as the reviewer expected, and this slower progression was not captured in our active- fluid hydrodynamic model because the model did not include the network melting mechanism (Supplementary Fig. 8c). These results demonstrate that the network melting mechanism plays a role in the mixing process of activity- nonuniform active fluid systems, and this mechanism has thus far been overlooked in the modeling of microtubule- kinesin active fluid systems. In the revised manuscript, we addressed the significance of the network rheology in mixing dynamics (lines 284- 297 and Supplementary Discussion 6) and suggested potential future experiments and simulations to further investigate the network melting dynamics (lines 297- 302). We believe that our current manuscript will encourage the active fluid community to investigate this topic further and inspire further exploration in self- organization of activity- nonuniform active fluid systems.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 208, 106]]<|/det|>
+## Reviewer 2:  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 882, 290]]<|/det|>
+The manuscript reports the results of a combined experimental and theoretical/numerical study of fluid mixing by active fluids. Ultraviolet light- activated caged ATP and fluorescent dyes are used to activate regions of a microtubule- kinesin active fluid. Both a system with an interface between an activated and a passive microtubule- kinesin mixture, and a checkerboard activation pattern are investigated. As activity is turned on, the interfaces broaden and move into the initially inactivated regions. t low activation levels, mixing is found to be governed by an active diffusion- like processes at the active- inactive interface, while at higher activity levels, superdiffusion- like processes dominate. Samples activated in a checkerboard pattern reach homogeneity faster than those with a single dividing interface. A model of active nematohydrodynamics coupled to ATP transport is employed to describe the coupled mixing process numerically. The results are in qualitative agreement, but differ on the quantitative level significantly.  
+
+<|ref|>text<|/ref|><|det|>[[115, 298, 882, 385]]<|/det|>
+Mixing of fluids on the microscale is difficult to achieve, because Reynolds number is low and mixing is dominated by molecular diffusion. Active fluids have the potential to generate a pronounced speed- up of the mixing process. The current study focuses on systems with spatiotemporally varying activities and the role of interfaces in the mixing process. This is a very interesting aspect, which has not received much attention so far.  
+
+<|ref|>text<|/ref|><|det|>[[115, 395, 881, 429]]<|/det|>
+We thank the reviewer for their comments on the intellectual merit of our work. Please find our point- by- point responses to the reviewer's feedback below.  
+
+<|ref|>text<|/ref|><|det|>[[115, 439, 555, 456]]<|/det|>
+The following questions and comments should be addressed:  
+
+<|ref|>text<|/ref|><|det|>[[115, 466, 882, 501]]<|/det|>
+(1) Why would a diffusive MSD of the interface position, which would naively be expected for an underlying diffusive process, give any hint about a super-diffusive process?  
+
+<|ref|>text<|/ref|><|det|>[[115, 510, 882, 616]]<|/det|>
+We thank the reviewer for their comment. We have revised the paper to clarify this point. In the previous version of the paper, we used a proof by contradiction: proving a point by demonstrating that its opposite is a contradiction. We sought to prove that \(\gamma = 1\) is the consequence of ATP diffusion \((a = 1)\) by first stating that \(\gamma = 1\) because \(a > 1\) (superdiffusion). Then we used the Fick's law model to show that this statement was incorrect; thus, the invalidity of the statement indicated that \(\gamma = 1\) from \(a = 1\) (or \(a \leq 1\) , more strictly).  
+
+<|ref|>text<|/ref|><|det|>[[115, 626, 882, 764]]<|/det|>
+We agree with the reviewer that this proof was too complicated and led to unnecessary confusion, and we removed the proof from the manuscript. As an alternative, we revised the Fick's law model section in the manuscript to directly show that modeling ATP dispersion with the diffusion equation leads to diffusion- like interface progression (lines 100- 133). To provide deeper insight into the connection between ATP dispersion and interface progression, we presented an algebraic proof that squared interface displacement is proportional to time: \(\Delta x^{2} \propto t\) (Eq. S6) and provided an analytical perspective on how diffusive ATP transport led to diffusion- like interface progression (Supplementary Discussion 3). We believe that the manuscript is now clearer and more understandable.  
+
+<|ref|>text<|/ref|><|det|>[[115, 774, 881, 809]]<|/det|>
+(2) How does the flow speed, which is determined by Michaelis-Menten kinetics, enter into model given by eqs. (1-3)?  
+
+<|ref|>text<|/ref|><|det|>[[115, 819, 882, 906]]<|/det|>
+We thank the reviewer for asking. The Michaelis- Menten kinetics equation is separate from Eqs. 1- 3. The reason we modeled the ATP- dependent flow speed with Michaelis- Menten kinetics is because our previous studies of the flow speed of active fluid for various ATP concentrations (Figs. 3A- C in Bate et al. Soft Matter 15, 5006 [2019]) showed that the flow speed of active fluid followed Michaelis- Menten kinetics for ATP concentrations greater than \(100 \mu \mathrm{M}\) . Additionally, we believe this choice of model is reasonable
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 883, 141]]<|/det|>
+because the active fluid is driven by the motion of microtubules, which are driven by kinesin motors whose stepping rates are known to follow the Michaelis- Menten kinetics (Schnitzer et al. Nature 388, 386 [1997]), and thus these kinetics could be scaled up to active fluid scale.  
+
+<|ref|>text<|/ref|><|det|>[[115, 151, 883, 272]]<|/det|>
+To clarify how our choice of flow speed- ATP relation affected the modeling results in terms of predicted interface progression exponents \((\gamma)\) and coefficients \((P_{l})\) , in the revised manuscript we explored ten other plausible relations to connect ATP and flow speed and found that the Michaelis- Menten equation led to the best match between the predictions and the experimental results (Supplementary Discussion 2, Supplementary Fig. 3d). We are now more confident that the Michaelis- Menten equation is an appropriate coarse- grained approach to model local flow speed in terms of ATP concentrations. We thank the reviewer for pointing out the weakness in our manuscript which drove us to make the manuscript more robust.  
+
+<|ref|>text<|/ref|><|det|>[[115, 282, 707, 299]]<|/det|>
+(3) What changes at the molecular scale at the velocity threshold of 5 micron/sec?  
+
+<|ref|>text<|/ref|><|det|>[[114, 310, 883, 672]]<|/det|>
+We thank the reviewer for asking this important question. In the previous version of manuscript, we stated that the active- inactive interface progression occurred at an active fluid flow speed of \(5 \mu \mathrm{m / s}\) . In the revised version of the manuscript, we adopted a new perspective on this issue: we found that the transition is in fact related to Péclet number rather than active fluid speed. Péclet number is defined as \(Pe \equiv \bar{v}_{ab} l_{c} / D\) , where \(\bar{v}_{ab}\) represents the mean flow speed of active fluid, \(D\) is the diffusion coefficient of ATP, and \(l_{c}\) is the correlation length of flow velocity. The physical interpretation of this quantity is the ratio of convective transport rate to diffusive transport rate. When the Péclet number is smaller than of order 1, the active transport is dominated by diffusion, whereas when the Péclet number is greater than of order 1, the active transport is dominated by convection. To explore how the behaviors of active transport affect mixing in an active- inactive fluid system, we first analyzed the interface progression exponent \(\gamma\) as a function of Péclet number (Fig. 4b) and found that as \(Pe \leq 3\) , \(\gamma \approx 1\) , which corresponds to the diffusion- like mixing captured in our Fick's law- based model (Fig. 3). Interestingly, as the Péclet number increased to \(Pe \geq 3\) , \(\gamma\) became greater than 1, which corresponded to the transition in interface progression from being diffusion- like to being superdiffusion- like. Our further analysis revealed that the transition from diffusion- like to superdiffusion- like progression of the active- inactive interface was not directly relevant to the \(5 \mu \mathrm{m / s}\) flow speed. Instead, it was due to the transition of active transport from diffusion- dominated to superdiffusion- dominated, as characterized by a dimensionless Péclet number increasing from of order 1 to of order greater than 1. We revised the manuscript to clarify the mechanism underlying the diffusion- superdiffusion transition of the active- inactive interface (lines 146- 161, Fig. 4b). We thank the reviewer for their question, which drove us to dive more deeply into our data, pursue a fuller understanding of interface progression, and make the manuscript more robust.  
+
+<|ref|>text<|/ref|><|det|>[[115, 684, 880, 717]]<|/det|>
+(4) Why does mixing time increase linearly with grid size (Fig. 7b), and extrapolate to zero mixing time at a grid size of about \(0.9 \mathrm{mm}\) ?  
+
+<|ref|>text<|/ref|><|det|>[[115, 728, 883, 900]]<|/det|>
+We thank the reviewer for the observations on our experimental checkerboard data (Fig. 7b). However, we think it was not clear that the mixing time varied linearly with grid size. We believe that as the grid size approached zero, the mixing time would vary more slowly and approach zero (see the figure below). To learn more about grid size- dependence of mixing time, we performed a corresponding checkerboard active fluid simulation (Fig. 8) and found that as the grid sizes decreased, the mixing time in active fluid system became indistinguishable with the mixing time in inactive fluid system where the mixing was driven only by molecular diffusion (Fig. 8b). We believe that this was because our active fluid system needed time to "warm up". In experiments, the system had a warm- up time caused by network melting (Supplementary Discussion 6). Although a network melting mechanism was not included in the model, the simulated active fluid flow still took dimensionless time to rise because the onset of the flows was triggered by the initial
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 278]]<|/det|>
+activity- driven instability in extensile \(Q\) field which took finite dimensionless time to develop ( \(\sim 1\) dimensionless time in this case; Supplementary Video 6), and during the warm- up time, molecular diffusion is the main driving force of mixing. Thus, the simulation showed that as the grid sizes approached zero, the mixing dynamics would also change from active flow- dominated to molecular diffusion- dominated, so the mixing time was not expected to vary linearly with all grid sizes. We thank the reviewer for the comment on our checkerboard data which motivated us to explore the checkerboard system more deeply. For readers who are interested in learning about checkerboard mixing dynamics in the limit of small grid sizes, we added our simulation checkerboard work in Results (lines 261- 268), Discussion (lines 307- 319), and Methods (lines 516- 529) along with accompanying figure (Fig. 8). We believe that our checkerboard work is now richer and can provide readers with deeper understanding of the mixing dynamics of nonuniform active fluid systems.  
+
+<|ref|>image<|/ref|><|det|>[[301, 292, 692, 494]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[114, 506, 883, 611]]<|/det|>
+(5) The theoretical model predicts a scaling of interface progression coefficient, which is consistent with the results from experiments. However, the coefficient magnitudes differed by a factor of 5 between the model and the experiments. Furthermore, the simulation does not show the transition from an active diffusion-like to a superdiffusion-like processes observed experimentally. The authors propose as a possible explanation that the rheology of microtubule network is not taken into account in the numerical model. Is this a plausible explanation, in particular for the factor 5?  
+
+<|ref|>text<|/ref|><|det|>[[114, 620, 883, 896]]<|/det|>
+We thank the reviewer for raising the concern about the five- factor discrepancy of interface progression coefficient \(P_{I}\) between the model and experiment. In response to this comment, we examined our data more closely and concluded that the discrepancy mainly resulted from using an incorrect ATP diffusion coefficient. In the previous version of this manuscript, the model used the diffusion coefficient of ATP in water, because the base of the microtubule- kinesin active fluid is water. However, Gagnon et al. showed that the microtubule network has a viscosity greater than water (PRL 125, 178003, [2020]), which implies that the actual diffusion coefficient of ATP should be lower than its value in water. To better estimate the diffusion coefficient of ATP in our microtubule network, we performed extensive additional studies to explore the diffusion coefficient in inactive microtubule- kinesin fluid. We were not able to track ATP directly, so as an alternative we tracked fluorescein, which we could visualize with fluorescent microscopy. We found that the diffusion coefficient of fluorescein in the inactive microtubule network was one- fifth the value reported for aqueous solution (Supplementary Discussion 1), which suggested that the diffusion coefficient of ATP in our active fluid is also 5 times lower than the value reported in pure water. After making this correction, we found that our model was consistent with experimental observation (with \(\sim 10\%\) relative difference; Fig. 3d). We revised the Fick’s law model section in the manuscript (lines 100- 133) and added a description of how we estimated the diffusion coefficient of ATP in our active fluid system
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 881, 123]]<|/det|>
+(Supplementary Discussion 1). We thank the reviewer for questioning about the weakness in our model which drove us to strengthen our model and make our finding more robust.  
+
+<|ref|>text<|/ref|><|det|>[[115, 133, 882, 255]]<|/det|>
+(6) The authors emphasize that the mixing efficacy of the nonuniform active fluid systems depends on the distribution of activity, with systems consisting of more small active areas evolve to a homogeneous state faster than systems with the same total active area distributed as one piece. They conclude that the activity-uniform active fluid has the highest mixing performance. However, this is exactly the behavior I would expect for a passive systems: As interfaces propagate diffusively, shorter distances can be covered more quickly. Thus, what is the significance of the active process in this general conclusion?  
+
+<|ref|>text<|/ref|><|det|>[[115, 264, 883, 540]]<|/det|>
+We thank the reviewer for questioning the significance of fluid activity in the checkerboard work. We performed the checkerboard experiments because throughout our investigation on active- inactive fluid system, we only focused on one configuration of activity distribution: One side activated and the other side inactive. Thus, we were curious about how the mixing process would have been different if the activity had been distributed in a different arrangement. In response to the reviewer's comment, we further explored the checkerboard mixing system via our active fluid simulation (Fig. 8) and compared the mixing of the active and inactive checkerboard- pattern systems, to delve more deeply into this topic and provide more relevant findings in the manuscript. Surprisingly, our simulation showed that the mixing time of the active fluid system depended less on the grid size than the mixing time of an inactive fluid system (Fig. 8b). Increasing the dimensionless grid size from 2 to 22 increased the mixing time of inactive fluid system by a factor of 40, whereas in the active fluid system, the mixing time was only increased by a factor of 3. Such drastic difference showed that the mixing efficacy of active fluid is less sensitive to changes in the initial condition of the mixture. We added these new checkerboard simulation results to Results (lines 261- 268), Discussion (lines 307- 319), and Methods (lines 516- 529) along with accompanied figure (Fig. 8). We thank the reviewer for questioning about the intellectual merits of our checkerboard work which drove us to delve into the checkerboard mixing systems and enriched the checkerboard work in our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 550, 880, 567]]<|/det|>
+(7) Arguments about maximizing system entropy are difficult because this a non-equilibrium active system.  
