Add files using upload-large-folder tool

peer_reviews/supplementary_0_Peer Review File__41f73bb80f7f4860b41cedbe7ceb9226dfa693d3d58be9b65f6123b0558820c4/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__41f73bb80f7f4860b41cedbe7ceb9226dfa693d3d58be9b65f6123b0558820c4/supplementary_0_Peer Review File__41f73bb80f7f4860b41cedbe7ceb9226dfa693d3d58be9b65f6123b0558820c4.mmd

@@ -0,0 +1,359 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Two gates mediate NMDA receptor activity and are under subunit- specific regulation  
+
+![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 --->
+
+Reviewers' Comments:  
+
+Reviewer #1: Remarks to the Author:  
+
+In the manuscript "Two gates mediate NMDA receptor activity and are under subunit- specific regulation" the authors are adding some new pieces into the puzzle of the NMDAR function and regulation.  
+
+Although the provided experimental evidence clearly demonstrates the effects of the G- to- A substitutions, the conclusions based on it are often overstated.  
+
+The authors suggest a redefinition of the common nomenclature, renaming what is usually known as the selectivity filter to a "(second) gate". In fact it could be even a "third" gate, after the M3 SYTAN and the neighboring "LILI gate". The LILI gate is in principle also consistent with the "GluN1 glycine mainly regulates single channel events within a cluster or bursts of activity, whereas the GluN2A glycine mainly regulates entry and exit from clusters".  
+
+To get a better idea about the GluN1 G- to- A- induced changes in the M2 gate/filter pore diameter experimentally, it would be useful to record the single channel properties also in presence of cations with different sizes.  
+
+In my opinion, the weakest part of the work is the molecular dynamics and I would suggest either significantly improving the MD setup and analysis, or removing the MD part as it in fact does not provide any additional details to the observed GluN1 G- to- A effects. It is questionable whether a model of the rat GluN1/GluN2B (why not GluN2A?) TMD- only based on an symmetrical AMPA "open" structure represents the GluN1/GluN2A system studied experimentally. The significant rearrangement between and within domains during the open to closed transition in GluN1/GluN2B full length model was studied by MD (10.3390/biom9100546), showing important non- symmetrical switching of N1 and N2B M3 helices induced by extracellular domains.  
+
+The receptor is sensitive to the presence of cholesterol in the membrane, using pure POPC could change the TMD behavior. The MD setup is using a rather short electrostatics cutoff of 8A, possibly lowering the accuracy of simulations especially when interested in ion permeation. There is also no transmembrane potential reported. While the experiments involve \(\text{Ca2 + }\) , the MD simulations contain only \(\text{Na + }\) (at least as described in the methods and "guessing" from descriptions using only "ion" without stating which one). Although the PTMs are somehow mentioned, it was described recently that N2 palmitoylation plays a significant role anchoring the base of M4 helices, however, it was not included in the model.  
+
+minor points:  
+
+Is there something missing in the "pathway that select ion(s) can cross"?  
+
+I don't really know what is meant by "membrane physiology".  
+
+In the "Gates' are barriers that occlude the flux of ions in the closed state.", if the above mentioned closed state model is correct, the gates you are talking about are actually relevant in the open state conformation. But this is probably just a missing definition of the closed state (I guess here you mean liganded (activated) receptor not permeable to ions).  
+
+I am not sure the upstream/downstream of transmembrane helices is clearly defined.  
+
+In the "A critical residue in the M4 helices", maybe a better would be one of the critical residues. In the "(eq. Popen) and to the same extent", "and" is not needed or something else is missing? In the "closed and opens states", just "open"?  
+
+I am not very familiar with the current data policy of Nature journals, but the ever more recognized FAIR principles would dictate that the data are deposited in a public repository and not just "available upon request". Although for example the methods section concerning the MD simulations is detailed enough to allow reproducibility, the actual trajectories could be useful to the community.
+
+<--- Page Split --->
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+The manuscript entitled "Two gates mediate NMDA receptor activity and are under subunit- specific regulation" is a solid study that provides some interesting data that is relevant to the function of the NMDA receptor. It is well written, and summarizes evidence that individual subunits control different aspects of gating, and also provides further support for functional relevance of the M4 in gating. The conclusions that the GluN1 M4 helix controls M2 helix of GluN2 and GluN2 M4 controls GluN1 M3 are important, and simulations suggest that mutations in the M4 region of different subunits can alter the pore diameter at distinct sites. I think it will be of interest to many working on ligand gated ion channels, glutamate receptors, and synaptic transmission. Below are several suggestions the authors should consider that could improve the clarity of the study and perhaps lead to more accurate and useful conclusions.  
+
+## Major  
+
+1. While the data is convincing in showing a change in the long-lived closed dwell time separating clusters of activity, I am not sure the stated conclusion is helpful (from the discussion) "...a primary gate at the M3 bundle helical crossing controls entry and exit from both long lived and short-lived closed states". That means it controls everything—is that the intended conclusion? Also, where are data supporting entry and exit from short-lived states? There are data in supplemental but they are not really discussed.  
+
+2. If the 12 sec dwell time reflected an important step in the pathway from agonist binding to channel opening (i.e. gating), then the rise times would be far slower than \(\sim 10\) ms reported by dozens of authors. However most kinetic studies conclude that these prolonged closed periods in between clusters of activity are some form of agonist-dependent desensitization, and I think a likely conclusion is that the M4 regulates recovery from a desensitized state. Seems like this should be raised or evaluated in terms of double pulse experiments that assess the recovery from the desensitized state.  
+
+3. Opening rates have been interpreted in early studies of frog neuromuscular nicotinic receptors from brief "nachschlag" closed states. The idea is the pore reverses a gating step during these brief gaps, and then proceeds forward to reopen. If the dwell time reflects a single visitation to one closed state, then the reciprocal of the duration is the rate associated with forward opening rate. Studies from Colquhoun, Auerbach, Traynelis, Gibb and others all showed three brief intra-cluster closed states that were interpreted as a very fast opening (perhaps pore dilation), and two other states with dwell times of \(\sim 1\) and \(\sim 5\) ms. If the author wants to conclude something about gating within a cluster, they should discuss some of the changes in various components of the histograms shown in the supplemental material) or provide some modelling results.  
+
+4. I think it is important to explicitly state whether the opening of the two gates occurs in any order or in a dependent order (e.g. slow step first, then fast step), as multiple Popescu papers have suggested. Dependent order requires slow step to be first, followed by a fast step and then opening, otherwise there would never be brief sojourns to the closed state in the single channel record if the brief step occurred first and then the slow step immediately preceded opening.  
+
+## Minor  
+
+5. I think the abstracts reference to "clusters" is vague, and bordering on jargon. How many readers interested in synaptic transmission will know what is meant by a cluster? I think the authors need to re-word to a conclusion that a wider audience can grasp and understand.
+
+<--- Page Split --->
+
+6. Minor suggested wording change: Page 4 "A variety of disease associated missense mutations have been identified at OR NEAR these conserved glycines...(could choose any of multiple refs to support)  
+
+7. Page 4 sentence starting "These data suggest that the M3 gate mediates the longer..." How can you rule out that both gates are important, and because the long dwell time dominates the interval, you are missing changes to the interval with the faster gate changes its rate. That is, both may be required, but you can only see changes in the slow gate (contributing the long delay). A short 10 ms change will hardly alter a 12 second interval.  
+
+8. Typo in text on page 6. Duration of long lived state should be sec, not ms in the last sentence.  
+
+9. Page 8, first sentence of the second paragraph -- Alasdair Gibb should get the credit for first showing unambiguously that there are five independent closed states in individual activations of a single native NMDA receptors in his clever low concentration experiment (Gibb et al., 1992). The references (40,41) confirmed his conclusions in heterologous systems—Schorge et al I think should be added.  
+
+10. There are several papers entirely devoted to the idea that the GluN1 M4 functionally interacts with the GluN2 M3, and the GluN2 M4 interacts with the GluN1 M3 (Chen et al. 2017 exploring actions of a pre-M4 mutation, also Gibb 2018). Chen concludes the interaction of GluN2-M4 with GluN1 M3 is critical for gating, consistent with the conclusions here. Seems as though the conclusions of these papers should be mentioned, which are consistent to some extent with what is being showing here?  
+
+11. I would recommend mentioning in the results that saturating agonists are used so a reader doesn't have to turn to the methods or scrutinize the legends.  
+
+12. I think it is important to comment that these results were obtained in the absence of divalents. Schorge et al recorded in divalent ions and was unable to reproduce some channel properties observed in the absence of divalents. Alternatively, one could perform a quick experiment in a patch with \(0.5 - 1 \text{mM Ca2 + }\) to demonstrate that the conclusions hold in a physiological context.  
+
+13. I suspect a subset of readers will want to know what Tcrit was used for cluster analysis when they read about the analysis in the results without stopping their train of thought to turn back to the methods to figure out how you separated clusters, especially for mutations that shortened the long closed duration.  
+
+14. Figure 1D: It looks like there are sublevels for the GluN1-G815A mutation, and possibly even the GluN2A-G819A mutation. This seems relevant for a paper discussing pore diameter. Why not show openings at a higher resolution and analyze the sublevels?  
+
+15. The K channel literature is fascinating because the authors in the structural era connect their results to work performed before structures were known. Seems like Banke's conclusion ten years ago (2003) before the structure of the NMDA receptor, "These data suggest that NR1 and NR2B subunits, respectively, undergo a fast and slow agonist-dependent conformational change that precedes opening of the pore", is relevant to the conclusions of the current manuscript. I believe that data in Jones et al 2002 with MTSEA also suggests different roles of GluN1 and GluN2 in gating. The current manuscript goes light years beyond these older studies, but it seems generally useful to recognize how early ideas about subunit dependent gating arose in the literature.  
+
+16. Near the middle of page 9, could the authors specify that 15 replicate simulations of 500ns each were run per construct? They specify 15 in the Methods section, but also specifying it here would save time from having to go to the Methods and help improve clarity for the reader.
+
+<--- Page Split --->
+
+17. In Figure 2 and the related text in Results, positions \(+2\) , \(+6\) , and \(+10\) aren't defined as being indexed from the S in SYTANLAAF, but it is clarified in the Supplementary info. It would be useful to define it in the text, either by verbally explaining this in the legend for Figure 2 or by moving panel 'a' from Supp Fig 3 to Fig 2 in the main text. I don't see that the magenta sphere is defined in the caption for Fig. 2a. Caption for Fig. 2g refers to left and right panels when the panels are stacked vertically  
+
+18. Has the model of the active state used in the MD simulations been used in previous publications? If so, a reference should be given as it would be helpful to see model quality metrics, as well as other information about the simulations such as RMSD plots, included in the supplemental information. If not, perhaps these quality metrics could be added as supplemental information.  
+
+19. Given that ion permeation was analyzed in the MD simulations, can the authors comment on why a membrane potential wasn't modeled using a method such as applying a constant electric field or using a polarizable force field? For reference, see https://doi.org/10.1021/acs.jctc.5b01202. The simulations as they are model the behavior of ions in the absence of any driving force. If one of these methods for the simulations had been used, I feel that the data presented in Figure 2d and e would be more meaningful and informative. However, it is obviously too much to re-do. Perhaps the authors could add a comment about this in the manuscript as a caveat?  
+
+20. Are the values in Figure 2e from a single trajectory, or averaged across the 15 replicates? If they are from a single trajectory, can you explain how that specific trajectory was selected? Can the authors provide an explanation of how individual trajectories were selected for the plots in Supplementary Figure 4?  
+
+21. Could the authors add to the Discussion their thoughts on why there are two long-lived closed states if the general claim is that the N2A M4s predominately regulate entry into and exit from long-lived closed states?  
+
+22. Page 18: "although evidence from our MD simulations and single channel recordings suggest that reductions in the pore diameter of the M2 gate correlate with reduced ion permeation...", I don't really see evidence of this in the supplementary or main figures at least in terms of single channels for permeation. Maybe I missed something.  
+
+23. Y-axes of Fig. 1h and 1i could be more clearly labeled (perhaps 'Closed duration ratio (G-A/wt)') Similarly, the word 'closed' should be added between 'mean' and 'durations' in the last sentence on pg. 8. Minor but might help keep the message clear.
+
+<--- Page Split --->
+
+## Response to Reviewers. Nature Communication  
+
+Manuscript#: NCOMMS- 22- 43338- T. "Two gates mediate NMDA receptor activity and are under subunit- specific regulation'  
+
+We greatly appreciate the Reviewers for their thoughtful comments. As outlined below, we have incorporated all comments/suggestions/requests into the revised version of the manuscript in some form. Overall, the comments and suggestions by the Reviewers have greatly improved the readability and clarity of the manuscript.  
+
+Please note that we have modified our figure set slightly, splitting original Figure 1 into two figures (new Figure 1 and 2). We did this because we added new information and panels to the original Figure 1, at the request of Reviewers#1 and #2, making it cumbersome as a single figure.  
+
+New Figure 1 (structure/cartoon figure) (original Figure 1a- 1c). We added new information to the structure/cartoon at the request of Reviewer#1. Making it its own figure we were able to expand its size somewhat making it easier to see all presented information.  
+
+New Figure 2 (single channel recordings) (original Figure 1d- 1i). We added new panels (Figure 2g- i), as requested by Reviewer#2, and having its own figure accommodated these new panels better.  
+
+We present our point- by- point responses to reviews below:  
+
+## Reviewer #1 (Remarks to the Author):  
+
+In the manuscript "Two gates mediate NMDA receptor activity and are under subunit specific regulation" the authors are adding some new pieces into the puzzle of the NMDAR function and regulation.  
+
+Although the provided experimental evidence clearly demonstrates the effects of the G to- A substitutions, the conclusions based on it are often overstated. The authors suggest a redefinition of the common nomenclature, renaming what is usually known as the selectivity filter to a "(second) gate". In fact it could be even a "third" gate, after the M3 SYTAN and the neighboring "LILI gate". The LILI gate is in principle also consistent with the "GluN1 glycine mainly regulates single channel events within a cluster or bursts of activity, whereas the GluN2A glycine mainly regulates entry and exit from clusters".  
+
+We apologize if we overstated our conclusions. We are very much aware of the rich history of the structure- function of NMDA receptors, and many aspects of our conclusions are certainly not novel. For example, the idea that the different subunits regulate different aspects of the gating process is widespread in the literature (see also comments of Reviewer#2) and even our earlier experiments indicated this (e.g., Sobolevsky et al., 2007, JGP). Still, we wanted to highlight the idea that the M2 loop can function as a 'gate' - that it can form a barrier for the flux of ions - and that the M3 gate (SYTAN + LILI) and M2 gate are under subunit specific regulation (though we do not believe this is absolute).  
+
+In addition, the LILI gate and the SYTAN motif were part of our thought process in trying to conceptualize the nature of the upper M3 gate. We would prefer to continue to lump LILI and SYTAN into the 'primary' gate since even in the original publication the
+
+<--- Page Split --->
+
+authors noted that '...the LILI motif...form a functional unit with the TTTT ring...' (Ladislav et al., 2018, FMN).  
+
+Nevertheless, we have added a more extensive discussion of the M3 gate – that it is not a single entity – in the Introduction (p. 4) and Discussion (p. 16) and also include the LILI motif in our new structural figure (Figure 1b). We also note that LILI gate is consistent with subunit- specific regulation in the Discussion (p. 16). We appreciate the suggestions.  
+
+To get a better idea about the GluN1 G- to- A- induced changes in the M2 gate/filter pore diameter experimentally, it would be useful to record the single channel properties also in presence of cations with different sizes.  
+
+The Reviewer brings up an interesting experiment. However, after consideration, we realized that these experiments would be extremely challenging (identifying different conducting states would be extremely hard), and their outcome would be ambiguous in that we are not looking at changes in pore diameter but rather that the M2 functions as a 'gate' – it can prevent the flow of permeant ions. The larger sized organics even in the M2 open confirmation would show slow permeation greatly reducing detectability of currents (these sorts of experiments which we have extensive experience with are invariably done in the whole cell mode). In addition, these experiments would more address that the pore dimension is changed, which we previously addressed by looking at changes in \(\mathrm{Ca^{2 + }}\) permeability (e.g., Amin et al., 2018, Nat. Comm.), than whether it acts as a gate.  
+
+In addition, these sorts of experiments would also be complicated in interpretation because they would start moving into questions about the nature of channel block. Unfortunately, because the mechanism of pore block is incompletely understood, interpreting the results of such an experiment would be difficult. Our goal in this manuscript was to outline the contribution of the M4s in influencing the activity of two separable gates in the ion channel pore.  
+
+In my opinion, the weakest part of the work is the molecular dynamics and I would suggest either significantly improving the MD setup and analysis, or removing the MD part as it in fact does not provide any additional details to the observed GluN1 G- to- A effects. It is questionable whether a model of the rat GluN1/GluN2B (why not GluN2A?) TMD- only based on an symmetrical AMPA "open" structure represents the GluN1/GluN2A system studied experimentally. The significant rearrangement between and within domains during the open to closed transition in GluN1/GluN2B full length model was studied by MD (10.3390/biom9100546), showing important non- symmetrical switching of N1 and N2B M3 helices induced by extracellular domains.  
+
+The receptor is sensitive to the presence of cholesterol in the membrane, using pure POPC could change the TMD behavior. The MD setup is using a rather short electrostatics cutoff of 8A, possibly lowering the accuracy of simulations especially when interested in ion permeation. There is also no transmembrane potential reported. While the experiments involve \(\mathrm{Ca2 + }\) , the MD simulations contain only \(\mathrm{Na + }\) (at least as described in the methods and "guessing" from descriptions using only "ion" without stating which one). Although the PTMs are somehow mentioned, it was described recently that N2 palmitoylation plays a significant role anchoring the base of M4 helices,
+
+<--- Page Split --->
+
+however, it was not included in the model.  
+
+We appreciate the concerns of the Reviewer. We certainly think the MD simulations are a key part of our conclusions, as we outline below.  
+
+First, we acknowledge the significant challenges in realistically simulating the NMDAR open and closed states. As correctly pointed out by the Reviewer, we neglected many details such as cholesterol and PTMs, focusing instead on the features that we felt were the most important for connecting with the electrophysiological studies. Issues about cholesterol and PTM are very important and one we are actively investigating in terms of our model, but these studies go beyond the present study since they add many new experimental manipulations.  
+
+Second, we used GluN1/GluN2B because we accumulated a significant number of simulations (some published in the 2018 NC paper but the vast number were unpublished). A large number of simulations is important for making sure the observations we report are robust. Note that our focus is the difference between the wild- type and the G- to- A mutants, and, as we now note (p. 8), the G- to- A mutation in GluN2B produces the same electrophysiological phenotype as in GluN2A (Amin et al., 2018, Nat. Comm.). In addition, we now state in the text (p. 8- 9) that the permeant ion is \(\mathrm{Na + }\) , and also note the recordings were done without \(\mathrm{Ca2 + }\) (p. 6, 9).  
+
+Third, the open model, based on homology with the AMPAR structure, is not symmetric. The diagonal distance of the GluN2 M3 helices are much longer than the counterpart of the GluN1 M3 helices, in line with electrophysiological data showing greater contribution of the GluN2 subunits to channel gating (e.g., Kazi et al., 2014, Nature Ns). Nevertheless, we did not include the extracellular domains as in Cerny et al., 2019 (new ref. 48). Again, we wanted to carry out as many simulations as possible to increase the number of replicates which was facilitated by using a reduced model.  
+
+Fourth, we initially refrained from carrying out simulations under a transmembrane potential because we wanted to isolate the differences between wild- type and the G- to- A mutants without the interference of external factors, especially considering that a large potential (e.g., \(>300 \mathrm{mV}\) ) might be required to achieve a meaningful number of ion permeation event.  
+
+Nevertheless, we agree with the Reviewer (as well as Reviewer 2) and have now made additional simulations with a transmembrane potential. In these simulations we also increased the nonbonded cutoff distance to 10 A. The outcome of these additional simulations further supports the idea that the M2 loop can form a gate, which is regulated by the GluN1 M4 segment. These new data are included in the Results section (p. 10 in the red- lined version).  
+
+## minor points:  
+
+Is there something missing in the "pathway that select ion(s) can cross"?  
+
+We have reworded this sentence.  
+
+I don't really know what is meant by "membrane physiology".
+
+<--- Page Split --->
+
+We have changed to 'membrane excitability'. Sorry for the ambiguity.  
+
+In the "'Gates' are barriers that occlude the flux of ions in the closed state.", if the above mentioned closed state model is correct, the gates you are talking about are actually relevant in the open state conformation. But this is probably just a missing definition of the closed state (I guess here you mean liganded (activated) receptor not permeable to ions).  
+
+We have added 'non- conducting' to help clarify 'closed state'. We hope this makes our meaning here clearer.  
+
+I am not sure the upstream/downstream of transmembrane helices is clearly defined.  
+
+We apologize for the lack of clarity. We now have replaced 'upstream' with 'N- terminal' and 'downstream' with 'C- terminal'.  
+
+In the "A critical residue in the M4 helices", maybe a better would be one of the critical residues.  
+
+As requested changed to the plural form.  
+
+In the "(eq. Popen) and to the same extent", "and" is not needed or something else is missing?  
+
+As requested, we have removed the extraneous word 'and' from the text.  
+
+In the "closed and opens states", just "open"?  
+
+We have corrected the typo. Thanks for catching.  
+
+I am not very familiar with the current data policy of Nature journals, but the ever more recognized FAIR principles would dictate that the data are deposited in a public repository and not just "available upon request". Although for example the methods section concerning the MD simulations is detailed enough to allow reproducibility, the actual trajectories could be useful to the community.  
+
+We appreciate the Reviewer for pointing out a more rigorous data transparency practice. In our original submission, we included considerable information in the Supplementary Information including extensive tables for many of our figures, figures with additional information as well as additional information on our MD simulations. In the resubmission we have added some more information here. We could post the Excel files associated with these figures/tables but am not sure they would provide anything beyond what is in the figure set and Supplementary Information. We have also followed as far as we know all of the guidelines prescribed by Nature Communications in terms of data transparency.
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+The manuscript entitled "Two gates mediate NMDA receptor activity and are under subunit- specific regulation" is a solid study that provides some interesting data that is relevant to the function of the NMDA receptor. It is well written, and summarizes evidence that individual subunits control different aspects of gating, and also provides further support for functional relevance of the M4 in gating. The conclusions that the GluN1 M4 helix controls M2 helix of GluN2 and GluN2 M4 controls GluN1 M3 are important, and simulations suggest that mutations in the M4 region of different subunits can alter the pore diameter at distinct sites. I think it will be of interest to many working on ligand gated ion channels, glutamate receptors, and synaptic transmission. Below are several suggestions the authors should consider that could improve the clarity of the study and perhaps lead to more accurate and useful conclusions.  
+
+We appreciate the Reviewer's positive and insightful comments about the manuscript and have tried to modify the manuscript and its presentation to accommodate the Reviewer's concern.  
+
+## Major  
+
+1. While the data is convincing in showing a change in the long-lived closed dwell time separating clusters of activity, I am not sure the stated conclusion is helpful (from the discussion) "...a primary gate at the M3 bundle helical crossing controls entry and exit from both long lived and short-lived closed states". That means it controls everything—is that the intended conclusion? Also, where are data supporting entry and exit from short-lived states? There are data in supplemental but they are not really discussed.  
+
+We apologize for not making our conclusions and the limitation of our conclusions clearer. Part of the challenge is that there remain many unanswered questions so for certain things we can give 'conclusions' and in other instances we can only give 'best guesses'. Clearly, we did not do a good job of clarifying these issues.  
+
+Based on circumstantial evidence (and we continue to work to resolve this issue but it is challenging), we believe that the M2 gate is the major regulator of C1, C2, and C3. At present, our data supports the idea that, at least under our conditions, that the M3 gate mainly regulates C4 and C5 and the M2 gate partially regulates C1, C2, and C3, but that none of these regulatory events are absolute: the M2 gate also impacts C4 and C5 and the M3 gate impacts C1, C2, and C3. Part of the challenge is that the energetics of these gates are most likely linked – the open/closed status of the M3 gate impacts the energetics of the M2 loop and vice-versa. A second challenge is that the M2 gate is 'behind' the M3 gate and cannot be studied independently (except for our efforts trying to lock 'open' the M3 gate, which has certain limitations). Finally, we believe that the fast energetic events with the initial opening event following initial agonist binding (what determines the activation rate of whole-cell currents) may or may not be structurally
+
+<--- Page Split --->
+
+related to the fast closed states measured at steady- state. Again we are trying to test these ideas (using outside- out patches, strategic mutations and various agonists) but these are very extensive experiments and go beyond the present manuscript.  
+
+In any case, we have added the data for C1, C2, and C3 to the new Figure 2 (panels g- i) (as well as Supplemental Figure 1) to emphasize that the GluN1 G- A has significant effects on C1, while GluN1 and GluN2A G- A have comparable effects on C2 and C3. We have modified the text, mainly in the Discussion (pp 14- 15) to suggest that the M3 gate mainly regulates C4 and C5 while the M2 gate mainly regulates C1, C2, and C3 but that these events are not absolute. We also try to emphasize the complex energetics of these gates. The energetics keeping the M3 gate open (and hence the cluster/burst length) is not only dependent upon the M3- S2/pre- M1/S2- M4, but also the status of the M2 loop and perhaps vice- versa. Hence, the duration of cluster/bursts are also influenced by the energetics of the M2 loop.  
+
+2. If the 12 sec dwell time reflected an important step in the pathway from agonist binding to channel opening (i.e. gating), then the rise times would be far slower than \(\sim 10\) ms reported by dozens of authors. However most kinetic studies conclude that these prolonged closed periods in between clusters of activity are some form of agonist-dependent desensitization, and I think a likely conclusion is that the M4 regulates recovery from a desensitized state. Seems like this should be raised or evaluated in terms of double pulse experiments that assess the recovery from the desensitized state.  
+
+We agree completely with the Reviewer – that the S2- M4/M4 is regulating aspects of desensitization, and we apologize for not making this clear in the original submission. We have now expanded on this point in the Discussion (p 15).  
+
+Note we did record recovery for desensitization for GluN2A G- A as requested but did not add to the manuscript. We do see an increase in the time required for recovery to occur \((1.7 \pm 0.2\) sec, \(n = 8\) ) for GluN2A G- A versus wild- type \((0.97 \pm 0.09\) sec, \(n = 7\) ). Clearly, the S2- M4/M4 is involved in regulating desensitization but this will require much more detailed investigations to resolve and goes beyond the present study.  
+
+3. Opening rates have been interpreted in early studies of frog neuromuscular nicotinic receptors from brief "nachschlag" closed states. The idea is the pore reverses a gating step during these brief gaps, and then proceeds forward to reopen. If the dwell time reflects a single visitation to one closed state, then the reciprocal of the duration is the rate associated with forward opening rate. Studies from Colquhoun, Auerbach, Traynelis, Gibb and others all showed three brief intra-cluster closed states that were interpreted as a very fast opening (perhaps pore dilation), and two other states with dwell times of \(\sim 1\) and \(\sim 5\) ms. If the author wants to conclude something about gating within a cluster, they should discuss some of the changes in various components of the histograms shown in the supplemental material) or provide some modelling results.  
+
+This is an outstanding point and one to be honest we struggle perhaps the most with. As indicated in comments to point 1, we added mean closed duration ratios for C1, C2, and C3 to Figure 2 to better analyze those brief states. We now mention that GluN1 G- A
+
+<--- Page Split --->
+
+has a somewhat greater influence on these brief closed states (specially C1) than GluN2A G- A but also note that for C1, C2, and C3 this is not very absolute (in contrast to C4 and C5). Importantly, we wish to emphasize (and have done so in the Discussion) that our interpretation of this data is limited by the high likelihood of energetic coupling between the two gates.  
+
+In any case the Reviewer raises some critical considerations that are ultimately essential to resolve to better define the gating mechanism in NMDARs. At present our data set cannot resolve because of numerous complications (see response to Point 1).  
+
+4. I think it is important to explicitly state whether the opening of the two gates occurs in any order or in a dependent order (e.g. slow step first, then fast step), as multiple Popescu papers have suggested. Dependent order requires slow step to be first, followed by a fast step and then opening, otherwise there would never be brief sojourns to the closed state in the single channel record if the brief step occurred first and then the slow step immediately preceded opening.  
+
+This comment brings up an interesting insight and one we cannot with the present data set directly answer. Nevertheless, we assume that the M3 gate must open first, and we speculate on this in the Discussion, though we note that our data does not directly address, and thus a definitive answer is better left for future studies.  
+
+## Minor  
+
+5. I think the abstracts reference to "clusters" is vague, and bordering on jargon. How many readers interested in synaptic transmission will know what is meant by a cluster? I think the authors need to re-word to a conclusion that a wider audience can grasp and understand.  
+
+We thank the Reviewer for this suggestion and agree that many physiologists will not understand the significance of the word. We have added a sentence to the Abstract to better clarify this term. We hope this will make better sense to synaptic physiologists.  
+
+6. Minor suggested wording change: Page 4 "A variety of disease associated missense mutations have been identified at OR NEAR these conserved glycines...(could choose any of multiple refs to support)  
+
+We thank the Reviewer for suggesting this change, which better acknowledges that mutations near this site may also act via the mechanisms we outline in this work. The change has been made and several citations added (Chen et al., 2017, Mol. Pharm.; Perszyk et al., 2020, JPhysiol).  
+
+7. Page 4 sentence starting "These data suggest that the M3 gate mediates the longer..." How can you rule out that both gates are important, and because the long dwell time dominates the interval, you are missing changes to the interval with the faster gate changes its rate. That is, both may be required, but you can only see changes in the slow gate (contributing the long delay). A short 10 ms change will hardly alter a 12
+
+<--- Page Split --->
+
+second interval.  
+
+As we discuss above and now in the Discussion, we think the gates are strongly energetically coupled. Indeed, C4 and C5 closed durations are significantly changed when we used GluN1 G- A to disrupt M2, just to a much less significant extend compared to GluN2.  
+
+8. Typo in text on page 6. Duration of long lived state should be sec, not ms in the last sentence.  
+
+We thank the Reviewer for catching this error! The text has now been changed.  
+
+9. Page 8, first sentence of the second paragraph - Alasdair Gibb should get the credit for first showing unambiguously that there are five independent closed states in individual activations of a single native NMDA receptors in his clever low concentration experiment (Gibb et al., 1992). The references (40,41) confirmed his conclusions in heterologous systems—Schorge et al I think should be added.  
+
+We appreciate the Reviewer's insights. As requested, we have added the suggested citations.  
+
+10. There are several papers entirely devoted to the idea that the GluN1 M4 functionally interacts with the GluN2 M3, and the GluN2 M4 interacts with the GluN1 M3 (Chen et al. 2017 exploring actions of a pre-M4 mutation, also Gibb 2018). Chen concludes the interaction of GluN2-M4 with GluN1 M3 is critical for gating, consistent with the conclusions here. Seems as though the conclusions of these papers should be mentioned, which are consistent to some extent with what is being showing here?  
+
+We thank the Reviewer for encouraging a broader discussion of the existing literature. These citations at multiple points have now been added to the Introduction and/or /Discussion.  
+
+11. I would recommend mentioning in the results that saturating agonists are used so a reader doesn't have to turn to the methods or scrutinize the legends.  
+
+We appreciate the suggestion from the Reviewer and have now added this information to the Results section.  
+
+12. I think it is important to comment that these results were obtained in the absence of divalents. Schorge et al recorded in divalent ions and was unable to reproduce some channel properties observed in the absence of divalents. Alternatively, one could perform a quick experiment in a patch with 0.5-1 mM Ca2+ to demonstrate that the conclusions hold in a physiological context.  