+
+<|ref|>text<|/ref|><|det|>[[115, 577, 697, 594]]<|/det|>
+We thank the reviewer for the correction. We removed arguments about entropy.  
+
+<|ref|>text<|/ref|><|det|>[[115, 605, 492, 621]]<|/det|>
+(8) The following references might also be relevant:  
+
+<|ref|>text<|/ref|><|det|>[[115, 632, 568, 649]]<|/det|>
+-- A. Doostmohammadi et al., Nat. Commun. 8, 15326 (2017);  
+
+<|ref|>text<|/ref|><|det|>[[115, 660, 531, 676]]<|/det|>
+-- G.A. Vliegenthart et al., Sci. Adv. 6, eaaw9975 (2020);  
+
+<|ref|>text<|/ref|><|det|>[[115, 688, 475, 704]]<|/det|>
+-- H. Reinken et al., Commun. Phys. 3, 76 (2020);  
+
+<|ref|>text<|/ref|><|det|>[[115, 715, 465, 731]]<|/det|>
+-- R. Alert et al., Nature Physics 16, 682 (2020);  
+
+<|ref|>text<|/ref|><|det|>[[115, 742, 434, 759]]<|/det|>
+-- K. Qi et al., Commun. Phys. 5, 49 (2022).  
+
+<|ref|>text<|/ref|><|det|>[[115, 769, 883, 803]]<|/det|>
+We thank the reviewer for bringing our attention to these important works. We have cited these works in relevant places in our manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 208, 107]]<|/det|>
+## Reviewer 3:  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 882, 289]]<|/det|>
+The authors have performed mixing experiments on microtubule- kinesin based active nematics by taking advantage of an interesting molecule called "caged ATP". This ATP can be introduced uniformly into the system in its inactive state and then activated remotely via a UV light to give spatial patterning of activity in the active nematic system. They observed diffusion- like and super- diffusive regimes dependent on the speed of the microtubule flows and compare results with some simulations of the system with a variety of quantitative experiments. Beautiful videos demonstrate the concepts very well. The data are fit to a fairly simple transport model that seems to describe the behaviors well. An additional and interesting experiment towards the end of the paper tests checker- board patterns of activation to look at how fast the system reaches homogeneity as a function of active/inactive interfacial length. This section of the paper could probably be a separate publication if explored in more detail.  
+
+<|ref|>text<|/ref|><|det|>[[115, 299, 882, 385]]<|/det|>
+Overall I think that the paper covers an exciting topic and one that should be focused on by the active nematic community. Self- mixing in active nematics represents an important direction in the evolution of this field. I like the hypothesis driven approach and the data is presented in a logical and interesting fashion. A large amount of data is shown with a focus largely on statistical measures such as 1D 'diffusion' constants (i.e. P in this case).  
+
+<|ref|>text<|/ref|><|det|>[[115, 395, 882, 430]]<|/det|>
+I found a few issues in the paper that should be corrected pre- publication but I think that the paper is appropriate for Nature Communications in scope and impact.  
+
+<|ref|>text<|/ref|><|det|>[[116, 440, 394, 457]]<|/det|>
+We thank the reviewer for the support.  
+
+<|ref|>text<|/ref|><|det|>[[115, 467, 882, 501]]<|/det|>
+1. The title of the paper can be improved for clarity - is it missing a comma or colon? I would use "activity" not "activities"  
+
+<|ref|>text<|/ref|><|det|>[[115, 511, 882, 546]]<|/det|>
+We thank for the reviewer's correction. We have changed the title to "Self-mixing in microtubule- kinesin active fluid from nonuniform to uniform distribution of activity."  
+
+<|ref|>text<|/ref|><|det|>[[115, 556, 882, 678]]<|/det|>
+2. I found that the description of the results at the start of that section to be a bit lacking in necessary detail. The authors need to be clearer about the nature of the "initially inactivated fluid" in the first results paragraph. Is it mixed Mts and kinesin but not formed into bundles (random filaments, or small bundles not yet aligned)? Presumably the solution has not yet been in the active nematic state? Are the kinesin clusters bound to MTs yet but the Mts are isotropic? It appears from the video that this is the case and I can guess, but clarification must be added. It seems that initiation of activity is not a reversible process - I.e. once bundled the system will never go back to the initial state (e.g. when ATP runs out). This can also be clarified.  
+
+<|ref|>text<|/ref|><|det|>[[115, 688, 882, 893]]<|/det|>
+We thank the reviewer for highlighting the need to clarify the initial, inactive state of our active fluid system. Our inactive fluid contained microtubules that spontaneously formed bundles by depletion; these bundles were further crosslinked by kinesin motor dimers, forming an elastic gel network. We prepared the inactive gel sample in a test tube where microtubules were expected to orient isotropically, but when we loaded the mixture into the flow cell, shear flow was induced and drove the microtubules to align along the flow cell, so microtubules initially had a preferred alignment along the long edge of the flow cell (Supplementary Fig. 1). After the fluid was activated by ultraviolet light, the microtubule network underwent an irreversible process of becoming a 3D self- rearranging isotropic active gel consisting of extensile microtubule bundles that buckled and annealed repeatedly until the ATP ran out. We have clarified the initial state of microtubule network before UV activation along with various details about the network dynamics such as bundle formation and the irreversibility of the network structure in the main text (lines 55- 77) and added a supplementary figure to show the microtubule alignment right after loading (Supplementary Fig. 1). We
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 883, 176]]<|/det|>
+noticed that Najma et al. recently posted an arXiv reporting the similar initial network structure (Fig. 1c in Najma et al. arXiv: 2112.11364 [2022])) so we also cited the arXiv for readers who wanted to know more details about the initial state of microtubule- kinesin active fluid. We believe that now the readers have access to sufficient information about the initial state of our inactive fluid. We thank the reviewer for the request that increased the clarity of our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 186, 883, 290]]<|/det|>
+3. In the first results paragraph we also need to know where the dyes are located to interpret the videos and Figures easily - is it on the MT? The kinesin or perhaps in the water. This needs to be added in the first results paragraph. How about the tracer particles? Are they coupled to the MTs? How are they confined to the active layer? I realize that there is plenty of detail in the methods section at the end of the paper but the most important points for figure interpretation should be in the main text otherwise it's too difficult to read the paper - especially for the uninitiated general reader.  
+
+<|ref|>text<|/ref|><|det|>[[115, 299, 883, 402]]<|/det|>
+We thank the reviewer for addressing the issue of important methods details in the main text. We added explicit description in the main text that the microtubules were labeled with Alexa 647 (Line 69- 70) and that tracers were freely suspended in solvent (lines 67- 69). Also, we clarified that our system was a 3D microtubule- based active fluid system consisting of self- rearranging isotropic active gel, so there were no active layers like there would be in a 2D active nematic system (lines 64- 67). We thank the reviewer for the suggestion, which increased the readability of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 412, 883, 499]]<|/det|>
+4. The authors don't address the concept of chaotic mixing and advective flows at all in the paper and I found that to be an omission - how does that concept relate to the measures for quantifying mixing in the submitted work? A short discussion might help put this work in context with that recent work cited as ref 24 where chaotic mixing was described for the same system. Some other groups are also considering advective flows in these systems. How would this be related to the present work?  
+
+<|ref|>text<|/ref|><|det|>[[115, 508, 883, 787]]<|/det|>
+We thank the reviewer for the suggestion on discussing the chaotic mixing and advective flows in our manuscript. We realize that our method of quantifying mixing by using the interface progression exponent, \(\gamma\) , and mixing time, \(t_0\) , is not a complete characterization of the mixing dynamics of the active fluid system. Tan et al. characterized the chaotic mixing by introducing topological entropy and Lyapunov exponents. We considered characterizing the mixing dynamics in our system the same way, but soon we realized that it was not practical because our system is different from the one used by Tan et al. Their system was a 2D active nematic system in which the embedded tracers could remain in one focal plane and thus could be tracked for a long period of time, whereas our experimental system was a 3D isotropic active gel where tracers frequently moved out of the focal plane, which prevented us from continuously tracking them. If the tracers could be imaged and tracked in 3D, it would be possible to measure Lyapunov exponents and topological entropies. We are working on developing 3D imaging and tracking; but we have not fully developed the technique at this time. Nevertheless, we agree with the reviewer that measuring Lyapunov exponents and topological entropies would have provided deeper insight into the mixing dynamics of our system from the perspective of system chaotic degree. In the revised Discussion section, we address the omission of chaotic characterization and suggest potential future work characterizing the chaotic degree in the nonuniform active fluid system to gain a deeper understanding of its mixing dynamics (lines 322- 325).  
+
+<|ref|>text<|/ref|><|det|>[[115, 796, 883, 899]]<|/det|>
+In response to reviewer's suggestion on exploring advective flows in our active- inactive fluid system, we adopted a dimensionless quantity, Péclet number, defined as \(Pe \equiv \bar{v}_{ab} l_c / D\) , where \(\bar{v}_{ab}\) represents the mean flow speed of active fluid, \(D\) is the diffusion coefficient of ATP, and \(l_c\) is the correlation length of flow velocity. The physical interpretation of this quantity is the ratio of convective transport rate to diffusive transport rate. When the Péclet number is greater than of order 1, the active transport is dominated by convection, whereas when the Péclet number is smaller than of order 1, the active transport is dominated
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 279]]<|/det|>
+by diffusion. To explore how the behaviors of active transport affect the mixing of an active- inactive fluid system, we first analyzed the interface progression exponent \(\gamma\) as a function of Péclet number (Fig. 4b) and found that as \(Pe \lesssim 3\) , \(\gamma \approx 1\) , which corresponded to the diffusion- like mixing captured in our Fick's law- based model (Fig. 3). Interestingly, as the Péclet number is increased to \(Pe \gtrsim 3\) , we observed \(\gamma\) became greater than 1, which corresponded to the transition in interface progression to being superdiffusion- like as the convection mechanisms started to emerge. Overall, introducing the concept of advection and quantifying it with the dimensionless Péclet number allowed us to understand the progression of active- inactive interface from a more fundamental perspective of material transport. We have revised the manuscript to include the concept of advection to interpret the observed transition of interface progression throughout the manuscript (e.g., lines 146- 161). We believe our manuscript is now more self- explanatory. We thank the reviewer for the suggestion that significantly improved the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[114, 290, 882, 376]]<|/det|>
+I also think that the idea that mixing is driven primarily by defects should be addressed. Can this idea be related to the present work? This was first introduced in theoretical works by Marchetti and shown experimentally in ref 24. Since the submitted paper is not the first to talk about self- mixing in active nematics, a discussion should be added to see how these different papers fit together and can be connected to the submitted work.  
+
+<|ref|>text<|/ref|><|det|>[[114, 386, 882, 524]]<|/det|>
+We thank the reviewer for pointing out a confusion in our manuscript. The work of Tan et al. and Marchetti (such as PRX 9, 041047 [2019]) mainly focuses on 2D active nematic systems with high nematic order in which defect dynamics play a dominant role in mixing. In contrast, our experimental system is a 3D isotropic active gel with nematic order parameter close to zero, in which mixing is mainly driven by extensile microtubule bundles. Because of this fundamental difference in mixing dynamics, we think that our work cannot be directly compared with theirs. To avoid readers misinterpreting our work as the results from a 2D active nematic system, we revised the first paragraph of Results section to clarify that our system is a 3D isotropic active gel.  
+
+<|ref|>text<|/ref|><|det|>[[114, 534, 882, 672]]<|/det|>
+With these being said, we think that it would be more elucidative to compare our work with previous research on 3D isotropic active gels. As such, we compared our works with experimental research by Sanchez et al. (Nature 491, 431 [2012]) and Henkin et al. (Philosophical Transactions. Series A, Mathematical, Physical, and Engineering Sciences 372, 20140142 [2014]) and modeling research by Varghese et al. (PRL 125, 268003 [2020]) and Saintillan and Shelly (Physics of Fluids 20, 123304 [2008]). We believe these comparisons better demonstrate how self- organization and mixing dynamics of 3D isotropic microtubule- kinesin active fluids with nonuniform activity are different from those with uniform activity.  
+
+<|ref|>text<|/ref|><|det|>[[114, 682, 882, 750]]<|/det|>
+5. The results presented in the paper focus on large length-scales (i.e. much larger that the active length scale. Can the authors discuss their results where the length scales approach the active length scale or even go below it? Transport measures may be different on small scales and should be ballistic-like. Can local flows o the scale of defects be tracked to get more detail on these scales?  
+
+<|ref|>text<|/ref|><|det|>[[114, 760, 882, 899]]<|/det|>
+We thank the reviewer for suggesting that we look into the smaller- scale kinematics. Indeed, our results for interface progression transitioning from diffusion- like to superdiffusion- like behaviors would be better elucidated if we could have measured the mean squared displacement (MSD) of tracers across the interface and extract the diffusion exponents (a), like previous studies by Sanchez et al. Nature 491, 431 (2012). We attempted to perform such experiments and analyses; however, we soon learned that it was difficult to realize because, unlike Sanchez et al.'s work where the active fluid had steady uniform activity, our system had a dynamic active- inactive interface whose position and width changed with time. Such a time- varying interface prevented us from collecting MSD of a tracer at a fixed activity level of the interface because the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 883, 194]]<|/det|>
+tracer initially at the diffusion zone (low ATP concentration) of the interface may later be in the superdiffusion zone (high ATP concentration) as the interfaces passed by so it would be difficult to distinguish between the diffusive and superdiffusive data, not to mention to analyze the corresponding diffusion exponents \((a)\) . Thus, we think that investigating the tracer motions at small length scales of the active- inactive interface in our system was impractical, at least with the tools and methods we currently have.  
+
+<|ref|>text<|/ref|><|det|>[[115, 203, 883, 375]]<|/det|>
+However, we agree with the reviewer that such a small length- scale study would provide a deeper insight into the active transport of active fluid at the active- inactive interface from the perspective of microscopic kinematics. Perhaps such a characterization can be better realized in an active fluid system whose active- inactive interface does not change with time, which would make it possible to measure tracer MSD at a fixed activity level of the interface and reveal how the tracer behaviors change across the interface. Such an experimental system could be established by adopting active fluid systems that are only activated upon light exposure and become inactive when the light is turned off (such as the systems developed by Ross et al. Nature 572, 224 [2019] and by Zhang et al. Nature Materials 20, 875 [2021]) because one can use such active fluid systems to create a steady, time- independent activity gradient by applying a fixed light intensity gradient.  
+
+<|ref|>text<|/ref|><|det|>[[115, 385, 883, 437]]<|/det|>
+We thank the reviewer for the insightful suggestions. We have revised the manuscript to clarify the limitations of our studies and suggested experiments for potential future work (lines 326- 338) for readers who are interested in learning about the microscopic kinematics of fluid flows at active- inactive interface.  
+
+<|ref|>text<|/ref|><|det|>[[115, 446, 883, 481]]<|/det|>
+6. It appears that Fig 3d might fit to Michaelis-Menten kinetics, tt looks like the trend is approximately there. Is that trend expected for this system? Comment on the shape of the curve and prior work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 491, 883, 789]]<|/det|>
+We thank the reviewer for sharing the keen observation. When we originally prepared the manuscript, we did not think the interface progression coefficient would follow the Michaelis- Menten trend because we thought it was irrelevant, but after we followed the reviewer's suggestion to fit \(P_{I}\) vs. \(C_{0}\) to the Michaelis- Menten equation (Supplementary Fig. 4a), we found out that we were wrong because \(P_{I}\) vs. \(C_{0}\) was well fit to the equation with goodness of fit \(R^{2} \geq 0.99\) (Supplementary Fig. 4b). In fact, we tried ten other flow speed- ATP relations and found that in each relation, \(P_{I}\) vs. \(C_{0}\) followed the corresponding ATP dependence (fit curves in Supplementary Fig. 4a). This is an unexpected result; we never thought that two such unrelated ATP dependences— \(P_{I}(C_{0})\) and \(\bar{\nu} (C)\) —were connected in our model. To explore whether there is an underlying algebra that connect \(P_{I}\) and \(C_{0}\) via a Michaelis- Menten equation, we derived an analytical expression for the interface progression coefficient as a function of initial ATP concentration \(P_{I}(C_{0})\) (Eq. S7 in Supplementary Discussion 3), which reproduced the numerical results (see magenta curve and red dots in Fig. 3d). This expression shows that \(P_{I}\) was connected to \(C_{0}\) via an inverse complementary error function. Coincidentally, this functional form is extremely well approximated by a Michaelis- Menten- type equation \((R^{2} \geq 0.99)\) . We have summarized this finding in Supplementary Discussions 2 and 3 for readers who are interested in how the choice of flow speed- ATP relation \(\bar{\nu} (C)\) affects the resulting interface progression coefficient \(P_{I}(C_{0})\) . We thank the reviewer for sharing with us this interesting perspective in our interface progression coefficient that made our manuscript more inspiring.  
+
+<|ref|>text<|/ref|><|det|>[[115, 799, 883, 833]]<|/det|>
+7. In all the Figures the captions need to make it much clearer which panels are calculations and which are experimental data. The reader should not have search around for this information.  
+
+<|ref|>text<|/ref|><|det|>[[115, 843, 883, 877]]<|/det|>
+We thank the reviewer for pointing out this lack of clarity in our figure captions. To address this concern, we explicitly stated whether each figure shows experimental or simulation results in the beginning of each
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 883, 124]]<|/det|>
+figure caption. In Fig. 3, where simulations and experiments were both presented for comparison, we specified in each panel whether the data shown was simulation or experimental results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 133, 883, 168]]<|/det|>
+Line 241 - You can't use "turbulences" - this should phrased better - do you mean vortices? Areas of "active turbulence". Please clarify.  
+
+<|ref|>text<|/ref|><|det|>[[115, 178, 883, 247]]<|/det|>
+We thank the reviewer for highlighting the confusion in our manuscript. We have changed the phrase "active turbulence" to "chaotic, turbulence- like mixing flows" to improve the readability of our manuscript. We thank the reviewer for the suggestion, which has made our manuscript more understandable and clearer to a wider range of readers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 257, 414, 274]]<|/det|>
+Fig S3 - the plural of spectrum is spectra.  
+
+<|ref|>text<|/ref|><|det|>[[115, 284, 884, 319]]<|/det|>
+We thank the reviewer for their correction. We have corrected the spelling mistakes in the caption of Supplementary Fig. 11.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 83, 348, 99]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 115, 450, 132]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 148, 776, 214]]<|/det|>
+In their rebuttal letter, the authors have responded in detail to all points raised in my previous report. They have modified and extended their manuscript accordingly. Thus, I support the publication of the manuscript in its present form.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 245, 450, 262]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 278, 780, 310]]<|/det|>
+I am satisfied with the changes, they have greatly improved the manuscript and I recommend this work be accepted.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 208, 106]]<|/det|>
+## Reviewer 2:  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 882, 168]]<|/det|>
+In their rebuttal letter, the authors have responded in detail to all points raised in my previous report. They have modified and extended their manuscript accordingly. Thus, I support the publication of the manuscript in its present form.  
+
+<|ref|>text<|/ref|><|det|>[[116, 179, 395, 196]]<|/det|>
+We thank the reviewer for the support.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 234, 208, 250]]<|/det|>
+## Reviewer 3:  
+
+<|ref|>text<|/ref|><|det|>[[115, 261, 882, 295]]<|/det|>
+I am satisfied with the changes, they have greatly improved the manuscript and I recommend this work be accepted.  
+
+<|ref|>text<|/ref|><|det|>[[115, 305, 462, 321]]<|/det|>
+We thank the reviewer for the recommendation.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a47fcbb94ebf731d9744ffcd0c2f044bbe9fa22088e2566de6ce29aba703cde8/images_list.json

@@ -0,0 +1,160 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure R1. Upgraded chemical encryption.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure R2. a-b The ridge-like morphology of the Au network. c The ridge-like polymer wrinkle (Adv. Funct. Mater. 31, 2106754, 2021).",
+    "footnote": [],
+    "bbox": [
+      [
+        240,
+        532,
+        803,
+        730
+      ]
+    ],
+    "page_idx": 4
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure R3. The full-grayscale distribution from 0-255.",
+    "footnote": [],
+    "bbox": [
+      [
+        234,
+        70,
+        800,
+        208
+      ]
+    ],
+    "page_idx": 5
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure R4. The response images with different sizes (left to right, 1000, 750, 500 and 250). One physical feature is extracted as a demonstration.",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        424,
+        825,
+        595
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Figure R5. a The training procedures of the base classification model (TRAIN_Set and VAL_Set). b The training time of the classification model as a function of the newly added 200 PUFs, showing a gradually stable training epoch below 40. The inset is the plot of training and validation accuracy of the classification model with the first newly added image. The training stops when the validation accuracy reaches at the threshold of 95%. \\(R_{c}\\) represents the accuracy. c Correct/Wrong validation ratio and invalidation ratio of the 37000 images as a function of similarity threshold. The ideal maximum of the correct accuracy is 0.7, as the whole test dataset contains 30% fake images. The 0% wrong validation can be achieved beyond the threshold of 0.5.",
+    "footnote": [],
+    "bbox": [
+      [
+        108,
+        518,
+        888,
+        661
+      ]
+    ],
+    "page_idx": 6
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Figure R6. a Heat map showing the cross-correlation values. b Histogram showing the distribution of cross-correlation values obtained from the heat map.",
+    "footnote": [],
+    "bbox": [
+      [
+        187,
+        73,
+        850,
+        262
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Figure R7. The 600 images for cross-correlation calculations.",
+    "footnote": [],
+    "bbox": [
+      [
+        97,
+        70,
+        900,
+        870
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Figure R8. a-c Three different kinds of percolation networks/clusters with fractal characterizations and d the approximate branching fractal model extracted from one percolation network-based PUF tag in our work.",
+    "footnote": [],
+    "bbox": [
+      [
+        213,
+        177,
+        780,
+        564
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Figure R9. The grayscale distribution before a and after b the image preprocessing of grayscale stretch.",
+    "footnote": [],
+    "bbox": [
+      [
+        225,
+        274,
+        805,
+        562
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Figure R5. a The training procedures of the base classification model (TRAIN_Set and VAL_Set). b The training time of the classification model as a function of the newly added 200 PUFs, showing a gradually stable training epoch below 40. The inset is the plot of training and validation accuracy of the classification model with the first newly added image. The training stops when the validation accuracy reaches at the threshold of 95%. \\(R_{c}\\) represents the accuracy. c Correct/Wrong validation ratio and invalidation ratio of the 37000 images (TEST_Set) as a function of similarity threshold. The ideal maximum of the correct accuracy is 0.7, as the whole test dataset contains 30% fake images. The 0% wrong validation can be achieved beyond the threshold of 0.5.",
+    "footnote": [],
+    "bbox": [
+      [
+        101,
+        500,
+        888,
+        642
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_10.jpg",
+    "caption": "Figure R1. Calculation of practical grayscale intensity level of the original PUF image. a Grayscale image of a typical PUF pattern. b Grayscale distribution histogram of a typical PUF pattern. c Grayscale distributions from 10 randomly selected PUF images in the basic database. d Maximums of feature grayscale and background grayscale from the 10 PUF images in c.",
+    "footnote": [],
+    "bbox": [
+      [
+        210,
+        66,
+        777,
+        355
+      ]
+    ],
+    "page_idx": 22
+  }
+]

peer_reviews/supplementary_0_Peer Review File__a47fcbb94ebf731d9744ffcd0c2f044bbe9fa22088e2566de6ce29aba703cde8/supplementary_0_Peer Review File__a47fcbb94ebf731d9744ffcd0c2f044bbe9fa22088e2566de6ce29aba703cde8.mmd

@@ -0,0 +1,578 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Random fractal- enabled physical unclonable functions with dynamic AI authentication  
+
+![](images/Figure_unknown_0.jpg)
+  
+
+Open Access This file is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. In the cases where the authors are anonymous, such as is the case for the reports of anonymous peer reviewers, author attribution should be to 'Anonymous Referee' followed by a clear attribution to the source work. The images or other third party material in this file are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
+
+<--- Page Split --->
+
+## Reviewers' comments:  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors present a random fractal- enabled PUF based on non- deterministically formed gold network patterns. Authentication is performed using graphical/imaging techniques aided by a dynamic deep- learning system and Raman signatures are shown as an added layer of uniqueness/complexity.  
+
+While I enjoyed reading this manuscript and find the investigation very thorough, and of high quality scientific rigor, several issues/concerns currently preclude recommending this work for publication in Nature Communications.  
+
+Regarding the PUF:  
+
+1A.) The authors claim "We have developed an unbreakable anti- counterfeiting system" which relies on a complex Au pattern/topography. However it's not clear why such a system is unclonable. It seems that an attacker could simply take an image of the tag and intentionally fabricate their own copy of that particular pattern. Assuming the Response images are binarized as indicated in Supplementary Figure 9, it seems there is a clear vulnerability to the proposed platform. Even if the response images are not binarized, but maintained in grayscale to maintain a high encoding capacity it's not clear if the proposed approach would reject a sophisticated forgery while still authenticating genuine devices.  
+
+1B.) There does not appear to be much "grayscale" information in each tag - - most of the features seem to be binary in nature (gold vs. no gold). From this perspective the present work does not seem like a significant advancement over prior literature (References 7- 11) where wrinkling/crumpling provide features with dynamic grayscale information. To merit publication in Nature Communications, this work would need to present a substantial advance relative to the prior art - - however, such an advance does not appear to be present or convincing.  
+
+2.) While the nanoscale roughness and Raman activity do add enhanced security - - and likely the Raman map could not be cloned - - the primary focus and most scalable implementation appears to be on the image based authentication. Despite the increased prevalence of handheld Raman readers - - a microsclared Raman mapping is significantly slower, more costly, and more challenging to implement in practice. The reliability of the Raman enhancement would also become a concern to address.  
+
+Regarding the AI authentication:  
+
+3.) The setting of the model training is not convincing. The model architecture shown in Supplementary Figure 19 is over- parameterized for a training dataset of 2700 images. The choice of this architecture needs to be justified.  
+
+4.) \(100\%\) validation accuracy is typically a red flag in machine learning applications. It is likely that the validation data do not separate well from training data. If the training accuracy is also high, the model is highly overfitted, which is actually expected under such an over- parameterized model. As a result, the model would not have high generalizability.  
+
+5.) The proposed dynamic database strategy is similar to the method in incremental learning or online learning. It is unclear whether the new images or the entire dataset is used to fine- tune the FC layers. Nevertheless, it is well known that such a method may suffer from catastrophic forgetting, especially when you start with only 20 PUF tags. The authors are recommended to show the test accuracy of the first 20 PUF tags after adding a large number of tags, e.g., another 20, to show the effectiveness of the proposed method.  
+
+6.) Conversely, using AI- based authentication indicates that the PUF can be modeled. Discussions
+
+<--- Page Split --->
+
+are required regarding how the designer can ensure the machine learning model cannot be captured by the adversary and then used to compromise the authentication systems, especially given that autoencoder can be easily used as a generative model.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The paper proposes a method to fabricate unclonable tags for anti- counterfeiting, based on the annealing of Au thin films of various thickness (being the major parameter to control the image properties of the tag). This work belongs to an ever increasing group of papers targeting at methods for fabrication of unclonable tags for object authentication. In fact one could classify these methods in those based on some kind of fluorescence after being irradiated with laser sources of proper wavelength and those exhibiting a static image with unique properties. The approach proposed by the paper belongs to the second category. The basic methodology for characterizing a "PUF" is statistical, targeting at evaluating its performance in terms of "robustness" and "unclonability" as discussed in many published papers on PUFs. Robustness is evaluated by interrogating the random structure under all different conditions that could take place in reality (illumination, temperature, surface dirt, any kind of surface degradation, etc.) and calculate the correlation of the obtained images, while unclonability is evaluated by calculation the cross correlation of different tags generated under different fabrication conditions.  
+
+Since it is a statistical analysis, it requires a very large number of samples (e.g. at least a few tens of thousands). This is my major objection with this paper. The authors fabricated only 20 samples, a number absolutely insufficient to evaluate their structures as "PUF" as it appears in fig. 3b of the paper.  