+
+Both Reviewers aptly point out that our original work could have been clearer about the absence of divalents in our experiments, which was done to give better resolution and to
+
+<--- Page Split --->
+
+remove confounding variables from our experiments. Text has now been added (p. 6, 8- 9) to the results that directly clarifies this. We also agree that testing \(\mathrm{Ca^{2 + }}\) is a critical question, but feel this is a whole set of new experiments to rigorously address (again which we are trying to do).  
+
+13. I suspect a subset of readers will want to know what Tcrit was used for cluster analysis when they read about the analysis in the results without stopping their train of thought to turn back to the methods to figure out how you separated clusters, especially for mutations that shortened the long closed duration.  
+
+We thank the Reviewer for encouraging clarification of our results and to make them more accessible to a wider audience. We now include this information in the Results section (p. 7).  
+
+14. Figure 1D: It looks like there are sublevels for the GluN1-G815A mutation, and possibly even the GluN2A-G819A mutation. This seems relevant for a paper discussing pore diameter. Why not show openings at a higher resolution and analyze the sublevels?  
+
+The Reviewer points out a possibility we considered previously. There are several problems here. First, the openings for GluN1- G815A are so brief, we are not able to properly and consistently identify any subconductance states. Second, at present it is not clear the nature of subconductance levels in NMDARs and we feel this would require a whole new (and extensive) study to define (which we are trying to do but with mutations that induce more robust subconductance levels).  
+
+15. The K channel literature is fascinating because the authors in the structural era connect their results to work performed before structures were known. Seems like Banke's conclusion ten years ago (2003) before the structure of the NMDA receptor, "These data suggest that NR1 and NR2B subunits, respectively, undergo a fast and slow agonist-dependent conformational change that precedes opening of the pore", is relevant to the conclusions of the current manuscript. I believe that data in Jones et al 2002 with MTSEA also suggests different roles of GluN1 and GluN2 in gating. The current manuscript goes light years beyond these older studies, but it seems generally useful to recognize how early ideas about subunit dependent gating arose in the literature.  
+
+We apologize for not recognizing these earlier works in terms of the subunit specific effects on gating. We now include them as well as several others in the Discussion.  
+
+16. Near the middle of page 9, could the authors specify that 15 replicate simulations of 500ns each were run per construct? They specify 15 in the Methods section, but also specifying it here would save time from having to go to the Methods and help improve clarity for the reader.  
+
+As requested, we now include the number of replicates in this section.
+
+<--- Page Split --->
+
+17. In Figure 2 and the related text in Results, positions \(+2\) , \(+6\) , and \(+10\) aren't defined as being indexed from the S in SYTANLAAF, but it is clarified in the Supplementary info. It would be useful to define it in the text, either by verbally explaining this in the legend for Figure 2 or by moving panel 'a' from Supp Fig 3 to Fig 2 in the main text. I don't see that the magenta sphere is defined in the caption for Fig. 2a. Caption for Fig. 2g refers to left and right panels when the panels are stacked vertically  
+
+These are very helpful suggestions and appreciate the Reviewer's efforts. We considered moving Supplementary Figure 3a to the main body but realized it would make this figure (Figure 2) even more complicated. We now explicitly state the referencing of \(+2\) , \(+6\) , and \(+10\) in the text (p. 9). We also now define the magenta sphere in the Figure legend. Finally, the figure legend has been altered to reflect the correct positions of the panels.  
+
+18. Has the model of the active state used in the MD simulations been used in previous publications? If so, a reference should be given as it would be helpful to see model quality metrics, as well as other information about the simulations such as RMSD plots, included in the supplemental information. If not, perhaps these quality metrics could be added as supplemental information.  
+
+We previously reported the model in Amin et al., 2018, Nature Communications, where many of these metrics, including RMSD plots, were included. This is now noted in the manuscript.  
+
+19. Given that ion permeation was analyzed in the MD simulations, can the authors comment on why a membrane potential wasn't modeled using a method such as applying a constant electric field or using a polarizable force field? For reference, see https://doi.org/10.1021/acs.jctc.5b01202. The simulations as they are model the behavior of ions in the absence of any driving force. If one of these methods for the simulations had been used, I feel that the data presented in Figure 2d and e would be more meaningful and informative. However, it is obviously too much to re-do. Perhaps the authors could add a comment about this in the manuscript as a caveat?  
+
+This is a very valid concern. We initially refrained from carrying out simulations under a transmembrane potential because we wanted to isolate the differences between wild- type and the G- to- A mutants without the interference of external factors, especially considering the fact that a large potential (e.g., \(>300 \text{mV}\) ) might be required to achieve a meaningful number of ion permeation event. Nevertheless, we agree with the Reviewer (as well as Reviewer 1) and have now made additional simulations with a transmembrane potential. The outcome of these additional simulations further supports the idea that the M2 loop can form a gate, which is regulated by the GluN1 M4 segment. These new data are included in the Results section (p. 10 in the red- lined version).  
+
+20. Are the values in Figure 2e from a single trajectory, or averaged across the 15 replicates? If they are from a single trajectory, can you explain how that specific trajectory was selected? Can the authors provide an explanation of how individual
+
+<--- Page Split --->
+
+trajectories were selected for the plots in Supplementary Figure 4?  
+
+The values in Figure 2e are the total crossing from 15 separate replicates, which now is explained better in the Figure 2 legend.  
+
+Trajectories in Figure 2d and Supplementary Figure 4 (they from the same simulation). We selected this example since they all came from the same simulation which showed the most transitions of all the replicates we made. We now explain this in the Supplementary Figure 4 legend.  
+
+21. Could the authors add to the Discussion their thoughts on why there are two long-lived closed states if the general claim is that the N2A M4s predominately regulate entry into and exit from long-lived closed states?  
+
+This is an outstanding question and one we have thought lots about. Our general idea, and we are certainly not the first to think of this, is that there are multiple structural elements regulating the opening of the M3 gate. The pre-M1 helix, the M3-S2 linkers, the S2-M4/M4 regions, among other structural elements. C4 and C5 presumably reflect to some extent all of these structural elements.  
+
+In short we speculate that a modifying process is occurring for our two long lived closed states, with the M4 movement being the primary driver, but with alternative structural confirmations bifurcating this process into two outcomes. However, we cannot delineate the culprit structural elements in this work without further experiments, the exploration of which is for another manuscript.  
+
+We have added some of this information to the Discussion (pp 14- 15).  
+
+22. Page 18: "although evidence from our MD simulations and single channel recordings suggest that reductions in the pore diameter of the M2 gate correlate with reduced ion permeation...", I don't really see evidence of this in the supplementary or main figures at least in terms of single channels for permeation. Maybe I missed something.  
+
+We have corrected this sentence [I assume what's meant by this sentence was correlation b/w pore radius and \(\mathrm{Ca2 + }\) permeability as shown in 2018NC].  
+
+23. Y-axes of Fig. 1h and 1i could be more clearly labeled (perhaps 'Closed duration ratio (G-A/wt)') Similarly, the word 'closed' should be added between 'mean' and 'durations' in the last sentence on pg. 8. Minor but might help keep the message clear.  
+
+We thank the Reviewer for their suggestion and have now altered Figure 1h and 1i to state "Closed duration ratio (G-A/wt)" and changed to 'mean closed durations'.
+
+<--- Page Split --->
+
+Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+Dear Authors, thank you for the revised version of the manuscript. I believe that the introduced changes sufficiently addressed my comments. As already said, the experimental evidence of G- to- A effects is solid and I understand that performing complicated new experiments or modified MD simulations at time scales comparable to the GluN2B ones would be impractical now.  
+
+You are right that the Supplementary material is rich and the data is treated according to the rules. My appeal was more like "it would be nice to have the raw trajectories as well". It would, like for example in structural biology field, allow reanalysis or even more importantly development or benchmarking of new data processing methods. Obviously the infrastructure allowing efficient storage of large datasets in an interoperable way is not very mature (neither zenodo nor say model archive are perfect for this purpose yet) but we are in a hen and egg situation here, without researchers offering their raw data, the repositories will evolve only slowly (if ever) into a useful and user- friendly data resource.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+The authors revised manuscript has adequately addressed the points raised. Congratulations on the completion of a very nice study.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__41f73bb80f7f4860b41cedbe7ceb9226dfa693d3d58be9b65f6123b0558820c4/supplementary_0_Peer Review File__41f73bb80f7f4860b41cedbe7ceb9226dfa693d3d58be9b65f6123b0558820c4_det.mmd

@@ -0,0 +1,502 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 155, 833, 211]]<|/det|>
+Two gates mediate NMDA receptor activity and are under subunit- specific regulation  
+
+<|ref|>image<|/ref|><|det|>[[93, 732, 262, 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, 286, 104]]<|/det|>
+Reviewers' Comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 120, 291, 149]]<|/det|>
+Reviewer #1: Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[116, 150, 825, 194]]<|/det|>
+In the manuscript "Two gates mediate NMDA receptor activity and are under subunit- specific regulation" the authors are adding some new pieces into the puzzle of the NMDAR function and regulation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 209, 802, 238]]<|/det|>
+Although the provided experimental evidence clearly demonstrates the effects of the G- to- A substitutions, the conclusions based on it are often overstated.  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 865, 314]]<|/det|>
+The authors suggest a redefinition of the common nomenclature, renaming what is usually known as the selectivity filter to a "(second) gate". In fact it could be even a "third" gate, after the M3 SYTAN and the neighboring "LILI gate". The LILI gate is in principle also consistent with the "GluN1 glycine mainly regulates single channel events within a cluster or bursts of activity, whereas the GluN2A glycine mainly regulates entry and exit from clusters".  
+
+<|ref|>text<|/ref|><|det|>[[115, 328, 864, 373]]<|/det|>
+To get a better idea about the GluN1 G- to- A- induced changes in the M2 gate/filter pore diameter experimentally, it would be useful to record the single channel properties also in presence of cations with different sizes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 387, 876, 507]]<|/det|>
+In my opinion, the weakest part of the work is the molecular dynamics and I would suggest either significantly improving the MD setup and analysis, or removing the MD part as it in fact does not provide any additional details to the observed GluN1 G- to- A effects. It is questionable whether a model of the rat GluN1/GluN2B (why not GluN2A?) TMD- only based on an symmetrical AMPA "open" structure represents the GluN1/GluN2A system studied experimentally. The significant rearrangement between and within domains during the open to closed transition in GluN1/GluN2B full length model was studied by MD (10.3390/biom9100546), showing important non- symmetrical switching of N1 and N2B M3 helices induced by extracellular domains.  
+
+<|ref|>text<|/ref|><|det|>[[115, 507, 880, 625]]<|/det|>
+The receptor is sensitive to the presence of cholesterol in the membrane, using pure POPC could change the TMD behavior. The MD setup is using a rather short electrostatics cutoff of 8A, possibly lowering the accuracy of simulations especially when interested in ion permeation. There is also no transmembrane potential reported. While the experiments involve \(\text{Ca2 + }\) , the MD simulations contain only \(\text{Na + }\) (at least as described in the methods and "guessing" from descriptions using only "ion" without stating which one). Although the PTMs are somehow mentioned, it was described recently that N2 palmitoylation plays a significant role anchoring the base of M4 helices, however, it was not included in the model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 641, 215, 655]]<|/det|>
+minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 670, 661, 686]]<|/det|>
+Is there something missing in the "pathway that select ion(s) can cross"?  
+
+<|ref|>text<|/ref|><|det|>[[115, 686, 576, 701]]<|/det|>
+I don't really know what is meant by "membrane physiology".  
+
+<|ref|>text<|/ref|><|det|>[[115, 701, 875, 760]]<|/det|>
+In the "Gates' are barriers that occlude the flux of ions in the closed state.", if the above mentioned closed state model is correct, the gates you are talking about are actually relevant in the open state conformation. But this is probably just a missing definition of the closed state (I guess here you mean liganded (activated) receptor not permeable to ions).  
+
+<|ref|>text<|/ref|><|det|>[[115, 760, 752, 775]]<|/det|>
+I am not sure the upstream/downstream of transmembrane helices is clearly defined.  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 833, 820]]<|/det|>
+In the "A critical residue in the M4 helices", maybe a better would be one of the critical residues. In the "(eq. Popen) and to the same extent", "and" is not needed or something else is missing? In the "closed and opens states", just "open"?  
+
+<|ref|>text<|/ref|><|det|>[[115, 835, 872, 894]]<|/det|>
+I am not very familiar with the current data policy of Nature journals, but the ever more recognized FAIR principles would dictate that the data are deposited in a public repository and not just "available upon request". Although for example the methods section concerning the MD simulations is detailed enough to allow reproducibility, the actual trajectories could be useful to the community.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 133, 218, 146]]<|/det|>
+Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 291, 162]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 163, 874, 313]]<|/det|>
+The manuscript entitled "Two gates mediate NMDA receptor activity and are under subunit- specific regulation" is a solid study that provides some interesting data that is relevant to the function of the NMDA receptor. It is well written, and summarizes evidence that individual subunits control different aspects of gating, and also provides further support for functional relevance of the M4 in gating. The conclusions that the GluN1 M4 helix controls M2 helix of GluN2 and GluN2 M4 controls GluN1 M3 are important, and simulations suggest that mutations in the M4 region of different subunits can alter the pore diameter at distinct sites. I think it will be of interest to many working on ligand gated ion channels, glutamate receptors, and synaptic transmission. Below are several suggestions the authors should consider that could improve the clarity of the study and perhaps lead to more accurate and useful conclusions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 328, 159, 342]]<|/det|>
+## Major  
+
+<|ref|>text<|/ref|><|det|>[[115, 356, 860, 446]]<|/det|>
+1. While the data is convincing in showing a change in the long-lived closed dwell time separating clusters of activity, I am not sure the stated conclusion is helpful (from the discussion) "...a primary gate at the M3 bundle helical crossing controls entry and exit from both long lived and short-lived closed states". That means it controls everything—is that the intended conclusion? Also, where are data supporting entry and exit from short-lived states? There are data in supplemental but they are not really discussed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 876, 550]]<|/det|>
+2. If the 12 sec dwell time reflected an important step in the pathway from agonist binding to channel opening (i.e. gating), then the rise times would be far slower than \(\sim 10\) ms reported by dozens of authors. However most kinetic studies conclude that these prolonged closed periods in between clusters of activity are some form of agonist-dependent desensitization, and I think a likely conclusion is that the M4 regulates recovery from a desensitized state. Seems like this should be raised or evaluated in terms of double pulse experiments that assess the recovery from the desensitized state.  
+
+<|ref|>text<|/ref|><|det|>[[115, 564, 880, 700]]<|/det|>
+3. Opening rates have been interpreted in early studies of frog neuromuscular nicotinic receptors from brief "nachschlag" closed states. The idea is the pore reverses a gating step during these brief gaps, and then proceeds forward to reopen. If the dwell time reflects a single visitation to one closed state, then the reciprocal of the duration is the rate associated with forward opening rate. Studies from Colquhoun, Auerbach, Traynelis, Gibb and others all showed three brief intra-cluster closed states that were interpreted as a very fast opening (perhaps pore dilation), and two other states with dwell times of \(\sim 1\) and \(\sim 5\) ms. If the author wants to conclude something about gating within a cluster, they should discuss some of the changes in various components of the histograms shown in the supplemental material) or provide some modelling results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 714, 879, 789]]<|/det|>
+4. I think it is important to explicitly state whether the opening of the two gates occurs in any order or in a dependent order (e.g. slow step first, then fast step), as multiple Popescu papers have suggested. Dependent order requires slow step to be first, followed by a fast step and then opening, otherwise there would never be brief sojourns to the closed state in the single channel record if the brief step occurred first and then the slow step immediately preceded opening.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 804, 158, 817]]<|/det|>
+## Minor  
+
+<|ref|>text<|/ref|><|det|>[[115, 848, 866, 893]]<|/det|>
+5. I think the abstracts reference to "clusters" is vague, and bordering on jargon. How many readers interested in synaptic transmission will know what is meant by a cluster? I think the authors need to re-word to a conclusion that a wider audience can grasp and understand.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 103, 875, 134]]<|/det|>
+6. Minor suggested wording change: Page 4 "A variety of disease associated missense mutations have been identified at OR NEAR these conserved glycines...(could choose any of multiple refs to support)  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 866, 223]]<|/det|>
+7. Page 4 sentence starting "These data suggest that the M3 gate mediates the longer..." How can you rule out that both gates are important, and because the long dwell time dominates the interval, you are missing changes to the interval with the faster gate changes its rate. That is, both may be required, but you can only see changes in the slow gate (contributing the long delay). A short 10 ms change will hardly alter a 12 second interval.  
+
+<|ref|>text<|/ref|><|det|>[[115, 237, 839, 253]]<|/det|>
+8. Typo in text on page 6. Duration of long lived state should be sec, not ms in the last sentence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 267, 879, 342]]<|/det|>
+9. Page 8, first sentence of the second paragraph -- Alasdair Gibb should get the credit for first showing unambiguously that there are five independent closed states in individual activations of a single native NMDA receptors in his clever low concentration experiment (Gibb et al., 1992). The references (40,41) confirmed his conclusions in heterologous systems—Schorge et al I think should be added.  
+
+<|ref|>text<|/ref|><|det|>[[115, 357, 878, 431]]<|/det|>
+10. There are several papers entirely devoted to the idea that the GluN1 M4 functionally interacts with the GluN2 M3, and the GluN2 M4 interacts with the GluN1 M3 (Chen et al. 2017 exploring actions of a pre-M4 mutation, also Gibb 2018). Chen concludes the interaction of GluN2-M4 with GluN1 M3 is critical for gating, consistent with the conclusions here. Seems as though the conclusions of these papers should be mentioned, which are consistent to some extent with what is being showing here?  
+
+<|ref|>text<|/ref|><|det|>[[115, 446, 825, 476]]<|/det|>
+11. I would recommend mentioning in the results that saturating agonists are used so a reader doesn't have to turn to the methods or scrutinize the legends.  
+
+<|ref|>text<|/ref|><|det|>[[115, 491, 865, 550]]<|/det|>
+12. I think it is important to comment that these results were obtained in the absence of divalents. Schorge et al recorded in divalent ions and was unable to reproduce some channel properties observed in the absence of divalents. Alternatively, one could perform a quick experiment in a patch with \(0.5 - 1 \text{mM Ca2 + }\) to demonstrate that the conclusions hold in a physiological context.  
+
+<|ref|>text<|/ref|><|det|>[[115, 565, 875, 625]]<|/det|>
+13. I suspect a subset of readers will want to know what Tcrit was used for cluster analysis when they read about the analysis in the results without stopping their train of thought to turn back to the methods to figure out how you separated clusters, especially for mutations that shortened the long closed duration.  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 870, 685]]<|/det|>
+14. Figure 1D: It looks like there are sublevels for the GluN1-G815A mutation, and possibly even the GluN2A-G819A mutation. This seems relevant for a paper discussing pore diameter. Why not show openings at a higher resolution and analyze the sublevels?  
+
+<|ref|>text<|/ref|><|det|>[[115, 700, 881, 819]]<|/det|>
+15. The K channel literature is fascinating because the authors in the structural era connect their results to work performed before structures were known. Seems like Banke's conclusion ten years ago (2003) before the structure of the NMDA receptor, "These data suggest that NR1 and NR2B subunits, respectively, undergo a fast and slow agonist-dependent conformational change that precedes opening of the pore", is relevant to the conclusions of the current manuscript. I believe that data in Jones et al 2002 with MTSEA also suggests different roles of GluN1 and GluN2 in gating. The current manuscript goes light years beyond these older studies, but it seems generally useful to recognize how early ideas about subunit dependent gating arose in the literature.  
+
+<|ref|>text<|/ref|><|det|>[[115, 834, 876, 879]]<|/det|>
+16. Near the middle of page 9, could the authors specify that 15 replicate simulations of 500ns each were run per construct? They specify 15 in the Methods section, but also specifying it here would save time from having to go to the Methods and help improve clarity for the reader.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 876, 179]]<|/det|>
+17. In Figure 2 and the related text in Results, positions \(+2\) , \(+6\) , and \(+10\) aren't defined as being indexed from the S in SYTANLAAF, but it is clarified in the Supplementary info. It would be useful to define it in the text, either by verbally explaining this in the legend for Figure 2 or by moving panel 'a' from Supp Fig 3 to Fig 2 in the main text. I don't see that the magenta sphere is defined in the caption for Fig. 2a. Caption for Fig. 2g refers to left and right panels when the panels are stacked vertically  
+
+<|ref|>text<|/ref|><|det|>[[115, 193, 872, 253]]<|/det|>
+18. Has the model of the active state used in the MD simulations been used in previous publications? If so, a reference should be given as it would be helpful to see model quality metrics, as well as other information about the simulations such as RMSD plots, included in the supplemental information. If not, perhaps these quality metrics could be added as supplemental information.  
+
+<|ref|>text<|/ref|><|det|>[[115, 267, 877, 372]]<|/det|>
+19. Given that ion permeation was analyzed in the MD simulations, can the authors comment on why a membrane potential wasn't modeled using a method such as applying a constant electric field or using a polarizable force field? For reference, see https://doi.org/10.1021/acs.jctc.5b01202. The simulations as they are model the behavior of ions in the absence of any driving force. If one of these methods for the simulations had been used, I feel that the data presented in Figure 2d and e would be more meaningful and informative. However, it is obviously too much to re-do. Perhaps the authors could add a comment about this in the manuscript as a caveat?  
+
+<|ref|>text<|/ref|><|det|>[[115, 386, 870, 445]]<|/det|>
+20. Are the values in Figure 2e from a single trajectory, or averaged across the 15 replicates? If they are from a single trajectory, can you explain how that specific trajectory was selected? Can the authors provide an explanation of how individual trajectories were selected for the plots in Supplementary Figure 4?  
+
+<|ref|>text<|/ref|><|det|>[[115, 460, 866, 506]]<|/det|>
+21. Could the authors add to the Discussion their thoughts on why there are two long-lived closed states if the general claim is that the N2A M4s predominately regulate entry into and exit from long-lived closed states?  
+
+<|ref|>text<|/ref|><|det|>[[115, 521, 875, 581]]<|/det|>
+22. Page 18: "although evidence from our MD simulations and single channel recordings suggest that reductions in the pore diameter of the M2 gate correlate with reduced ion permeation...", I don't really see evidence of this in the supplementary or main figures at least in terms of single channels for permeation. Maybe I missed something.  
+
+<|ref|>text<|/ref|><|det|>[[115, 595, 870, 640]]<|/det|>
+23. Y-axes of Fig. 1h and 1i could be more clearly labeled (perhaps 'Closed duration ratio (G-A/wt)') Similarly, the word 'closed' should be added between 'mean' and 'durations' in the last sentence on pg. 8. Minor but might help keep the message clear.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 564, 108]]<|/det|>
+## Response to Reviewers. Nature Communication  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 881, 143]]<|/det|>
+Manuscript#: NCOMMS- 22- 43338- T. "Two gates mediate NMDA receptor activity and are under subunit- specific regulation'  
+
+<|ref|>text<|/ref|><|det|>[[115, 159, 881, 231]]<|/det|>
+We greatly appreciate the Reviewers for their thoughtful comments. As outlined below, we have incorporated all comments/suggestions/requests into the revised version of the manuscript in some form. Overall, the comments and suggestions by the Reviewers have greatly improved the readability and clarity of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 236, 881, 307]]<|/det|>
+Please note that we have modified our figure set slightly, splitting original Figure 1 into two figures (new Figure 1 and 2). We did this because we added new information and panels to the original Figure 1, at the request of Reviewers#1 and #2, making it cumbersome as a single figure.  
+
+<|ref|>text<|/ref|><|det|>[[115, 306, 881, 358]]<|/det|>
+New Figure 1 (structure/cartoon figure) (original Figure 1a- 1c). We added new information to the structure/cartoon at the request of Reviewer#1. Making it its own figure we were able to expand its size somewhat making it easier to see all presented information.  
+
+<|ref|>text<|/ref|><|det|>[[115, 358, 881, 410]]<|/det|>
+New Figure 2 (single channel recordings) (original Figure 1d- 1i). We added new panels (Figure 2g- i), as requested by Reviewer#2, and having its own figure accommodated these new panels better.  
+
+<|ref|>text<|/ref|><|det|>[[115, 417, 626, 436]]<|/det|>
+We present our point- by- point responses to reviews below:  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 443, 466, 460]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 461, 850, 513]]<|/det|>
+In the manuscript "Two gates mediate NMDA receptor activity and are under subunit specific regulation" the authors are adding some new pieces into the puzzle of the NMDAR function and regulation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 530, 870, 670]]<|/det|>
+Although the provided experimental evidence clearly demonstrates the effects of the G to- A substitutions, the conclusions based on it are often overstated. The authors suggest a redefinition of the common nomenclature, renaming what is usually known as the selectivity filter to a "(second) gate". In fact it could be even a "third" gate, after the M3 SYTAN and the neighboring "LILI gate". The LILI gate is in principle also consistent with the "GluN1 glycine mainly regulates single channel events within a cluster or bursts of activity, whereas the GluN2A glycine mainly regulates entry and exit from clusters".  
+
+<|ref|>text<|/ref|><|det|>[[115, 686, 882, 826]]<|/det|>
+We apologize if we overstated our conclusions. We are very much aware of the rich history of the structure- function of NMDA receptors, and many aspects of our conclusions are certainly not novel. For example, the idea that the different subunits regulate different aspects of the gating process is widespread in the literature (see also comments of Reviewer#2) and even our earlier experiments indicated this (e.g., Sobolevsky et al., 2007, JGP). Still, we wanted to highlight the idea that the M2 loop can function as a 'gate' - that it can form a barrier for the flux of ions - and that the M3 gate (SYTAN + LILI) and M2 gate are under subunit specific regulation (though we do not believe this is absolute).  
+
+<|ref|>text<|/ref|><|det|>[[115, 826, 881, 878]]<|/det|>
+In addition, the LILI gate and the SYTAN motif were part of our thought process in trying to conceptualize the nature of the upper M3 gate. We would prefer to continue to lump LILI and SYTAN into the 'primary' gate since even in the original publication the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 881, 125]]<|/det|>
+authors noted that '...the LILI motif...form a functional unit with the TTTT ring...' (Ladislav et al., 2018, FMN).  
+
+<|ref|>text<|/ref|><|det|>[[115, 126, 881, 213]]<|/det|>
+Nevertheless, we have added a more extensive discussion of the M3 gate – that it is not a single entity – in the Introduction (p. 4) and Discussion (p. 16) and also include the LILI motif in our new structural figure (Figure 1b). We also note that LILI gate is consistent with subunit- specific regulation in the Discussion (p. 16). We appreciate the suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 229, 872, 281]]<|/det|>
+To get a better idea about the GluN1 G- to- A- induced changes in the M2 gate/filter pore diameter experimentally, it would be useful to record the single channel properties also in presence of cations with different sizes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 299, 882, 473]]<|/det|>
+The Reviewer brings up an interesting experiment. However, after consideration, we realized that these experiments would be extremely challenging (identifying different conducting states would be extremely hard), and their outcome would be ambiguous in that we are not looking at changes in pore diameter but rather that the M2 functions as a 'gate' – it can prevent the flow of permeant ions. The larger sized organics even in the M2 open confirmation would show slow permeation greatly reducing detectability of currents (these sorts of experiments which we have extensive experience with are invariably done in the whole cell mode). In addition, these experiments would more address that the pore dimension is changed, which we previously addressed by looking at changes in \(\mathrm{Ca^{2 + }}\) permeability (e.g., Amin et al., 2018, Nat. Comm.), than whether it acts as a gate.  
+
+<|ref|>text<|/ref|><|det|>[[115, 474, 882, 578]]<|/det|>
+In addition, these sorts of experiments would also be complicated in interpretation because they would start moving into questions about the nature of channel block. Unfortunately, because the mechanism of pore block is incompletely understood, interpreting the results of such an experiment would be difficult. Our goal in this manuscript was to outline the contribution of the M4s in influencing the activity of two separable gates in the ion channel pore.  
+
+<|ref|>text<|/ref|><|det|>[[115, 595, 876, 730]]<|/det|>
+In my opinion, the weakest part of the work is the molecular dynamics and I would suggest either significantly improving the MD setup and analysis, or removing the MD part as it in fact does not provide any additional details to the observed GluN1 G- to- A effects. It is questionable whether a model of the rat GluN1/GluN2B (why not GluN2A?) TMD- only based on an symmetrical AMPA "open" structure represents the GluN1/GluN2A system studied experimentally. The significant rearrangement between and within domains during the open to closed transition in GluN1/GluN2B full length model was studied by MD (10.3390/biom9100546), showing important non- symmetrical switching of N1 and N2B M3 helices induced by extracellular domains.  
+
+<|ref|>text<|/ref|><|det|>[[115, 752, 872, 891]]<|/det|>
+The receptor is sensitive to the presence of cholesterol in the membrane, using pure POPC could change the TMD behavior. The MD setup is using a rather short electrostatics cutoff of 8A, possibly lowering the accuracy of simulations especially when interested in ion permeation. There is also no transmembrane potential reported. While the experiments involve \(\mathrm{Ca2 + }\) , the MD simulations contain only \(\mathrm{Na + }\) (at least as described in the methods and "guessing" from descriptions using only "ion" without stating which one). Although the PTMs are somehow mentioned, it was described recently that N2 palmitoylation plays a significant role anchoring the base of M4 helices,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 479, 107]]<|/det|>
+however, it was not included in the model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 124, 881, 160]]<|/det|>
+We appreciate the concerns of the Reviewer. We certainly think the MD simulations are a key part of our conclusions, as we outline below.  
+
+<|ref|>text<|/ref|><|det|>[[115, 159, 882, 282]]<|/det|>
+First, we acknowledge the significant challenges in realistically simulating the NMDAR open and closed states. As correctly pointed out by the Reviewer, we neglected many details such as cholesterol and PTMs, focusing instead on the features that we felt were the most important for connecting with the electrophysiological studies. Issues about cholesterol and PTM are very important and one we are actively investigating in terms of our model, but these studies go beyond the present study since they add many new experimental manipulations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 281, 882, 421]]<|/det|>
+Second, we used GluN1/GluN2B because we accumulated a significant number of simulations (some published in the 2018 NC paper but the vast number were unpublished). A large number of simulations is important for making sure the observations we report are robust. Note that our focus is the difference between the wild- type and the G- to- A mutants, and, as we now note (p. 8), the G- to- A mutation in GluN2B produces the same electrophysiological phenotype as in GluN2A (Amin et al., 2018, Nat. Comm.). In addition, we now state in the text (p. 8- 9) that the permeant ion is \(\mathrm{Na + }\) , and also note the recordings were done without \(\mathrm{Ca2 + }\) (p. 6, 9).  
+
+<|ref|>text<|/ref|><|det|>[[115, 420, 860, 543]]<|/det|>
+Third, the open model, based on homology with the AMPAR structure, is not symmetric. The diagonal distance of the GluN2 M3 helices are much longer than the counterpart of the GluN1 M3 helices, in line with electrophysiological data showing greater contribution of the GluN2 subunits to channel gating (e.g., Kazi et al., 2014, Nature Ns). Nevertheless, we did not include the extracellular domains as in Cerny et al., 2019 (new ref. 48). Again, we wanted to carry out as many simulations as possible to increase the number of replicates which was facilitated by using a reduced model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 882, 630]]<|/det|>
+Fourth, we initially refrained from carrying out simulations under a transmembrane potential because we wanted to isolate the differences between wild- type and the G- to- A mutants without the interference of external factors, especially considering that a large potential (e.g., \(>300 \mathrm{mV}\) ) might be required to achieve a meaningful number of ion permeation event.  