+
+Moreover, the authors provide a number of claims to support the random nature of the fabricated tags. Starting with the major claim of "fractal structures" of the formed patterns (called "networks") in the paper, from the figures of the fabricated patterns they comprise of dendrite structures with different number of branches but to my opinion these patterns do not exhibit statistical self similarities between local and whole geometries appearing at different scales as defined by Mandelbrot in his pioneering work (ref. 27 of the paper). And in any case no mention should be made on "Chaos" and "Chaotic traits" (page 64,65 of the paper.  
+
+Going to the calculation of the encoding capacity of 10630 (line 243 of the paper) this comes from the assumption that two different "PUFs" can differentiate by one pixel (out of 10 6) and one grayscale tone (out of 255). Obviously, this number is totally unrealistic taking into account the actual properties of the imaging unit (optical interrogator) in terms of modulation transfer function of the optics or the noise of the detection unit, etc. Moreover such minor differences would result in very high cross correlation between the corresponding patterns, reducing radically the number of usable uncorrelated tags.  
+
+Concerning the authentication method, as the authors mention, a conventional image processing analysis could be used to extract the image features of the tag. The authors have developed a modified deep learning scheme for this purpose. Here also the paper suffers from the negligible number of samples. For example, for the generation of the training set they used fifteen out of the twenty sample patterns and they enriched them with additional patterns coming from rotation of the initial 15 patterns by one degree. I'm not sure if this is the best training approach but again a deep learning system based on a total of 20 samples is out of the question. Moreover, I'm wondering what are the benefits of the ML method compared to the conventional image processing methods. Concluding, I cannot recommend publication of the submitted paper to Nature Communications.
+
+<--- Page Split --->
+
+## Response Letter  
+
+We are grateful for the positive comments and helpful suggestions from the reviewers. Our responses to the reviewers' comments and the changes made in our manuscript are as follows. For the sake of readability, our responses are highlighted by blue font and the changes to our manuscript are presented in green.  
+
+## Reviewer #1:  
+
+The authors present a random fractal- enabled PUF based on non- deterministically formed gold network patterns. Authentication is performed using graphical/imaging techniques aided by a dynamic deep- learning system and Raman signatures are shown as an added layer of uniqueness/complexity.  
+
+While I enjoyed reading this manuscript and find the investigation very thorough, and of high quality scientific rigor, several issues/concerns currently preclude recommending this work for publication in Nature Communications.  
+
+We truly appreciate the reviewer for the careful consideration the enthusiastic support of our manuscript.  
+
+## C1: Regarding the PUF:  
+
+1A.) The authors claim "We have developed an unbreakable anti- counterfeiting system" which relies on a complex Au pattern/topography. However it's not clear why such a system is unclonable. It seems that an attacker could simply take an image of the tag and intentionally fabricate their own copy of that particular pattern. Assuming the Response images are binarized as indicated in Supplementary Figure 9, it seems there is a clear vulnerability to the proposed platform. Even if the response images are not binarized, but maintained in grayscale to maintain a high encoding capacity it's not clear if the proposed approach would reject a sophisticated forgery while still authenticating genuine devices.  
+
+R1: We thank the reviewer for proposing this general issue. We really agree with the reviewer that the attacker can pose a great threat to PUF security. First and foremost, due to the trait of mass production in the proposed PUF fabrication, each "fingerprint" tag is allowed to only protect and identify one particular good. Microfabrication (e.g. micro/nano 3D printing technique) of the Au network structures is theoretically possible, but the time and equipment needed to duplicate a mass of anticounterfeiting labels eliminate the risk posed from such a laborious endeavor. Compared with the economic value of the protected mass- market products, the cost spent on the duplication technology makes the duplication itself a bad bargain.  
+
+Additionally, the three- dimensional height information of the network encoded into the PUFs can also be used for anti- clone, while the height information can't be simply obtained by general imaging. The random surface texture and height fluctuation at the nanoscale also increase the difficulty of the full duplication of grayscale details. Although remarkable progress has been made
+
+<--- Page Split --->
+
+in 3- dimensional micro/nano fabrication technologies from now, elaborate architecture at the nanoscale and flexible employment of host materials still remain a big challenge. Noting that the binary images in Supplementary Figure 9 are only used to extract the physical feature and count the filling ratios for encoding capability, rather than to be regarded as the security key for authenticating the goods. The security key can be maintained in grayscale and carried more information, more details are shown in Comment 2.  
+
+Moreover, the conceptually demonstrated chemical tag orthogonal to the physical feature in our manuscript enables the network to be a multidimensional PUF tag with a higher non- replicability. The fabrication of the plasmonic platform can also increase the difficulty of the duplication for attackers. Potentially, an upgraded encryption strategy can be carried out by introducing a random loss of signal hid in the chemical pattern, as shown in Figure R1. By artificially blocking random positions of the physical pattern, the orthogonal chemical pattern would be incomplete. Therefore, even if the physical object has been accurately duplicated by a sophisticated forgery, the corresponding chemical identifier still remains an unpredictable variable (intrinsic information preserved by manufacturers), which can be also responsible for anticounterfeiting. In other words, an accurately replicated chemical pattern can be regarded as the fake instead, guaranteeing an almost unbreakable anticounterfeiting system. Finally, it's worth noting that the developed PUF is not only a 3D physical object in visual, but also a hierarchical complex structure extending from microscale to nanoscale and integrated with multiple responses. This makes the duplication of the hierarchical PUFs an almost impossible project.  
+
+![](images/Figure_unknown_1.jpg)
+
+<center>Figure R1. Upgraded chemical encryption. </center>  
+
+Regarding the authentication system, we can implement the one- time authentication strategy at the user side. After one authentication, the security key can be permanently invalid in the secure server and the security key can be removed from the database, preventing a cloned PUF from being reused.  
+
+Actually, sophisticated forgery remains a common issue for all kinds of PUF labels. Our article mainly demonstrates the concept and methodology of the fractal- enabled PUF system, so we do not make many assumptions about possible loopholes in real- world applications.  
+
+"Microfabrication of similar network structures is theoretically possible, but the cost of a mass duplication of "fingerprint" labels and the intricate manufacturing technology with nanoscale accuracy eliminate the risk posed from such a laborious endeavor."  
+
+"The conceptual presentation of the SERS- based encoding mode reveals that the Au network- based PUF can be employed as the universal plasmonic platform and has the potential for carrying more information and further improving the security level of the label. Multidimensional
+
+<--- Page Split --->
+
+chemical encoding can strongly fight against the sophisticated forgery..."  
+
+C2: 1B.) There does not appear to be much "grayscale" information in each tag — most of the features seem to be binary in nature (gold vs. no gold). From this perspective the present work does not seem like a significant advancement over prior literature (References 7- 11) where wrinkling/crumpling provide features with dynamic grayscale information. To merit publication in Nature Communications, this work would need to present a substantial advance relative to the prior art — however, such an advance does not appear to be present or convincing.  
+
+R2: Thank the reviewer for the important and detailed comments. The response image of \(\mathrm{Si / SiO_2}\) support is truly in a single grayscale, which has been excluded from the encoding capacity calculation. However, our PUFs are not the regular structures, different from the structures produced by the template- assisted film deposition. Oppositely, the PUFs are based on the three- dimensional Au networks spontaneously shrinking from the Au film, in which the curved edges, height fluctuation of the rugged surface (Fig. 2e, f and g in the manuscript), random textures and other topography difference can generate rich information in contrast (grayscale) on the response images due to the different reflection and refraction of light. Therefore, the ridge- like morphology of the Au network is actually similar to the polymer wrinkle, as compared in Figure R2. Moreover, from the grayscale distribution (Figure R3) of one Au network- based PUF, we can see that the response image dose presents a uniform full- grayscale distribution from 0 to 255. As mentioned, it seems to be only two grayscale values (gold vs. no gold) in the response image with the naked eyes. Therefore, we carried out the presentation of grayscale distribution through algorithm. Although the subtle grayscale distribution is difficult to be distinguished with the naked eyes, this can be actually recognized by the algorithm accurately and also be responsible for the credible encoding capacity calculation.  
+
+![](images/Figure_unknown_2.jpg)
+
+<center>Figure R2. a-b The ridge-like morphology of the Au network. c The ridge-like polymer wrinkle (Adv. Funct. Mater. 31, 2106754, 2021). </center>
+
+<--- Page Split --->
+![](images/Figure_unknown_3.jpg)
+
+<center>Figure R3. The full-grayscale distribution from 0-255. </center>  
+
+Besides, we also realize that the \(1000 \times 1000\) pixels of the PUF feature image may be relatively high so that the grayscale information will be unconspicuous. We can flexibly change the different size of the captured images to highlight the grayscale information, although the corresponding encoding capacity will be decreased. The contrast of response images with different sizes (sizes from \(1000 \times 1000\) to \(250 \times 250\) ) is shown in Figure R4. Here we set the size of a single PUF feature image to \(750 \times 750\) for showing a more evident and rational response image in grayscale. Meanwhile, the encoding capacity proportional to the image size can be decreased correspondingly (from \(10^{630}\) to \(10^{353}\) ), while is still beyond the standard encoding value of \(10^{20}\) . Notably, a lower size of image is not desirable, because it will obscure some feature details.  
+
+![](images/Figure_unknown_4.jpg)
+
+<center>Figure R4. The response images with different sizes (left to right, 1000, 750, 500 and 250). One physical feature is extracted as a demonstration. </center>  
+
+We really agree with the reviewer that the wrinkling/crumpling due to the compressive stresses in layered systems can provide more abundant topography feature information than our network- based PUF tags, which can provide a higher encoding capacity. However, a PUF with extremely large encoding capacity can reduce the probability of fabricating two tags with same configuration by a stochastic process and resulting in the same response, which can cause a low reproducibility and readout robustness. Besides, the PUF with dense grayscale information is also hard to be applied to cryptographic primitives owing to their ultrahigh complexity and poor stability.  
+
+The encoding capacity is not the only concern of a PUF system and the focus should not be blindly pursuing a high encoding capacity. The robustness of the PUF system is also the major consideration. In fact, our PUF tag is specifically designed aiming at the shortcomings existing in conventional graphical PUFs, especially the wrinkling/crumpling PUF system. The advances can be shown as follows,
+
+<--- Page Split --->
+
+1) our Au network-based PUF has inherited the high configurability in tag complexity of the flexible wrinkling system, such as the on-demand design on dimensions of physical features and external geometric of the single tag, while having a higher physical robustness in extreme conditions (such as the high temperature durability reaching \(800^{\circ}\mathrm{C}\) ) benefits from the inert-metal's intrinsic stability. The elastomeric polymer often suffers from the physical aging in high temperature, humidity/water, or oxygen. This breaks the trade-off between the high code configurability and relatively low stability in general PUF taggants, which is the main advances compared with the prior art.  
+
+2) the high-throughput production process combined with the mask-assisted UV lithography and film deposition ensures the mass production of the labels. Besides, a better compatibility with the microelectronic process along with a high code configurability and miniaturization broadens the application fields of the PUFs in hardware products.  
+
+3) our PUF can be regarded as the inherent and homogeneous plasmonic platform for multiple response and multiple-level security, which is independent of chemical-synthesis nanoparticles and complex surface modifications necessary for other wrinkling/crumpling system (ACS Appl. Mater. Interfaces 8, 4031-4041, 2016. ACS Appl. Mater. Interfaces 13, 11247-11259, 2021).  
+
+4) the fractal-guided manufacturing strategy also provides a new perspective into the concept design of PUFs and can be universally extended to various material systems at multiple scales.  
+
+"The 3-dimensional feature information can be encoded into every pixel of the PUF pattern, while each pixel is utilized as variables in different grayscale levels (grayscale levels in the range of 0-255, which is presented in a uniform full-grayscale distribution in Supplementary Fig. 8) derived from the structure height-dependent different light reflection and refraction."  
+
+"The encoding capacity of the PUFs is not dominant compared to wrinkling/crumpling system with dynamic grayscale information or random distributed stimuli-responsive taggants with multiple responses. However, a PUF with extremely large encoding capacity can reduce the probability of fabricating two tags with same configuration by a stochastic process (i.e., a low reproducibility) and resulting in the same response. The ultrahigh complexity and poor stability also cause the difficulty of the application in cryptography."  
+
+C3: 2. ) While the nanoscale roughness and Raman activity do add enhanced security — and likely the Raman map could not be cloned — the primary focus and most scalable implementation appears to be on the image based authentication. Despite the increased prevalence of handheld Raman readers — a microscaled Raman mapping is significantly slower, more costly, and more challenging to implement in practice. The reliability of the Raman enhancement would also become a concern to address.  
+
+R3: Thanks for pointing out this concern. As the reviewer said, Raman detection especially the micro- scaled Raman mapping dose has shortcomings in convenience and efficiency. Importantly, the readout time and corresponding readout hardware, as well as the security level of the security key should be commensurate with the economic value and purpose of the product that the tag protects. Actually, this additional chemical encoding mode can be potentially applied to the tailored products, such as the jewelry or antiques. The owner of a luxury product can have a customized authentication during purchase by a staff with handheld Raman readers or even sending to a laboratory.  
+
+Recently, many studies have reported about the ever increasing spectral PUFs based on the
+
+<--- Page Split --->
+
+Raman mapping (Nano Today 41, 101324, 2021. ACS Appl. Mater. Interfaces 13, 11247- 11259, 2021. Nanoscale 12, 9471- 9480, 2020. Nat. Commun. 11, 516, 2020. ACS Appl. Mater. Interfaces 8, 4031- 4041, 2016). The concern of the reading speed has been paid more and more attention by researchers. For example, Yuqing Gu et al. reported a gap- enhanced Raman PUF tags using a high- speed Raman mapping method, which can achieve a Raman mapping- based PUF tag ( \(100 \times 100 \mu \mathrm{m}^2\) ) readout in DuoScan mode in only 6 s with a resolution of \(50 \times 50\) pixels (Nat. Commun. 11, 516, 2020). Nevertheless, the scanning speed of the lab- based confocal Raman system will be further continuously improved for practical use with many strategies and even the Raman mapping can be potentially realized on a hand- held device in the future. Apart from the synthesis of Raman tags with better performance, new Raman imaging modes could be introduced, such as the application of line- shaped (Proc. Natl Acad. Sci. USA. 110, 12408- 12413, 2013), multipoint laser (Anal. Chem. 88, 1281- 1285, 2015) or direct Raman imaging with a narrow- band filter (Nat. Protoc. 8, 677, 2013), or a mode, where the stage movement, light collection and data readout occur continuously and synchronously (Proc. Natl Acad. Sci. USA. 110, 12408- 12413, 2013). In brief, the engineers and industries can help apply it to the products. Actually, our article mainly demonstrates the concept and feasibility of multiple- level security strategies, so we do not have many discussions on the improvements and key problem tackling of general Raman imaging technique.  