+
+<|ref|>text<|/ref|><|det|>[[115, 630, 882, 735]]<|/det|>
+Nevertheless, we agree with the Reviewer (as well as Reviewer 2) and have now made additional simulations with a transmembrane potential. In these simulations we also increased the nonbonded cutoff distance to 10 A. The outcome of these additional simulations further supports the idea that the M2 loop can form a gate, which is regulated by the GluN1 M4 segment. These new data are included in the Results section (p. 10 in the red- lined version).  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 752, 242, 769]]<|/det|>
+## minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 787, 745, 805]]<|/det|>
+Is there something missing in the "pathway that select ion(s) can cross"?  
+
+<|ref|>text<|/ref|><|det|>[[116, 822, 407, 839]]<|/det|>
+We have reworded this sentence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 857, 641, 875]]<|/det|>
+I don't really know what is meant by "membrane physiology".
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 712, 108]]<|/det|>
+We have changed to 'membrane excitability'. Sorry for the ambiguity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 124, 877, 213]]<|/det|>
+In the "'Gates' are barriers that occlude the flux of ions in the closed state.", if the above mentioned closed state model is correct, the gates you are talking about are actually relevant in the open state conformation. But this is probably just a missing definition of the closed state (I guess here you mean liganded (activated) receptor not permeable to ions).  
+
+<|ref|>text<|/ref|><|det|>[[115, 229, 881, 264]]<|/det|>
+We have added 'non- conducting' to help clarify 'closed state'. We hope this makes our meaning here clearer.  
+
+<|ref|>text<|/ref|><|det|>[[115, 281, 850, 298]]<|/det|>
+I am not sure the upstream/downstream of transmembrane helices is clearly defined.  
+
+<|ref|>text<|/ref|><|det|>[[115, 315, 880, 350]]<|/det|>
+We apologize for the lack of clarity. We now have replaced 'upstream' with 'N- terminal' and 'downstream' with 'C- terminal'.  
+
+<|ref|>text<|/ref|><|det|>[[115, 367, 864, 402]]<|/det|>
+In the "A critical residue in the M4 helices", maybe a better would be one of the critical residues.  
+
+<|ref|>text<|/ref|><|det|>[[115, 420, 470, 438]]<|/det|>
+As requested changed to the plural form.  
+
+<|ref|>text<|/ref|><|det|>[[115, 455, 857, 490]]<|/det|>
+In the "(eq. Popen) and to the same extent", "and" is not needed or something else is missing?  
+
+<|ref|>text<|/ref|><|det|>[[115, 507, 746, 525]]<|/det|>
+As requested, we have removed the extraneous word 'and' from the text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 513, 560]]<|/det|>
+In the "closed and opens states", just "open"?  
+
+<|ref|>text<|/ref|><|det|>[[115, 577, 543, 595]]<|/det|>
+We have corrected the typo. Thanks for catching.  
+
+<|ref|>text<|/ref|><|det|>[[115, 612, 864, 699]]<|/det|>
+I am not very familiar with the current data policy of Nature journals, but the ever more recognized FAIR principles would dictate that the data are deposited in a public repository and not just "available upon request". Although for example the methods section concerning the MD simulations is detailed enough to allow reproducibility, the actual trajectories could be useful to the community.  
+
+<|ref|>text<|/ref|><|det|>[[115, 716, 882, 855]]<|/det|>
+We appreciate the Reviewer for pointing out a more rigorous data transparency practice. In our original submission, we included considerable information in the Supplementary Information including extensive tables for many of our figures, figures with additional information as well as additional information on our MD simulations. In the resubmission we have added some more information here. We could post the Excel files associated with these figures/tables but am not sure they would provide anything beyond what is in the figure set and Supplementary Information. We have also followed as far as we know all of the guidelines prescribed by Nature Communications in terms of data transparency.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 141, 465, 160]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 875, 369]]<|/det|>
+The manuscript entitled "Two gates mediate NMDA receptor activity and are under subunit- specific regulation" is a solid study that provides some interesting data that is relevant to the function of the NMDA receptor. It is well written, and summarizes evidence that individual subunits control different aspects of gating, and also provides further support for functional relevance of the M4 in gating. The conclusions that the GluN1 M4 helix controls M2 helix of GluN2 and GluN2 M4 controls GluN1 M3 are important, and simulations suggest that mutations in the M4 region of different subunits can alter the pore diameter at distinct sites. I think it will be of interest to many working on ligand gated ion channels, glutamate receptors, and synaptic transmission. Below are several suggestions the authors should consider that could improve the clarity of the study and perhaps lead to more accurate and useful conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 385, 881, 438]]<|/det|>
+We appreciate the Reviewer's positive and insightful comments about the manuscript and have tried to modify the manuscript and its presentation to accommodate the Reviewer's concern.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 456, 168, 473]]<|/det|>
+## Major  
+
+<|ref|>text<|/ref|><|det|>[[115, 489, 868, 594]]<|/det|>
+1. While the data is convincing in showing a change in the long-lived closed dwell time separating clusters of activity, I am not sure the stated conclusion is helpful (from the discussion) "...a primary gate at the M3 bundle helical crossing controls entry and exit from both long lived and short-lived closed states". That means it controls everything—is that the intended conclusion? Also, where are data supporting entry and exit from short-lived states? There are data in supplemental but they are not really discussed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 880, 681]]<|/det|>
+We apologize for not making our conclusions and the limitation of our conclusions clearer. Part of the challenge is that there remain many unanswered questions so for certain things we can give 'conclusions' and in other instances we can only give 'best guesses'. Clearly, we did not do a good job of clarifying these issues.  
+
+<|ref|>text<|/ref|><|det|>[[115, 682, 882, 890]]<|/det|>
+Based on circumstantial evidence (and we continue to work to resolve this issue but it is challenging), we believe that the M2 gate is the major regulator of C1, C2, and C3. At present, our data supports the idea that, at least under our conditions, that the M3 gate mainly regulates C4 and C5 and the M2 gate partially regulates C1, C2, and C3, but that none of these regulatory events are absolute: the M2 gate also impacts C4 and C5 and the M3 gate impacts C1, C2, and C3. Part of the challenge is that the energetics of these gates are most likely linked – the open/closed status of the M3 gate impacts the energetics of the M2 loop and vice-versa. A second challenge is that the M2 gate is 'behind' the M3 gate and cannot be studied independently (except for our efforts trying to lock 'open' the M3 gate, which has certain limitations). Finally, we believe that the fast energetic events with the initial opening event following initial agonist binding (what determines the activation rate of whole-cell currents) may or may not be structurally
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 882, 143]]<|/det|>
+related to the fast closed states measured at steady- state. Again we are trying to test these ideas (using outside- out patches, strategic mutations and various agonists) but these are very extensive experiments and go beyond the present manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 143, 882, 317]]<|/det|>
+In any case, we have added the data for C1, C2, and C3 to the new Figure 2 (panels g- i) (as well as Supplemental Figure 1) to emphasize that the GluN1 G- A has significant effects on C1, while GluN1 and GluN2A G- A have comparable effects on C2 and C3. We have modified the text, mainly in the Discussion (pp 14- 15) to suggest that the M3 gate mainly regulates C4 and C5 while the M2 gate mainly regulates C1, C2, and C3 but that these events are not absolute. We also try to emphasize the complex energetics of these gates. The energetics keeping the M3 gate open (and hence the cluster/burst length) is not only dependent upon the M3- S2/pre- M1/S2- M4, but also the status of the M2 loop and perhaps vice- versa. Hence, the duration of cluster/bursts are also influenced by the energetics of the M2 loop.  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 880, 457]]<|/det|>
+2. If the 12 sec dwell time reflected an important step in the pathway from agonist binding to channel opening (i.e. gating), then the rise times would be far slower than \(\sim 10\) ms reported by dozens of authors. However most kinetic studies conclude that these prolonged closed periods in between clusters of activity are some form of agonist-dependent desensitization, and I think a likely conclusion is that the M4 regulates recovery from a desensitized state. Seems like this should be raised or evaluated in terms of double pulse experiments that assess the recovery from the desensitized state.  
+
+<|ref|>text<|/ref|><|det|>[[115, 472, 882, 526]]<|/det|>
+We agree completely with the Reviewer – that the S2- M4/M4 is regulating aspects of desensitization, and we apologize for not making this clear in the original submission. We have now expanded on this point in the Discussion (p 15).  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 882, 630]]<|/det|>
+Note we did record recovery for desensitization for GluN2A G- A as requested but did not add to the manuscript. We do see an increase in the time required for recovery to occur \((1.7 \pm 0.2\) sec, \(n = 8\) ) for GluN2A G- A versus wild- type \((0.97 \pm 0.09\) sec, \(n = 7\) ). Clearly, the S2- M4/M4 is involved in regulating desensitization but this will require much more detailed investigations to resolve and goes beyond the present study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 647, 872, 822]]<|/det|>
+3. Opening rates have been interpreted in early studies of frog neuromuscular nicotinic receptors from brief "nachschlag" closed states. The idea is the pore reverses a gating step during these brief gaps, and then proceeds forward to reopen. If the dwell time reflects a single visitation to one closed state, then the reciprocal of the duration is the rate associated with forward opening rate. Studies from Colquhoun, Auerbach, Traynelis, Gibb and others all showed three brief intra-cluster closed states that were interpreted as a very fast opening (perhaps pore dilation), and two other states with dwell times of \(\sim 1\) and \(\sim 5\) ms. If the author wants to conclude something about gating within a cluster, they should discuss some of the changes in various components of the histograms shown in the supplemental material) or provide some modelling results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 840, 881, 892]]<|/det|>
+This is an outstanding point and one to be honest we struggle perhaps the most with. As indicated in comments to point 1, we added mean closed duration ratios for C1, C2, and C3 to Figure 2 to better analyze those brief states. We now mention that GluN1 G- A
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 878, 177]]<|/det|>
+has a somewhat greater influence on these brief closed states (specially C1) than GluN2A G- A but also note that for C1, C2, and C3 this is not very absolute (in contrast to C4 and C5). Importantly, we wish to emphasize (and have done so in the Discussion) that our interpretation of this data is limited by the high likelihood of energetic coupling between the two gates.  
+
+<|ref|>text<|/ref|><|det|>[[115, 178, 868, 230]]<|/det|>
+In any case the Reviewer raises some critical considerations that are ultimately essential to resolve to better define the gating mechanism in NMDARs. At present our data set cannot resolve because of numerous complications (see response to Point 1).  
+
+<|ref|>text<|/ref|><|det|>[[115, 246, 878, 352]]<|/det|>
+4. I think it is important to explicitly state whether the opening of the two gates occurs in any order or in a dependent order (e.g. slow step first, then fast step), as multiple Popescu papers have suggested. Dependent order requires slow step to be first, followed by a fast step and then opening, otherwise there would never be brief sojourns to the closed state in the single channel record if the brief step occurred first and then the slow step immediately preceded opening.  
+
+<|ref|>text<|/ref|><|det|>[[115, 368, 881, 438]]<|/det|>
+This comment brings up an interesting insight and one we cannot with the present data set directly answer. Nevertheless, we assume that the M3 gate must open first, and we speculate on this in the Discussion, though we note that our data does not directly address, and thus a definitive answer is better left for future studies.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 456, 168, 472]]<|/det|>
+## Minor  
+
+<|ref|>text<|/ref|><|det|>[[115, 490, 880, 560]]<|/det|>
+5. I think the abstracts reference to "clusters" is vague, and bordering on jargon. How many readers interested in synaptic transmission will know what is meant by a cluster? I think the authors need to re-word to a conclusion that a wider audience can grasp and understand.  
+
+<|ref|>text<|/ref|><|det|>[[115, 577, 864, 630]]<|/det|>
+We thank the Reviewer for this suggestion and agree that many physiologists will not understand the significance of the word. We have added a sentence to the Abstract to better clarify this term. We hope this will make better sense to synaptic physiologists.  
+
+<|ref|>text<|/ref|><|det|>[[115, 647, 872, 700]]<|/det|>
+6. Minor suggested wording change: Page 4 "A variety of disease associated missense mutations have been identified at OR NEAR these conserved glycines...(could choose any of multiple refs to support)  
+
+<|ref|>text<|/ref|><|det|>[[115, 716, 882, 786]]<|/det|>
+We thank the Reviewer for suggesting this change, which better acknowledges that mutations near this site may also act via the mechanisms we outline in this work. The change has been made and several citations added (Chen et al., 2017, Mol. Pharm.; Perszyk et al., 2020, JPhysiol).  
+
+<|ref|>text<|/ref|><|det|>[[115, 803, 881, 891]]<|/det|>
+7. Page 4 sentence starting "These data suggest that the M3 gate mediates the longer..." How can you rule out that both gates are important, and because the long dwell time dominates the interval, you are missing changes to the interval with the faster gate changes its rate. That is, both may be required, but you can only see changes in the slow gate (contributing the long delay). A short 10 ms change will hardly alter a 12
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 91, 252, 107]]<|/det|>
+second interval.  
+
+<|ref|>text<|/ref|><|det|>[[115, 124, 882, 195]]<|/det|>
+As we discuss above and now in the Discussion, we think the gates are strongly energetically coupled. Indeed, C4 and C5 closed durations are significantly changed when we used GluN1 G- A to disrupt M2, just to a much less significant extend compared to GluN2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 211, 866, 247]]<|/det|>
+8. Typo in text on page 6. Duration of long lived state should be sec, not ms in the last sentence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 263, 803, 283]]<|/det|>
+We thank the Reviewer for catching this error! The text has now been changed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 298, 874, 387]]<|/det|>
+9. Page 8, first sentence of the second paragraph - Alasdair Gibb should get the credit for first showing unambiguously that there are five independent closed states in individual activations of a single native NMDA receptors in his clever low concentration experiment (Gibb et al., 1992). The references (40,41) confirmed his conclusions in heterologous systems—Schorge et al I think should be added.  
+
+<|ref|>text<|/ref|><|det|>[[115, 402, 881, 439]]<|/det|>
+We appreciate the Reviewer's insights. As requested, we have added the suggested citations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 454, 880, 561]]<|/det|>
+10. There are several papers entirely devoted to the idea that the GluN1 M4 functionally interacts with the GluN2 M3, and the GluN2 M4 interacts with the GluN1 M3 (Chen et al. 2017 exploring actions of a pre-M4 mutation, also Gibb 2018). Chen concludes the interaction of GluN2-M4 with GluN1 M3 is critical for gating, consistent with the conclusions here. Seems as though the conclusions of these papers should be mentioned, which are consistent to some extent with what is being showing here?  
+
+<|ref|>text<|/ref|><|det|>[[115, 576, 881, 630]]<|/det|>
+We thank the Reviewer for encouraging a broader discussion of the existing literature. These citations at multiple points have now been added to the Introduction and/or /Discussion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 647, 875, 682]]<|/det|>
+11. I would recommend mentioning in the results that saturating agonists are used so a reader doesn't have to turn to the methods or scrutinize the legends.  
+
+<|ref|>text<|/ref|><|det|>[[115, 699, 881, 734]]<|/det|>
+We appreciate the suggestion from the Reviewer and have now added this information to the Results section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 750, 875, 840]]<|/det|>
+12. I think it is important to comment that these results were obtained in the absence of divalents. Schorge et al recorded in divalent ions and was unable to reproduce some channel properties observed in the absence of divalents. Alternatively, one could perform a quick experiment in a patch with 0.5-1 mM Ca2+ to demonstrate that the conclusions hold in a physiological context.  
+
+<|ref|>text<|/ref|><|det|>[[115, 856, 882, 892]]<|/det|>
+Both Reviewers aptly point out that our original work could have been clearer about the absence of divalents in our experiments, which was done to give better resolution and to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 882, 161]]<|/det|>
+remove confounding variables from our experiments. Text has now been added (p. 6, 8- 9) to the results that directly clarifies this. We also agree that testing \(\mathrm{Ca^{2 + }}\) is a critical question, but feel this is a whole set of new experiments to rigorously address (again which we are trying to do).  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 879, 247]]<|/det|>
+13. I suspect a subset of readers will want to know what Tcrit was used for cluster analysis when they read about the analysis in the results without stopping their train of thought to turn back to the methods to figure out how you separated clusters, especially for mutations that shortened the long closed duration.  
+
+<|ref|>text<|/ref|><|det|>[[115, 264, 882, 317]]<|/det|>
+We thank the Reviewer for encouraging clarification of our results and to make them more accessible to a wider audience. We now include this information in the Results section (p. 7).  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 877, 403]]<|/det|>
+14. Figure 1D: It looks like there are sublevels for the GluN1-G815A mutation, and possibly even the GluN2A-G819A mutation. This seems relevant for a paper discussing pore diameter. Why not show openings at a higher resolution and analyze the sublevels?  
+
+<|ref|>text<|/ref|><|det|>[[115, 420, 882, 525]]<|/det|>
+The Reviewer points out a possibility we considered previously. There are several problems here. First, the openings for GluN1- G815A are so brief, we are not able to properly and consistently identify any subconductance states. Second, at present it is not clear the nature of subconductance levels in NMDARs and we feel this would require a whole new (and extensive) study to define (which we are trying to do but with mutations that induce more robust subconductance levels).  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 865, 717]]<|/det|>
+15. The K channel literature is fascinating because the authors in the structural era connect their results to work performed before structures were known. Seems like Banke's conclusion ten years ago (2003) before the structure of the NMDA receptor, "These data suggest that NR1 and NR2B subunits, respectively, undergo a fast and slow agonist-dependent conformational change that precedes opening of the pore", is relevant to the conclusions of the current manuscript. I believe that data in Jones et al 2002 with MTSEA also suggests different roles of GluN1 and GluN2 in gating. The current manuscript goes light years beyond these older studies, but it seems generally useful to recognize how early ideas about subunit dependent gating arose in the literature.  
+
+<|ref|>text<|/ref|><|det|>[[115, 734, 881, 768]]<|/det|>
+We apologize for not recognizing these earlier works in terms of the subunit specific effects on gating. We now include them as well as several others in the Discussion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 785, 876, 855]]<|/det|>
+16. Near the middle of page 9, could the authors specify that 15 replicate simulations of 500ns each were run per construct? They specify 15 in the Methods section, but also specifying it here would save time from having to go to the Methods and help improve clarity for the reader.  
+
+<|ref|>text<|/ref|><|det|>[[115, 873, 721, 891]]<|/det|>
+As requested, we now include the number of replicates in this section.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 106, 880, 213]]<|/det|>
+17. In Figure 2 and the related text in Results, positions \(+2\) , \(+6\) , and \(+10\) aren't defined as being indexed from the S in SYTANLAAF, but it is clarified in the Supplementary info. It would be useful to define it in the text, either by verbally explaining this in the legend for Figure 2 or by moving panel 'a' from Supp Fig 3 to Fig 2 in the main text. I don't see that the magenta sphere is defined in the caption for Fig. 2a. Caption for Fig. 2g refers to left and right panels when the panels are stacked vertically  
+
+<|ref|>text<|/ref|><|det|>[[115, 228, 881, 317]]<|/det|>
+These are very helpful suggestions and appreciate the Reviewer's efforts. We considered moving Supplementary Figure 3a to the main body but realized it would make this figure (Figure 2) even more complicated. We now explicitly state the referencing of \(+2\) , \(+6\) , and \(+10\) in the text (p. 9). We also now define the magenta sphere in the Figure legend. Finally, the figure legend has been altered to reflect the correct positions of the panels.  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 875, 421]]<|/det|>
+18. Has the model of the active state used in the MD simulations been used in previous publications? If so, a reference should be given as it would be helpful to see model quality metrics, as well as other information about the simulations such as RMSD plots, included in the supplemental information. If not, perhaps these quality metrics could be added as supplemental information.  
+
+<|ref|>text<|/ref|><|det|>[[115, 437, 881, 491]]<|/det|>
+We previously reported the model in Amin et al., 2018, Nature Communications, where many of these metrics, including RMSD plots, were included. This is now noted in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 506, 863, 647]]<|/det|>
+19. Given that ion permeation was analyzed in the MD simulations, can the authors comment on why a membrane potential wasn't modeled using a method such as applying a constant electric field or using a polarizable force field? For reference, see https://doi.org/10.1021/acs.jctc.5b01202. The simulations as they are model the behavior of ions in the absence of any driving force. If one of these methods for the simulations had been used, I feel that the data presented in Figure 2d and e would be more meaningful and informative. However, it is obviously too much to re-do. Perhaps the authors could add a comment about this in the manuscript as a caveat?  
+
+<|ref|>text<|/ref|><|det|>[[115, 662, 882, 821]]<|/det|>
+This is a very valid concern. We initially refrained from carrying out simulations under a transmembrane potential because we wanted to isolate the differences between wild- type and the G- to- A mutants without the interference of external factors, especially considering the fact that a large potential (e.g., \(>300 \text{mV}\) ) might be required to achieve a meaningful number of ion permeation event. Nevertheless, we agree with the Reviewer (as well as Reviewer 1) and have now made additional simulations with a transmembrane potential. The outcome of these additional simulations further supports the idea that the M2 loop can form a gate, which is regulated by the GluN1 M4 segment. These new data are included in the Results section (p. 10 in the red- lined version).  
+
+<|ref|>text<|/ref|><|det|>[[115, 837, 828, 891]]<|/det|>
+20. Are the values in Figure 2e from a single trajectory, or averaged across the 15 replicates? If they are from a single trajectory, can you explain how that specific trajectory was selected? Can the authors provide an explanation of how individual
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 693, 108]]<|/det|>
+trajectories were selected for the plots in Supplementary Figure 4?  
+
+<|ref|>text<|/ref|><|det|>[[115, 124, 881, 160]]<|/det|>
+The values in Figure 2e are the total crossing from 15 separate replicates, which now is explained better in the Figure 2 legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 177, 881, 247]]<|/det|>
+Trajectories in Figure 2d and Supplementary Figure 4 (they from the same simulation). We selected this example since they all came from the same simulation which showed the most transitions of all the replicates we made. We now explain this in the Supplementary Figure 4 legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 264, 880, 317]]<|/det|>
+21. Could the authors add to the Discussion their thoughts on why there are two long-lived closed states if the general claim is that the N2A M4s predominately regulate entry into and exit from long-lived closed states?  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 881, 421]]<|/det|>
+This is an outstanding question and one we have thought lots about. Our general idea, and we are certainly not the first to think of this, is that there are multiple structural elements regulating the opening of the M3 gate. The pre-M1 helix, the M3-S2 linkers, the S2-M4/M4 regions, among other structural elements. C4 and C5 presumably reflect to some extent all of these structural elements.  
+
+<|ref|>text<|/ref|><|det|>[[115, 438, 881, 525]]<|/det|>
+In short we speculate that a modifying process is occurring for our two long lived closed states, with the M4 movement being the primary driver, but with alternative structural confirmations bifurcating this process into two outcomes. However, we cannot delineate the culprit structural elements in this work without further experiments, the exploration of which is for another manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 542, 722, 560]]<|/det|>
+We have added some of this information to the Discussion (pp 14- 15).  
+
+<|ref|>text<|/ref|><|det|>[[115, 576, 881, 648]]<|/det|>
+22. Page 18: "although evidence from our MD simulations and single channel recordings suggest that reductions in the pore diameter of the M2 gate correlate with reduced ion permeation...", I don't really see evidence of this in the supplementary or main figures at least in terms of single channels for permeation. Maybe I missed something.  
+
+<|ref|>text<|/ref|><|det|>[[115, 664, 881, 700]]<|/det|>
+We have corrected this sentence [I assume what's meant by this sentence was correlation b/w pore radius and \(\mathrm{Ca2 + }\) permeability as shown in 2018NC].  
+
+<|ref|>text<|/ref|><|det|>[[115, 716, 863, 769]]<|/det|>
+23. Y-axes of Fig. 1h and 1i could be more clearly labeled (perhaps 'Closed duration ratio (G-A/wt)') Similarly, the word 'closed' should be added between 'mean' and 'durations' in the last sentence on pg. 8. Minor but might help keep the message clear.  
+
+<|ref|>text<|/ref|><|det|>[[115, 786, 881, 821]]<|/det|>
+We thank the Reviewer for their suggestion and have now altered Figure 1h and 1i to state "Closed duration ratio (G-A/wt)" and changed to 'mean closed durations'.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 90, 286, 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, 864, 208]]<|/det|>
+Dear Authors, thank you for the revised version of the manuscript. I believe that the introduced changes sufficiently addressed my comments. As already said, the experimental evidence of G- to- A effects is solid and I understand that performing complicated new experiments or modified MD simulations at time scales comparable to the GluN2B ones would be impractical now.  
+
+<|ref|>text<|/ref|><|det|>[[115, 208, 875, 327]]<|/det|>
+You are right that the Supplementary material is rich and the data is treated according to the rules. My appeal was more like "it would be nice to have the raw trajectories as well". It would, like for example in structural biology field, allow reanalysis or even more importantly development or benchmarking of new data processing methods. Obviously the infrastructure allowing efficient storage of large datasets in an interoperable way is not very mature (neither zenodo nor say model archive are perfect for this purpose yet) but we are in a hen and egg situation here, without researchers offering their raw data, the repositories will evolve only slowly (if ever) into a useful and user- friendly data resource.  
+
+<|ref|>text<|/ref|><|det|>[[115, 373, 216, 386]]<|/det|>
+Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[116, 388, 291, 401]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 402, 860, 432]]<|/det|>
+The authors revised manuscript has adequately addressed the points raised. Congratulations on the completion of a very nice study.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42151a82c0281305539beb0a5a086a26c865e127f2c665b3801e9dc9d28b53f2/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__42151a82c0281305539beb0a5a086a26c865e127f2c665b3801e9dc9d28b53f2/supplementary_0_Peer Review File__42151a82c0281305539beb0a5a086a26c865e127f2c665b3801e9dc9d28b53f2.mmd

@@ -0,0 +1,487 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Shaping Lightwaves in Time and Frequency for Optical Fiber Communication  
+
+![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 --->
+
+# Review of NCOMMS-21-15776: "Shaping Lightwaves in Time and Frequency for Optical Fiber Communication"  
+
+The authors of the paper investigate constellation shaping (sphere shaping in particular) for optical communication systems. They focus on the effect of (1) the block length of sphere shaping and (2) the symbol rate of the signaling on the amount of nonlinear interference (NLI) generated during propagation. The dependency of NLI on the shaping block length has been investigate in the literature. The authors propose a new metric (windowed central moment) to study this dependency. This metric measures the high- order deviations of the energy of the optical field from its average over a finite time interval. They claim that properly selecting the width of this window is important to properly compare different shaping schemes from NLI generation perspective and to predict their performance. Furthermore, they claim that the nonlinearities depend on the energy structure of the optical field in absolute time rather than in number of symbols. This leads to a complex relation between the shaping block length, symbol rate and NLI. Using the values of the window length at which the correlation between the windowed moment and the effective SNR is maximized, authors claim that the EGN model predicts the effective SNR (using windowed moments) more accurately than its traditional form (using regular moments) (Fig. 3(b)).  
+
+The paper is well- written, findings are interesting, the topic is timely/relevant, and the explanation of the experimental setup is very clear and detailed. I believe the paper can be accepted after a round of revision/polishing. The structure of the paper is different than I am used to from other journals in the literature and this structure may be leading to a different reading experience than I am used to. However I must say that the introduction can be improved by stating (1) the problem, (2) what is done, and (3) what is found more clearly with short and direct sentences. At some points, I had trouble following the train of reasoning. I believe the findings of the paper are quite interesting, and they should be stated more clearly to increase their impact. In the following, I will list my comments which I hope would help the authors to improve the presentation in the paper.  
+
+- Firstly, I find it a bit complicated to understand the fundamental motivation and findings of the paper from the abstract and the introduction. I suggest a clearer statement of the fact that (1) the paper investigates the optimal combination of the shaping block length and the symbol rate that maximizes the performance, (2) this is achieved using a new metric—similar to a metric proposed in [1]—that has high correlation with the effective SNR. At some parts of the paper, it is implied that a NLI-optimized sphere shaping approach is proposed. However, as far as I understand, only the optimum system parameters are investigated (which is also a valuable contribution) for regular sphere shaping.  
+
+- In the traditional EGN model, higher order moments of the channel input distribution are used in the expressions that compute the NLI variance. Here, they are replaced with their windowed versions and the reported results match the actual performance more accurately than the original EGN model. Can the authors comment on the use of the windowed moments in the EGN model? Is there a theoretical background for this, or is it just heuristic? Can this approach be used to propose an enhanced EGN model that works for finite-length shaping schemes and captures the temporal structure of the shaped waveforms in the analytical model?  
+
+- The improvement in rate (or in reach) due to shaping (and sphere shaping in particular) has been demonstrated in the literature repeatedly. The paper provides an optimization of shaping over (i) block length and (ii) symbol rate. There are also other works in the literature which discusses the effect of block length and symbol rate on performance. They should at least be cited, e.g., [2], [3], [4], [5], etc.  
+
+- In p.5, it is stated that refs. 7 and 15 show that if \(n\) is small as in Fig. 1(c), sphere shaping reduces NLI. Firstly, ref. 15 is not about sphere shaping, but about constant composition sequences. Secondly, as previously observed in [6] and re-stated in ref. 7, (AWGN-optimal) shaping leads to increased NLI (decreased effective SNR). This is attributed to the increased kurtosis in [6], [7]. Therefore, I am not sure I see the contradiction that you mention here. Furthermore, the explanation concerning Fig. 1(d) is not clear. Are we focusing on the effect of \(n\) on NLI here or the importance of selecting \(w\) to predict the performance? Because regardless of \(w\) , smaller \(n\) implies smaller \(\bar{\mu}_{2}\) in Fig. 1(d) (for most of the region). I would suggest a careful revision of the paragraph in p.5 that starts with "Fig. 1(c)" (which I believe should be "Figure 1(c)") with clear explanations (of the claims and figures) and as much connections to the existing literature as possible. I believe this paragraph is extremely important to motivate your work.  
+
+- As far as I know, the effective SNR in EGN model depends on the average power of the input as well as its regular fourth and sixth order moments. In p.8, how do you use the EGN model with \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) to compute the effective SNR? Figure 3(b) reports that this computation (EGN with windowed moments) is far more accurate than the regular EGN model in predicting the effective SNR. If this is the case, would you claim that you're proposing an enhanced EGN model?
+
+<--- Page Split --->
+
+Furthermore, is it possible to obtain a further more accurate relation if (somehow) some other higher- order windowed moments are included in the model? Also, when you say "optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) " here, do you mean their values computed with the optimum window length?  
+
+In the following, I list some minor comments.  
+
+1) p.2 paragraph 1: "ever-increasing demand for communication capacity" \(\rightarrow\) "ever-increasing demand for higher data rates" 
+2) p.2 paragraph 2: Refs. 8 and 9 are given as references for sphere shaping, but they are not. I would suggest instead [8], [7].  
+
+3) p.2 paragraph 2: "block of amplitudes" is rather vague, it should be related to "communication (QAM) symbols" (mentioned in the previous paragraph) better.  
+
+4) p.3 paragraph 2: 4 consecutive amplitudes \(\rightarrow\) one DP symbol: This is discussed extensively in [4], I would suggest referencing it.  
+
+5) p.3: \(||\cdot ||^{2}\) is not defined  
+
+6) p.3: Explanation of sphere shaping can be connected to the related literature more strongly  
+
+7) p.3 paragraph 1: "(...) if every shaped block is plotted as a point in \(4n\) -dimensional signal space, with the \(i\) -th amplitude being the position of the point on the \(i\) -th coordinate axis, the points uniformly fill a \(4n\) -dimensional (hyper-)sphere (...) \(\rightarrow\) Since the amplitudes are drawn from a discrete set, this is not correct. The points are located in a \(4n\) -dimensional spherical region of a \(4n\) -dimensional rectangular lattice. I would avoid the use of the phrase "uniformly fill".  