+
+We really agree with the reviewers that the reliability of the Raman signal is also the main concern in Raman detection. Except for the influence of the Raman detection device, the uniform and repeatability of the SERS substrate are the prerequisites deciding the reliability of the Raman signal (J. Phys. Chem. C 120, 16946- 16953, 2016). Regarding the nanoparticle- based SERS substrate, the spatial inhomogeneity of the nanoparticle morphology and of the nanoparticle (hot- spots) distribution are the main reasons of fluctuations in SERS intensity, which can cause the loss of the information and influence the robustness of the readout. In our work we introduce a uniform and large- area ion bombardment technique to fabricate the homogeneous SERS substrate. As shown in Fig. 3e in the manuscript, an amount of dense and uniform convex Au nanostructures are generated on the flat surface (Supplementary Fig. 11), which can be used as the "hot spots" of the electromagnetic field enhancement (Supplementary Fig. 12). The Raman mapping in Fig. 3g provides a direct insight into the uniform spatial distribution of SERS response, which shows the reliability of the Raman signal on our SERS substrate. The SERS substrate also has a high fabrication repeatability under the same ion bombardment power and time, ensuring the identical Raman signal intensity and distribution.  
+
+In particular, we would like to highlight that the additional Raman- based chemical encoding in the manuscript can be regarded as a proof- of- concept, which reveals that the Au network- based PUFs can be employed as the universal plasmonic platform and have the potential for carrying more information and further improving the security level of the label. More progress remains to be explored.  
+
+"The conceptual presentation of the SERS- based encoding mode reveals that the Au network- based PUF can be employed as the universal plasmonic platform and has the potential for carrying more information and further improving the security level of the label. Multidimensional chemical encoding can strongly fight against the sophisticated forgery but requires a long readout time and special readout condition. However, with the rapid development of the handheld Raman system with a high- speed readout \(^{23}\) , Raman characterization is potential to be a convenient readout way in the near future."
+
+<--- Page Split --->
+
+C4: Regarding the AI authentication:  
+
+3. ) The setting of the model training is not convincing. The model architecture shown in Supplementary Figure 19 is over-parameterized for a training dataset of 2700 images. The choice of this architecture needs to be justified.  
+
+R4: We thank the reviewer for pointing out this concern. In order to provide a convincing model, we changed two important parts of our AI authentication system:  
+
+1) The architecture of the model is changed to a ResNet50 based classification model pretrained by ImageNet (3.2 million images). The input image is preprocessed in grayscale by modification of the pixel distribution and sent to a ResNet50 based classification model. Some random Gaussian noise was added to the training image in order to limit overfitting and also for covering the influence from the image noise. To reduce the training time and guarantee the generalizability of the model, the pretrained model parameters on ImageNet is used to initialize the ResNet50 and a base model is trained via 1100 PUFs. With a new PUF, only one parameter is added to the last classification layer of the base model, and only this final layer is updated during the training of the new PUF. Noting that both the original PUFs in the database and newly added PUFs are used to update the model to avoid the model forgetting the previous database. The update of only the last layer ensures a fast training procedure. When the fifth new image was added, the training can be finished in 5.48 s (total 40 epochs, 0.137s for each epoch, Figure R5b) in the database expansion test. The results of the database expansion test are shown below.  
+
+2) A lot more PUF images are captured (1850 PUFs): 1) 1300 PUFs for the establishment of the database with 1100 PUFs for building the base database and 200 PUFs for the expansion of the database; 2) 550 PUFs as the fake PUFs out of the database and augmented for training a general model. The TEST_Set contains 37000 images augmented from the original 1850 PUFs, which considers different conditions (brightness, rotations, and noise) that can occur in practice, covering the influence from different capturing conditions and imaging unit. We truly know that not all conditions can be considered in this test, but our work aims to provide a benchmark to prove that our model and technique have the potential application in the PUF authentication.  
+
+The corresponding training/validation/testing results are shown below. According to the results, we can achieve a \(0\%\) wrong validation ratio of the genuine images ( \(0\%\) false positives and \(5\%\) false positives). Noting that the high image invalidation rate comes from the large ratio of the fake image ( \(30\%\) ), which is tested as the invalidation in our model setup. Of course, the further optimization of the model for obtaining a higher accuracy is being implemented. Besides, we have made a list of the different database sizes for AI model training in other PUF works.  
+
+Two added references related to the modified model are as follows,  
+
+[1] ImageNet:  
+
+J. Deng, 
+W. Dong, 
+R. Socher, 
+L. 
+J. Li, Kai Li and Li Fei-Fei, "ImageNet: A large-scale hierarchical image database," 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248-255, doi: 10.1109/CVPR.2009.5206848.  
+
+[2] ResNet50:  
+
+K. He, 
+X. Zhang, 
+S. Ren and 
+J. Sun, "Deep Residual Learning for Image Recognition," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770-778, doi: 10.1109/CVPR.2016.90.
+
+<--- Page Split --->
+
+
+Table 1. The parameters for the AI model training/validation/testing.   
+
+<table><tr><td>Dataset</td><td>Size</td><td>Augmentation</td><td>Labels</td></tr><tr><td>Original images (ORI_Set)</td><td>1850 PUFs</td><td></td><td></td></tr><tr><td>Training set (TRAIN_Set)</td><td>1100 PUFs from ORI_Set</td><td>Rotation: 0, 30, 60, 90, ..., 330°</td><td>Corresponding classes: 0-1099</td></tr><tr><td>Validation set (VAL_Set)</td><td>The same PUFs with TRAIN_Set</td><td>Rotation: 1, 3, 5, 7, ..., 359°</td><td>Corresponding classes: 0-1099</td></tr><tr><td>Candidate set (CAND_Set)</td><td>200 PUFs from ORI_Set</td><td>Rotation: 0, 30, 60, 90, ..., 330° for training and 1, 3, 5, 7, ..., 359° for validation</td><td>Corresponding classes: 1100-1299</td></tr><tr><td>Fake set (FAKE_Set)</td><td>550 PUFs from ORI_Set</td><td rowspan="2">1) Rotation: 10 angles are randomly selected from those that are not considered in the TRAIN_Set and VAL_Set.<br>2) Gaussian noise for each image<br>3) 10 different brightness conditions applied to each image</td><td rowspan="2">All labelled as “FAKE”<br>Corresponding classes: 0-1299 for the first 1300 PUFs in the database, and “FAKE” for the left 550 PUFs</td></tr><tr><td>Test set (TEST_Set)</td><td>1850 PUFs of ORI_Set with augmentations that generate 37000 images</td></tr></table>  
+
+![](images/Figure_unknown_5.jpg)
+
+<center>Figure R5. a The training procedures of the base classification model (TRAIN_Set and VAL_Set). b The training time of the classification model as a function of the newly added 200 PUFs, showing a gradually stable training epoch below 40. The inset is the plot of training and validation accuracy of the classification model with the first newly added image. The training stops when the validation accuracy reaches at the threshold of 95%. \(R_{c}\) represents the accuracy. c Correct/Wrong validation ratio and invalidation ratio of the 37000 images as a function of similarity threshold. The ideal maximum of the correct accuracy is 0.7, as the whole test dataset contains 30% fake images. The 0% wrong validation can be achieved beyond the threshold of 0.5. </center>
+
+<--- Page Split --->
+
+
+Table 2 The database size for AI model training in different works of PUFs.   
+
+<table><tr><td>PUF type</td><td>Database size for AI model training</td><td>Reference</td></tr><tr><td>Crumpling of 2D Materials</td><td>500</td><td>Matter 3, 2160–2180, 2020.</td></tr><tr><td>Drop-casting nanoparticle tags</td><td>1020</td><td>ACS Nano 15, 2901–2910, 2021.</td></tr><tr><td>Laser-induced polymer wrinkle</td><td>112</td><td>Adv. Mater. 32, 2003032, 2020.</td></tr><tr><td>Multi-functional nanoinks</td><td>100</td><td>Nano Today 41, 101324, 2021.</td></tr><tr><td>Random wrinkles</td><td>6</td><td>ACS Appl. Mater. Interfaces 13, 27548–27556, 2021.</td></tr><tr><td>Inkjet-printed unclonable quantum dot</td><td>6</td><td>Nat. Commun. 10, 2409, 2019.</td></tr><tr><td>Inkjet-printed unclonable quantum dot</td><td>10</td><td>ACS Appl. Mater. Interfaces 13, 15701–15708, 2021.</td></tr><tr><td>Unclonable photonic crystal hydrogels</td><td>10</td><td>ACS Appl. Mater. Interfaces 14, 2369–2380, 2022.</td></tr><tr><td>Our work</td><td>1300</td><td></td></tr></table>  
+
+The further optimization of the model (for getting a higher accuracy) and testing of more images with different conditions are underway (images with different scaling and blurring are to be tested). As for the database expansion test, more new PUFs (another 300 images) are to be supplemented. This part of the works is expected to be finished in two weeks.  
+
+"Figure 4a schematically demonstrates the deep learning- based authentication system of the security label. Firstly, each PUF pattern is captured by the manufacturers by using the microscope. The different images are then preprocessed by modification of the pixel distribution and devoted to a ResNet50 based classification model \(^{47}\) for learning the characteristics of the PUF patterns. To reduce the training time and guarantee the generalizability of the model, the pretrained model parameters on ImageNet \(^{48}\) is used to initialize the ResNet50. Next, the images are classified in a general manner and stored in the database for subsequent authentication."  
+
+"For experimentally demonstrating the above authentication system, 1300 different PUF tags were randomly captured to establish the security key database (parts of PUFs shown in Supplementary Fig. 22). 1100 PUFs of them were used for the establishment of the basic database, and the rest 200 PUFs were used for the key expansion test of the database. Each of the 1100 PUFs were rotated with different angles, forming 211200 images as the dataset for deep learning model training (13200 images with less rotations for training and 198000 images for validation). Every input image is preprocessed by grayscale stretch and added random noise to avoid overfitting."
+
+<--- Page Split --->
+
+"For testing our authentication system, the above 1300 PUF tags (in the database) in different conditions (brightness, rotation angle, and random noise, total of 26000 images, Supplementary Fig. 24) and 11000 images from 550 new PUF tags (not in the database) were captured and uploaded to the trained AI for a test. We investigated the validation ratio as a function of the similarity threshold by testing multiple similarity indexes for obtaining the best threshold of validation. Figure 4c reveals that when the similarity threshold reaches 0.5, we can achieve a 0% wrong validation ratio of the genuine images (that is, the rate of false positives is 0%) and 5% false negatives. In this work, we set the similarity threshold at a value of 0.5, and the authentication of a PUF tag with an encoding capacity of about \(10^{353}\) can be finished in 6.36 seconds."  
+
+C5: 4. ) \(100\%\) validation accuracy is typically a red flag in machine learning applications. It is likely that the validation data do not separate well from training data. If the training accuracy is also high, the model is highly overfitted, which is actually expected under such an over- parameterized model. As a result, the model would not have high generalizability.  
+
+R5: We thank the reviewer for pointing out this problem. The \(100\%\) accuracy came from that the choice of training and validation sets are too similar to each other. Therefore, we consider to make the dataset larger as shown in the above table 1. Additionally, we use less rotation angles in the training set compared with the validation set and apply the gaussian noise during training to avoid the overfitting of the model. We also use the pretrained ResNet on the ImageNet (3.2 million images) to do the initialization of the model parameters for a transfer learning. Based on those modifications and the result in Figure R5a, we believe that our model has a higher generalizability. As shown in Figure R5a, this training process was repeated until convergency (totally 2500 epochs, 51.4 s for each epoch) and the model at the epoch (2250) with the highest validation accuracy \((99.63\%)\) is selected as the final base model.  
+
+"Every input image is preprocessed by grayscale stretch and added random noise to avoid overfitting."  
+
+"This training process was repeated until convergency (totally 2500 epochs) and the model at the epoch of 2250 with the highest validation accuracy \((99.63\%)\) , Supplementary Fig. 23) is selected as the final base model."  
+
+C6: 5. ) The proposed dynamic database strategy is similar to the method in incremental learning or online learning. It is unclear whether the new images or the entire dataset is used to fine- tune the FC layers. Nevertheless, it is well known that such a method may suffer from catastrophic forgetting, especially when you start with only 20 PUF tags. The authors are recommended to show the test accuracy of the first 20 PUF tags after adding a large number of tags, e.g., another 20, to show the effectiveness of the proposed method.  
+
+R6: We totally agree with the reviewer about the forgetting problem of incremental learning. In such case, our previous dataset is not sufficient to verify whether that the model is still effective for largely increasing the size of the database. Therefore, we built a larger dataset (totally 1850 PUFs: 1100 PUFs for the establishment of the base model, 200 PUFs for the expansion of the database and 550 PUFs as the fake image out of the database) with a huge TEST_Set containing 37000 images from the 1850 different original PUFs. With a new PUF, only one parameter is added to the last classification layer of the base model, and only this final layer is updated during the training of the
+
+<--- Page Split --->
+
+newly added PUF. Noting that both the original PUFs in the database and newly added PUFs are used to update the model to avoid the model forgetting the previous database. Here we added 200 new PUFs as the database expansion test, and the expansion results and test accuracy on the TEST_Set are shown in Figure R5b and c.  
+
+"When the fifth new image was added, the training can be finished in 5.48 s (totally 40 epochs, 0.137s for each epoch, Fig. 4b) in the database expansion test. Another 195 images were further added to the deep learning model successively and a gradually stable training epoch below 40 is demonstrated (Fig. 4b),.."  
+
+"With a new PUF, only one parameter is added to the last classification layer, and only this final layer is updated during the training of the new PUF. Noting that both the original PUFs in the database and newly added PUFs are used to update the model to avoid the model forgetting the previous database, and the training of the update model stops until a validation accuracy threshold \(T_{add}\) is reached."  