+
+8) p.4: I would recommend the use of \(R\) instead of \(H\) for the information rate, since \(H\) is almost always used to denote the entropy  
+
+9) p.4 Fig. 1(a): Sphere shaping indeed decreases the maximum energy substantially. However, what is more important is the decrease in average energy. I would recommend indicating the decrease in average energy, and even relating it to the "linear" shaping gain discussed, e.g., in [9].  
+
+10) p.4 Fig. 1(b): I do not understand what this figure tries to explain, or how Fig. 1(a) relates to this.  
+
+11) p.4: sentence before (1) and (2): "(...) \(\eta\) increases as the central moment (...) \(\rightarrow \eta\) or \(P_{NLI}\) ?  
+
+12) p.5: "(...) deviation by the instantaneous power (...) \(\rightarrow\) "(...) deviation of the instantaneous power (...) from average \(< p > = 1\) (...)"?  
+
+13) p.5: "(...) appears to be more spread around the average (...) \(\rightarrow\) I think this can be related to the difference in their \(\bar{\mu}_{2}\) , right? If so, please refer to these values.  
+
+14) Figures 1(c-d): What is the shaping rate and the resulting distribution?  
+
+15) p.6: I do not understand what "With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods in all conditions except for \(R_{Sym} = 88\) GBd in Link D." exactly means.  
+
+16) p.6: I am not sure how the observation (iii) is explained with "(...) since increasing \(R_{Sym}\) reduces the shaping block duration in absolute time for the same \(n\) ".  
+
+17) p.7: I understand the observation "Areas far from the circled areas have a negligible impact on \(SNR_{Eff}\) . But I believe this can be related to Figs. 2(a,c) more strongly and quantitatively.  
+
+18) Fig. 3(a): How many spans?  
+
+19) p.8: You say "optimal sphere shaping" more than once in the paper. What exactly do you mean by "optimal"? The combination of \(n\) and \(R_{Sym}\) that maximizes the performance? There is an optimal window length \(w\) that maximizes the correlation \(b/w\) the windowed moment and the effective SNR. But I am not sure what you mean by "optimal" sphere shaping. Clarification is needed.  
+
+20) p.9: I believe the conclusion "In general, \(n^{*}\) increases as \(|D_{Total}|\) (green solid line) increases, (...) is a bit strong to be made from Fig. 5(a). The optimal block length here is mostly constant around 10-20, increasing around 195 THz, but then decreasing again. If the authors can provide some additional explanation for this behaviour, it would make the paper stronger.  
+
+21) p.9: At which block length Fig. 5(b) is plotted? Also, similar to my previous comment, the conclusion "(...) \(R_{Sym}^{*}\) decreases (...) when \(|D_{Total}|\) increases (...) is a bit strong to be made from Fig. 5(b). Instead of simply saying Figs. 5(a,b) agree with Figs. 2(b,a), possible explanations for the peculiar behaviours in Figs. 5(a,b) should be provided.  
+
+22) p.9, third line: \(SNR_{Eff}\) of Fig. 4(c), or of Fig. 5(c)?  
+
+23) p.11: CCDM is not a sphere shaping algorithm.  
+
+24) Similar gains in data rate/SNR/AIR to the ones that are reported at the end manuscript exist in the literature. A short review of such works would make the paper stronger.  
+
+## REFERENCES  
+
+[1] K. Wu, G. Liga, A. Sheikh, F. M. J. Willems, and A. Alvarado, "Temporal energy analysis of symbol sequences for fiber nonlinear interference modelling via energy dispersion index," Feb. 2021. [Online]. Available: https://arxiv.org/abs/2102.124114 [2] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike-Akino, K. Kojima, K. Parsons, and D. S. Millar, "Nonlinearity tolerant LUT-based probabilistic shaping for extended-reach single-span links," IEEE Photon. Technol. Lett., vol. 32, no. 16, pp. 967- 970, 2020.
+
+<--- Page Split --->
+
+[3] T. Fehenberger, "On the impact of finite-length probabilistic shaping on fiber nonlinear interference," in Proc. Signal Process. in Photon. Commun. (SPPCom), Washington, D.C., United States, July 2020.  
+
+[4] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike-Akino, K. Kojima, K. Parsons, and D. S. Millar, "Huffman-coded sphere shaping for extended-reach single-span links," IEEE J. Sel. Topics Quantum Electron., vol. 27, no. 3: 3500215, May-June 2021.  
+
+[5] S. Civelli, E. Forestieri, and M. Secondini, "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation," in Proc. Eur. Conf. Opt. Commun. (ECOC), Brussels, Belgium, Dec. 2020.  
+
+[6] T. Fehenberger, A. Alvarado, G. Böcherer, and N. Hanik, "On probabilistic shaping of quadrature amplitude modulation for the nonlinear fiber channel," J. Lightw. Technol., vol. 34, no. 21, pp. 5063- 5073, Nov. 2016.  
+
+[7] Y. C. Gültekin, A. Alvarado, O. Vassilieva, I. Kim, P. Palacharla, C. M. Okonkwo, and F. M. Willems, "Kurtosis-limited sphere shaping for nonlinear interference noise reduction in optical channels," June 2021. [Online]. Available: https://arxiv.org/abs/2105.14794  
+
+[8] P. Schulte and F. Steiner, "Divergence-optimal fixed- to- fixed length distribution matching with shell mapping," IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 620- 623, Apr. 2019.  
+
+[9] Y. C. Gültekin, W. J. van Houtum, A. Koppelaar, and F. M. J. Willems, "Enumerative sphere shaping for wireless communications with short packets," IEEE Trans. Wireless Commun., vol. 19, no. 2, pp. 1098- 1112, Feb. 2020.
+
+<--- Page Split --->
+
+This work studies the optimization of the time and bandwidth over which sphere shaping should be performed to minimize the impact of nonlinear effects and maximize performance in optical fiber communication.  
+
+The work is timely, well written, and reports some novel interesting results.  
+
+The optimization of the shaping block length in the nonlinear regime has been already investigated in some other papers but, to my knowledge, the simultaneous optimization of block length and symbol rate (i.e., time and bandwidth) is studied here for the first time. Moreover, this study encompasses several scenarios and a wide range of parameters, both numerically and experimentally.  
+
+The methodology is sound and the methods are described with enough detail. Moreover, the conclusions are supported both by numerical and experimental results.  
+
+In general, I believe that the paper deserves publication. However, there are some weaknesses and some issues that should be considered and possibly addressed, as detailed below.  
+
+1) As a general note, while reading the paper, I was expecting to find a theoretical analysis which explains and predicts the results observed numerically and experimentally. This expectation is not fully meet by the paper, which provides only some insight about the observed phenomena and some empirical laws to model them - e.g., (7), (8) and the use of the windowed moments in the EGN model. In fact, there are a number of perturbation models in the literature which could be used to attempt such an analysis, for instance by removing the i.i.d. assumption in the EGN model. An example can be found in Liga et al "Extending Fibre Nonlinear Interference Power Modelling to Account for General Dual-Polarisation 4D Modulation Formats", Entropy, 2020. I recommend discussing this issue and highlighting the empirical nature of the proposed model.  
+
+2) As a second general comment, the analysis practically neglects the impact of carrier recovery. Indeed, this might be substantial, as a sufficiently fast carrier recovery algorithm can mitigate the impact of nonlinear phase noise and change the dependence of performance on block length (see for instance Civelli et al. "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation", ECOC 2020). With carrier recovery, there might be no advantage in using short (optimized) block lengths. It seems that carrier recovery is not present in the simulations and is quite slow in the experiments (using only 1 pilot every 48 symbols), possibly overestimating the advantage of using a short optimized block length. Please comment about that. As a side note, for the sake of transparency, I am a coauthor of the above mentioned paper. I do not want to push my own work, therefore feel free to decide if the paper is relevant or not to your work and if it deserves to be cited or not. This decision will not affect my final recommendation.  
+
+3) Abstract: The sentence "In optical fiber, however, the sphere shaping induces Kerr nonlinearity in a peculiar way that makes analysis of transmission performance difficult, potentially lowering the communications capacity" sounds a bit odd. It suggests that it is sphere shaping that makes the
+
+<--- Page Split --->
+
+analysis difficult, with Kerr nonlinearity just playing an indirect role. In fact, I think the opposite is true. Moreover, it suggests that the difficulty of the analysis may play a role in the capacity reduction.  
+
+4) Introduction: The idea of using sphere shaping on an optimized block length to minimize the impact of nonlinear interference was first proposed in [10] and then further investigated by some of the same authors in Geller et al. "A Shaping Algorithm for Mitigating Inter-Channel Nonlinear Phase-Noise in Nonlinear Fiber Systems", JLT 2016. Your work (as well as some other recent works) heavily relies on this idea, extending the optimization to the frequency domain. I think that it would be fair to better acknowledge this fact and the role of these two papers.  
+
+5) Page 3, Results: you state that "The symmetry of the probability allows for legitimate analysis with only positive amplitudes, and hence we omit the sign throughout this article". I guess that the presence of a random sign with uniform distribution is implicitly assumed in the paper and neglected only in the description, whereas it is included both in the simulations and in the experiments. Is that correct? I am not sure that transmitting only positive amplitudes would indeed give the same results. I think that two consecutive symbols with the same sign or with opposite signs induce a fundamentally different response in the nonlinear channel. Please be more explicit.  
+
+6) Page 3, Results: the sentence "one dual-polarization symbol that maintains the value over a period of T_sym (s) and can change the value at a rate of R_sym=1/T_sym (Bd)", suggests a rectangular pulse shape. I suggest revising the sentence. Moreover, the indication of the units of measure in this way is neither standard nor required.  
+
+7) Page 5, 4 lines before the end of the subsection: The sentence "As w increases, \mu - 2 remains the same for all types of i.i.d. symbols" is misleading. It is not "the same" for all types of symbols but, rather, it remains "constant" as w increases for a given modulation (type of symbols).  
+
+8) Equation (3): for sphere shaping, the transmitted signal is a ciclostationary process with period equal to the length of the shaped blocks (unless a uniform random delay is assumed). Therefore, I guess that the windowed moments depend on the time at which the window is centered, unless they are averaged over it. Please comment about that and explain if and how you have considered this time dependence in your analysis.  
+
+9) Page 5, Section "Optimization of Sphere Shaping in the Time-Frequency Plane": Please specify what QAM format is considered in this section. I guess from the given data that it is 16QAM (as in the previous section). Yet, an explicit mention here would be useful.  
+
+10) Page 6: You state that "With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods
+
+<--- Page Split --->
+
+in all conditions except for R_sym=88 GBd in Link D" and that "like the i.i.d. uniform QAM, the optimal R_sym for i.i.d. shaping to maximize SNR_eff (yellow stars) decreases as the total net dispersion D_Total increases". I think that this discussion is a bit weak and that the analogy with the symbol rate optimization for uniform i.i.d. QAM is misleading for two reasons: 1) The dependence of SNR_eff on symbol rate for the uniform i.i.d. QAM case vanishes for Gaussian symbols (the EGN and GN model converge to the same equations, see for instance [23]). Therefore, I expect a very little dependence on the symbol rate for shaped i.i.d. symbols, as they approximate i.i.d. Gaussian symbols. In fact, the dependence on symbol rate in Fig. 2a for the case \(n = 5120\) is very weak (and would be probably even weaker for shaped 64- QAM or 256- QAM) and the peak SNR is not so different from the surrounding values in the simulation range; 2) assuming that \(n = 5120\) represents the i.i.d. case from a dispersion perspective is an approximation whose accuracy decreases as the symbol rate increases; therefore, it cannot be used when explaining the (weak) dependence of SNR_eff on symbol rate. In fact, for \(n = 5120\) , this dependence becomes a little more relevant exactly where the approximation becomes looser. In my opinion, it seems more likely that the dependence of SNR_eff on symbol rate is mainly explained, even for \(n = 5120\) , by the same reason that explains it for lower n.  
+
+11) Page 6, you state that "the optimal R_sym is higher for finite-length shaping than for i.i.d. shaping (compare the yellow and red stars), since increasing R_sym reduces the shaping block duration in absolute time for the same n". Following up on the previous comment, can we better (and more simply) say that the optimal symbol rate increases when n decreases?  
+
+12) Still on the same matter: you refer many times in the results to the i.i.d. case, considering \(n = 5120\) as an approximation to it. Why not directly simulating the true i.i.d. case? It should be sufficient to draw i.i.d. QAM symbols from a Maxwell-Boltzmann distribution. That would be a more meaningful benchmark against which to compare the results for finite block lengths and, for instance, to compute \(\backslash \mathrm{eta}^{\wedge}\backslash \mathrm{infly}\) in (6).  
+
+13) Page 8, Demonstration of Optimal Sphere Shaping through Full C-Band Transmission Experiment: The experiment is performed on a link that is substantially different from the 4 links considered in the simulation. Why? In particular, each channel sees a different dispersion profile, making the interpretation of the results harder. Moreover, the important dispersion-unmanaged case is not considered in the experiments. Is there a particular reason for this choice? Is it instrumental to highlight some effect or behavior? Please explain and motivate this choice.  
+
+14) Page 11, Methods: you state that "The digital sphere shaping encoder is implemented by enumerative sphere shaping (ESS) for \(n = 5, 10, 20, 40, 80\) , and by constant composition distribution matching (CCDM) for \(N = 320, 1280, 5120\) ." and "...consume 0.583, 0.346, 0.209, 0.137, 0.099, 0.025, 0.009, 0.002 dB more average symbol energy than ideal shaping, respectively". Strictly speaking, CCDM does not implement sphere shaping (as all constant composition sequences lie on the surface of the sphere, partially covering it) and, in fact, its energy loss is higher than that of ESS for the same blocklength. Of course, they both asymptotically converge to i.i.d. Maxwell-Boltzmann symbols, but as you are studying how performance changes with block length, you cannot assume that such an
+
+<--- Page Split --->
+
+asymptotic regime has been achieved. In fact, I am also a bit surprised that CCDM with \(n = 320\) performs significantly better than ESS with \(n = 80\) (see, e.g., the comparison in [7]). Please comment about that and double check the numbers (I have not verified them, it is just a feeling and I might be wrong).  
+
+Marco Secondini
+
+<--- Page Split --->
+
+# Response to Reviewers' Comments on NCOMM-21-15776 "Shaping Lightwaves in Time and Frequency for Optical Fiber Communication"  
+
+Junho Cho, Xi Chen, Greg Raybon, Di Che, Ellsworth Burrows, Samuel Olsson, and Robert Tkach  
+
+Dear Reviewers,  
+
+We sincerely thank all reviewers for carefully reading and commenting on the paper. We addressed in the revised manuscript all the suggestions and comments, which we believe has significantly improved the paper. The changes made in the revised manuscript are highlighted in yellow. Please find our point- by- point response to the comments below, where the reviewers' comments are reproduced verbatim in blue italic type, and our response is written in black roman type. Note that the figure numbers, table numbers and reference numbers below indicate those of the revised manuscript, unless specified otherwise.  
+
+## Response to Reviewer 1:  
+
+The authors of the paper investigate constellation shaping (sphere shaping in particular) for optical communication systems. They focus on the effect of (1) the block length of sphere shaping and (2) the symbol rate of the signaling on the amount of nonlinear interference (NLI) generated during propagation. The dependency of NLI on the shaping block length has been investigate in the literature. The authors propose a new metric (windowed central moment) to study this dependency. This metric measures the high- order deviations of the energy of the optical field from its average over a finite time interval. They claim that properly selecting the width of this window is important to properly compare different shaping schemes from NLI generation perspective and to predict their performance. Furthermore, they claim that the nonlinearities depend on the energy structure of the optical field in absolute time rather than in number of symbols. This leads to a complex relation between the shaping block length, symbol rate and NLI. Using the values of the window length at which the correlation between the windowed moment and the effective SNR is maximized, authors claim that the EGN model predicts the effective SNR (using windowed moments) more accurately than its traditional form (using regular moments) (Fig. 3(b)).  
+
+The paper is well- written, findings are interesting, the topic is timely/relevant, and the explanation of the experimental setup is very clear and detailed. I believe the paper can be accepted after a round of revision/polishing. The structure of the paper is different than I am used to from other journals in the literature and this structure may be leading to a different reading experience than I am used to. However I must say that the introduction can be improved by stating (1) the problem, (2) what is done, and (3) what is found more clearly with short and direct sentences. At some points, I had trouble following the train of reasoning. I believe the findings of the paper are quite interesting, and they should be stated more clearly to increase their impact. In the following, I will list my comments which I hope would help the authors to improve the presentation in the paper.
+
+<--- Page Split --->
+
+Response: Thank you for all the constructive comments here and in what follows. We have tried to address your concerns by improving the presentation as much as we can. We hope that the ambiguities found in the previous manuscript have completely been removed in the revised manuscript and the content of the paper is now much clearer.  
+
+Firstly, I find it a bit complicated to understand the fundamental motivation and findings of the paper from the abstract and the introduction. I suggest a clearer statement of the fact that (1) the paper investigates the optimal combination of the shaping block length and the symbol rate that maximizes the performance, (2) this is achieved using a new metric—similar to a metric proposed in [1]—that has high correlation with the effective SNR. At some parts of the paper, it is implied that a NLI- optimized sphere shaping approach is proposed. However, as far as I understand, only the optimum system parameters are investigated (which is also a valuable contribution) for regular sphere shaping.  
+
+Response: As the reviewer suggested, we made the motivation and contribution of this paper clearer in the abstract and the introduction. The changed sentence in the abstract is "In this article, we show that the impact of sphere shaping on Kerr nonlinearity varies with chromatic dispersion, shaping block length and symbol rate, and that this impact can be predicted using a novel statistical measure of light energy." In the abstract, a longer explanation on the motivation and contribution of this work was avoided to meet the suggested length of 150 words, but we tried to make it as clear as possible within the given length. In the introduction, several sentences have been changed to clarify the motivation and contribution of this paper, as highlighted in yellow in the upper half of Page 3. In particular, we emphasized that the previous studies have focused only on the block length optimization, neglecting the influence of the symbol rate and chromatic dispersion. It is only by knowing the findings of this work that the results can be consistently explained between several independently performed experiments that used different settings. To clarify the contribution of this work, we added "While the previous works<sup>8,11,12,16,23</sup> optimized only \(n\) to observe some gains in \(SNR_{Eff}\) and NDR over specific links (e.g., for single- span links), joint optimization of \(n\) and \(R_{Sym}\) in this work produces significantly larger gains and allows these gains to be achieved over a much wider variety of links." at the end of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section. We also made it clear throughout the abstract, introduction, and main text that it is not the shaping algorithm that we optimize to minimize the NLI, but the parameters of the regular sphere shaping.  
+
+In the traditional EGN model, higher order moments of the channel input distribution are used in the expressions that compute the NLI variance. Here, they are replaced with their windowed versions and the reported results match the actual performance more accurately than the original EGN model. Can the authors comment on the use of the windowed moments in the EGN model? Is there a theoretical background for this, or is it just heuristic? Can this approach be used to propose an enhanced EGN model that works for finite- length shaping schemes and captures the temporal structure of the shaped waveforms in the analytical model?  
+
+Response: Thanks for this valuable comment. At a recent conference, we learned from an author of Ref. 26 that an extended EGN model was published last year to deal with structures (or correlations) present in four amplitudes (i.e., in one symbol period). A mathematically unequivocal approach to modeling the propagation of structured lightwaves would be to further extend Ref. 26 (see also Ref. 27)
+
+<--- Page Split --->
+
+for structures spanning more than one symbol period. However, given the enormous complexity of the mathematical expansion to account for only four correlated amplitudes, it can easily be inferred that it is mathematically daunting to find an accurate analytical model to handle tens to thousands of correlated amplitudes that we cover in this paper. Also, as seen from Refs. 26- 27, the resulting equations of the extended EGN model for many correlated amplitudes may have too many complex terms to be practically useful.  
+
+In this work, we allowed for model inaccuracies caused by the assumptions on i.i.d. amplitudes, and attempted to solve the problem with manageable complexity by substituting the windowed moments that can characterize structured lightwaves statistically and quantitatively into the traditional EGN model, whether the structure is short or long. We have clarified this in the revised manuscript with properly added citations, as found in the first paragraph of Page 9. It may be possible to apply our approach to the extended EGN model of Ref. 26 to see if the prediction accuracy is even more improved, which we would like to leave for future work.  
+
+The improvement in rate (or in reach) due to shaping (and sphere shaping in particular) has been demonstrated in the literature repeatedly. The paper provides an optimization of shaping over (i) block length and (ii) symbol rate. There are also other works in the literature which discusses the effect of block length and symbol rate on performance. They should at least be cited, e.g., [2], [3], [4], [5], etc.  
+
+Response: We thank you for letting us reference the missing important prior studies. By citing Refs. 15- 17, 23 (which are [2]- [5] in your reference numbers) at the end of the second and third paragraphs of the revised introduction, we noted that there are several recent works addressing the effect of the shaping block length. However, those papers are cited only in the context of block length, since there are no existing studies on the effect of the symbol rate on Kerr nonlinearity. For example, in Refs. 8, 11, 15- 17, only single symbol rates of 45 GBd, 100 GBd, 56 GBd, 56 GBd, and 50 GBd are used, respectively. In Ref. 23, signal is modulated with two different symbol rates of 42 GBd and 64 GBd, but from this no systematic analysis of the effect of symbol rate can be made. We therefore cited the prior studies in the context of the shaping block length only.  
+
+In p.5, it is stated that refs. 7 and 15 show that if n is small as in Fig. 1(c), sphere shaping reduces NLI. Firstly, ref. 15 is not about sphere shaping, but about constant composition sequences. Secondly, as previously observed in [6] and re- stated in ref. 7, (AWGN- optimal) shaping leads to increased NLI (decreased effective SNR). This is attributed to the increased kurtosis in [6], [7]. Therefore, I am not sure I see the contradiction that you mention here. Furthermore, the explanation concerning Fig. 1(d) is not clear. Are we focusing on the effect of n on NLI here or the importance of selecting w to predict the performance? Because regardless of w, smaller n implies smaller \(\bar{\mu}_{2}\) in Fig. 1(d) (for most of the region). I would suggest a careful revision of the paragraph in p.5 that starts with “Fig. 1(c)” (which I believe should be “Figure 1(c)”) with clear explanations (of the claims and figures) and as much connections to the existing literature as possible. I believe this paragraph is extremely important to motivate your work.  
+
+Response: As the reviewer pointed out, Ref. 22 (Ref. 15 in our previous manuscript) deals with constant composition sequences. As the block length increases, the constant composition sequences can approximately realize sphere shaping with gradually decreasing rate loss, as a consequence of the sphere
+
+<--- Page Split --->
+
+hardening phenomenon. However, we admit that for short block lengths covered in Pages 5- 6, constant composition sequences produce a large difference from sphere shaping. Therefore, we replaced Ref. 22 with proper citations in the sentence you referred above, and added citations to Refs. 17 and 34 (which are [6] and [7] in your reference numbers). We also added "Note that CCDM can only approximately realize sphere shaping for finite block lengths, but it converges to ideal sphere shaping with a decreasing approximation error as the block length increases" to the first paragraph in the Methods section.  
+
+As far as the "contradiction" is concerned, it is observed between the analytical model for i.i.d. symbols and the empirical results obtained using sphere shaping with short block lengths; namely, if we calculate the statistical moments of short sphere- shaped symbols with \(w = 1\) as conventionally done, it is much larger than that of the unshaped symbols (see, e.g., Fig. 1(d)), and if we use the EGN model neglecting the required i.i.d. properties of symbols, these large statistical moments imply that NLI of sphere- shaped symbols is much greater than that of unshaped symbols. On the other hand, the NLI that is empirically obtained with short sphere shaping can be much smaller than that of the unshaped symbols (see, e.g., Fig. 8 of Ref. 8). In the paragraph that you mentioned, we have tried to convey the meaning of "contradiction" more clearly by rewriting several sentences, with citations to the associated prior works.  
+
+We have also paraphrased several sentences in this paragraph to avoid duplication of content or confusion between what is presented here and what is presented later sections.  
+
+Regarding the text citation of figures, some journals such as IEEE journals use the abbreviation "Fig." even when it begins a sentence. However, for Nature Communications, we were not able to find whether "Fig. 1(c)" or "Figure 1(c)" conforms to the editorial style, so will check with the journal's editorial team during the proofreading stage (if the paper is accepted).  
+
+As far as I know, the effective SNR in EGN model depends on the average power of the input as well as its regular fourth and sixth order moments. In p.8, how do you use the EGN model with \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) to compute the effective SNR? Figure 3(b) reports that this computation (EGN with windowed moments) is far more accurate than the regular EGN model in predicting the effective SNR. If this is the case, would you claim that you're proposing an enhanced EGN model? Furthermore, is it possible to obtain a further more accurate relation if (somehow) some other higher- order windowed moments are included in the model? Also, when you say "optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) " here, do you mean their values computed with the optimum window length?  
+
+Response: You are correct that the EGN model requires the average power as well as the fourth and sixth standardized moments. To clarify this, we corrected a phrase in Page 9 as "Plugging the average symbol power \((||x||^2)\) and the windowed central moments \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) obtained with \(w_{SPM}^*\) and \(w_{XPM}^*\) into a state- of- the- art analytic model known as the enhanced Gaussian noise (EGN) model". We also added a small subsection "EGN simulation" in the Method section to explain how to convert the windowed central moments into the fourth and sixth standardized moments. We could approach mathematically as to how this conversion can be derived and why this conversion is necessary; however, we believe that this mathematical detail is beyond the scope of Nature Communications articles and that the physical rationale as addressed in the current article is more significant and relevant to this journal. Therefore, we would like to omit the mathematical details in this article.
+
+<--- Page Split --->
+
+As mentioned in an earlier response above, long mathematical derivations are needed to extend the EGN model to account for only 4 correlated amplitudes (see Ref. 26), and improving the EGN model to account for larger structures seems mathematically daunting. In this article, we are not proposing an enhanced EGN model; rather, while allowing for model mismatch by using the classical EGN model, we improve the accuracy of evaluating structured lightwaves by replacing parameters of the EGN model with improved ones (i.e., by replacing \(\mu_{2}\) and \(\mu_{3}\) with optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) ). It is uncertain whether the existing approach to improving the model and our approach to using improved parameters can converge in the future, but regardless of the approach researchers take, we believe that our work will inspire people who have worked in this field.  
+
+Regarding your last point, you're right that "optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) " means that the values of \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) are obtained with the optimum window length. We clarified this by rephrasing it as "by replacing \(\mu_{2}\) and \(\mu_{3}\) with optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) (i.e., obtained with \(W_{SPM}^{*}\) and \(W_{XPM}^{*}\) ).  
+
+In the following, I list some minor comments.  
+
+1) \(p.2\) paragraph 1: "ever-increasing demand for communication capacity" \(\rightarrow\) "ever-increasing demand for higher data rates"  
+
+Response: We have modified this phrase as you suggested.  
+
+2) \(p.2\) paragraph 2: Refs. 8 and 9 are given as references for sphere shaping, but they are not. I would suggest instead [8],[7].  
+
+Response: We believe that this paragraph should provide readers with extensive prior work on sphere shaping. Since Refs. 9 and 10 realize sphere shaping for long block lengths, we have left citations to Refs. 9 and 10 and added more citations as you suggested.  
+
+3) \(p.2\) paragraph 2: "block of amplitudes" is rather vague, it should be related to "communication (QAM) symbols" (mentioned in the previous paragraph) better.  
+
+Response: We have modified this phrase as you suggested.  
+
+4) \(p.3\) paragraph 2: 4 consecutive amplitudes \(\rightarrow\) one DP symbol: This is discussed extensively in [4], I would suggest referencing it.  
+
+Response: We have added citations to related previous works.  
+
+5) \(p.3\) : \(\| \cdot \|^{2}\) is not defined  
+
+Response: Its definition has been added to the revised manuscript.
+
+<--- Page Split --->
+
+6) p.3: Explanation of sphere shaping can be connected to the related literature more strongly  
+
+Response: We have improved the explanation of sphere shaping in various places with added citations to previous works. Specifically, we added explanations of sphere shaping in relation to existing works in the second paragraph of Page 2, at the end of Page 4, and in the "Sphere shaping" section in Page 13.  
+
+7) p.3 paragraph 1: "(...) if every shaped block is plotted as a point in 4n-dimensional signal space, with the i-th amplitude being the position of the point on the i-th coordinate axis, the points uniformly fill a 4n-dimensional (hyper-)sphere (...) \(\rightarrow\) Since the amplitudes are drawn from a discrete set, this is not correct. The points are located in a 4n-dimensional spherical region of a 4n-dimensional rectangular lattice. I would avoid the use of the phrase "uniformly fill".  
+
+Response: Thanks for pointing out the inaccuracies in the description. We have improved the description of sphere shaping as follows: "the points are distributed uniformly over a set of 4n- dimensional square  
+
+lattice points that lie on or contained in a 4n- dimensional (hyper-) sphere of radius \(\sqrt{E_{\mathrm{shaped}}^{*}}\) (due to the symmetry by equiprobable signs)"  
+
+8) p.4: I would recommend the use of R instead of H for the information rate, since H is almost always used to denote the entropy  
+
+Response: We have changed the notation from \(H\) to \(R\) throughout the paper.  
+
+9) p.4 Fig. 1(a): Sphere shaping indeed decreases the maximum energy substantially. However, what is more important is the decrease in average energy. I would recommend indicating the decrease in average energy, and even relating it to the "linear" shaping gain discussed, e.g., in [9].  
+
+Response: To address this concern, we added two sentences to Page 4 with proper citations as "The average energy of \(\pmb{a}\) to achieve \(R\) with sphere shaping decreases with increasing block length \(4n\) (see, e.g., \(^{8,16}\) ), achieving a theoretical minimum average energy as \(n\rightarrow \infty\) . We refer to the reduction in average energy of \(\pmb{a}\) by shaping as the fundamental shaping efficiency in this article." The fundamental shaping efficiency is mentioned in Page 7 when explaining the NGMI of Fig. 2(b), and in Page 11 when analyzing Figs. 5(c) and (d). The reduction in average energy by sphere shaping is important (as is well known for communications over linear transmission media), but for nonlinear optical fiber communications that is the focus of this study, we would like to stress that the reduction in maximum energy (more precisely, the reduction in high- order statistical moments) is just as important as the average energy. Compared to infinite- length sphere shaping, the sphere shaping with \(n = 5\) can produce 0.9 to 1.0 dB higher \(SNR_{Eff}\) after nonlinear fiber propagation (see Fig. 2(a)) while achieving 0.6 dB lower fundamental shaping efficiency.  
+
+10) p.4 Fig. 1(b): I do not understand what this figure tries to explain, or how Fig. 1(a) relates to this.
+
+<--- Page Split --->
+
+Response: Fig. 1(a) shows only the probability distribution of total energy in unshaped versus shaped blocks, without the context of communications over a physical medium. Fig. 1(b) shows in the context of communications over a physical medium, in which rectangular regions of the time- frequency plane such probability distributions can be measured. The size of the rectangular regions is determined by \(R_{Sym}\) and \(n\) , and the distribution within this region is determined by \(n\) for the given \(\mathcal{A}\) and \(R\) . We believe that how Fig. 1(a) relates to Fig. 1(b) is expressed in the current statement, which reads “In a densely packed WDM system with identical channel configurations, such probabilistic energy distributions as in Fig. 1(a) are observed within each rectangular block that divides lightwaves in the time- frequency plane as shown in Fig. 1(b), where the width and height of the block are determined by both \(R_{Sym}\) and \(n\) .” To further clarify why this relation matters for communications, we added “While the shaping block length \(n\) or the distribution of energy (cf. Fig. 1(a)) has been optimized in existing studies \(^{8,11 - 12,15 - 17,23}\) to mitigate nonlinear interference (NLI), the spectro- temporal region where the distribution is found (cf. Fig. 1(b)) has never been noted previously. In the following sections, we will see that it is the distribution of light energy in all aspects of probability, time, and frequency that determines the manifestation of Kerr nonlinearity as NLI, and thus \(R_{Sym}\) and \(n\) must be controlled simultaneously to minimize NLI.”  
+
+11) p.4: sentence before (1) and (2): “(...) \(\eta\) increases as the central moment (...)” \(\rightarrow \eta\) or \(P_{NLI}\) ?  
+
+Response: We believe that the current statement is correct as is. \(\eta\) increases as the central moment \(\mu_{n}\) of \(\mathcal{P}\) increases. By (1), \(\langle P_{NLI} \rangle\) also increases as \(\mu_{n}\) increases.  
+
+12) p.5: “(...) deviation by the instantaneous power (...)” \(\rightarrow\) “(...) deviation of the instantaneous power (...) from average \(< \mathcal{P} > = 1\) (...)”?  
+
+Response: We corrected this phrase as “deviation using the instantaneous power”.  
+
+13) p.5: “(...) appears to be more spread around the average (...)” \(\rightarrow\) I think this can be related to the difference in their \(\bar{\mu}_{2}\) , right? If so, please refer to these values.  
+
+Response: We rephrased this sentence as “sphere shaping appears to make the instantaneous power more spread out around the average power (note the increase of \(\mu_{2}\) from 0.32 to 0.687 after sphere shaping)”.  
+
+14) Figures 1(c-d): What is the shaping rate and the resulting distribution?  
+
+Response: We added “All shaped symbols in Fig. 1 achieve \(R = 6.4\) ” to the caption of Fig. 1. The probability distribution of energy in a block of 5 symbols is shown in Fig. 1(a). The probability distribution of normalized energy in each symbol is shown in Fig. 1(c).  
+
+15) p.6: I do not understand what “With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods in all conditions except for \(R_{Sym} = 88\) GBd in Link D.” exactly means.
+
+<--- Page Split --->
+
+Response: It meant that the channel memory due to chromatic dispersion is shorter than the shaping block length, except for \(R_{Sym} = 88\) GBd in Link D. However, in the revised manuscript, we additionally provided results obtained with i.i.d. random symbols drawn from a Maxwell- Boltzmann distribution. Accordingly, the sentence you mentioned has been deleted.  
+
+16) p.6: I am not sure how the observation (iii) is explained with "(...) since increasing \(R_{Sym}\) reduces the shaping block duration in absolute time for the same \(n\) ".  
+
+Response: To address your concern (and as Reviewer 2 suggested), this sentence has been modified to "(iii) the optimal \(R_{Sym}\) increases as \(n\) decreases (compare, e.g., the yellow and red stars)."  
+
+17) p.7: I understand the observation "Areas far from the circled areas have a negligible impact on \(SNR_{Eff}\) . But I believe this can be related to Figs. 2(a,c) more strongly and quantitatively.  
+
+Response: We are not sure if you wanted to refer to Figs. 2(c- 5) and (c- 8). It is apparent from these figures that the low symbol rate region in Fig. 2(c- 5) and high symbol rate region in Fig. 2(c- 8) have a negligible impact on \(SNR_{Eff}\) . Taking \(\eta^{\infty}\) of these figures for i.i.d. shaping as baselines, \(\Delta \eta\) in Figs. 2(c- 6) and (c- 9) is added as offsets for finite- length shaping, and here we want to say that how significant the effect of such offsets \(\Delta \eta\) on \(SNR_{Eff}\) is determined by the baseline \(\eta^{\infty}\) . The meaning of the sentence is apparent from its preceding sentence, which reads "the influence of \(\Delta \eta_{SPM}\) and \(\Delta \eta_{XPM}\) on \(SNR_{Eff}\) is prominent only near the red circled areas, where their base coefficients \(\eta_{SPM}^{\infty}\) and \(\eta_{XPM}^{\infty}\) are large."  
+
+18) Fig. 3(a): How many spans?  
+
+Response: Thanks for pointing out the missing information. We added "at 240 spans in Link D" to the figure description in the text and "obtained at 240 spans in Link D" to the figure caption.  
+
+19) p.8: You say "optimal sphere shaping" more than once in the paper. What exactly do you mean by "optimal"? The combination of \(n\) and \(R_{Sym}\) that maximizes the performance? There is an optimal window length \(w\) that maximizes the correlation \(b/w\) the windowed moment and the effective SNR. But I am not sure what you mean by "optimal" sphere shaping. Clarification is needed.  
+
+Response: We made sure that the meaning of the optimality is clear in all places. In the fist sentence of the subsection "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" in Page 9, we clarified that optimal sphere shaping implies maximizing NDR. We also clarified this in Fig. 5 caption. In all other places, we believe that the meaning of optimality for sphere shaping or system parameters is explicitly stated or apparent from its surrounding sentences.  
+
+20) p.9: I believe the conclusion "In general, \(n^*\) increases as \(|D_{Total}|\) (green solid line) increases, (...) is a bit strong to be made from Fig. 5(a). The optimal block length here is mostly constant around 10-20,
+
+<--- Page Split --->
+
+increasing around 195 THz, but then decreasing again. If the authors can provide some additional explanation for this behaviour, it would make the paper stronger.  
+
+Response: We relaxed the conclusion by saying "In general, \(n^{*}\) tends to increase as \(|D_{Total}|\) (green solid line) increases." We attribute the unsmooth curves and deviation from theoretic prediction in Fig. 5 mostly to the difficulty of creating an ideal experimental setup with limited equipment. In particular, the recirculating loop that is the only experimental method of long- haul transmission involves many variations of real- world components. The most important challenge for us was minimizing the optical power excursion across the C- band using the DGEs that have only 0.1 dB nominal attenuation granularity. In true straight- line long- haul systems, the gain tilts and ripples of inline EDFAs can nearly perfectly be flattened by passive optical attenuators with continuous attenuation profiles that were predetermined for a fixed total optical power. On the other hand, in the recirculating loops operated with various total optical powers, the gain tilts and ripples should be flattened by the DGEs with coarse attenuation granularity. A small residual power excursion due to this limitation increases as it accumulates over the number of loops, leading to deviations from the ideal flat optical power spectral density. To address this, we added "The power excursions due to experimental constraints (e.g., a small power excursion caused by coarse attenuation granularity of the DGEs results in an increasing power excursion as the number of loops increases) are considered to be the most important contributor to discrepancy in validation of theory" at the end of Page 10.  
+
+21) p.9: At which block length Fig. 5(b) is plotted? Also, similar to my previous comment, the conclusion "(...) \(R_{Sym}\) decreases (...) when \(|D_{Total}|\) increases (...) is a bit strong to be made from Fig. 5(b). Instead of simply saying Figs. 5(a,b) agree with Figs. 2(b,a), possible explanations for the peculiar behaviours in Figs. 5(a,b) should be provided.  
+
+Response: For clarification, we combined the descriptions of Figs. 5(a) and (b) in the text as "Figs. 5(a) and (b) show, respectively, the optimal \(n^{*}\) and \(R_{Sym}^{*}\) that jointly maximize NDR in each channel," and added "for maximum NDR" to the captions of Figs. 5(a) and (b). We also relaxed the conclusion by saying " \(R_{Sym}^{*}\) tends to decrease." The non- smooth change of optimal parameters is attributed to experimental limitations, and the added sentence for Fig. 5(a) mentioned above also provides an explanation for Fig. 5(b).  
+
+22) p.9, third line: \(SNR_{Eff}\) of Fig. 4(c), or of Fig. 5(c)?  
+
+Response: We corrected this typo.  
+
+23) p.11: CCDM is not a sphere shaping algorithm.  
+
+Response: We added "Note that CCDM can only approximately realize sphere shaping for finite block lengths, but it converges to ideal sphere shaping with a decreasing approximation error as the block length increases" to the first paragraph in the Methods section.
+
+<--- Page Split --->
+
+24) Similar gains in data rate/SNR/AIR to the ones that are reported at the end manuscript exist in the literature. A short review of such works would make the paper stronger.  
+
+Response: At the end of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section, we added brief explanation about the previous works and emphasized the difference of this work compared to the previous works as "While the previous works<sup>8,11,12,16,23</sup> optimized only \(n\) to observe some gains in \(SNR_{Eff}\) and NDR over specific links (e.g., for single- span links), joint optimization of \(n\) and \(R_{Sym}\) in this work produces significantly larger gains and allows these gains to be achieved over a much wider variety of links." Due to the word count limit of Nature Communications, this level of explanation seems to be the best we can do.  
+
+## REFERENCES  
+
+[1] K. Wu, G. Liga, A. Sheikh, F. M. J. Willems, and A. Alvarado, "Temporal energy analysis of symbol sequences for fiber nonlinear interference modelling via energy dispersion index," Feb. 2021. [Online]. Available: https://arxiv.org/abs/2102.124114 [2] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike- Akino, K. Kojima, K. Parsons, and D. S. Millar, "Nonlinearity tolerant LUT- based probabilistic shaping for extended- reach single- span links," IEEE Photon. Technol. Lett., vol. 32, no. 16, pp. 967- 970, 2020. [3] T. Fehenberger, "On the impact of finite- length probabilistic shaping on fiber nonlinear interference," in Proc. Signal Process. in Photon. Commun. (SPPCom), Washington, D.C., United States, July 2020. [4] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike- Akino, K. Kojima, K. Parsons, and D. S. Millar, "Huffman- coded sphere shaping for extended- reach single- span links," IEEE J. Sel. Topics Quantum Electron., vol. 27, no. 3: 3500215, May- June 2021. [5] S. Civelli, E. Forestieri, and M. Secondini, "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation," in Proc. Eur. Conf. Opt. Commun. (ECOC), Brussels, Belgium, Dec. 2020. [6] T. Fehenberger, A. Alvarado, G. B'oherer, and N. Hanik, "On probabilistic shaping of quadrature amplitude modulation for the nonlinear fiber channel," J. Lightw. Technol., vol. 34, no. 21, pp. 5063- 5073, Nov. 2016. [7] Y. C. Gültekin, A. Alvarado, O. Vassilieva, I. Kim, P. Palacharla, C. M. Okonkwo, and F. M. Willems, "Kurtosis- limited sphere shaping for nonlinear interference noise reduction in optical channels," June 2021. [Online]. Available: https://arxiv.org/abs/2105.14794 [8] P. Schulte and F. Steiner, "Divergence- optimal fixed- to- fixed length distribution matching with shell mapping," IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 620- 623, Apr. 2019. [9] Y. C. Gültekin, W. J. van Houtum, A. Koppelaar, and F. M. J. Willems, "Enumerative sphere shaping for wireless communications with short packets," IEEE Trans. Wireless Commun., vol. 19, no. 2, pp. 1098- 1112, Feb. 2020.
+
+<--- Page Split --->
+
+## Response to Reviewer 2:  
+
+This work studies the optimization of the time and bandwidth over which sphere shaping should be performed to minimize the impact of nonlinear effects and maximize performance in optical fiber communication.  
+
+The work is timely, well written, and reports some novel interesting results.  
+
+The optimization of the shaping block length in the nonlinear regime has been already investigated in some other papers but, to my knowledge, the simultaneous optimization of block length and symbol rate (i.e., time and bandwidth) is studied here for the first time. Moreover, this study encompasses several scenarios and a wide range of parameters, both numerically and experimentally.  
+
+The methodology is sound and the methods are described with enough detail. Moreover, the conclusions are supported both by numerical and experimental results.  
+
+In general, I believe that the paper deserves publication. However, there are some weaknesses and some issues that should be considered and possibly addressed, as detailed below.  
+
+Response: Thank you very much for the positive comments and suggestions for improving the paper. We tried to address all of your concerns in the revised manuscript, and hope that the weaknesses and issues found in the previous manuscript have been removed by this revision.  
+
+1) As a general note, while reading the paper, I was expecting to find a theoretical analysis which explains and predicts the results observed numerically and experimentally. This expectation is not fully meet by the paper, which provides only some insight about the observed phenomena and some empirical laws to model them - e.g., (7), (8) and the use of the windowed moments in the EGN model. In fact, there are a number of perturbation models in the literature which could be used to attempt such an analysis, for instance by removing the i.i.d. assumption in the EGN model. An example can be found in Liga et al "Extending Fibre Nonlinear Interference Power Modelling to Account for General Dual-Polarisation 4D Modulation Formats", Entropy, 2020. I recommend discussing this issue and highlighting the empirical nature of the proposed model.  
+
+Response: Thank you for referring us to the previous work on propagation modeling of structured lightwaves. In fact, after submitting the first manuscript, we learned from an author of Ref. 26 (which is the paper that you mentioned above) at a conference that an extended EGN model was published last year for the presence of structures (or correlations) in four amplitudes, i.e., in one symbol period. To the best of our knowledge, this (and its simplification in Ref. 27) is the only published work on propagation modeling of structured lightwaves. However, given the enormous complexity of the mathematical expansion to account for only four correlated amplitudes in Ref. 26, it can be inferred that it is daunting to find a mathematically accurate analytical model to handle tens to thousands of correlated amplitudes that we cover in this paper. In this regard, we addressed your concern by adding a sentence to the beginning of Page 3 as "Analytical approaches to take into account the structure of lightwaves have so far been successful up to one symbol<sup>26,27</sup>, but extending the analysis to structures spanning many symbols seems mathematically daunting. To quantify the effect of large temporal structures of lightwave on Kerr nonlinearity, empirical approaches are being taken in rapidly growing recent studies<sup>8,15- 17,23</sup>. However, there has been no study on whether or how the symbol rate affects this quantification." We also stated on Page 9 that "The EGN model assumes i.i.d. amplitudes and phases of symbols, and hence is not accurate
+
+<--- Page Split --->
+
+for lightwaves with local energy structures. There is a recently developed analytical model<sup>26,27</sup> that extends the EGN model to account for energy structures present over one symbol period, but extending this further to energy structures spanning tens to thousands of symbol periods that we deal with in this work seems mathematically intractable. Therefore, we allow for model mismatch by using the classical EGN model, but improve the accuracy of evaluating structured lightwaves (green solid lines in the figure) by replacing \(\mu_{2}\) and \(\mu_{3}\) with optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) (i.e., obtained with \(w_{SPM}^{*}\) and \(w_{XPM}^{*}\) )."  
+
+2) As a second general comment, the analysis practically neglects the impact of carrier recovery. Indeed, this might be substantial, as a sufficiently fast carrier recovery algorithm can mitigate the impact of nonlinear phase noise and change the dependence of performance on block length (see for instance Civelli et al. "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation", ECOC 2020). With carrier recovery, there might be no advantage in using short (optimized) block lengths. It seems that carrier recovery is not present in the simulations and is quite slow in the experiments (using only 1 pilot every 48 symbols), possibly overestimating the advantage of using a short optimized block length. Please comment about that. As a side note, for the sake of transparency, I am a coauthor of the above mentioned paper. I do not want to push my own work, therefore feel free to decide if the paper is relevant or not to your work and if it deserves to be cited or not. This decision will not affect my final recommendation.  
+
+Response: Thanks for pointing out another prior work related to this manuscript and for providing transparency. As reported in the suggested reference paper, the phase rotation due to NLI can be mitigated by the BPS algorithm under certain conditions. To check this, we performed the BPS on our simulation data, but no appreciable effect was observed in our settings, especially over long distances with low SNR as shown in the table below for Link D. The number of symbols, \(2N_{BPS} + 1\) , that are used for averaging the phase rotation is determined such that the resulting \(SNR_{Eff}\) is approximately maximized, and a launch power of 1 dBm is used that is close to optimal.  
+
+<table><tr><td rowspan="2">Number of Spans</td><td rowspan="2">2NBP5 + 1</td><td colspan="2">NSC = 1</td><td colspan="2">NSC = 32</td></tr><tr><td>Without BPS</td><td>With BPS</td><td>Without BPS</td><td>With BPS</td></tr><tr><td>15</td><td>17</td><td>20.41</td><td>20.57</td><td>20.88</td><td>20.94</td></tr><tr><td>240</td><td>513</td><td>7.34</td><td>7.38</td><td>8.39</td><td>8.31</td></tr></table>  
+
+Similarly, when we performed the BPS on our experimental data, we did not observe noticeable change in \(SNR_{Eff}\) . Note that for carrier recovery of the experimental data, we used linear interpolation to obtain the phase of 47 payload symbols between 2 consecutive pilot symbols. We briefly discussed about the BPS at the end of the Discussion section as "Also, the use of advanced carrier recovery algorithms such as the maximum- likelihood blind phase search (BPS) may influence the impact of sphere shaping on NLI under certain conditions<sup>46</sup>, but in this work at transmission distances that match the sphere- shaped 16- QAM format, no noticeable effect was observed using the BPS."  
+
+3) Abstract: The sentence "In optical fiber, however, the sphere shaping induces Kerr nonlinearity in a peculiar way that makes analysis of transmission performance difficult, potentially lowering the communications capacity" sounds a bit odd. It suggests that it is sphere shaping that makes the analysis
+
+<--- Page Split --->
+
+difficult, with Kerr nonlinearity just playing an indirect role. In fact, I think the opposite is true. Moreover, it suggests that the difficulty of the analysis may play a role in the capacity reduction.  
+
+Response: To emphasize that it is Kerr nonlinearity that is hard to understand, we rephrased the sentence as "However, when shaped lightwaves are transmitted through optical fiber, Kerr nonlinearity manifests itself as nonlinear interference in a peculiar way, potentially lowering communications capacity."  
+
+4) Introduction: The idea of using sphere shaping on an optimized block length to minimize the impact of nonlinear interference was first proposed in [10] and then further investigated by some of the same authors in Geller et al. "A Shaping Algorithm for Mitigating Inter-Channel Nonlinear Phase-Noise in Nonlinear Fiber Systems", JLT 2016. Your work (as well as some other recent works) heavily relies on this idea, extending the optimization to the frequency domain. I think that it would be fair to better acknowledge this fact and the role of these two papers.  
+
+Response: We clarified what has been done in the previous works and what is the difference of this work compared to the previous works by adding several sentences to the paper as follows. In the second paragraph of the Introduction section, we added "For this reason, there have been several recent approaches to optimizing the shaping block length to mitigate Kerr nonlinearity<sup>8,11-12,15-16,23</sup>." At the end of the "Sphere Shaping of Lightwaves" section, we added "While the shaping block length \(n\) or the distribution of energy (cf. Fig. 1(a)) has been optimized in existing studies<sup>8,11-12,15-17,23</sup> to mitigate nonlinear interference (NLI), the spectro-temporal region where the distribution is found (cf. Fig. 1(b)) has never been noted previously." At the end of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section, we added "While the previous works<sup>8,11,12,16,23</sup> optimized only \(n\) to observe some gains in \(SNR_{eff}\) and NDR over specific links (e.g., for single- span links), joint optimization of \(n\) and \(R_{Sym}\) in this work produces significantly larger gains and allows these gains to be achieved over a much wider variety of links."  
+
+As for the last sentence added above, the findings of this article suggest that for symbol rates greater than 10 GBd, as is commonly used in recent experiments, optimization of the shaping block length is only effective over very short distances on dispersion- unmanaged SSMF links where net dispersion is small (see equations (7), (8) and Supplementary Fig. 4). This perhaps explains why most of the previous works on optimizing the shaping block length used single- span links. For long distances on dispersion- unmanaged SSMF links, we are able to observe a noticeable benefit by shaping block length optimization only when using a low symbol rate of a few GBd. The last sentence above emphasizes this briefly.  
+
+5) Page 3, Results: you state that "The symmetry of the probability allows for legitimate analysis with only positive amplitudes, and hence we omit the sign throughout this article". I guess that the presence of a random sign with uniform distribution is implicitly assumed in the paper and neglected only in the description, whereas it is included both in the simulations and in the experiments. Is that correct? I am not sure that transmitting only positive amplitudes would indeed give the same results. I think that two consecutive symbols with the same sign or with opposite signs induce a fundamentally different response in the nonlinear channel. Please be more explicit.
+
+<--- Page Split --->
+
+Response: You're correct that random signs were used with equal probability in simulations and experiments. We have clarified this by rewriting the sentence as "we omit the sign throughout this article for descriptive purposes (but in simulations and experiments, equally distributed positive and negative signs are used)."  
+
+6) Page 3, Results: the sentence "one dual-polarization symbol that maintains the value over a period of \(T\_ sym (s)\) and can change the value at a rate of \(R\_ sym = 1 / T\_ sym (Bd)\) ", suggests a rectangular pulse shape. I suggest revising the sentence. Moreover, the indication of the units of measure in this way is neither standard nor required.  
+
+Response: To eliminate the confusion that the pulse shape is rectangular, we have rephrased the sentence as "In our system, four consecutive amplitudes constitute one dual-polarization symbol, as has been done, e.g., in \(^{16,17}\) , that is transmitted with a symbol period of \(T_{Sym}\) and a symbol rate of \(R_{Sym} = 1 / T_{Sym}\) ." We also removed the unnecessary unit of \(T_{Sym}\) and \(R_{Sym}\) here.  
+
+7) Page 5, 4 lines before the end of the subsection: The sentence "As \(w\) increases, \(\forall m\_ 2\) remains the same for all types of i.i.d. symbols" is misleading. It is not "the same" for all types of symbols but, rather, it remains "constant" as \(w\) increases for a given modulation (type of symbols).  
+
+Response: We fixed the incorrect sentence as " \(\bar{\mu}_{2}\) remains constant for i.i.d. symbols."  
+
+8) Equation (3): for sphere shaping, the transmitted signal is a ciclostationary process with period equal to the length of the shaped blocks (unless a uniform random delay is assumed). Therefore, I guess that the windowed moments depend on the time at which the window is centered, unless they are averaged over it. Please comment about that and explain if and how you have considered this time dependence in your analysis.  
+
+Response: You're correct that sphere- shaped signal, and hence \(p\) , is a cyclostationary process with a period equal to the \(n\) symbol periods. If the sliding step size of the moving average filter \((\cdot)_{w}\) is greater than one symbol, it is also true that the windowed moments \(\bar{\mu}_{n}\) in equation (3) can have different values depending on where the window starts sliding in the stream of the shaped symbol blocks. However, we use a sliding step size of one symbol, as is typically done in moving average filters. In this case, \(\bar{\mu}_{n}\) can have only a single fixed value. To avoid potential confusion, we explicitly stated this fact under equation (3) as " \((\cdot)_{w}\) denotes a moving average filter with a sliding window of length \(w\) symbols (with a sliding step size of one symbol)."  
+
+9) Page 5, Section "Optimization of Sphere Shaping in the Time-Frequency Plane": Please specify what QAM format is considered in this section. I guess from the given data that it is 16QAM (as in the previous section). Yet, an explicit mention here would be useful.  
+
+Response: In the "Optimization of Sphere Shaping Parameters in the Time- Frequency Plane" section, we explicitly stated the modulation format by modifying a sentence to "Sphere shaping is performed in each
+
+<--- Page Split --->
+
+channel with \(n = 5\) , 10, 20, 40, 80, 320, 1280, 5120, with a fixed \(R = 6.4\) bits per dual- polarization symbol using 16- QAM." Also, in the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section, we specified the modulation format by modifying a sentence to "For each pair of \(n\) and \(R_{sym}\) , the NDR achieved by sphere shaping of 16- QAM is determined by...".  
+
+10) Page 6: You state that "With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods in all conditions except for \(R_{sym} = 88\) GBd in Link D" and that "like the i.i.d. uniform QAM, the optimal \(R_{sym}\) for i.i.d. shaping to maximize SNR_eff (yellow stars) decreases as the total net dispersion \(D_{Total}\) increases". I think that this discussion is a bit weak and that the analogy with the symbol rate optimization for uniform i.i.d. QAM is misleading for two reasons: 1) The dependence of SNR_eff on symbol rate for the uniform i.i.d. QAM case vanishes for Gaussian symbols (the EGN and GN model converge to the same equations, see for instance [23]). Therefore, I expect a very little dependence on the symbol rate for shaped i.i.d. symbols, as they approximate i.i.d. Gaussian symbols. In fact, the dependence on symbol rate in Fig. 2a for the case \(n = 5120\) is very weak (and would be probably even weaker for shaped 64-QAM or 256-QAM) and the peak SNR is not so different from the surrounding values in the simulation range; 2) assuming that \(n = 5120\) represents the i.i.d. case from a dispersion perspective is an approximation whose accuracy decreases as the symbol rate increases; therefore, it cannot be used when explaining the (weak) dependence of SNR_eff on symbol rate. In fact, for \(n = 5120\) , this dependence becomes a little more relevant exactly where the approximation becomes looser. In my opinion, it seems more likely that the dependence of SNR_eff on symbol rate is mainly explained, even for \(n = 5120\) , by the same reason that explains it for lower \(n\) .  
+
+Response: Thank you very much for this valuable insight. To address your concern, we added split- step simulation results of i.i.d. shaping, in which shaped symbols are drawn from a 16- QAM alphabet according to a Maxwell- Boltzmann distribution. Inclusion of i.i.d. shaping posed difficulties in visualizing \(SNR_{Eff}\) in contour plots of Fig. 2 and Supplementary Figs. 1- 4, since it is equivalent to infinite- length sphere shaping in principle. In fact, as mentioned in the Methods section, the number of transmitted symbols in each WDM channel is limited to \(2^{18}\) , \(2^{17}\) , \(2^{16}\) , \(2^{15}\) , \(2^{14}\) , \(2^{14}\) , \(2^{14}\) dual- polarization symbols in each channel, respectively, for \(N_{Ch} = 1\) , 2, 4, 8, 16, 32, 64. Nevertheless, we kept the format of the contour plots the same as in the first manuscript, adding one point at the top of each \(R_{Sym}\) . This visualization would be acceptable as we see that the difference in \(SNR_{Eff}\) between \(n = 5120\) and i.i.d. shaping is very small in all figures. Accordingly, below equation (4), we explained the figure as "The top points at each \(R_{Sym}\) represent i.i.d. shaping, so their \(y\) - axis values are not exact values but merely represent very large numbers. The \(y\) - axis values for all other points are exact".  
+
+We agree that \(SNR_{Eff}\) is not changed by \(R_{Sym}\) if the modulation format is continuous Gaussian, and that the dependence of \(SNR_{Eff}\) on \(R_{Sym}\) will be very weak in the case of sphere shaping with high- order QAM. However, with relatively small i.i.d. shaped 16- QAM, we observe about 1 dB change in \(\eta^{\infty}\) due to \(R_{Sym}\) , which is not much different from the change in \(\eta\) due to \(R_{Sym}\) in i.i.d. uniform QAM (see the lower figure in Fig. 4 of Ref. 25). To emphasize the small QAM order, we specified the modulation format as "(i) like the i.i.d. uniform 16- QAM \(^{25}\) , the optimal \(R_{Sym}\) for i.i.d. shaping of 16- QAM to maximize \(SNR_{Eff}\) (yellow stars) decreases as the total net dispersion \(D_{Total}\) increases". Also, we wrote "The
+
+<--- Page Split --->
+
+influence of \(R_{Sym}\) on \(\eta^{\infty}\) is expected to decrease further as the modulation order increases and the shaped signal approaches continuous Gaussian" in the first paragraph of Page 8, and "It also remains for future work to see how the dependence of \(\eta\) on \(R_{Sym}\) and \(\eta\) changes as the sphere-shaped QAM modulation order increases to approach continuous Gaussian signaling in terms of the time- averaged probability distribution." in the first paragraph of the Discussion section.  
+
+11) Page 6, you state that "the optimal \(R_{-}sym\) is higher for finite-length shaping than for i.i.d. shaping (compare the yellow and red stars), since increasing \(R_{-}sym\) reduces the shaping block duration in absolute time for the same \(n\) ". Following up on the previous comment, can we better (and more simply) say that the optimal symbol rate increases when \(n\) decreases?  
+
+Response: Thanks for the great suggestion. Following your suggestion, we have modified the sentence as "(iii) the optimal \(R_{Sym}\) increases as \(n\) decreases (compare, e.g., the yellow and red stars)."  
+
+12) Still on the same matter: you refer many times in the results to the i.i.d. case, considering \(n = 5120\) as an approximation to it. Why not directly simulating the true i.i.d. case? It should be sufficient to draw i.i.d. QAM symbols from a Maxwell-Boltzmann distribution. That would be a more meaningful benchmark against which to compare the results for finite block lengths and, for instance, to compute \(\forall \mathrm{eta}^{\infty}\) (infy) in (6).  
+
+Response: We agree that the inclusion of i.i.d. shaping would greatly improve the paper, and as mentioned above, we added split- step simulation results of i.i.d. shaping. Figures and many parts of the text have been modified accordingly.  
+
+13) Page 8, Demonstration of Optimal Sphere Shaping through Full C-Band Transmission Experiment: The experiment is performed on a link that is substantially different from the 4 links considered in the simulation. Why? In particular, each channel sees a different dispersion profile, making the interpretation of the results harder. Moreover, the important dispersion-unmanaged case is not considered in the experiments. Is there a particular reason for this choice? Is it instrumental to highlight some effect or behavior? Please explain and motivate this choice.  
+
+Response: Although we aim to estimate the benefits of shaping optimization in the full C- band transmission, our simulation setup is limited to a total bandwidth of 100 GHz due to the long simulation time. The experiment is much easier to implement the full C- band transmission setup, but it is very difficult to test all 4 links through the experiment due to the availability of different types of fiber and the enormous effort required to construct a link (e.g., custom gain flattening for each link and for each launch power is a huge effort). On one hand, performing the experiment in a single link close to one of the 4 links used in simulation is good for validating the findings of simulation. But it has some weaknesses as follows. (i) The validation is limited to a specific dispersion coefficient of the selected link. (ii) To evaluate the impact of dispersion on NLI, the transmission distance needs to be changed as shown in, e.g., Supplementary Fig. 4. However, as the distance changes, 16- QAM becomes no longer a suitable modulation format that matches the underlying \(SNR_{eff}\) , making it difficult to evaluate the NDR increase
+
+<--- Page Split --->
+
+achieved by optimizing the sphere shaping parameters with the same modulation format as in the simulation.  
+
+On the other hand, allowing for mismatches of the setup between simulation and experiment, performing the experiment on a dispersion- managed link near zero- dispersion frequency with a dispersion slope allows us to see how NLI varies with the sphere shaping parameters under various chromatic dispersion coefficients, even at a fixed distance. This allows for qualitative validation of simulation results, as briefly mentioned in the first paragraph of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section as "The dependence of the optimal sphere shaping on dispersion is conveniently verified in a recirculation loop that accumulates varying dispersions over frequency." While we agree that the dispersion- unmanaged system is an important use case, we would like to leave the experiment in dispersion- unmanaged systems for future work.  
+
+14) Page 11, Methods: you state that "The digital sphere shaping encoder is implemented by enumerative sphere shaping (ESS) for \(n = 5\) , 10, 20, 40, 80, and by constant composition distribution matching (CCDM) for \(N = 320\) , 1280, 5120." and "...consume 0.583, 0.346, 0.209, 0.137, 0.099, 0.025, 0.009, 0.002 dB more average symbol energy than ideal shaping, respectively". Strictly speaking, CCDM does not implement sphere shaping (as all constant composition sequences lie on the surface of the sphere, partially covering it) and, in fact, its energy loss is higher than that of ESS for the same blocklength. Of course, they both asymptotically converge to i.i.d. Maxwell-Boltzmann symbols, but as you are studying how performance changes with block length, you cannot assume that such an asymptotic regime has been achieved. In fact, I am also a bit surprised that CCDM with \(n = 320\) performs significantly better than ESS with \(n = 80\) (see, e.g., the comparison in [7]). Please comment about that and double check the numbers (I have not verified them, it is just a feeling and I might be wrong).  