+
+C7: 6.) Conversely, using AI- based authentication indicates that the PUF can be modeled. Discussions are required regarding how the designer can ensure the machine learning model cannot be captured by the adversary and then used to compromise the authentication systems, especially given that autoencoder can be easily used as a generative model.  
+
+R7: Thank the reviewer for pointing out the issue. Apparently, the autoencoder is super similar to some generative models like variational autoencoder (VAE), and we have also the same concern. Therefore, we changed our model architecture to a ResNet50 based classification model without autoencoder. We truly know that the generalizability of the model can be decreased without the basic autoencoder training, so we built a much larger dataset and add the augmentation strategies to ensure the model is not over- parameterized. As for the concern caused by the adversary, our model has the capacity of filtering the PUFs that are out of the database. Although the adversary could capture the model architecture, it would be difficult to train the same model with the same parameters. Besides, all the PUFs captured by the users are only processed by our own trained model instead of being processed by the model from the adversary. We understand that the adversary has many ways to compromise the authentication system, but the method proposed in our manuscript is expected to be a benchmark to demonstrate the pipeline, so a more accurate and robust model would be trained to handle the problems that occur in practice.  
+
+Finally, we truly thank the reviewer for the helpful suggestions that help us change the architecture of the model.  
+
+## Reviewer #2:  
+
+The paper proposes a method to fabricate unclonable tags for anti- counterfeiting, based on the annealing of Au thin films of various thickness (being the major parameter to control the image properties of the tag). This work belongs to an ever increasing group of papers targeting at methods for fabrication of unclonable tags for object authentication. In fact one could classify these methods in those based on some kind of fluorescence after being irradiated with laser sources of proper
+
+<--- Page Split --->
+
+wavelength and those exhibiting a static image with unique properties. The approach proposed by the paper belongs to the second category. The basic methodology for characterizing a "PUF" is statistical, targeting at evaluating its performance in terms of "robustness" and "unclonability" as discussed in many published papers on PUFs. Robustness is evaluated by interrogating the random structure under all different conditions that could take place in reality (illumination, temperature, surface dirt, any kind of surface degradation, etc.) and calculate the correlation of the obtained images, while unclonability is evaluated by calculation the cross correlation of different tags generated under different fabrication conditions.  
+
+We really appreciate the reviewer for the careful consideration of our manuscript and the significant guidance of the basic concept and research progress of the physical unclonable functions (PUFs). In the current tendency of excessively pursuing particular characteristics (e.g., encoding capacity) or employing diverse taggants, we mainly focus on the comprehensive performance of the PUF system, not just on the propose of a new methodology. We really agree with the reviewer that the statistical property is a key factor in evaluating PUF effectiveness and thank the reviewer for proposing this concern. Therefore, we mainly focused on this concern and revised our manuscript through sample collection, cross correlation calculation and AI model training/validating based on a large sample volume.  
+
+C1: Since it is a statistical analysis, it requires a very large number of samples (e.g. at least a few tens of thousands). This is my major objection with this paper. The authors fabricated only 20 samples, a number absolutely insufficient to evaluate their structures as "PUF" as it appears in fig. 3b of the paper.  
+
+R1: Thank the reviewer for pointing out this basic issue. We are very sorry that we haven't employed enough samples for the calculation of cross correlation values by FSIM, which causing a poor evaluation of the unclonability of PUFs. The random fractal- enabled PUF fabrication process actually has the traits of high randomness and predictability, and the uniqueness has been examined by comparing their topography with each other through optical imaging. The similarity statistics of different PUFs is also very important in quantitative evaluation of the unclonability. Therefore, we have supplemented the fabrication batch and fabricated a larger amount of PUFs. We selected 600 (30 times larger than before) different PUFs for the calculation of cross correlation values to verify the uniqueness of the PUF tag. The corresponding heat map, distribution histogram of the similarity and corresponding PUF patterns are shown below. The statistical results further prove that our PUFs have an absolute uniqueness.  
+
+We have made a comparation of the sample number for similarity calculation in other related researches. It can be revealed that the several hundred samples are sufficient for evaluating the unclonability of the PUFs. Nevertheless, our sample number of 600 is undoubtedly the highest among the current work.
+
+<--- Page Split --->
+![](images/Figure_unknown_6.jpg)
+
+<center>Figure R6. a Heat map showing the cross-correlation values. b Histogram showing the distribution of cross-correlation values obtained from the heat map.</center>  
+
+Table 3 The sample numbers for cross-correlation value calculation in different works of PUFs.   
+
+<table><tr><td>PUF type</td><td>Sample number for similarity comparison</td><td>Reference</td></tr><tr><td>Crumpling of 2D Materials</td><td>296</td><td>Matter 3, 2160–2180, 2020.</td></tr><tr><td>Drop-casting nanoparticle tags</td><td>256</td><td>ACS Nano 15, 2901–2910, 2021.</td></tr><tr><td>Laser-induced polymer wrinkle</td><td>200</td><td>Adv. Mater. 32, 2003032, 2020.</td></tr><tr><td>Multi-functional nanoinks</td><td>20</td><td>Nano Today 41, 101324, 2021.</td></tr><tr><td>Polymer wrinkle</td><td>200</td><td>Adv. Mater. 27, 2083–2089, 2015.</td></tr><tr><td>Chaotic organic crystal phosphorescent patterns</td><td>480</td><td>Adv. Mater. 33, 2102542, 2021.</td></tr><tr><td>Random fluorescent proteins</td><td>30</td><td>Nat. Commun. 11, 328, 2020.</td></tr><tr><td>Randomly distributed fibers</td><td>30</td><td>Nat. Commun. 13, 247, 2022.</td></tr><tr><td>Drop-casting Raman tag</td><td>10</td><td>Nat. Commun. 11, 5543, 2020.</td></tr><tr><td>Folding of plasmonic gel</td><td>100</td><td>ACS Appl. Mater. Interfaces 8, 4031–4041, 2016.</td></tr><tr><td>Our work</td><td>600</td><td></td></tr></table>
+
+<--- Page Split --->
+![](images/Figure_unknown_7.jpg)
+
+<center>Figure R7. The 600 images for cross-correlation calculations. </center>
+
+<--- Page Split --->
+
+C2: Moreover, the authors provide a number of claims to support the random nature of the fabricated tags. Starting with the major claim of "fractal structures" of the formed patterns (called "networks") in the paper, from the figures of the fabricated patterns they comprise of dendrite structures with different number of branches but to my opinion these patterns do not exhibit statistical self similarities between local and whole geometries appearing at different scales as defined by Mandelbrot in his pioneering work (ref. 27 of the paper). And in any case no mention should be made on "Chaos" and "Chaotic traits" (page 64,65 of the paper).  
+
+R2: Thank the reviewer for the insightful consideration. Actually, fractal structures have a wide range from earth science (coastline, river and mountain) to experiment synthesis (electrochemical deposition, metal- induced crystallization and dielectric breakdown). One most widely studied class of fractal surface structures is that of deposited coatings. Agglomeration by diffusion, ballistic impact, and chemical or electrochemical processes often result in self- similar structures.  
+
+Percolation network/cluster is also a typical fractal structure with statistical self- similarities, such as the self- assembled fractal gold nanostructure via wet chemistry synthesis method (Nano Lett. 18, 3593- 3599, 2018) (Figure R8a), fractal clusters in thin gold films through film deposition (Phys. Rev. Lett. 49, 1441- 1444, 1982. Phys. Rev. Lett. 49, 1444- 1447, 1982) (Figure R8b) and Au film annealed- induced random islands after de- percolation (Nano Lett. 20, 3291- 3298, 2020) (Figure R8c). Percolation is a process during which unconnected clusters grow and fill a system. Percolation theory describes connectivity of objects within a network structure and the effects of this connectivity on the macroscale properties of the system, which is widely used to describe the stochastic geometry system like fractals.  
+
+Random fractal Au islands have been commonly prepared by annealing of Au films with different thicknesses below the percolation threshold (a given filling fraction of gold reaching the whole connection of the gold islands, J. Phys. Chem. C 117, 11337- 11346, 2013. Nano Lett. 20, 3291- 3298, 2020). The annealing of the Au film is the process of de- percolation and the isolated ramified Au networks described in our manuscript (Figure R8d) belongs to the isolated percolation clusters/networks with different percolation correlation lengths (percolation cluster dimension, L). In percolation, any clusters of the order of L or larger is argued to be a self- similar object up to length L (Phys. Rev. Lett. 49, 1444- 1447, 1982), which means that the different percolation networks with different fractal orders and percolation correlation lengths (Fig. 1d in the manuscript) in a single PUF are self- similarities.  
+
+The construction of a Cayley tree is by straightforward iterative function. Upon each iteration, two new branches are added to each terminal branch. Similar to the Cayley tree model, the network can be regarded as an iterated quasi- Y shape object, and the branching fractal- Au networks are self- similar to itself in bifurcations (from main bifurcation to sub- bifurcation). The difference is that the percolation networks are not in precise self- similarities, which means that the branch orientations and lengths are random varied and the symmetry is broken (Fig. 1d in the manuscript). However, the multiple iterated bifurcations are in statistical self- similarities. For example, in Figure R8d, the part 1 (whole) and the derived part 2 (part) have the similar quasi- Y shape feature and iterative trend. Besides, the parts of the network have the same complexity with the whole network, both showing high randomness.  
+
+Thank the reviewer again for pointing out that the use of "chaos" or "chaotic traits" is wrong
+
+<--- Page Split --->
+
+here. The chaos in physics may refer to the dynamic process, while the fractal in mathematics pay more attention to the description of morphology and geometry. Therefore, we deleted this corresponding expression in our manuscript.  
+
+We are very sorry that we did not illustrate clearly about the concept of the percolation network, as a typical fractal structure, in the initial manuscript. We have further elaborated this theory in the revised manuscript.  
+
+![](images/Figure_unknown_8.jpg)
+
+<center>Figure R8. a-c Three different kinds of percolation networks/clusters with fractal characterizations and d the approximate branching fractal model extracted from one percolation network-based PUF tag in our work. </center>  
+
+"Fractal theory is also used to elucidate the complex surface morphology evolution of diverse thin film systems, such as the fractal- guided percolation networks/clusters of films<sup>28,29,30</sup>. Percolation network/cluster refers to a system in global connectivity through a continuous "chain" of locally connected objects, such as the self- assembled Au nanoframeworks<sup>28</sup>, Au clusters through film deposition<sup>29</sup>, and film annealed- induced Au islands<sup>31</sup>."  
+
+"The branching fractal model is exhibited in Fig. 1d. A single PUF tag is composed of distinct fractal objects, and several iterated bifurcations make up one fractal. Upon each iteration, two new branches are added to each terminal branch. The branch orientations and lengths of the fractals are randomly varied and the symmetry is broken, but exhibiting statistical feature similarity in different bifurcations, as shown in one quasi- Y shape iterated network in Fig. 1d."  
+
+C3: Going to the calculation of the encoding capacity of 10630 (line 243 of the paper) this comes from the assumption that two different "PUFs" can differentiate by one pixel (out of 10 6) and one grayscale tone (out of 255). Obviously, this number is totally unrealistic taking into account the
+
+<--- Page Split --->
+
+actual properties of the imaging unit (optical interrogator) in terms of modulation transfer function of the optics or the noise of the detection unit, etc. Moreover such minor differences would result in very high cross correlation between the corresponding patterns, reducing radically the number of usable uncorrelated tags.  
+
+R3: Thank the review for pointing out this concern. However, signal interference from the imaging unit is a common issue for nearly all kinds of image- based PUF labels, which has been gradually improved through the optimization of current imaging technologies. The signal interference (decrease of image pixel number or value, gaussian noise, blurring, etc.) could be caused by the instability of the CCD or CMOS sensor in the imaging unit, which can't be completely avoided. But here we would like to respectively illustrate the coping strategy from the label itself and authentication system design.  
+
+Firstly, we realize that the image sizes of \(1000 \times 1000\) can't properly match with the actual response image. Therefore, we rationally decrease the sizes to \(750 \times 750\) with an encoding capacity of \(10^{353}\) (the basic encoding capacity shall be \(10^{20}\) or larger for PUF system). It's worth noting that the encoding capacity of \(10^{353}\) is indeed an ideal maximum without the consideration of the imaging interference.  
+
+The image pixel number, or the image resolution is the key factor in encoding capacity calculation. With a decreased image resolution caused by the external interference, an encoding capacity can be decreased if the physical object size exceeds pattern resolution (Nat. Rev. Chem. 1, 0031, 2017). First and foremost, the tag must be physically robust and the feature information must be sufficiently dense so that each PUF key is truly unique and immune to the decrease of resolution caused by the imaging unit. Firstly, our Au- based PUF tag is physically robust against extreme condition, such as high temperature, mechanical abrasion, and contamination. More importantly, we can flexibly improve the encoding capacity by increasing the Au network density in a single PUF, as shown in Fig. 2a- d and 3c in the manuscript. The unlimited density of the physical feature can meet the challenge of the loss of pixels.  
+
+Besides, the image noise and contrast can influence the feature extraction of our PUFs, which can finally influence the pattern filling ratio in encoding capacity calculation. In terms of the authentication system, those conditions can be implemented via image processing methods before training and validating. Specifically, deep learning methods have the capability of automatic denoising and deblurring, like the Nonlinear Activation Free Network (NAFNet) concept proposed by Sun, J. et al. For the end- user, too extreme influence form those conditions can scarcely occur, as our system only generates authentication result until a desirable image is acquired. A prompt asking for the end- user to keep the phone steady may be necessary. In our authentication system, if the uploaded image is invalidated, the client is reminded by the software to capture the original PUF again with a better imaging quality for further identification. Importantly, we have improved the AI training model to cover the influence of gaussian noise and further blurring through image preprocessing, the corresponding test results are shown in Comment 4. Besides, in order to avoid the loss of grayscale value (0- 255) and the decrease of image contrast caused by the imaging unit, we preprocessed the PUF image via grayscale stretch before the AI model training (shown in Figure R9), so that we can fully utilize the grayscale values from 0 to 255 in the encoding capacity calculation. This can also cover the influence of different brightness and contrast of the images in the following PUF authentication.  