+
+## Marco Secondini  
+
+Response: As you pointed out, CCDM can only approximately realize sphere shaping. Its rate loss decreases as the block length increases, and approaches to ideal sphere shaping as the block length tends to infinity, as a result of the sphere hardening phenomenon. To address your concern, in the first paragraph of the Methods section, we added the sentence "Note that CCDM can only approximately realize sphere shaping for finite block lengths, but it converges to ideal sphere shaping with a decreasing approximation error as the block length increases."  
+
+To explain the performance of CCDM as good as that of ESS at \(n = 320\) , we have reproduced Fig. 3 of Ref. 8 (which is [7] in your reference number) in the figure on the left below with \(R = 1.6\) bit/amp and 64- QAM. However, unlike Ref. 8, this paper uses 16- QAM, and in this case the increase in rate loss of CCDM compared to ESS is not large even at a fairly short block length, as shown on the right figure below. This explains the good performance of CCDM at \(n = 320\) .
+
+<--- Page Split --->
+![PLACEHOLDER_25_0]
+  
+
+![PLACEHOLDER_25_1]
+
+
+<--- Page Split --->
+
+REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+All my comments and suggestions seem to be taken into account. I can suggest the publication of the paper following a proofreading.  
+
+Reviewer #2 (Remarks to the Author):  
+
+All my comments have been properly addressed.  
+
+Marco Secondini
+
+<--- Page Split --->
+
+# Response to Reviewers' Comments on NCOMM-21-15776 "Shaping Lightwaves in Time and Frequency for Optical Fiber Communication"  
+
+Junho Cho, Xi Chen, Greg Raybon, Di Che, Ellsworth Burrows, Samuel Olsson, and Robert Tkach  
+
+Dear Reviewers,  
+
+We sincerely thank all reviewers for carefully reading and commenting on our paper again. Please find our point- by- point response to the comments below, where the reviewers' comments are reproduced verbatim in blue italic type, and our response is written in black roman type.  
+
+## Response to Reviewer 1:  
+
+All my comments and suggestions seem to be taken into account. I can suggest the publication of the paper following a proofreading.  
+
+Response: Your comments have greatly improved our manuscript. Thank you again very much for this.  
+
+## Response to Reviewer 2:  
+
+All my comments have been properly addressed.  
+
+Response: Thank you very much for your comments during the last two reviews. They advanced our understanding of the subject area and significantly improved the manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42151a82c0281305539beb0a5a086a26c865e127f2c665b3801e9dc9d28b53f2/supplementary_0_Peer Review File__42151a82c0281305539beb0a5a086a26c865e127f2c665b3801e9dc9d28b53f2_det.mmd

@@ -0,0 +1,669 @@
+<|ref|>title<|/ref|><|det|>[[61, 41, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[70, 111, 362, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 154, 890, 211]]<|/det|>
+Shaping Lightwaves in Time and Frequency for Optical Fiber Communication  
+
+<|ref|>image<|/ref|><|det|>[[57, 733, 240, 783]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 733, 912, 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|>title<|/ref|><|det|>[[120, 61, 877, 162]]<|/det|>
+# Review of NCOMMS-21-15776: "Shaping Lightwaves in Time and Frequency for Optical Fiber Communication"  
+
+<|ref|>text<|/ref|><|det|>[[77, 209, 920, 376]]<|/det|>
+The authors of the paper investigate constellation shaping (sphere shaping in particular) for optical communication systems. They focus on the effect of (1) the block length of sphere shaping and (2) the symbol rate of the signaling on the amount of nonlinear interference (NLI) generated during propagation. The dependency of NLI on the shaping block length has been investigate in the literature. The authors propose a new metric (windowed central moment) to study this dependency. This metric measures the high- order deviations of the energy of the optical field from its average over a finite time interval. They claim that properly selecting the width of this window is important to properly compare different shaping schemes from NLI generation perspective and to predict their performance. Furthermore, they claim that the nonlinearities depend on the energy structure of the optical field in absolute time rather than in number of symbols. This leads to a complex relation between the shaping block length, symbol rate and NLI. Using the values of the window length at which the correlation between the windowed moment and the effective SNR is maximized, authors claim that the EGN model predicts the effective SNR (using windowed moments) more accurately than its traditional form (using regular moments) (Fig. 3(b)).  
+
+<|ref|>text<|/ref|><|det|>[[77, 377, 920, 496]]<|/det|>
+The paper is well- written, findings are interesting, the topic is timely/relevant, and the explanation of the experimental setup is very clear and detailed. I believe the paper can be accepted after a round of revision/polishing. The structure of the paper is different than I am used to from other journals in the literature and this structure may be leading to a different reading experience than I am used to. However I must say that the introduction can be improved by stating (1) the problem, (2) what is done, and (3) what is found more clearly with short and direct sentences. At some points, I had trouble following the train of reasoning. I believe the findings of the paper are quite interesting, and they should be stated more clearly to increase their impact. In the following, I will list my comments which I hope would help the authors to improve the presentation in the paper.  
+
+<|ref|>text<|/ref|><|det|>[[92, 498, 920, 580]]<|/det|>
+- Firstly, I find it a bit complicated to understand the fundamental motivation and findings of the paper from the abstract and the introduction. I suggest a clearer statement of the fact that (1) the paper investigates the optimal combination of the shaping block length and the symbol rate that maximizes the performance, (2) this is achieved using a new metric—similar to a metric proposed in [1]—that has high correlation with the effective SNR. At some parts of the paper, it is implied that a NLI-optimized sphere shaping approach is proposed. However, as far as I understand, only the optimum system parameters are investigated (which is also a valuable contribution) for regular sphere shaping.  
+
+<|ref|>text<|/ref|><|det|>[[92, 589, 920, 679]]<|/det|>
+- In the traditional EGN model, higher order moments of the channel input distribution are used in the expressions that compute the NLI variance. Here, they are replaced with their windowed versions and the reported results match the actual performance more accurately than the original EGN model. Can the authors comment on the use of the windowed moments in the EGN model? Is there a theoretical background for this, or is it just heuristic? Can this approach be used to propose an enhanced EGN model that works for finite-length shaping schemes and captures the temporal structure of the shaped waveforms in the analytical model?  
+
+<|ref|>text<|/ref|><|det|>[[92, 680, 920, 740]]<|/det|>
+- The improvement in rate (or in reach) due to shaping (and sphere shaping in particular) has been demonstrated in the literature repeatedly. The paper provides an optimization of shaping over (i) block length and (ii) symbol rate. There are also other works in the literature which discusses the effect of block length and symbol rate on performance. They should at least be cited, e.g., [2], [3], [4], [5], etc.  
+
+<|ref|>text<|/ref|><|det|>[[92, 741, 920, 875]]<|/det|>
+- In p.5, it is stated that refs. 7 and 15 show that if \(n\) is small as in Fig. 1(c), sphere shaping reduces NLI. Firstly, ref. 15 is not about sphere shaping, but about constant composition sequences. Secondly, as previously observed in [6] and re-stated in ref. 7, (AWGN-optimal) shaping leads to increased NLI (decreased effective SNR). This is attributed to the increased kurtosis in [6], [7]. Therefore, I am not sure I see the contradiction that you mention here. Furthermore, the explanation concerning Fig. 1(d) is not clear. Are we focusing on the effect of \(n\) on NLI here or the importance of selecting \(w\) to predict the performance? Because regardless of \(w\) , smaller \(n\) implies smaller \(\bar{\mu}_{2}\) in Fig. 1(d) (for most of the region). I would suggest a careful revision of the paragraph in p.5 that starts with "Fig. 1(c)" (which I believe should be "Figure 1(c)") with clear explanations (of the claims and figures) and as much connections to the existing literature as possible. I believe this paragraph is extremely important to motivate your work.  
+
+<|ref|>text<|/ref|><|det|>[[92, 876, 920, 937]]<|/det|>
+- As far as I know, the effective SNR in EGN model depends on the average power of the input as well as its regular fourth and sixth order moments. In p.8, how do you use the EGN model with \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) to compute the effective SNR? Figure 3(b) reports that this computation (EGN with windowed moments) is far more accurate than the regular EGN model in predicting the effective SNR. If this is the case, would you claim that you're proposing an enhanced EGN model?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[108, 68, 920, 113]]<|/det|>
+Furthermore, is it possible to obtain a further more accurate relation if (somehow) some other higher- order windowed moments are included in the model? Also, when you say "optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) " here, do you mean their values computed with the optimum window length?  
+
+<|ref|>text<|/ref|><|det|>[[95, 116, 405, 131]]<|/det|>
+In the following, I list some minor comments.  
+
+<|ref|>text<|/ref|><|det|>[[92, 133, 920, 165]]<|/det|>
+1) p.2 paragraph 1: "ever-increasing demand for communication capacity" \(\rightarrow\) "ever-increasing demand for higher data rates" 
+2) p.2 paragraph 2: Refs. 8 and 9 are given as references for sphere shaping, but they are not. I would suggest instead [8], [7].  
+
+<|ref|>text<|/ref|><|det|>[[92, 183, 920, 212]]<|/det|>
+3) p.2 paragraph 2: "block of amplitudes" is rather vague, it should be related to "communication (QAM) symbols" (mentioned in the previous paragraph) better.  
+
+<|ref|>text<|/ref|><|det|>[[92, 212, 920, 240]]<|/det|>
+4) p.3 paragraph 2: 4 consecutive amplitudes \(\rightarrow\) one DP symbol: This is discussed extensively in [4], I would suggest referencing it.  
+
+<|ref|>text<|/ref|><|det|>[[92, 241, 285, 255]]<|/det|>
+5) p.3: \(||\cdot ||^{2}\) is not defined  
+
+<|ref|>text<|/ref|><|det|>[[92, 256, 730, 271]]<|/det|>
+6) p.3: Explanation of sphere shaping can be connected to the related literature more strongly  
+
+<|ref|>text<|/ref|><|det|>[[92, 272, 920, 330]]<|/det|>
+7) p.3 paragraph 1: "(...) if every shaped block is plotted as a point in \(4n\) -dimensional signal space, with the \(i\) -th amplitude being the position of the point on the \(i\) -th coordinate axis, the points uniformly fill a \(4n\) -dimensional (hyper-)sphere (...) \(\rightarrow\) Since the amplitudes are drawn from a discrete set, this is not correct. The points are located in a \(4n\) -dimensional spherical region of a \(4n\) -dimensional rectangular lattice. I would avoid the use of the phrase "uniformly fill".  
+
+<|ref|>text<|/ref|><|det|>[[92, 330, 920, 360]]<|/det|>
+8) p.4: I would recommend the use of \(R\) instead of \(H\) for the information rate, since \(H\) is almost always used to denote the entropy  
+
+<|ref|>text<|/ref|><|det|>[[92, 360, 920, 404]]<|/det|>
+9) p.4 Fig. 1(a): Sphere shaping indeed decreases the maximum energy substantially. However, what is more important is the decrease in average energy. I would recommend indicating the decrease in average energy, and even relating it to the "linear" shaping gain discussed, e.g., in [9].  
+
+<|ref|>text<|/ref|><|det|>[[92, 405, 780, 420]]<|/det|>
+10) p.4 Fig. 1(b): I do not understand what this figure tries to explain, or how Fig. 1(a) relates to this.  
+
+<|ref|>text<|/ref|><|det|>[[92, 421, 750, 436]]<|/det|>
+11) p.4: sentence before (1) and (2): "(...) \(\eta\) increases as the central moment (...) \(\rightarrow \eta\) or \(P_{NLI}\) ?  
+
+<|ref|>text<|/ref|><|det|>[[92, 437, 920, 466]]<|/det|>
+12) p.5: "(...) deviation by the instantaneous power (...) \(\rightarrow\) "(...) deviation of the instantaneous power (...) from average \(< p > = 1\) (...)"?  
+
+<|ref|>text<|/ref|><|det|>[[92, 467, 920, 496]]<|/det|>
+13) p.5: "(...) appears to be more spread around the average (...) \(\rightarrow\) I think this can be related to the difference in their \(\bar{\mu}_{2}\) , right? If so, please refer to these values.  
+
+<|ref|>text<|/ref|><|det|>[[92, 497, 593, 511]]<|/det|>
+14) Figures 1(c-d): What is the shaping rate and the resulting distribution?  
+
+<|ref|>text<|/ref|><|det|>[[92, 512, 920, 556]]<|/det|>
+15) p.6: I do not understand what "With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods in all conditions except for \(R_{Sym} = 88\) GBd in Link D." exactly means.  
+
+<|ref|>text<|/ref|><|det|>[[92, 557, 920, 586]]<|/det|>
+16) p.6: I am not sure how the observation (iii) is explained with "(...) since increasing \(R_{Sym}\) reduces the shaping block duration in absolute time for the same \(n\) ".  
+
+<|ref|>text<|/ref|><|det|>[[92, 587, 920, 616]]<|/det|>
+17) p.7: I understand the observation "Areas far from the circled areas have a negligible impact on \(SNR_{Eff}\) . But I believe this can be related to Figs. 2(a,c) more strongly and quantitatively.  
+
+<|ref|>text<|/ref|><|det|>[[92, 618, 312, 632]]<|/det|>
+18) Fig. 3(a): How many spans?  
+
+<|ref|>text<|/ref|><|det|>[[92, 633, 920, 692]]<|/det|>
+19) p.8: You say "optimal sphere shaping" more than once in the paper. What exactly do you mean by "optimal"? The combination of \(n\) and \(R_{Sym}\) that maximizes the performance? There is an optimal window length \(w\) that maximizes the correlation \(b/w\) the windowed moment and the effective SNR. But I am not sure what you mean by "optimal" sphere shaping. Clarification is needed.  
+
+<|ref|>text<|/ref|><|det|>[[92, 692, 920, 752]]<|/det|>
+20) p.9: I believe the conclusion "In general, \(n^{*}\) increases as \(|D_{Total}|\) (green solid line) increases, (...) is a bit strong to be made from Fig. 5(a). The optimal block length here is mostly constant around 10-20, increasing around 195 THz, but then decreasing again. If the authors can provide some additional explanation for this behaviour, it would make the paper stronger.  
+
+<|ref|>text<|/ref|><|det|>[[92, 753, 920, 797]]<|/det|>
+21) p.9: At which block length Fig. 5(b) is plotted? Also, similar to my previous comment, the conclusion "(...) \(R_{Sym}^{*}\) decreases (...) when \(|D_{Total}|\) increases (...) is a bit strong to be made from Fig. 5(b). Instead of simply saying Figs. 5(a,b) agree with Figs. 2(b,a), possible explanations for the peculiar behaviours in Figs. 5(a,b) should be provided.  
+
+<|ref|>text<|/ref|><|det|>[[92, 798, 490, 813]]<|/det|>
+22) p.9, third line: \(SNR_{Eff}\) of Fig. 4(c), or of Fig. 5(c)?  
+
+<|ref|>text<|/ref|><|det|>[[92, 814, 444, 828]]<|/det|>
+23) p.11: CCDM is not a sphere shaping algorithm.  
+
+<|ref|>text<|/ref|><|det|>[[92, 829, 920, 858]]<|/det|>
+24) Similar gains in data rate/SNR/AIR to the ones that are reported at the end manuscript exist in the literature. A short review of such works would make the paper stronger.  
+
+<|ref|>sub_title<|/ref|><|det|>[[450, 875, 545, 888]]<|/det|>
+## REFERENCES  
+
+<|ref|>text<|/ref|><|det|>[[78, 895, 920, 940]]<|/det|>
+[1] K. Wu, G. Liga, A. Sheikh, F. M. J. Willems, and A. Alvarado, "Temporal energy analysis of symbol sequences for fiber nonlinear interference modelling via energy dispersion index," Feb. 2021. [Online]. Available: https://arxiv.org/abs/2102.124114 [2] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike-Akino, K. Kojima, K. Parsons, and D. S. Millar, "Nonlinearity tolerant LUT-based probabilistic shaping for extended-reach single-span links," IEEE Photon. Technol. Lett., vol. 32, no. 16, pp. 967- 970, 2020.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[75, 70, 920, 99]]<|/det|>
+[3] T. Fehenberger, "On the impact of finite-length probabilistic shaping on fiber nonlinear interference," in Proc. Signal Process. in Photon. Commun. (SPPCom), Washington, D.C., United States, July 2020.  
+
+<|ref|>text<|/ref|><|det|>[[75, 96, 920, 120]]<|/det|>
+[4] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike-Akino, K. Kojima, K. Parsons, and D. S. Millar, "Huffman-coded sphere shaping for extended-reach single-span links," IEEE J. Sel. Topics Quantum Electron., vol. 27, no. 3: 3500215, May-June 2021.  
+
+<|ref|>text<|/ref|><|det|>[[75, 118, 920, 141]]<|/det|>
+[5] S. Civelli, E. Forestieri, and M. Secondini, "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation," in Proc. Eur. Conf. Opt. Commun. (ECOC), Brussels, Belgium, Dec. 2020.  
+
+<|ref|>text<|/ref|><|det|>[[75, 139, 920, 162]]<|/det|>
+[6] T. Fehenberger, A. Alvarado, G. Böcherer, and N. Hanik, "On probabilistic shaping of quadrature amplitude modulation for the nonlinear fiber channel," J. Lightw. Technol., vol. 34, no. 21, pp. 5063- 5073, Nov. 2016.  
+
+<|ref|>text<|/ref|><|det|>[[75, 160, 920, 184]]<|/det|>
+[7] Y. C. Gültekin, A. Alvarado, O. Vassilieva, I. Kim, P. Palacharla, C. M. Okonkwo, and F. M. Willems, "Kurtosis-limited sphere shaping for nonlinear interference noise reduction in optical channels," June 2021. [Online]. Available: https://arxiv.org/abs/2105.14794  
+
+<|ref|>text<|/ref|><|det|>[[75, 183, 920, 206]]<|/det|>
+[8] P. Schulte and F. Steiner, "Divergence-optimal fixed- to- fixed length distribution matching with shell mapping," IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 620- 623, Apr. 2019.  
+
+<|ref|>text<|/ref|><|det|>[[75, 205, 920, 229]]<|/det|>
+[9] Y. C. Gültekin, W. J. van Houtum, A. Koppelaar, and F. M. J. Willems, "Enumerative sphere shaping for wireless communications with short packets," IEEE Trans. Wireless Commun., vol. 19, no. 2, pp. 1098- 1112, Feb. 2020.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 135, 867, 186]]<|/det|>
+This work studies the optimization of the time and bandwidth over which sphere shaping should be performed to minimize the impact of nonlinear effects and maximize performance in optical fiber communication.  
+
+<|ref|>text<|/ref|><|det|>[[119, 198, 686, 214]]<|/det|>
+The work is timely, well written, and reports some novel interesting results.  
+
+<|ref|>text<|/ref|><|det|>[[118, 225, 864, 310]]<|/det|>
+The optimization of the shaping block length in the nonlinear regime has been already investigated in some other papers but, to my knowledge, the simultaneous optimization of block length and symbol rate (i.e., time and bandwidth) is studied here for the first time. Moreover, this study encompasses several scenarios and a wide range of parameters, both numerically and experimentally.  
+
+<|ref|>text<|/ref|><|det|>[[119, 321, 825, 354]]<|/det|>
+The methodology is sound and the methods are described with enough detail. Moreover, the conclusions are supported both by numerical and experimental results.  
+
+<|ref|>text<|/ref|><|det|>[[118, 365, 866, 398]]<|/det|>
+In general, I believe that the paper deserves publication. However, there are some weaknesses and some issues that should be considered and possibly addressed, as detailed below.  
+
+<|ref|>text<|/ref|><|det|>[[117, 434, 878, 590]]<|/det|>
+1) As a general note, while reading the paper, I was expecting to find a theoretical analysis which explains and predicts the results observed numerically and experimentally. This expectation is not fully meet by the paper, which provides only some insight about the observed phenomena and some empirical laws to model them - e.g., (7), (8) and the use of the windowed moments in the EGN model. In fact, there are a number of perturbation models in the literature which could be used to attempt such an analysis, for instance by removing the i.i.d. assumption in the EGN model. An example can be found in Liga et al "Extending Fibre Nonlinear Interference Power Modelling to Account for General Dual-Polarisation 4D Modulation Formats", Entropy, 2020. I recommend discussing this issue and highlighting the empirical nature of the proposed model.  
+
+<|ref|>text<|/ref|><|det|>[[117, 625, 878, 814]]<|/det|>
+2) As a second general comment, the analysis practically neglects the impact of carrier recovery. Indeed, this might be substantial, as a sufficiently fast carrier recovery algorithm can mitigate the impact of nonlinear phase noise and change the dependence of performance on block length (see for instance Civelli et al. "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation", ECOC 2020). With carrier recovery, there might be no advantage in using short (optimized) block lengths. It seems that carrier recovery is not present in the simulations and is quite slow in the experiments (using only 1 pilot every 48 symbols), possibly overestimating the advantage of using a short optimized block length. Please comment about that. As a side note, for the sake of transparency, I am a coauthor of the above mentioned paper. I do not want to push my own work, therefore feel free to decide if the paper is relevant or not to your work and if it deserves to be cited or not. This decision will not affect my final recommendation.  
+
+<|ref|>text<|/ref|><|det|>[[118, 852, 870, 903]]<|/det|>
+3) Abstract: The sentence "In optical fiber, however, the sphere shaping induces Kerr nonlinearity in a peculiar way that makes analysis of transmission performance difficult, potentially lowering the communications capacity" sounds a bit odd. It suggests that it is sphere shaping that makes the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 850, 135]]<|/det|>
+analysis difficult, with Kerr nonlinearity just playing an indirect role. In fact, I think the opposite is true. Moreover, it suggests that the difficulty of the analysis may play a role in the capacity reduction.  
+
+<|ref|>text<|/ref|><|det|>[[118, 171, 876, 275]]<|/det|>
+4) Introduction: The idea of using sphere shaping on an optimized block length to minimize the impact of nonlinear interference was first proposed in [10] and then further investigated by some of the same authors in Geller et al. "A Shaping Algorithm for Mitigating Inter-Channel Nonlinear Phase-Noise in Nonlinear Fiber Systems", JLT 2016. Your work (as well as some other recent works) heavily relies on this idea, extending the optimization to the frequency domain. I think that it would be fair to better acknowledge this fact and the role of these two papers.  
+
+<|ref|>text<|/ref|><|det|>[[118, 311, 880, 431]]<|/det|>
+5) Page 3, Results: you state that "The symmetry of the probability allows for legitimate analysis with only positive amplitudes, and hence we omit the sign throughout this article". I guess that the presence of a random sign with uniform distribution is implicitly assumed in the paper and neglected only in the description, whereas it is included both in the simulations and in the experiments. Is that correct? I am not sure that transmitting only positive amplitudes would indeed give the same results. I think that two consecutive symbols with the same sign or with opposite signs induce a fundamentally different response in the nonlinear channel. Please be more explicit.  
+
+<|ref|>text<|/ref|><|det|>[[118, 466, 848, 535]]<|/det|>
+6) Page 3, Results: the sentence "one dual-polarization symbol that maintains the value over a period of T_sym (s) and can change the value at a rate of R_sym=1/T_sym (Bd)", suggests a rectangular pulse shape. I suggest revising the sentence. Moreover, the indication of the units of measure in this way is neither standard nor required.  
+
+<|ref|>text<|/ref|><|det|>[[118, 571, 878, 623]]<|/det|>
+7) Page 5, 4 lines before the end of the subsection: The sentence "As w increases, \mu - 2 remains the same for all types of i.i.d. symbols" is misleading. It is not "the same" for all types of symbols but, rather, it remains "constant" as w increases for a given modulation (type of symbols).  
+
+<|ref|>text<|/ref|><|det|>[[118, 659, 861, 745]]<|/det|>
+8) Equation (3): for sphere shaping, the transmitted signal is a ciclostationary process with period equal to the length of the shaped blocks (unless a uniform random delay is assumed). Therefore, I guess that the windowed moments depend on the time at which the window is centered, unless they are averaged over it. Please comment about that and explain if and how you have considered this time dependence in your analysis.  
+
+<|ref|>text<|/ref|><|det|>[[118, 782, 853, 832]]<|/det|>
+9) Page 5, Section "Optimization of Sphere Shaping in the Time-Frequency Plane": Please specify what QAM format is considered in this section. I guess from the given data that it is 16QAM (as in the previous section). Yet, an explicit mention here would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[118, 870, 874, 904]]<|/det|>
+10) Page 6: You state that "With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 83, 875, 357]]<|/det|>
+in all conditions except for R_sym=88 GBd in Link D" and that "like the i.i.d. uniform QAM, the optimal R_sym for i.i.d. shaping to maximize SNR_eff (yellow stars) decreases as the total net dispersion D_Total increases". I think that this discussion is a bit weak and that the analogy with the symbol rate optimization for uniform i.i.d. QAM is misleading for two reasons: 1) The dependence of SNR_eff on symbol rate for the uniform i.i.d. QAM case vanishes for Gaussian symbols (the EGN and GN model converge to the same equations, see for instance [23]). Therefore, I expect a very little dependence on the symbol rate for shaped i.i.d. symbols, as they approximate i.i.d. Gaussian symbols. In fact, the dependence on symbol rate in Fig. 2a for the case \(n = 5120\) is very weak (and would be probably even weaker for shaped 64- QAM or 256- QAM) and the peak SNR is not so different from the surrounding values in the simulation range; 2) assuming that \(n = 5120\) represents the i.i.d. case from a dispersion perspective is an approximation whose accuracy decreases as the symbol rate increases; therefore, it cannot be used when explaining the (weak) dependence of SNR_eff on symbol rate. In fact, for \(n = 5120\) , this dependence becomes a little more relevant exactly where the approximation becomes looser. In my opinion, it seems more likely that the dependence of SNR_eff on symbol rate is mainly explained, even for \(n = 5120\) , by the same reason that explains it for lower n.  
+
+<|ref|>text<|/ref|><|det|>[[118, 395, 850, 463]]<|/det|>
+11) Page 6, you state that "the optimal R_sym is higher for finite-length shaping than for i.i.d. shaping (compare the yellow and red stars), since increasing R_sym reduces the shaping block duration in absolute time for the same n". Following up on the previous comment, can we better (and more simply) say that the optimal symbol rate increases when n decreases?  
+
+<|ref|>text<|/ref|><|det|>[[118, 500, 873, 587]]<|/det|>
+12) Still on the same matter: you refer many times in the results to the i.i.d. case, considering \(n = 5120\) as an approximation to it. Why not directly simulating the true i.i.d. case? It should be sufficient to draw i.i.d. QAM symbols from a Maxwell-Boltzmann distribution. That would be a more meaningful benchmark against which to compare the results for finite block lengths and, for instance, to compute \(\backslash \mathrm{eta}^{\wedge}\backslash \mathrm{infly}\) in (6).  
+
+<|ref|>text<|/ref|><|det|>[[118, 622, 877, 725]]<|/det|>
+13) Page 8, Demonstration of Optimal Sphere Shaping through Full C-Band Transmission Experiment: The experiment is performed on a link that is substantially different from the 4 links considered in the simulation. Why? In particular, each channel sees a different dispersion profile, making the interpretation of the results harder. Moreover, the important dispersion-unmanaged case is not considered in the experiments. Is there a particular reason for this choice? Is it instrumental to highlight some effect or behavior? Please explain and motivate this choice.  
+
+<|ref|>text<|/ref|><|det|>[[118, 761, 878, 900]]<|/det|>
+14) Page 11, Methods: you state that "The digital sphere shaping encoder is implemented by enumerative sphere shaping (ESS) for \(n = 5, 10, 20, 40, 80\) , and by constant composition distribution matching (CCDM) for \(N = 320, 1280, 5120\) ." and "...consume 0.583, 0.346, 0.209, 0.137, 0.099, 0.025, 0.009, 0.002 dB more average symbol energy than ideal shaping, respectively". Strictly speaking, CCDM does not implement sphere shaping (as all constant composition sequences lie on the surface of the sphere, partially covering it) and, in fact, its energy loss is higher than that of ESS for the same blocklength. Of course, they both asymptotically converge to i.i.d. Maxwell-Boltzmann symbols, but as you are studying how performance changes with block length, you cannot assume that such an
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 874, 152]]<|/det|>
+asymptotic regime has been achieved. In fact, I am also a bit surprised that CCDM with \(n = 320\) performs significantly better than ESS with \(n = 80\) (see, e.g., the comparison in [7]). Please comment about that and double check the numbers (I have not verified them, it is just a feeling and I might be wrong).  
+
+<|ref|>text<|/ref|><|det|>[[118, 190, 246, 204]]<|/det|>
+Marco Secondini
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[190, 90, 808, 157]]<|/det|>
+# Response to Reviewers' Comments on NCOMM-21-15776 "Shaping Lightwaves in Time and Frequency for Optical Fiber Communication"  
+
+<|ref|>text<|/ref|><|det|>[[140, 175, 857, 193]]<|/det|>
+Junho Cho, Xi Chen, Greg Raybon, Di Che, Ellsworth Burrows, Samuel Olsson, and Robert Tkach  
+
+<|ref|>text<|/ref|><|det|>[[115, 253, 237, 269]]<|/det|>
+Dear Reviewers,  
+
+<|ref|>text<|/ref|><|det|>[[115, 278, 877, 389]]<|/det|>
+We sincerely thank all reviewers for carefully reading and commenting on the paper. We addressed in the revised manuscript all the suggestions and comments, which we believe has significantly improved the paper. The changes made in the revised manuscript are highlighted in yellow. Please find our point- by- point response to the comments below, where the reviewers' comments are reproduced verbatim in blue italic type, and our response is written in black roman type. Note that the figure numbers, table numbers and reference numbers below indicate those of the revised manuscript, unless specified otherwise.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 449, 320, 467]]<|/det|>
+## Response to Reviewer 1:  
+
+<|ref|>text<|/ref|><|det|>[[114, 476, 880, 715]]<|/det|>
+The authors of the paper investigate constellation shaping (sphere shaping in particular) for optical communication systems. They focus on the effect of (1) the block length of sphere shaping and (2) the symbol rate of the signaling on the amount of nonlinear interference (NLI) generated during propagation. The dependency of NLI on the shaping block length has been investigate in the literature. The authors propose a new metric (windowed central moment) to study this dependency. This metric measures the high- order deviations of the energy of the optical field from its average over a finite time interval. They claim that properly selecting the width of this window is important to properly compare different shaping schemes from NLI generation perspective and to predict their performance. Furthermore, they claim that the nonlinearities depend on the energy structure of the optical field in absolute time rather than in number of symbols. This leads to a complex relation between the shaping block length, symbol rate and NLI. Using the values of the window length at which the correlation between the windowed moment and the effective SNR is maximized, authors claim that the EGN model predicts the effective SNR (using windowed moments) more accurately than its traditional form (using regular moments) (Fig. 3(b)).  
+
+<|ref|>text<|/ref|><|det|>[[114, 722, 882, 888]]<|/det|>
+The paper is well- written, findings are interesting, the topic is timely/relevant, and the explanation of the experimental setup is very clear and detailed. I believe the paper can be accepted after a round of revision/polishing. The structure of the paper is different than I am used to from other journals in the literature and this structure may be leading to a different reading experience than I am used to. However I must say that the introduction can be improved by stating (1) the problem, (2) what is done, and (3) what is found more clearly with short and direct sentences. At some points, I had trouble following the train of reasoning. I believe the findings of the paper are quite interesting, and they should be stated more clearly to increase their impact. In the following, I will list my comments which I hope would help the authors to improve the presentation in the paper.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 858, 163]]<|/det|>
+Response: Thank you for all the constructive comments here and in what follows. We have tried to address your concerns by improving the presentation as much as we can. We hope that the ambiguities found in the previous manuscript have completely been removed in the revised manuscript and the content of the paper is now much clearer.  
+
+<|ref|>text<|/ref|><|det|>[[114, 193, 879, 322]]<|/det|>
+Firstly, I find it a bit complicated to understand the fundamental motivation and findings of the paper from the abstract and the introduction. I suggest a clearer statement of the fact that (1) the paper investigates the optimal combination of the shaping block length and the symbol rate that maximizes the performance, (2) this is achieved using a new metric—similar to a metric proposed in [1]—that has high correlation with the effective SNR. At some parts of the paper, it is implied that a NLI- optimized sphere shaping approach is proposed. However, as far as I understand, only the optimum system parameters are investigated (which is also a valuable contribution) for regular sphere shaping.  
+
+<|ref|>text<|/ref|><|det|>[[113, 337, 883, 673]]<|/det|>
+Response: As the reviewer suggested, we made the motivation and contribution of this paper clearer in the abstract and the introduction. The changed sentence in the abstract is "In this article, we show that the impact of sphere shaping on Kerr nonlinearity varies with chromatic dispersion, shaping block length and symbol rate, and that this impact can be predicted using a novel statistical measure of light energy." In the abstract, a longer explanation on the motivation and contribution of this work was avoided to meet the suggested length of 150 words, but we tried to make it as clear as possible within the given length. In the introduction, several sentences have been changed to clarify the motivation and contribution of this paper, as highlighted in yellow in the upper half of Page 3. In particular, we emphasized that the previous studies have focused only on the block length optimization, neglecting the influence of the symbol rate and chromatic dispersion. It is only by knowing the findings of this work that the results can be consistently explained between several independently performed experiments that used different settings. To clarify the contribution of this work, we added "While the previous works<sup>8,11,12,16,23</sup> optimized only \(n\) to observe some gains in \(SNR_{Eff}\) and NDR over specific links (e.g., for single- span links), joint optimization of \(n\) and \(R_{Sym}\) in this work produces significantly larger gains and allows these gains to be achieved over a much wider variety of links." at the end of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section. We also made it clear throughout the abstract, introduction, and main text that it is not the shaping algorithm that we optimize to minimize the NLI, but the parameters of the regular sphere shaping.  
+
+<|ref|>text<|/ref|><|det|>[[114, 679, 880, 807]]<|/det|>
+In the traditional EGN model, higher order moments of the channel input distribution are used in the expressions that compute the NLI variance. Here, they are replaced with their windowed versions and the reported results match the actual performance more accurately than the original EGN model. Can the authors comment on the use of the windowed moments in the EGN model? Is there a theoretical background for this, or is it just heuristic? Can this approach be used to propose an enhanced EGN model that works for finite- length shaping schemes and captures the temporal structure of the shaped waveforms in the analytical model?  
+
+<|ref|>text<|/ref|><|det|>[[115, 823, 864, 895]]<|/det|>
+Response: Thanks for this valuable comment. At a recent conference, we learned from an author of Ref. 26 that an extended EGN model was published last year to deal with structures (or correlations) present in four amplitudes (i.e., in one symbol period). A mathematically unequivocal approach to modeling the propagation of structured lightwaves would be to further extend Ref. 26 (see also Ref. 27)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 880, 198]]<|/det|>
+for structures spanning more than one symbol period. However, given the enormous complexity of the mathematical expansion to account for only four correlated amplitudes, it can easily be inferred that it is mathematically daunting to find an accurate analytical model to handle tens to thousands of correlated amplitudes that we cover in this paper. Also, as seen from Refs. 26- 27, the resulting equations of the extended EGN model for many correlated amplitudes may have too many complex terms to be practically useful.  
+
+<|ref|>text<|/ref|><|det|>[[114, 207, 881, 335]]<|/det|>
+In this work, we allowed for model inaccuracies caused by the assumptions on i.i.d. amplitudes, and attempted to solve the problem with manageable complexity by substituting the windowed moments that can characterize structured lightwaves statistically and quantitatively into the traditional EGN model, whether the structure is short or long. We have clarified this in the revised manuscript with properly added citations, as found in the first paragraph of Page 9. It may be possible to apply our approach to the extended EGN model of Ref. 26 to see if the prediction accuracy is even more improved, which we would like to leave for future work.  
+
+<|ref|>text<|/ref|><|det|>[[114, 365, 882, 438]]<|/det|>
+The improvement in rate (or in reach) due to shaping (and sphere shaping in particular) has been demonstrated in the literature repeatedly. The paper provides an optimization of shaping over (i) block length and (ii) symbol rate. There are also other works in the literature which discusses the effect of block length and symbol rate on performance. They should at least be cited, e.g., [2], [3], [4], [5], etc.  
+
+<|ref|>text<|/ref|><|det|>[[114, 453, 874, 618]]<|/det|>
+Response: We thank you for letting us reference the missing important prior studies. By citing Refs. 15- 17, 23 (which are [2]- [5] in your reference numbers) at the end of the second and third paragraphs of the revised introduction, we noted that there are several recent works addressing the effect of the shaping block length. However, those papers are cited only in the context of block length, since there are no existing studies on the effect of the symbol rate on Kerr nonlinearity. For example, in Refs. 8, 11, 15- 17, only single symbol rates of 45 GBd, 100 GBd, 56 GBd, 56 GBd, and 50 GBd are used, respectively. In Ref. 23, signal is modulated with two different symbol rates of 42 GBd and 64 GBd, but from this no systematic analysis of the effect of symbol rate can be made. We therefore cited the prior studies in the context of the shaping block length only.  
+
+<|ref|>text<|/ref|><|det|>[[114, 648, 880, 833]]<|/det|>
+In p.5, it is stated that refs. 7 and 15 show that if n is small as in Fig. 1(c), sphere shaping reduces NLI. Firstly, ref. 15 is not about sphere shaping, but about constant composition sequences. Secondly, as previously observed in [6] and re- stated in ref. 7, (AWGN- optimal) shaping leads to increased NLI (decreased effective SNR). This is attributed to the increased kurtosis in [6], [7]. Therefore, I am not sure I see the contradiction that you mention here. Furthermore, the explanation concerning Fig. 1(d) is not clear. Are we focusing on the effect of n on NLI here or the importance of selecting w to predict the performance? Because regardless of w, smaller n implies smaller \(\bar{\mu}_{2}\) in Fig. 1(d) (for most of the region). I would suggest a careful revision of the paragraph in p.5 that starts with “Fig. 1(c)” (which I believe should be “Figure 1(c)”) with clear explanations (of the claims and figures) and as much connections to the existing literature as possible. I believe this paragraph is extremely important to motivate your work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 848, 872, 904]]<|/det|>
+Response: As the reviewer pointed out, Ref. 22 (Ref. 15 in our previous manuscript) deals with constant composition sequences. As the block length increases, the constant composition sequences can approximately realize sphere shaping with gradually decreasing rate loss, as a consequence of the sphere
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 881, 200]]<|/det|>
+hardening phenomenon. However, we admit that for short block lengths covered in Pages 5- 6, constant composition sequences produce a large difference from sphere shaping. Therefore, we replaced Ref. 22 with proper citations in the sentence you referred above, and added citations to Refs. 17 and 34 (which are [6] and [7] in your reference numbers). We also added "Note that CCDM can only approximately realize sphere shaping for finite block lengths, but it converges to ideal sphere shaping with a decreasing approximation error as the block length increases" to the first paragraph in the Methods section.  
+
+<|ref|>text<|/ref|><|det|>[[114, 207, 874, 374]]<|/det|>
+As far as the "contradiction" is concerned, it is observed between the analytical model for i.i.d. symbols and the empirical results obtained using sphere shaping with short block lengths; namely, if we calculate the statistical moments of short sphere- shaped symbols with \(w = 1\) as conventionally done, it is much larger than that of the unshaped symbols (see, e.g., Fig. 1(d)), and if we use the EGN model neglecting the required i.i.d. properties of symbols, these large statistical moments imply that NLI of sphere- shaped symbols is much greater than that of unshaped symbols. On the other hand, the NLI that is empirically obtained with short sphere shaping can be much smaller than that of the unshaped symbols (see, e.g., Fig. 8 of Ref. 8). In the paragraph that you mentioned, we have tried to convey the meaning of "contradiction" more clearly by rewriting several sentences, with citations to the associated prior works.  
+
+<|ref|>text<|/ref|><|det|>[[115, 381, 880, 417]]<|/det|>
+We have also paraphrased several sentences in this paragraph to avoid duplication of content or confusion between what is presented here and what is presented later sections.  
+
+<|ref|>text<|/ref|><|det|>[[115, 424, 876, 498]]<|/det|>
+Regarding the text citation of figures, some journals such as IEEE journals use the abbreviation "Fig." even when it begins a sentence. However, for Nature Communications, we were not able to find whether "Fig. 1(c)" or "Figure 1(c)" conforms to the editorial style, so will check with the journal's editorial team during the proofreading stage (if the paper is accepted).  
+
+<|ref|>text<|/ref|><|det|>[[114, 527, 880, 675]]<|/det|>
+As far as I know, the effective SNR in EGN model depends on the average power of the input as well as its regular fourth and sixth order moments. In p.8, how do you use the EGN model with \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) to compute the effective SNR? Figure 3(b) reports that this computation (EGN with windowed moments) is far more accurate than the regular EGN model in predicting the effective SNR. If this is the case, would you claim that you're proposing an enhanced EGN model? Furthermore, is it possible to obtain a further more accurate relation if (somehow) some other higher- order windowed moments are included in the model? Also, when you say "optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) " here, do you mean their values computed with the optimum window length?  
+
+<|ref|>text<|/ref|><|det|>[[114, 690, 880, 874]]<|/det|>
+Response: You are correct that the EGN model requires the average power as well as the fourth and sixth standardized moments. To clarify this, we corrected a phrase in Page 9 as "Plugging the average symbol power \((||x||^2)\) and the windowed central moments \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) obtained with \(w_{SPM}^*\) and \(w_{XPM}^*\) into a state- of- the- art analytic model known as the enhanced Gaussian noise (EGN) model". We also added a small subsection "EGN simulation" in the Method section to explain how to convert the windowed central moments into the fourth and sixth standardized moments. We could approach mathematically as to how this conversion can be derived and why this conversion is necessary; however, we believe that this mathematical detail is beyond the scope of Nature Communications articles and that the physical rationale as addressed in the current article is more significant and relevant to this journal. Therefore, we would like to omit the mathematical details in this article.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 880, 255]]<|/det|>
+As mentioned in an earlier response above, long mathematical derivations are needed to extend the EGN model to account for only 4 correlated amplitudes (see Ref. 26), and improving the EGN model to account for larger structures seems mathematically daunting. In this article, we are not proposing an enhanced EGN model; rather, while allowing for model mismatch by using the classical EGN model, we improve the accuracy of evaluating structured lightwaves by replacing parameters of the EGN model with improved ones (i.e., by replacing \(\mu_{2}\) and \(\mu_{3}\) with optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) ). It is uncertain whether the existing approach to improving the model and our approach to using improved parameters can converge in the future, but regardless of the approach researchers take, we believe that our work will inspire people who have worked in this field.  
+
+<|ref|>text<|/ref|><|det|>[[115, 262, 875, 320]]<|/det|>
+Regarding your last point, you're right that "optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) " means that the values of \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) are obtained with the optimum window length. We clarified this by rephrasing it as "by replacing \(\mu_{2}\) and \(\mu_{3}\) with optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) (i.e., obtained with \(W_{SPM}^{*}\) and \(W_{XPM}^{*}\) ).  
+
+<|ref|>text<|/ref|><|det|>[[115, 349, 441, 366]]<|/det|>
+In the following, I list some minor comments.  
+
+<|ref|>text<|/ref|><|det|>[[114, 375, 874, 410]]<|/det|>
+1) \(p.2\) paragraph 1: "ever-increasing demand for communication capacity" \(\rightarrow\) "ever-increasing demand for higher data rates"  
+
+<|ref|>text<|/ref|><|det|>[[115, 426, 544, 444]]<|/det|>
+Response: We have modified this phrase as you suggested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 474, 860, 511]]<|/det|>
+2) \(p.2\) paragraph 2: Refs. 8 and 9 are given as references for sphere shaping, but they are not. I would suggest instead [8],[7].  
+
+<|ref|>text<|/ref|><|det|>[[115, 526, 880, 581]]<|/det|>
+Response: We believe that this paragraph should provide readers with extensive prior work on sphere shaping. Since Refs. 9 and 10 realize sphere shaping for long block lengths, we have left citations to Refs. 9 and 10 and added more citations as you suggested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 612, 833, 648]]<|/det|>
+3) \(p.2\) paragraph 2: "block of amplitudes" is rather vague, it should be related to "communication (QAM) symbols" (mentioned in the previous paragraph) better.  
+
+<|ref|>text<|/ref|><|det|>[[115, 664, 544, 682]]<|/det|>
+Response: We have modified this phrase as you suggested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 713, 857, 749]]<|/det|>
+4) \(p.3\) paragraph 2: 4 consecutive amplitudes \(\rightarrow\) one DP symbol: This is discussed extensively in [4], I would suggest referencing it.  
+
+<|ref|>text<|/ref|><|det|>[[115, 765, 566, 783]]<|/det|>
+Response: We have added citations to related previous works.  
+
+<|ref|>text<|/ref|><|det|>[[115, 814, 305, 832]]<|/det|>
+5) \(p.3\) : \(\| \cdot \|^{2}\) is not defined  
+
+<|ref|>text<|/ref|><|det|>[[115, 848, 595, 865]]<|/det|>
+Response: Its definition has been added to the revised manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 797, 108]]<|/det|>
+6) p.3: Explanation of sphere shaping can be connected to the related literature more strongly  
+
+<|ref|>text<|/ref|><|det|>[[115, 122, 879, 178]]<|/det|>
+Response: We have improved the explanation of sphere shaping in various places with added citations to previous works. Specifically, we added explanations of sphere shaping in relation to existing works in the second paragraph of Page 2, at the end of Page 4, and in the "Sphere shaping" section in Page 13.  
+
+<|ref|>text<|/ref|><|det|>[[115, 200, 879, 293]]<|/det|>
+7) p.3 paragraph 1: "(...) if every shaped block is plotted as a point in 4n-dimensional signal space, with the i-th amplitude being the position of the point on the i-th coordinate axis, the points uniformly fill a 4n-dimensional (hyper-)sphere (...) \(\rightarrow\) Since the amplitudes are drawn from a discrete set, this is not correct. The points are located in a 4n-dimensional spherical region of a 4n-dimensional rectangular lattice. I would avoid the use of the phrase "uniformly fill".  
+
+<|ref|>text<|/ref|><|det|>[[115, 308, 875, 345]]<|/det|>
+Response: Thanks for pointing out the inaccuracies in the description. We have improved the description of sphere shaping as follows: "the points are distributed uniformly over a set of 4n- dimensional square  
+
+<|ref|>text<|/ref|><|det|>[[115, 348, 864, 400]]<|/det|>
+lattice points that lie on or contained in a 4n- dimensional (hyper-) sphere of radius \(\sqrt{E_{\mathrm{shaped}}^{*}}\) (due to the symmetry by equiprobable signs)"  
+
+<|ref|>text<|/ref|><|det|>[[115, 429, 859, 466]]<|/det|>
+8) p.4: I would recommend the use of R instead of H for the information rate, since H is almost always used to denote the entropy  
+
+<|ref|>text<|/ref|><|det|>[[115, 482, 667, 500]]<|/det|>
+Response: We have changed the notation from \(H\) to \(R\) throughout the paper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 530, 881, 586]]<|/det|>
+9) p.4 Fig. 1(a): Sphere shaping indeed decreases the maximum energy substantially. However, what is more important is the decrease in average energy. I would recommend indicating the decrease in average energy, and even relating it to the "linear" shaping gain discussed, e.g., in [9].  
+
+<|ref|>text<|/ref|><|det|>[[115, 600, 882, 824]]<|/det|>
+Response: To address this concern, we added two sentences to Page 4 with proper citations as "The average energy of \(\pmb{a}\) to achieve \(R\) with sphere shaping decreases with increasing block length \(4n\) (see, e.g., \(^{8,16}\) ), achieving a theoretical minimum average energy as \(n\rightarrow \infty\) . We refer to the reduction in average energy of \(\pmb{a}\) by shaping as the fundamental shaping efficiency in this article." The fundamental shaping efficiency is mentioned in Page 7 when explaining the NGMI of Fig. 2(b), and in Page 11 when analyzing Figs. 5(c) and (d). The reduction in average energy by sphere shaping is important (as is well known for communications over linear transmission media), but for nonlinear optical fiber communications that is the focus of this study, we would like to stress that the reduction in maximum energy (more precisely, the reduction in high- order statistical moments) is just as important as the average energy. Compared to infinite- length sphere shaping, the sphere shaping with \(n = 5\) can produce 0.9 to 1.0 dB higher \(SNR_{Eff}\) after nonlinear fiber propagation (see Fig. 2(a)) while achieving 0.6 dB lower fundamental shaping efficiency.  
+
+<|ref|>text<|/ref|><|det|>[[115, 855, 852, 873]]<|/det|>
+10) p.4 Fig. 1(b): I do not understand what this figure tries to explain, or how Fig. 1(a) relates to this.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 879, 370]]<|/det|>
+Response: Fig. 1(a) shows only the probability distribution of total energy in unshaped versus shaped blocks, without the context of communications over a physical medium. Fig. 1(b) shows in the context of communications over a physical medium, in which rectangular regions of the time- frequency plane such probability distributions can be measured. The size of the rectangular regions is determined by \(R_{Sym}\) and \(n\) , and the distribution within this region is determined by \(n\) for the given \(\mathcal{A}\) and \(R\) . We believe that how Fig. 1(a) relates to Fig. 1(b) is expressed in the current statement, which reads “In a densely packed WDM system with identical channel configurations, such probabilistic energy distributions as in Fig. 1(a) are observed within each rectangular block that divides lightwaves in the time- frequency plane as shown in Fig. 1(b), where the width and height of the block are determined by both \(R_{Sym}\) and \(n\) .” To further clarify why this relation matters for communications, we added “While the shaping block length \(n\) or the distribution of energy (cf. Fig. 1(a)) has been optimized in existing studies \(^{8,11 - 12,15 - 17,23}\) to mitigate nonlinear interference (NLI), the spectro- temporal region where the distribution is found (cf. Fig. 1(b)) has never been noted previously. In the following sections, we will see that it is the distribution of light energy in all aspects of probability, time, and frequency that determines the manifestation of Kerr nonlinearity as NLI, and thus \(R_{Sym}\) and \(n\) must be controlled simultaneously to minimize NLI.”  
+
+<|ref|>text<|/ref|><|det|>[[115, 401, 816, 421]]<|/det|>
+11) p.4: sentence before (1) and (2): “(...) \(\eta\) increases as the central moment (...)” \(\rightarrow \eta\) or \(P_{NLI}\) ?  
+
+<|ref|>text<|/ref|><|det|>[[113, 436, 872, 474]]<|/det|>
+Response: We believe that the current statement is correct as is. \(\eta\) increases as the central moment \(\mu_{n}\) of \(\mathcal{P}\) increases. By (1), \(\langle P_{NLI} \rangle\) also increases as \(\mu_{n}\) increases.  
+
+<|ref|>text<|/ref|><|det|>[[113, 503, 860, 541]]<|/det|>
+12) p.5: “(...) deviation by the instantaneous power (...)” \(\rightarrow\) “(...) deviation of the instantaneous power (...) from average \(< \mathcal{P} > = 1\) (...)”?  
+
+<|ref|>text<|/ref|><|det|>[[115, 555, 710, 574]]<|/det|>
+Response: We corrected this phrase as “deviation using the instantaneous power”.  
+
+<|ref|>text<|/ref|><|det|>[[115, 604, 852, 641]]<|/det|>
+13) p.5: “(...) appears to be more spread around the average (...)” \(\rightarrow\) I think this can be related to the difference in their \(\bar{\mu}_{2}\) , right? If so, please refer to these values.  
+
+<|ref|>text<|/ref|><|det|>[[115, 656, 879, 694]]<|/det|>
+Response: We rephrased this sentence as “sphere shaping appears to make the instantaneous power more spread out around the average power (note the increase of \(\mu_{2}\) from 0.32 to 0.687 after sphere shaping)”.  
+
+<|ref|>text<|/ref|><|det|>[[115, 723, 660, 742]]<|/det|>
+14) Figures 1(c-d): What is the shaping rate and the resulting distribution?  
+
+<|ref|>text<|/ref|><|det|>[[115, 757, 828, 813]]<|/det|>
+Response: We added “All shaped symbols in Fig. 1 achieve \(R = 6.4\) ” to the caption of Fig. 1. The probability distribution of energy in a block of 5 symbols is shown in Fig. 1(a). The probability distribution of normalized energy in each symbol is shown in Fig. 1(c).  
+
+<|ref|>text<|/ref|><|det|>[[115, 843, 849, 900]]<|/det|>
+15) p.6: I do not understand what “With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods in all conditions except for \(R_{Sym} = 88\) GBd in Link D.” exactly means.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 870, 164]]<|/det|>
+Response: It meant that the channel memory due to chromatic dispersion is shorter than the shaping block length, except for \(R_{Sym} = 88\) GBd in Link D. However, in the revised manuscript, we additionally provided results obtained with i.i.d. random symbols drawn from a Maxwell- Boltzmann distribution. Accordingly, the sentence you mentioned has been deleted.  
+
+<|ref|>text<|/ref|><|det|>[[114, 194, 874, 234]]<|/det|>
+16) p.6: I am not sure how the observation (iii) is explained with "(...) since increasing \(R_{Sym}\) reduces the shaping block duration in absolute time for the same \(n\) ".  
+
+<|ref|>text<|/ref|><|det|>[[114, 249, 860, 287]]<|/det|>
+Response: To address your concern (and as Reviewer 2 suggested), this sentence has been modified to "(iii) the optimal \(R_{Sym}\) increases as \(n\) decreases (compare, e.g., the yellow and red stars)."  
+
+<|ref|>text<|/ref|><|det|>[[114, 318, 839, 355]]<|/det|>
+17) p.7: I understand the observation "Areas far from the circled areas have a negligible impact on \(SNR_{Eff}\) . But I believe this can be related to Figs. 2(a,c) more strongly and quantitatively.  
+
+<|ref|>text<|/ref|><|det|>[[114, 371, 880, 507]]<|/det|>
+Response: We are not sure if you wanted to refer to Figs. 2(c- 5) and (c- 8). It is apparent from these figures that the low symbol rate region in Fig. 2(c- 5) and high symbol rate region in Fig. 2(c- 8) have a negligible impact on \(SNR_{Eff}\) . Taking \(\eta^{\infty}\) of these figures for i.i.d. shaping as baselines, \(\Delta \eta\) in Figs. 2(c- 6) and (c- 9) is added as offsets for finite- length shaping, and here we want to say that how significant the effect of such offsets \(\Delta \eta\) on \(SNR_{Eff}\) is determined by the baseline \(\eta^{\infty}\) . The meaning of the sentence is apparent from its preceding sentence, which reads "the influence of \(\Delta \eta_{SPM}\) and \(\Delta \eta_{XPM}\) on \(SNR_{Eff}\) is prominent only near the red circled areas, where their base coefficients \(\eta_{SPM}^{\infty}\) and \(\eta_{XPM}^{\infty}\) are large."  
+
+<|ref|>text<|/ref|><|det|>[[116, 537, 352, 555]]<|/det|>
+18) Fig. 3(a): How many spans?  
+
+<|ref|>text<|/ref|><|det|>[[114, 570, 857, 607]]<|/det|>
+Response: Thanks for pointing out the missing information. We added "at 240 spans in Link D" to the figure description in the text and "obtained at 240 spans in Link D" to the figure caption.  
+
+<|ref|>text<|/ref|><|det|>[[114, 637, 865, 714]]<|/det|>
+19) p.8: You say "optimal sphere shaping" more than once in the paper. What exactly do you mean by "optimal"? The combination of \(n\) and \(R_{Sym}\) that maximizes the performance? There is an optimal window length \(w\) that maximizes the correlation \(b/w\) the windowed moment and the effective SNR. But I am not sure what you mean by "optimal" sphere shaping. Clarification is needed.  
+
+<|ref|>text<|/ref|><|det|>[[114, 729, 863, 820]]<|/det|>
+Response: We made sure that the meaning of the optimality is clear in all places. In the fist sentence of the subsection "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" in Page 9, we clarified that optimal sphere shaping implies maximizing NDR. We also clarified this in Fig. 5 caption. In all other places, we believe that the meaning of optimality for sphere shaping or system parameters is explicitly stated or apparent from its surrounding sentences.  
+
+<|ref|>text<|/ref|><|det|>[[114, 851, 870, 888]]<|/det|>
+20) p.9: I believe the conclusion "In general, \(n^*\) increases as \(|D_{Total}|\) (green solid line) increases, (...) is a bit strong to be made from Fig. 5(a). The optimal block length here is mostly constant around 10-20,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 830, 126]]<|/det|>
+increasing around 195 THz, but then decreasing again. If the authors can provide some additional explanation for this behaviour, it would make the paper stronger.  
+
+<|ref|>text<|/ref|><|det|>[[113, 140, 872, 435]]<|/det|>
+Response: We relaxed the conclusion by saying "In general, \(n^{*}\) tends to increase as \(|D_{Total}|\) (green solid line) increases." We attribute the unsmooth curves and deviation from theoretic prediction in Fig. 5 mostly to the difficulty of creating an ideal experimental setup with limited equipment. In particular, the recirculating loop that is the only experimental method of long- haul transmission involves many variations of real- world components. The most important challenge for us was minimizing the optical power excursion across the C- band using the DGEs that have only 0.1 dB nominal attenuation granularity. In true straight- line long- haul systems, the gain tilts and ripples of inline EDFAs can nearly perfectly be flattened by passive optical attenuators with continuous attenuation profiles that were predetermined for a fixed total optical power. On the other hand, in the recirculating loops operated with various total optical powers, the gain tilts and ripples should be flattened by the DGEs with coarse attenuation granularity. A small residual power excursion due to this limitation increases as it accumulates over the number of loops, leading to deviations from the ideal flat optical power spectral density. To address this, we added "The power excursions due to experimental constraints (e.g., a small power excursion caused by coarse attenuation granularity of the DGEs results in an increasing power excursion as the number of loops increases) are considered to be the most important contributor to discrepancy in validation of theory" at the end of Page 10.  
+
+<|ref|>text<|/ref|><|det|>[[115, 464, 878, 540]]<|/det|>
+21) p.9: At which block length Fig. 5(b) is plotted? Also, similar to my previous comment, the conclusion "(...) \(R_{Sym}\) decreases (...) when \(|D_{Total}|\) increases (...) is a bit strong to be made from Fig. 5(b). Instead of simply saying Figs. 5(a,b) agree with Figs. 2(b,a), possible explanations for the peculiar behaviours in Figs. 5(a,b) should be provided.  
+
+<|ref|>text<|/ref|><|det|>[[114, 555, 870, 673]]<|/det|>
+Response: For clarification, we combined the descriptions of Figs. 5(a) and (b) in the text as "Figs. 5(a) and (b) show, respectively, the optimal \(n^{*}\) and \(R_{Sym}^{*}\) that jointly maximize NDR in each channel," and added "for maximum NDR" to the captions of Figs. 5(a) and (b). We also relaxed the conclusion by saying " \(R_{Sym}^{*}\) tends to decrease." The non- smooth change of optimal parameters is attributed to experimental limitations, and the added sentence for Fig. 5(a) mentioned above also provides an explanation for Fig. 5(b).  
+
+<|ref|>text<|/ref|><|det|>[[115, 704, 520, 723]]<|/det|>
+22) p.9, third line: \(SNR_{Eff}\) of Fig. 4(c), or of Fig. 5(c)?  
+
+<|ref|>text<|/ref|><|det|>[[115, 740, 366, 756]]<|/det|>
+Response: We corrected this typo.  
+
+<|ref|>text<|/ref|><|det|>[[115, 788, 490, 805]]<|/det|>
+23) p.11: CCDM is not a sphere shaping algorithm.  
+
+<|ref|>text<|/ref|><|det|>[[115, 821, 881, 875]]<|/det|>
+Response: We added "Note that CCDM can only approximately realize sphere shaping for finite block lengths, but it converges to ideal sphere shaping with a decreasing approximation error as the block length increases" to the first paragraph in the Methods section.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 857, 127]]<|/det|>
+24) Similar gains in data rate/SNR/AIR to the ones that are reported at the end manuscript exist in the literature. A short review of such works would make the paper stronger.  
+
+<|ref|>text<|/ref|><|det|>[[114, 141, 876, 274]]<|/det|>
+Response: At the end of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section, we added brief explanation about the previous works and emphasized the difference of this work compared to the previous works as "While the previous works<sup>8,11,12,16,23</sup> optimized only \(n\) to observe some gains in \(SNR_{Eff}\) and NDR over specific links (e.g., for single- span links), joint optimization of \(n\) and \(R_{Sym}\) in this work produces significantly larger gains and allows these gains to be achieved over a much wider variety of links." Due to the word count limit of Nature Communications, this level of explanation seems to be the best we can do.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 304, 228, 320]]<|/det|>
+## REFERENCES  
+
+<|ref|>text<|/ref|><|det|>[[111, 320, 880, 799]]<|/det|>
+[1] K. Wu, G. Liga, A. Sheikh, F. M. J. Willems, and A. Alvarado, "Temporal energy analysis of symbol sequences for fiber nonlinear interference modelling via energy dispersion index," Feb. 2021. [Online]. Available: https://arxiv.org/abs/2102.124114 [2] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike- Akino, K. Kojima, K. Parsons, and D. S. Millar, "Nonlinearity tolerant LUT- based probabilistic shaping for extended- reach single- span links," IEEE Photon. Technol. Lett., vol. 32, no. 16, pp. 967- 970, 2020. [3] T. Fehenberger, "On the impact of finite- length probabilistic shaping on fiber nonlinear interference," in Proc. Signal Process. in Photon. Commun. (SPPCom), Washington, D.C., United States, July 2020. [4] P. Skvortcov, I. Phillips, W. Forysiak, T. Koike- Akino, K. Kojima, K. Parsons, and D. S. Millar, "Huffman- coded sphere shaping for extended- reach single- span links," IEEE J. Sel. Topics Quantum Electron., vol. 27, no. 3: 3500215, May- June 2021. [5] S. Civelli, E. Forestieri, and M. Secondini, "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation," in Proc. Eur. Conf. Opt. Commun. (ECOC), Brussels, Belgium, Dec. 2020. [6] T. Fehenberger, A. Alvarado, G. B'oherer, and N. Hanik, "On probabilistic shaping of quadrature amplitude modulation for the nonlinear fiber channel," J. Lightw. Technol., vol. 34, no. 21, pp. 5063- 5073, Nov. 2016. [7] Y. C. Gültekin, A. Alvarado, O. Vassilieva, I. Kim, P. Palacharla, C. M. Okonkwo, and F. M. Willems, "Kurtosis- limited sphere shaping for nonlinear interference noise reduction in optical channels," June 2021. [Online]. Available: https://arxiv.org/abs/2105.14794 [8] P. Schulte and F. Steiner, "Divergence- optimal fixed- to- fixed length distribution matching with shell mapping," IEEE Wireless Commun. Lett., vol. 8, no. 2, pp. 620- 623, Apr. 2019. [9] Y. C. Gültekin, W. J. van Houtum, A. Koppelaar, and F. M. J. Willems, "Enumerative sphere shaping for wireless communications with short packets," IEEE Trans. Wireless Commun., vol. 19, no. 2, pp. 1098- 1112, Feb. 2020.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 320, 107]]<|/det|>
+## Response to Reviewer 2:  
+
+<|ref|>text<|/ref|><|det|>[[115, 116, 830, 170]]<|/det|>
+This work studies the optimization of the time and bandwidth over which sphere shaping should be performed to minimize the impact of nonlinear effects and maximize performance in optical fiber communication.  