+
+Moreover, the feature information in one PUF is extremely dense due to the indeterministic
+
+<--- Page Split --->
+
+production process. Feature similarity algorithm (FSIM) is based on feature point matching and very sensitive to the image pixel distribution, brightness, contrast, rotation, etc. Under the basic condition of the image resolution that makes the network structure distinguishable, it's almost impossible to generate a high cross correlation between two entirely different PUFs by FSIM, although one captured image may have local pixel distortion or loss due to the poor imaging quality. This is because the PUF pattern has dense pixel distribution and rich grayscale information. Besides, the mass production trait of the PUFs can ensure the tolerance of the poor image quality and keep a high yield.  
+
+"Noting that the captured image size and contrast may be decreased by inevitable interference from imaging unit, which can be alleviated by employing PUFs with denser feature information or image preprocessing of denoising and grayscale stretch."  
+
+![](images/Figure_unknown_9.jpg)
+
+<center>Figure R9. The grayscale distribution before a and after b the image preprocessing of grayscale stretch. </center>  
+
+C4: Concerning the authentication method, as the authors mention, a conventional image processing analysis could be used to extract the image features of the tag. The authors have developed a modified deep learning scheme for this purpose. Here also the paper suffers from the negligible number of samples. For example, for the generation of the training set they used fifteen out of the twenty sample patterns and they enriched them with additional patterns coming from rotation of the initial 15 patterns by one degree. I'm not sure if this is the best training approach but again a deep learning system based on a total of 20 samples is out of the question. Moreover, I'm wondering what are the benefits of the ML method compared to the conventional image processing methods.  
+
+R4: We agree with the reviewer about this concern. In order to provide a convincing model, we changed two important parts of our work:  
+
+1) The architecture of the model is changed to a ResNet50 based classification model pretrained by ImageNet (3.2 million images). The input image is preprocessed in grayscale by modification of the pixel distribution and sent to a ResNet50 based classification model. Some random Gaussian noise was added to the training image in order to limit overfitting and also for
+
+<--- Page Split --->
+
+covering the influence from the image noise. To reduce the training time and guarantee the generalizability of the model, the pretrained model parameters on ImageNet is used to initialize the ResNet50 and a base model is trained via 1100 PUFs. With a new PUF, only one parameter is added to the last classification layer of the base model, and only this final layer is updated during the training of the new PUF. Noting that both the original PUFs in the database and newly added PUFs are used to update the model to avoid the model forgetting the previous database. The update of only the last layer ensures a fast training procedure. When the fifth new image was added, the training can be finished in 5.48 s (total 40 epochs, 0.137s for each epoch, Figure R5b) in the database expansion test. The results of the database expansion test are shown below.  
+
+2) A lot more PUF images are captured (1850 PUFs): 1) 1300 PUFs for the establishment of the database with 1100 PUFs for building the base database and 200 PUFs for the expansion of the database; 2) 550 PUFs as the fake PUFs out of the database and augmented for training a general model. The TEST_Set contains 37000 images augmented from the original 1850 PUFs, which considers different conditions (brightness, rotations, and noise) that can occur in practice, covering the influence from different capturing conditions and imaging unit. We truly know that not all conditions can be considered in this test, but our work aims to provide a benchmark to prove that our model and technique have the potential application in the PUF authentication.  
+
+The corresponding training/validation/testing results are shown below. According to the results, we can achieve a \(0\%\) wrong validation ratio of the genuine images ( \(0\%\) false positives and \(5\%\) false positives). Noting that the high image invalidation rate comes from the large ratio of the fake image ( \(30\%\) ), which is tested as the invalidation in our model setup. Of course, the further optimization of the model for obtaining a higher accuracy is being implemented. Besides, we have made a list of the different database sizes for AI model training in other PUF works.  
+
+Two added references related to the modified model are as follows,  
+
+[1] ImageNet:  
+
+J. Deng, W. Dong, R. Socher, L. - J. Li, Kai Li and Li Fei-Fei, "ImageNet: A large-scale hierarchical image database," 2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248-255, doi: 10.1109/CVPR.2009.5206848.  
+
+[2] ResNet50:  
+
+K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770-778, doi: 10.1109/CVPR.2016.90.
+
+<--- Page Split --->
+
+
+Table 1. The parameters for the AI model training/validation/testing.   
+
+<table><tr><td>Dataset</td><td>Size</td><td>Augmentation</td><td>Labels</td></tr><tr><td>Original images (ORI_Set)</td><td>1850 PUFs</td><td></td><td></td></tr><tr><td>Training set (TRAIN_Set)</td><td>1100 PUFs from ORI_Set</td><td>Rotation: 0, 30, 60, 90, ..., 330°</td><td>Corresponding classes: 0-1099</td></tr><tr><td>Validation set (VAL_Set)</td><td>The same PUFs with TRAIN_Set</td><td>Rotation: 1, 3, 5, 7, ..., 359°</td><td>Corresponding classes: 0-1099</td></tr><tr><td>Candidate set (CAND_Set)</td><td>200 PUFs from ORI_Set</td><td>Rotation: 0, 30, 60, 90, ..., 330° for training and 1, 3, 5, 7, ..., 359° for validation</td><td>Corresponding classes: 1100-1299</td></tr><tr><td>Fake set (FAKE_Set)</td><td>550 PUFs from ORI_Set</td><td rowspan="2">1) Rotation: 10 angles are randomly selected from those that are not considered in the TRAIN_Set and VAL_Set.<br>2) Gaussian noise for each image<br>3) 10 different brightness conditions applied to each image</td><td rowspan="2">All labelled as “FAKE”<br>Corresponding classes: 0-1299 for the first 1300 PUFs in the database, and “FAKE” for the left 550 PUFs</td></tr><tr><td>Test set (TEST_Set)</td><td>1850 PUFs of ORI_Set with augmentations that generate 37000 images</td></tr></table>  
+
+![](images/Figure_unknown_10.jpg)
+
+<center>Figure R5. a The training procedures of the base classification model (TRAIN_Set and VAL_Set). b The training time of the classification model as a function of the newly added 200 PUFs, showing a gradually stable training epoch below 40. The inset is the plot of training and validation accuracy of the classification model with the first newly added image. The training stops when the validation accuracy reaches at the threshold of 95%. \(R_{c}\) represents the accuracy. c Correct/Wrong validation ratio and invalidation ratio of the 37000 images (TEST_Set) as a function of similarity threshold. The ideal maximum of the correct accuracy is 0.7, as the whole test dataset contains 30% fake images. The 0% wrong validation can be achieved beyond the threshold of 0.5. </center>
+
+<--- Page Split --->
+
+
+Table 2 The database size for AI model training in different works of PUFs.   
+
+<table><tr><td>PUF type</td><td>Database size for AI model training</td><td>Reference</td></tr><tr><td>Crumpling of 2D Materials</td><td>500</td><td>Matter 3, 2160–2180, 2020.</td></tr><tr><td>Drop-casting nanoparticle tags</td><td>1020</td><td>ACS Nano 15, 2901–2910, 2021.</td></tr><tr><td>Laser-induced polymer wrinkle</td><td>112</td><td>Adv. Mater. 32, 2003032, 2020.</td></tr><tr><td>Multi-functional nanoinks</td><td>100</td><td>Nano Today 41, 101324, 2021.</td></tr><tr><td>Random wrinkles</td><td>6</td><td>ACS Appl. Mater. Interfaces 13, 27548–27556, 2021.</td></tr><tr><td>Inkjet-printed unclonable quantum dot</td><td>6</td><td>Nat. Commun. 10, 2409, 2019.</td></tr><tr><td>Inkjet-printed unclonable quantum dot</td><td>10</td><td>ACS Appl. Mater. Interfaces 13, 15701–15708, 2021.</td></tr><tr><td>Unclonable photonic crystal hydrogels</td><td>10</td><td>ACS Appl. Mater. Interfaces 14, 2369–2380, 2022.</td></tr><tr><td>Our work</td><td>1300</td><td></td></tr></table>  
+
+The further optimization of the model (for getting a higher accuracy) and testing of more images with different conditions are underway (images with different scaling and blurring are to be tested). As for the database expansion test, more new PUFs (another 300 images) are to be supplemented. This part of the works is expected to be finished in two weeks.  
+
+The deep learning method has many advantages compared to the conventional image processing methods, as follows,  
+
+1) the deep neural networks can record the information of unclonable features so that the pattern verification can be completed only through the neural network, considerably improving the speed and accuracy. On the contrary, the conventional image processing method (e.g. the similarity methods) is time-consuming especially when the database is huge as all the images in the database need to be compared with the input image from the user during the authentication.  
+
+2) deep learning, as a black box with a few explanations of the functioning mechanism, is an advantage for unclonable anti-counterfeiting technique because it is a tamper proof.  
+
+3) as for the conventional image processing, a rectangular coordinate system is required to define the position and orientation of each physical feature, e.g. the input image usually needs to be rotated in different angles to search the matching PUF in the database. However, any rotation of the security label results in a completely different code (Fig. S28; Sci. Adv. 4, e1701384, 2018). In other word, a real security label will be recognized as a fake one if the end-user rotates or zooms in/out it
+
+<--- Page Split --->
+
+during the pattern readout process. Alternatively, a mark on the security label is used to define xy axis, which will rotate as the security label rotates (Adv. Mater. 28, 2330- 2336, 2016). This guarantees only one code for each security label. By contrast, the deep learning- based authentication system allows the end- user to readout the patterns with different image brightness, amplifications and rotations, as AI can generate the scale and rotation invariant features of the security labels. Although, it is time- consuming to train the deep learning machine to learn the characteristic features of the security labels and the repeated model training is required for each new PUF pattern. In our work, mass of new images are further added to the deep learning model successively and a gradually shortened training time is demonstrated, allowing for a potentially unlimited new image amount with a low time cost. This strategy helps to break the trade- off between the large key database of the PUF system and long training time of the deep learning model in practice.  
+
+"Figure 4a schematically demonstrates the deep learning- based authentication system of the security label. Firstly, each PUF pattern is captured by the manufacturers by using the microscope. The different images are then preprocessed by modification of the pixel distribution and devoted to a ResNet50 based classification model<sup>47</sup> for learning the characteristics of the PUF patterns. To reduce the training time and guarantee the generalizability of the model, the pretrained model parameters on ImageNet<sup>48</sup> is used to initialize the ResNet50. Next, the images are classified in a general manner and stored in the database for subsequent authentication."  
+
+"For experimentally demonstrating the above authentication system, 1300 different PUF tags were randomly captured to establish the security key database (parts of PUFs shown in Supplementary Fig. 22). 1100 PUFs of them were used for the establishment of the basic database, and the rest 200 PUFs were used for the key expansion test of the database. Each of the 1100 PUFs were rotated with different angles, forming 211200 images as the dataset for deep learning model training (13200 images with less rotations for training and 198000 images for validation). Every input image is preprocessed by grayscale stretch and added random noise to avoid overfitting."  
+
+"For testing our authentication system, the above 1300 PUF tags (in the database) in different conditions (brightness, rotation angle, and random noise, total of 26000 images, Supplementary Fig. 24) and 11000 images from 550 new PUF tags (not in the database) were captured and uploaded to the trained AI for a test. We investigated the validation ratio as a function of the similarity threshold by testing multiple similarity indexes for obtaining the best threshold of validation. Figure 4c reveals that when the similarity threshold reaches 0.5, we can achieve a 0% wrong validation ratio of the genuine images (that is, the rate of false positives is 0%) and 5% false negatives. In this work, we set the similarity threshold at a value of 0.5, and the authentication of a PUF tag with an encoding capacity of about \(10^{353}\) can be finished in 6.36 seconds."  
+
+"The conventional image processing algorithms are based on the matching of feature points in library files one by one<sup>1,37</sup>. Apart from the relatively tedious matching time, their performance also strongly depends on the image orientation and quality<sup>36,38</sup>. Deep learning (DL)<sup>11,13,17</sup>, as an artificial intelligence (AI) technique, has been popularly used to validate the security key through the trained neural networks with high authentication efficiency and accuracy as well as high readout toleration in different conditions."
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have submitted a revised version of their manuscript entitled "Random fractal- enabled physical unclonable functions with dynamic AI authentication". The authors have made numerous revisions in response to a prior review.  
+
+The first innovation of this work stems from the development of a fractal based graphical PUF with large grayscale encoding capacity. This is contrasted to wrinkling based and luminescent- type graphical PUFs in the recent literature which are argued to exhibit poor robustness and/or durability.  
+
+The second innovation of this work is the implementation of a dynamic database strategy which is developed and employed to train a deep- learning authentication system. This strategy avoids the need to re- train with the entire database when new images or graphical PUF data are generated. Hence, new devices/products can be continuously added to the database with limited training time/cost.  
+
+Thirdly a Raman mapping is shown as a proof- of- concept pathway to potentially adding additional information/complexity to the graphical PUF by adding data that is in principal orthogonal to the PUF image (relying on nanoscale roughness).  
+
+Regarding prior Review comments:  
+
+The authors have adequately addressed the majority of my prior concerns/comments. Including those related to the potential Raman aspect, convincingness of the AI authentication protocol etc. However, one critical item, regarding the reliance/usage of grayscale information has only been partially suitably addressed in response to Reviewer 1 Comments 1 and 2. The remaining concern is regarding the heavy reliance on 0- 255 levels in computing the encoding capacity, as echoed by Reviewer 2 and detailed below.  
+
+## Comments/concerns:  
+
+(1) Reviewer 2 also previously raised an excellent point regarding encoding capacity (comment C3). The reported encoding capacity indeed appears somewhat optimistic. The authors have made some acknowledgments and discussion of the practical nature of this point, but it is not clear if they have made satisfactory edits to the main text. According to reference #1 (Carro-Temboury et al) of the supporting information, the encoding capacity is generally less than the upper theoretical limit based on the image capacity (# pixel values)^(# number of pixels) and a master equation is provided to estimate the practical encoding capacity. In the supporting information Supplementary Note #1 the authors utilize the Carro-Temboury method/equation to estimate their encoding capacity. However, Carro-Temboury utilized C = 2, 4, or 7 based on their particular PUF. Meanwhile the present work assumes C = 255. Per reviewer 2's comment, it is not clear if using C = 255 represents a fair estimate of the color resolution and hence the encoding capacity, especially when considering practical intensity resolution factors and the pre/post-processing histograms of Supplementary Fig 8. Perhaps the authors could either: (1) better justify the use of this C = 255 value and reiterate its assumption in the calculation in the main text (near key lines 224-226 of page 9), or (2) could revise the calculation using a more conservative estimate for C, perhaps based on a conservative estimate of the practical grayscale resolution.  
+
+## New comments:  
+
+(NC1) Upon close inspection, the literature introduction and citations need improvement.  
+
+"Electronic PUF" misses a whole body of literature, see relevant review articles that can be cited, for
+
+<--- Page Split --->
+
+example:  
+
+(A) Gao, Y., Al-Sarawi, S.F. & Abbott, D. Physical unclonable functions. Nat Electron 3, 81-91 (2020).  
+
+(B) C. -H. Chang, Y. Zheng and L. Zhang, "A Retrospective and a Look Forward: Fifteen Years of Physical Unclonable Function Advancement," in IEEE Circuits and Systems Magazine, vol. 17, no. 3, pp. 32-62, thirdquarter 2017, doi: 10.1109/MCAS.2017.2713305.  
+
+"Spectral PUF" misses some recent photonics-based examples in the literature that could be cited, e.g.:  
+
+(A) Brian C. Grubel, Bryan T. Bosworth, Michael R. Kossey, Hongcheng Sun, A. Brinton Cooper, Mark A. Foster, and Amy C. Foster, "Silicon photonic physical unclonable function," Opt. Express 25, 12710-12721 (2017)  
+
+(B) Tarik, F.B., Famili, A., Lao, Y. et al. Scalable and CMOS compatible silicon photonic physical unclonable functions for supply chain assurance. Sci Rep 12, 15653 (2022).  
+
+(C) "Harnessing disorder for photonic device applications" [Appl. Phys. Rev. 9, 011309 (2022)]  
+
+(NC2) English/grammar needs improvement in various areas, below are some examples needing improvement along side possible suggestion or interpretation in parenthesis:  
+
+Page 1, line 17 - "often causes it difficult" (often makes it difficult)  
+
+Page 1, line 24 - "can be served as the" (can serve as)  
+
+Page 2, line 30 - "Anti- fake labels as the certification of products retain an increasing challenge in authenticity" (unclear)  
+
+Page 6, line 167- 169 (unclear)  
+
+Page 15, line 367 - "Raman characterization is potential" (has potential)  
+
+[1]  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors reacted in a satisfying way to the concerns and comment of this reviewer and I think the paper can be accepted for publication to the Nature Communications journal
+
+<--- Page Split --->
+
+## Response Letter  
+
+We are grateful for the positive comments and helpful suggestions from the reviewers. Our responses to the reviewers' comments and the changes made in our manuscript are as follows. For the sake of readability, our responses are highlighted by blue font and the changes to our manuscript are presented in green.  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The authors have submitted a revised version of their manuscript entitled "Random fractal- enabled physical unclonable functions with dynamic AI authentication". The authors have made numerous revisions in response to a prior review.  
+
+The first innovation of this work stems from the development of a fractal based graphical PUF with large grayscale encoding capacity. This is contrasted to wrinkling based and luminescent- type graphical PUFs in the recent literature which are argued to exhibit poor robustness and/or durability.  
+
+The second innovation of this work is the implementation of a dynamic database strategy which is developed and employed to train a deep- learning authentication system. This strategy avoids the need to re- train with the entire database when new images or graphical PUF data are generated. Hence, new devices/products can be continuously added to the database with limited training time/cost.  
+
+Thirdly a Raman mapping is shown as a proof- of- concept pathway to potentially adding additional information/complexity to the graphical PUF by adding data that is in principal orthogonal to the PUF image (relying on nanoscale roughness).  
+
+Regarding prior Review comments:  
+
+The authors have adequately addressed the majority of my prior concerns/comments. Including those related to the potential Raman aspect, convincingness of the AI authentication protocol etc. However, one critical item, regarding the reliance/usage of grayscale information has only been partially suitably addressed in response to Reviewer 1 Comments 1 and 2. The remaining concern is regarding the heavy reliance on 0- 255 levels in computing the encoding capacity, as echoed by Reviewer 2 and detailed below.  
+
+We truly appreciate the reviewer for the enthusiastic support and precious comments of our manuscript. We have carefully revised the manuscript based on the constructive suggestions.  
+
+C1: Reviewer 2 also previously raised an excellent point regarding encoding capacity (comment C3). The reported encoding capacity indeed appears somewhat optimistic. The authors have made some acknowledgments and discussion of the practical nature of this point, but it is not clear if they have made satisfactory edits to the main text. According to reference #1 (Carro- Temboury et al) of the supporting information, the encoding capacity is generally less than the upper theoretical limit
+
+<--- Page Split --->
+
+based on the image capacity (# pixel values) \(^x\) (# number of pixels) and a master equation is provided to estimate the practical encoding capacity. In the supporting information Supplementary Note #1 the authors utilize the Carro- Temboury method/equation to estimate their encoding capacity. However, Carro- Temboury utilized \(\mathrm{C} = 2\) , 4, or 7 based on their particular PUF. Meanwhile the present work assumes \(\mathrm{C} = 255\) . Per reviewer 2's comment, it is not clear if using \(\mathrm{C} = 255\) represents a fair estimate of the color resolution and hence the encoding capacity, especially when considering practical intensity resolution factors and the pre/post- processing histograms of Supplementary Fig 8. Perhaps the authors could either: (1) better justify the use of this \(\mathrm{C} = 255\) value and reiterate its assumption in the calculation in the main text (near key lines 224- 226 of page 9), or (2) could revise the calculation using a more conservative estimate for \(\mathrm{C}\) , perhaps based on a conservative estimate of the practical grayscale resolution.  
+
+R1: We thank the reviewer for proposing this critical issue and the specific suggestions. We really agree with the reviewer that the calculation of the encoding capacity should be based on a more practical grayscale resolution, considering that the practically captured PUF image cannot possess a theoretical grayscale distribution from 0 to 255 due to the limitation of the imaging unit (shortage in image brightness/contrast/color resolution). Particularly, the grayscale level of 255 can be achieved if the imaging unit provides a desirable contrast between the feature structure and background (substrate), which puts forward a high requirement for the imaging device and image quality.  
+
+Here we followed the second suggestions from the reviewer. Based on the practical grayscale distribution of the captured PUF image calculated by the histograms in Supplementary Fig 8, we tried to define a conservative level of grayscale intensity in one pixel as the final "C" in the encoding capacity calculation. In order to obtain a more practical "C", we randomly selected 10 PUF images from the basic database (all of them were captured under normal imaging conditions) and calculated their grayscale distributions, as shown in Figure R1. We defined the peak of grayscale histogram as the maximum of background grayscale, which means that the background loads the largest number of pixels with consistent grayscale intensity and the above grayscale intensities are all from the PUF feature structures. After subtracting the invalid grayscale information from the background and the high grayscale intensities that the practical images cannot reach, we defined the grayscale range of the feature structures from 74 (average maximum of background grayscale) to 214 (average maximum of feature grayscale), and the value of "C" is set as the difference, i.e., 140. The corresponding encoding capacity calculations have been revised in the manuscript (Figure 3c) and Supplementary Information (Supplementary Note 1). For example, after the correcting of "C", the encoding capacity of PUF tag in Supplementary Fig 9 has changed from \(10^{353}\) to \(10^{348}\) , showing a little decrease in capacity.
+
+<--- Page Split --->
+![PLACEHOLDER_29_0]
+
+<center>Figure R1. Calculation of practical grayscale intensity level of the original PUF image. a Grayscale image of a typical PUF pattern. b Grayscale distribution histogram of a typical PUF pattern. c Grayscale distributions from 10 randomly selected PUF images in the basic database. d Maximums of feature grayscale and background grayscale from the 10 PUF images in c. </center>  
+
+"However, due to the inevitable limitation of the imaging unit in image contrast or color resolution, the physical features in practically captured PUF images usually cannot possess a theorical grayscale distribution from 0 to 255. To present a more general level of grayscale encoded in physical features, we calculated the grayscale histograms from 10 randomly selected PUF images in the basic database, subtracted the grayscale information that the feature structure cannot reach, and counted an average value of 140 as the valid level of grayscale in a practical PUF image (valid grayscale distribution from 74 to 214, as shown in Supplementary Fig. 8)."  
+
+C2: Upon close inspection, the literature introduction and citations need improvement.  
+
+"Electronic PUF" misses a whole body of literature, see relevant review articles that can be cited, for example:  
+
+(A) Gao, Y., Al-Sarawi, S.F. & Abbott, D. Physical unclonable functions. Nat Electron 3, 81-91 (2020).  
+
+(B) C. -H. Chang, Y. Zheng and L. Zhang, "A Retrospective and a Look Forward: Fifteen Years of Physical Unclonable Function Advancement," in IEEE Circuits and Systems Magazine, vol. 17, no. 3, pp. 32-62, thirdquarter 2017, doi: 10.1109/MCAS.2017.2713305.  
+
+"Spectral PUF" misses some recent photonics-based examples in the literature that could be cited, e.g.:  
+
+(A) Brian C. Grubel, Bryan T. Bosworth, Michael R. Kossey, Hongcheng Sun, A. Brinton Cooper,
+
+<--- Page Split --->
+
+Mark A. Foster, and Amy C. Foster, "Silicon photonic physical unclonable function," Opt. Express 25, 12710- 12721 (2017)  
+
+(B) Tarik, F.B., Famili, A., Lao, Y. et al. Scalable and CMOS compatible silicon photonic physical unclonable functions for supply chain assurance. Sci Rep 12, 15653 (2022).  
+
+(C) "Harnessing disorder for photonic device applications" [Appl. Phys. Rev. 9, 011309 (2022)]  
+
+R2: We thank the review for recommending the relevant references about electronic PUFs and silicon photonic PUFs. As suggested, we have made a careful research on the development of electronic and silicon photonic PUFs and included the relevant descriptions and references into the Introduction in the manuscript.  
+
+"(ii) spectral PUFs...irregular texture \(^{25}\) /matrix \(^{5,26}\) linear scattering-based speckle patterns, and chaotic nonlinear silicon photonic devices \(^{27,28}\) ; and (iii) complex electronic PUFs with diverse disorders and inherent imperfections (e.g., graphene \(^{29}\) or randomly distributed carbon nanotube-based \(^{30}\) field-effect transistors, oxide or halide-based memristors with intrinsic entropy sources \(^{6,31,32}\) )."  
+
+25. Kim, M. S. et al. Revisiting silk: a lens-free optical physical unclonable function. Nat. Commun. 13, 247 (2022).  
+
+26. Cao, H. & Eliezer, Y. Harnessing disorder for photonic device applications. Appl. Phys. Rev. 9, 011309 (2022).  
+
+27. Grubel, B. C. et al. Silicon photonic physical unclonable function. Opt. Express 25, 12710 (2017).  
+
+28. Tarik, F. B. et al. Scalable and CMOS compatible silicon photonic physical unclonable functions for supply chain assurance. Sci. Rep. 12, 15653 (2022).  
+
+29. Dodda, A. et al. Graphene-based physically unclonable functions that are reconfigurable and resilient to machine learning attacks. Nat. Electron. 4, 364-374 (2021).  
+
+30. Zhong, D. et al. Twin physically unclonable functions based on aligned carbon nanotube arrays. Nat. Electron. 5, 424-432 (2022).  
+
+31. Chang, C., Zheng, Y. & Zhang, L. A retrospective and a look forward: Fifteen years of physical unclonable function advancement. IEEE Circuits Syst. Mag. 17, 32-62 (2017).  
+
+32. John, R. A. et al. Halide perovskite memristors as flexible and reconfigurable physical unclonable functions. Nat. Commun. 12, 3681 (2021).  
+
+C3: English/grammar needs improvement in various areas, below are some examples needing improvement along side possible suggestion or interpretation in parenthesis:  
+
+Page 1, line 17 - "often causes it difficult" (often makes it difficult)  
+
+Page 1, line 24 - "can be served as the" (can serve as)  
+
+Page 2, line 30 - "Anti-fake labels as the certification of products retain an increasing challenge in authenticity" (unclear)  
+
+Page 6, line 167- 169 (unclear)  
+
+Page 15, line 367 - "Raman characterization is potential" (has potential)
+
+<--- Page Split --->
+
+R3: We thank the reviewer for the thoughtful comments. As suggested, we have revised the whole manuscript and carefully proof- read the manuscript to improve English and clarity. We also sent the manuscript to a language polishing service for English language editing during the revision process. The corresponding Editing Certificate has uploaded as a supplementary file for editor and reviewer only. We believe that the language is now acceptable for the review process and publication. All the revisions in grammar/English have been highlighted in green in the Revised and marked manuscript, and the revision will not change the original meanings of the text.  
+
+Two proposed examples:  
+
+"Anti- fake labels as the certification of products retain an increasing challenge in authenticity" has changed to  
+
+"Anti- fake labels as the authentication tools of product authenticity face an increasing challenge in security"  
+
+"The branch orientations and lengths of the fractals are randomly varied and the symmetry is broken, but exhibiting statistical feature similarity in different bifurcations, as shown in one quasi "Y- shape" iterated network in Fig. 1d." has changed to  
+
+"The extension directions and lengths of the branches are randomly varied but exhibit statistical feature similarity in different bifurcations. For example, in Fig. 1d, the bifurcations of part 1 and the derived part 2 have the similar quasi "Y- shape" feature."  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The authors reacted in a satisfying way to the concerns and comment of this reviewer and I think the paper can be accepted for publication to the Nature Communications journal.  
+
+Thanks so much for the reviewer's recognition and putting so many efforts in reviewing our manuscript.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have suitably addressed my remaining concerns and I now recommend this work for timely publication.
+
+<--- Page Split --->
+
+## Reviewer #1:  
+
+The authors have suitably addressed my remaining concerns and I now recommend this work for timely publication.  
+
+## Our response:  
+
+Thanks so much for the reviewer's recognition and putting so many efforts in reviewing our manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__a48f1ac55880665449aa1921bf621716bbf5200b5a4fbf5a6396269a1d36f934/images_list.json

@@ -0,0 +1 @@
+[]