+
+<|ref|>text<|/ref|><|det|>[[115, 173, 660, 190]]<|/det|>
+The work is timely, well written, and reports some novel interesting results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 192, 866, 265]]<|/det|>
+The optimization of the shaping block length in the nonlinear regime has been already investigated in some other papers but, to my knowledge, the simultaneous optimization of block length and symbol rate (i.e., time and bandwidth) is studied here for the first time. Moreover, this study encompasses several scenarios and a wide range of parameters, both numerically and experimentally.  
+
+<|ref|>text<|/ref|><|det|>[[115, 265, 870, 301]]<|/det|>
+The methodology is sound and the methods are described with enough detail. Moreover, the conclusions are supported both by numerical and experimental results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 302, 870, 337]]<|/det|>
+In general, I believe that the paper deserves publication. However, there are some weaknesses and some issues that should be considered and possibly addressed, as detailed below.  
+
+<|ref|>text<|/ref|><|det|>[[115, 353, 877, 408]]<|/det|>
+Response: Thank you very much for the positive comments and suggestions for improving the paper. We tried to address all of your concerns in the revised manuscript, and hope that the weaknesses and issues found in the previous manuscript have been removed by this revision.  
+
+<|ref|>text<|/ref|><|det|>[[114, 437, 881, 602]]<|/det|>
+1) As a general note, while reading the paper, I was expecting to find a theoretical analysis which explains and predicts the results observed numerically and experimentally. This expectation is not fully meet by the paper, which provides only some insight about the observed phenomena and some empirical laws to model them - e.g., (7), (8) and the use of the windowed moments in the EGN model. In fact, there are a number of perturbation models in the literature which could be used to attempt such an analysis, for instance by removing the i.i.d. assumption in the EGN model. An example can be found in Liga et al "Extending Fibre Nonlinear Interference Power Modelling to Account for General Dual-Polarisation 4D Modulation Formats", Entropy, 2020. I recommend discussing this issue and highlighting the empirical nature of the proposed model.  
+
+<|ref|>text<|/ref|><|det|>[[114, 618, 883, 893]]<|/det|>
+Response: Thank you for referring us to the previous work on propagation modeling of structured lightwaves. In fact, after submitting the first manuscript, we learned from an author of Ref. 26 (which is the paper that you mentioned above) at a conference that an extended EGN model was published last year for the presence of structures (or correlations) in four amplitudes, i.e., in one symbol period. To the best of our knowledge, this (and its simplification in Ref. 27) is the only published work on propagation modeling of structured lightwaves. However, given the enormous complexity of the mathematical expansion to account for only four correlated amplitudes in Ref. 26, it can be inferred that it is daunting to find a mathematically accurate analytical model to handle tens to thousands of correlated amplitudes that we cover in this paper. In this regard, we addressed your concern by adding a sentence to the beginning of Page 3 as "Analytical approaches to take into account the structure of lightwaves have so far been successful up to one symbol<sup>26,27</sup>, but extending the analysis to structures spanning many symbols seems mathematically daunting. To quantify the effect of large temporal structures of lightwave on Kerr nonlinearity, empirical approaches are being taken in rapidly growing recent studies<sup>8,15- 17,23</sup>. However, there has been no study on whether or how the symbol rate affects this quantification." We also stated on Page 9 that "The EGN model assumes i.i.d. amplitudes and phases of symbols, and hence is not accurate
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 877, 201]]<|/det|>
+for lightwaves with local energy structures. There is a recently developed analytical model<sup>26,27</sup> that extends the EGN model to account for energy structures present over one symbol period, but extending this further to energy structures spanning tens to thousands of symbol periods that we deal with in this work seems mathematically intractable. Therefore, we allow for model mismatch by using the classical EGN model, but improve the accuracy of evaluating structured lightwaves (green solid lines in the figure) by replacing \(\mu_{2}\) and \(\mu_{3}\) with optimized \(\bar{\mu}_{2}\) and \(\bar{\mu}_{3}\) (i.e., obtained with \(w_{SPM}^{*}\) and \(w_{XPM}^{*}\) )."  
+
+<|ref|>text<|/ref|><|det|>[[113, 230, 879, 431]]<|/det|>
+2) As a second general comment, the analysis practically neglects the impact of carrier recovery. Indeed, this might be substantial, as a sufficiently fast carrier recovery algorithm can mitigate the impact of nonlinear phase noise and change the dependence of performance on block length (see for instance Civelli et al. "Interplay of probabilistic shaping and carrier phase recovery for nonlinearity mitigation", ECOC 2020). With carrier recovery, there might be no advantage in using short (optimized) block lengths. It seems that carrier recovery is not present in the simulations and is quite slow in the experiments (using only 1 pilot every 48 symbols), possibly overestimating the advantage of using a short optimized block length. Please comment about that. As a side note, for the sake of transparency, I am a coauthor of the above mentioned paper. I do not want to push my own work, therefore feel free to decide if the paper is relevant or not to your work and if it deserves to be cited or not. This decision will not affect my final recommendation.  
+
+<|ref|>text<|/ref|><|det|>[[114, 446, 884, 577]]<|/det|>
+Response: Thanks for pointing out another prior work related to this manuscript and for providing transparency. As reported in the suggested reference paper, the phase rotation due to NLI can be mitigated by the BPS algorithm under certain conditions. To check this, we performed the BPS on our simulation data, but no appreciable effect was observed in our settings, especially over long distances with low SNR as shown in the table below for Link D. The number of symbols, \(2N_{BPS} + 1\) , that are used for averaging the phase rotation is determined such that the resulting \(SNR_{Eff}\) is approximately maximized, and a launch power of 1 dBm is used that is close to optimal.  
+
+<|ref|>table<|/ref|><|det|>[[267, 584, 728, 680]]<|/det|>
+
+<table><tr><td rowspan="2">Number of Spans</td><td rowspan="2">2NBP5 + 1</td><td colspan="2">NSC = 1</td><td colspan="2">NSC = 32</td></tr><tr><td>Without BPS</td><td>With BPS</td><td>Without BPS</td><td>With BPS</td></tr><tr><td>15</td><td>17</td><td>20.41</td><td>20.57</td><td>20.88</td><td>20.94</td></tr><tr><td>240</td><td>513</td><td>7.34</td><td>7.38</td><td>8.39</td><td>8.31</td></tr></table>  
+
+<|ref|>text<|/ref|><|det|>[[114, 690, 882, 819]]<|/det|>
+Similarly, when we performed the BPS on our experimental data, we did not observe noticeable change in \(SNR_{Eff}\) . Note that for carrier recovery of the experimental data, we used linear interpolation to obtain the phase of 47 payload symbols between 2 consecutive pilot symbols. We briefly discussed about the BPS at the end of the Discussion section as "Also, the use of advanced carrier recovery algorithms such as the maximum- likelihood blind phase search (BPS) may influence the impact of sphere shaping on NLI under certain conditions<sup>46</sup>, but in this work at transmission distances that match the sphere- shaped 16- QAM format, no noticeable effect was observed using the BPS."  
+
+<|ref|>text<|/ref|><|det|>[[114, 851, 865, 906]]<|/det|>
+3) Abstract: The sentence "In optical fiber, however, the sphere shaping induces Kerr nonlinearity in a peculiar way that makes analysis of transmission performance difficult, potentially lowering the communications capacity" sounds a bit odd. It suggests that it is sphere shaping that makes the analysis
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 812, 127]]<|/det|>
+difficult, with Kerr nonlinearity just playing an indirect role. In fact, I think the opposite is true. Moreover, it suggests that the difficulty of the analysis may play a role in the capacity reduction.  
+
+<|ref|>text<|/ref|><|det|>[[115, 141, 874, 197]]<|/det|>
+Response: To emphasize that it is Kerr nonlinearity that is hard to understand, we rephrased the sentence as "However, when shaped lightwaves are transmitted through optical fiber, Kerr nonlinearity manifests itself as nonlinear interference in a peculiar way, potentially lowering communications capacity."  
+
+<|ref|>text<|/ref|><|det|>[[115, 226, 877, 337]]<|/det|>
+4) Introduction: The idea of using sphere shaping on an optimized block length to minimize the impact of nonlinear interference was first proposed in [10] and then further investigated by some of the same authors in Geller et al. "A Shaping Algorithm for Mitigating Inter-Channel Nonlinear Phase-Noise in Nonlinear Fiber Systems", JLT 2016. Your work (as well as some other recent works) heavily relies on this idea, extending the optimization to the frequency domain. I think that it would be fair to better acknowledge this fact and the role of these two papers.  
+
+<|ref|>text<|/ref|><|det|>[[113, 352, 880, 576]]<|/det|>
+Response: We clarified what has been done in the previous works and what is the difference of this work compared to the previous works by adding several sentences to the paper as follows. In the second paragraph of the Introduction section, we added "For this reason, there have been several recent approaches to optimizing the shaping block length to mitigate Kerr nonlinearity<sup>8,11-12,15-16,23</sup>." At the end of the "Sphere Shaping of Lightwaves" section, we added "While the shaping block length \(n\) or the distribution of energy (cf. Fig. 1(a)) has been optimized in existing studies<sup>8,11-12,15-17,23</sup> to mitigate nonlinear interference (NLI), the spectro-temporal region where the distribution is found (cf. Fig. 1(b)) has never been noted previously." At the end of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section, we added "While the previous works<sup>8,11,12,16,23</sup> optimized only \(n\) to observe some gains in \(SNR_{eff}\) and NDR over specific links (e.g., for single- span links), joint optimization of \(n\) and \(R_{Sym}\) in this work produces significantly larger gains and allows these gains to be achieved over a much wider variety of links."  
+
+<|ref|>text<|/ref|><|det|>[[114, 584, 879, 712]]<|/det|>
+As for the last sentence added above, the findings of this article suggest that for symbol rates greater than 10 GBd, as is commonly used in recent experiments, optimization of the shaping block length is only effective over very short distances on dispersion- unmanaged SSMF links where net dispersion is small (see equations (7), (8) and Supplementary Fig. 4). This perhaps explains why most of the previous works on optimizing the shaping block length used single- span links. For long distances on dispersion- unmanaged SSMF links, we are able to observe a noticeable benefit by shaping block length optimization only when using a low symbol rate of a few GBd. The last sentence above emphasizes this briefly.  
+
+<|ref|>text<|/ref|><|det|>[[114, 742, 878, 870]]<|/det|>
+5) Page 3, Results: you state that "The symmetry of the probability allows for legitimate analysis with only positive amplitudes, and hence we omit the sign throughout this article". I guess that the presence of a random sign with uniform distribution is implicitly assumed in the paper and neglected only in the description, whereas it is included both in the simulations and in the experiments. Is that correct? I am not sure that transmitting only positive amplitudes would indeed give the same results. I think that two consecutive symbols with the same sign or with opposite signs induce a fundamentally different response in the nonlinear channel. Please be more explicit.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 875, 163]]<|/det|>
+Response: You're correct that random signs were used with equal probability in simulations and experiments. We have clarified this by rewriting the sentence as "we omit the sign throughout this article for descriptive purposes (but in simulations and experiments, equally distributed positive and negative signs are used)."  
+
+<|ref|>text<|/ref|><|det|>[[114, 193, 872, 266]]<|/det|>
+6) Page 3, Results: the sentence "one dual-polarization symbol that maintains the value over a period of \(T\_ sym (s)\) and can change the value at a rate of \(R\_ sym = 1 / T\_ sym (Bd)\) ", suggests a rectangular pulse shape. I suggest revising the sentence. Moreover, the indication of the units of measure in this way is neither standard nor required.  
+
+<|ref|>text<|/ref|><|det|>[[114, 282, 877, 356]]<|/det|>
+Response: To eliminate the confusion that the pulse shape is rectangular, we have rephrased the sentence as "In our system, four consecutive amplitudes constitute one dual-polarization symbol, as has been done, e.g., in \(^{16,17}\) , that is transmitted with a symbol period of \(T_{Sym}\) and a symbol rate of \(R_{Sym} = 1 / T_{Sym}\) ." We also removed the unnecessary unit of \(T_{Sym}\) and \(R_{Sym}\) here.  
+
+<|ref|>text<|/ref|><|det|>[[114, 388, 875, 444]]<|/det|>
+7) Page 5, 4 lines before the end of the subsection: The sentence "As \(w\) increases, \(\forall m\_ 2\) remains the same for all types of i.i.d. symbols" is misleading. It is not "the same" for all types of symbols but, rather, it remains "constant" as \(w\) increases for a given modulation (type of symbols).  
+
+<|ref|>text<|/ref|><|det|>[[114, 460, 740, 478]]<|/det|>
+Response: We fixed the incorrect sentence as " \(\bar{\mu}_{2}\) remains constant for i.i.d. symbols."  
+
+<|ref|>text<|/ref|><|det|>[[114, 508, 880, 599]]<|/det|>
+8) Equation (3): for sphere shaping, the transmitted signal is a ciclostationary process with period equal to the length of the shaped blocks (unless a uniform random delay is assumed). Therefore, I guess that the windowed moments depend on the time at which the window is centered, unless they are averaged over it. Please comment about that and explain if and how you have considered this time dependence in your analysis.  
+
+<|ref|>text<|/ref|><|det|>[[114, 615, 874, 762]]<|/det|>
+Response: You're correct that sphere- shaped signal, and hence \(p\) , is a cyclostationary process with a period equal to the \(n\) symbol periods. If the sliding step size of the moving average filter \((\cdot)_{w}\) is greater than one symbol, it is also true that the windowed moments \(\bar{\mu}_{n}\) in equation (3) can have different values depending on where the window starts sliding in the stream of the shaped symbol blocks. However, we use a sliding step size of one symbol, as is typically done in moving average filters. In this case, \(\bar{\mu}_{n}\) can have only a single fixed value. To avoid potential confusion, we explicitly stated this fact under equation (3) as " \((\cdot)_{w}\) denotes a moving average filter with a sliding window of length \(w\) symbols (with a sliding step size of one symbol)."  
+
+<|ref|>text<|/ref|><|det|>[[114, 794, 880, 847]]<|/det|>
+9) Page 5, Section "Optimization of Sphere Shaping in the Time-Frequency Plane": Please specify what QAM format is considered in this section. I guess from the given data that it is 16QAM (as in the previous section). Yet, an explicit mention here would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[114, 863, 877, 900]]<|/det|>
+Response: In the "Optimization of Sphere Shaping Parameters in the Time- Frequency Plane" section, we explicitly stated the modulation format by modifying a sentence to "Sphere shaping is performed in each
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 87, 870, 164]]<|/det|>
+channel with \(n = 5\) , 10, 20, 40, 80, 320, 1280, 5120, with a fixed \(R = 6.4\) bits per dual- polarization symbol using 16- QAM." Also, in the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section, we specified the modulation format by modifying a sentence to "For each pair of \(n\) and \(R_{sym}\) , the NDR achieved by sphere shaping of 16- QAM is determined by...".  
+
+<|ref|>text<|/ref|><|det|>[[113, 195, 880, 507]]<|/det|>
+10) Page 6: You state that "With our system settings, \(n = 5120\) approximately represents i.i.d. shaping from the dispersion perspective, since each symbol is dispersed over less than 5120 symbol periods in all conditions except for \(R_{sym} = 88\) GBd in Link D" and that "like the i.i.d. uniform QAM, the optimal \(R_{sym}\) for i.i.d. shaping to maximize SNR_eff (yellow stars) decreases as the total net dispersion \(D_{Total}\) increases". I think that this discussion is a bit weak and that the analogy with the symbol rate optimization for uniform i.i.d. QAM is misleading for two reasons: 1) The dependence of SNR_eff on symbol rate for the uniform i.i.d. QAM case vanishes for Gaussian symbols (the EGN and GN model converge to the same equations, see for instance [23]). Therefore, I expect a very little dependence on the symbol rate for shaped i.i.d. symbols, as they approximate i.i.d. Gaussian symbols. In fact, the dependence on symbol rate in Fig. 2a for the case \(n = 5120\) is very weak (and would be probably even weaker for shaped 64-QAM or 256-QAM) and the peak SNR is not so different from the surrounding values in the simulation range; 2) assuming that \(n = 5120\) represents the i.i.d. case from a dispersion perspective is an approximation whose accuracy decreases as the symbol rate increases; therefore, it cannot be used when explaining the (weak) dependence of SNR_eff on symbol rate. In fact, for \(n = 5120\) , this dependence becomes a little more relevant exactly where the approximation becomes looser. In my opinion, it seems more likely that the dependence of SNR_eff on symbol rate is mainly explained, even for \(n = 5120\) , by the same reason that explains it for lower \(n\) .  
+
+<|ref|>text<|/ref|><|det|>[[113, 521, 874, 752]]<|/det|>
+Response: Thank you very much for this valuable insight. To address your concern, we added split- step simulation results of i.i.d. shaping, in which shaped symbols are drawn from a 16- QAM alphabet according to a Maxwell- Boltzmann distribution. Inclusion of i.i.d. shaping posed difficulties in visualizing \(SNR_{Eff}\) in contour plots of Fig. 2 and Supplementary Figs. 1- 4, since it is equivalent to infinite- length sphere shaping in principle. In fact, as mentioned in the Methods section, the number of transmitted symbols in each WDM channel is limited to \(2^{18}\) , \(2^{17}\) , \(2^{16}\) , \(2^{15}\) , \(2^{14}\) , \(2^{14}\) , \(2^{14}\) dual- polarization symbols in each channel, respectively, for \(N_{Ch} = 1\) , 2, 4, 8, 16, 32, 64. Nevertheless, we kept the format of the contour plots the same as in the first manuscript, adding one point at the top of each \(R_{Sym}\) . This visualization would be acceptable as we see that the difference in \(SNR_{Eff}\) between \(n = 5120\) and i.i.d. shaping is very small in all figures. Accordingly, below equation (4), we explained the figure as "The top points at each \(R_{Sym}\) represent i.i.d. shaping, so their \(y\) - axis values are not exact values but merely represent very large numbers. The \(y\) - axis values for all other points are exact".  
+
+<|ref|>text<|/ref|><|det|>[[113, 759, 881, 899]]<|/det|>
+We agree that \(SNR_{Eff}\) is not changed by \(R_{Sym}\) if the modulation format is continuous Gaussian, and that the dependence of \(SNR_{Eff}\) on \(R_{Sym}\) will be very weak in the case of sphere shaping with high- order QAM. However, with relatively small i.i.d. shaped 16- QAM, we observe about 1 dB change in \(\eta^{\infty}\) due to \(R_{Sym}\) , which is not much different from the change in \(\eta\) due to \(R_{Sym}\) in i.i.d. uniform QAM (see the lower figure in Fig. 4 of Ref. 25). To emphasize the small QAM order, we specified the modulation format as "(i) like the i.i.d. uniform 16- QAM \(^{25}\) , the optimal \(R_{Sym}\) for i.i.d. shaping of 16- QAM to maximize \(SNR_{Eff}\) (yellow stars) decreases as the total net dispersion \(D_{Total}\) increases". Also, we wrote "The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 88, 875, 185]]<|/det|>
+influence of \(R_{Sym}\) on \(\eta^{\infty}\) is expected to decrease further as the modulation order increases and the shaped signal approaches continuous Gaussian" in the first paragraph of Page 8, and "It also remains for future work to see how the dependence of \(\eta\) on \(R_{Sym}\) and \(\eta\) changes as the sphere-shaped QAM modulation order increases to approach continuous Gaussian signaling in terms of the time- averaged probability distribution." in the first paragraph of the Discussion section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 215, 863, 288]]<|/det|>
+11) Page 6, you state that "the optimal \(R_{-}sym\) is higher for finite-length shaping than for i.i.d. shaping (compare the yellow and red stars), since increasing \(R_{-}sym\) reduces the shaping block duration in absolute time for the same \(n\) ". Following up on the previous comment, can we better (and more simply) say that the optimal symbol rate increases when \(n\) decreases?  
+
+<|ref|>text<|/ref|><|det|>[[115, 303, 877, 341]]<|/det|>
+Response: Thanks for the great suggestion. Following your suggestion, we have modified the sentence as "(iii) the optimal \(R_{Sym}\) increases as \(n\) decreases (compare, e.g., the yellow and red stars)."  
+
+<|ref|>text<|/ref|><|det|>[[115, 371, 882, 464]]<|/det|>
+12) Still on the same matter: you refer many times in the results to the i.i.d. case, considering \(n = 5120\) as an approximation to it. Why not directly simulating the true i.i.d. case? It should be sufficient to draw i.i.d. QAM symbols from a Maxwell-Boltzmann distribution. That would be a more meaningful benchmark against which to compare the results for finite block lengths and, for instance, to compute \(\forall \mathrm{eta}^{\infty}\) (infy) in (6).  
+
+<|ref|>text<|/ref|><|det|>[[115, 480, 863, 535]]<|/det|>
+Response: We agree that the inclusion of i.i.d. shaping would greatly improve the paper, and as mentioned above, we added split- step simulation results of i.i.d. shaping. Figures and many parts of the text have been modified accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[115, 565, 872, 675]]<|/det|>
+13) Page 8, Demonstration of Optimal Sphere Shaping through Full C-Band Transmission Experiment: The experiment is performed on a link that is substantially different from the 4 links considered in the simulation. Why? In particular, each channel sees a different dispersion profile, making the interpretation of the results harder. Moreover, the important dispersion-unmanaged case is not considered in the experiments. Is there a particular reason for this choice? Is it instrumental to highlight some effect or behavior? Please explain and motivate this choice.  
+
+<|ref|>text<|/ref|><|det|>[[114, 690, 880, 892]]<|/det|>
+Response: Although we aim to estimate the benefits of shaping optimization in the full C- band transmission, our simulation setup is limited to a total bandwidth of 100 GHz due to the long simulation time. The experiment is much easier to implement the full C- band transmission setup, but it is very difficult to test all 4 links through the experiment due to the availability of different types of fiber and the enormous effort required to construct a link (e.g., custom gain flattening for each link and for each launch power is a huge effort). On one hand, performing the experiment in a single link close to one of the 4 links used in simulation is good for validating the findings of simulation. But it has some weaknesses as follows. (i) The validation is limited to a specific dispersion coefficient of the selected link. (ii) To evaluate the impact of dispersion on NLI, the transmission distance needs to be changed as shown in, e.g., Supplementary Fig. 4. However, as the distance changes, 16- QAM becomes no longer a suitable modulation format that matches the underlying \(SNR_{eff}\) , making it difficult to evaluate the NDR increase
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 816, 125]]<|/det|>
+achieved by optimizing the sphere shaping parameters with the same modulation format as in the simulation.  
+
+<|ref|>text<|/ref|><|det|>[[114, 134, 881, 300]]<|/det|>
+On the other hand, allowing for mismatches of the setup between simulation and experiment, performing the experiment on a dispersion- managed link near zero- dispersion frequency with a dispersion slope allows us to see how NLI varies with the sphere shaping parameters under various chromatic dispersion coefficients, even at a fixed distance. This allows for qualitative validation of simulation results, as briefly mentioned in the first paragraph of the "Demonstration of Optimal Sphere Shaping through Full C- Band Transmission Experiment" section as "The dependence of the optimal sphere shaping on dispersion is conveniently verified in a recirculation loop that accumulates varying dispersions over frequency." While we agree that the dispersion- unmanaged system is an important use case, we would like to leave the experiment in dispersion- unmanaged systems for future work.  
+
+<|ref|>text<|/ref|><|det|>[[113, 329, 880, 532]]<|/det|>
+14) Page 11, Methods: you state that "The digital sphere shaping encoder is implemented by enumerative sphere shaping (ESS) for \(n = 5\) , 10, 20, 40, 80, and by constant composition distribution matching (CCDM) for \(N = 320\) , 1280, 5120." and "...consume 0.583, 0.346, 0.209, 0.137, 0.099, 0.025, 0.009, 0.002 dB more average symbol energy than ideal shaping, respectively". Strictly speaking, CCDM does not implement sphere shaping (as all constant composition sequences lie on the surface of the sphere, partially covering it) and, in fact, its energy loss is higher than that of ESS for the same blocklength. Of course, they both asymptotically converge to i.i.d. Maxwell-Boltzmann symbols, but as you are studying how performance changes with block length, you cannot assume that such an asymptotic regime has been achieved. In fact, I am also a bit surprised that CCDM with \(n = 320\) performs significantly better than ESS with \(n = 80\) (see, e.g., the comparison in [7]). Please comment about that and double check the numbers (I have not verified them, it is just a feeling and I might be wrong).  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 550, 242, 566]]<|/det|>
+## Marco Secondini  
+
+<|ref|>text<|/ref|><|det|>[[114, 575, 872, 686]]<|/det|>
+Response: As you pointed out, CCDM can only approximately realize sphere shaping. Its rate loss decreases as the block length increases, and approaches to ideal sphere shaping as the block length tends to infinity, as a result of the sphere hardening phenomenon. To address your concern, in the first paragraph of the Methods section, we added the sentence "Note that CCDM can only approximately realize sphere shaping for finite block lengths, but it converges to ideal sphere shaping with a decreasing approximation error as the block length increases."  
+
+<|ref|>text<|/ref|><|det|>[[114, 693, 867, 787]]<|/det|>
+To explain the performance of CCDM as good as that of ESS at \(n = 320\) , we have reproduced Fig. 3 of Ref. 8 (which is [7] in your reference number) in the figure on the left below with \(R = 1.6\) bit/amp and 64- QAM. However, unlike Ref. 8, this paper uses 16- QAM, and in this case the increase in rate loss of CCDM compared to ESS is not large even at a fairly short block length, as shown on the right figure below. This explains the good performance of CCDM at \(n = 320\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[160, 95, 470, 268]]<|/det|>  
+
+<|ref|>image<|/ref|><|det|>[[490, 95, 816, 268]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 309, 100]]<|/det|>
+REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[118, 137, 405, 153]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 191, 852, 224]]<|/det|>
+All my comments and suggestions seem to be taken into account. I can suggest the publication of the paper following a proofreading.  
+
+<|ref|>text<|/ref|><|det|>[[118, 290, 405, 305]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 342, 486, 358]]<|/det|>
+All my comments have been properly addressed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 395, 247, 410]]<|/det|>
+Marco Secondini
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[190, 89, 808, 157]]<|/det|>
+# Response to Reviewers' Comments on NCOMM-21-15776 "Shaping Lightwaves in Time and Frequency for Optical Fiber Communication"  
+
+<|ref|>text<|/ref|><|det|>[[140, 174, 857, 193]]<|/det|>
+Junho Cho, Xi Chen, Greg Raybon, Di Che, Ellsworth Burrows, Samuel Olsson, and Robert Tkach  
+
+<|ref|>text<|/ref|><|det|>[[115, 253, 237, 269]]<|/det|>
+Dear Reviewers,  
+
+<|ref|>text<|/ref|><|det|>[[115, 279, 859, 333]]<|/det|>
+We sincerely thank all reviewers for carefully reading and commenting on our paper again. Please find our point- by- point response to the comments below, where the reviewers' comments are reproduced verbatim in blue italic type, and our response is written in black roman type.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 367, 320, 385]]<|/det|>
+## Response to Reviewer 1:  
+
+<|ref|>text<|/ref|><|det|>[[115, 395, 841, 431]]<|/det|>
+All my comments and suggestions seem to be taken into account. I can suggest the publication of the paper following a proofreading.  
+
+<|ref|>text<|/ref|><|det|>[[115, 444, 863, 462]]<|/det|>
+Response: Your comments have greatly improved our manuscript. Thank you again very much for this.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 519, 320, 536]]<|/det|>
+## Response to Reviewer 2:  
+
+<|ref|>text<|/ref|><|det|>[[115, 546, 464, 563]]<|/det|>
+All my comments have been properly addressed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 580, 845, 616]]<|/det|>
+Response: Thank you very much for your comments during the last two reviews. They advanced our understanding of the subject area and significantly improved the manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42236cdce814b6f16452541b829274fa16205f7d595121945ec587de3c00c67c/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__42236cdce814b6f16452541b829274fa16205f7d595121945ec587de3c00c67c/supplementary_0_Peer Review File__42236cdce814b6f16452541b829274fa16205f7d595121945ec587de3c00c67c.mmd

@@ -0,0 +1,165 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+# Long-term Doppler imaging of the star XX Trianguli indicates chaotic nonperiodic dynamo  
+
+![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 --->
+
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors' responses to the comments posed were largely acceptable and addressed the concerns through clarifications and additional text. All of the attached movie files are also now accessible.  
+
+There are some concerns that remain.  
+
+- While the explanation about the north-south degeneracy is acceptable, more details on the star's inclination need to be given. The authors gave a thorough explanation in response to Referee #2's question of whether the inclination from previous study had been improved upon or not. The inclination is given in the text as though it was measured and not inferred through fits to the data. Please clarify this in the text. Additionally, on Page 3, inclination is included in a list of parameters that are considered "known with high precision," which is misleading given the inclination is not directly measured.  
+
+- In response to my comment in Section 1.3 APT, the authors suggest that it would be damaging to the data and results to provide a qualitative analysis comparing their surfaces to the photometric data. I disagree.  
+
+As the authors state, the non- cyclic, long- term trend seen in the photometry cannot yet be explained. While one approach taken in past studies was to eliminate this feature with a simple long- term trend and assess at the rotations individually, it is a stretch to consider this brutal when it is often largely a median subtraction from each observing period that occurs.  
+
+It should be possible to just assess consistency of the Doppler images with the photometry- - - even something as basic as the brightness in the Doppler surface goes up when photometry also indicate a brightness increase showing evidence that these are real structures. While the information to be garnered from photometry is limited, pointing out these consistencies makes the Doppler results more believable. Ignoring this in the text while including the photometry makes the reader suspicious that there are substantial disagreements that are being intentionally left out. Given the
+
+<--- Page Split --->
+
+timespan of the data included in Figures 3c and 7 compressing the data so individual rotations cannot be seen, it is difficult for the reader to assess this on their own.  
+
+- The mass of the star is mentioned multiple times, but only the mass function can be known, by the authors' own words. While this is explored in Section 2.3 General Assumptions, the paragraph that starts on Page 2 asserts the primary star's mass is "1.2 Msun, which in 2.3 is not given as the most probable mass. This may be resolved by including citations for the stellar parameters in that paragraph on Page 2 when the parameters are not going to be those determined by this work. For the value given in Table 1 and the description given in Section 2.3, it would be useful to understand what led to 1.1 Msun being the most probable value.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Review of the manuscript entitled:  
+
+"A 16- yr movie of the spotted surface of the K- giant XX Trianguli: evidence for a chaotic stellar dynamo"  
+
+by Strassmeier et al.  
+
+This revised manuscript describes the mapping of the surface spot features on the K0III star XX Tri, covering "16 years of observations. The authors have used the technique of Doppler Imaging to recreate 99 images of the stellar surface over this time making it the most heavily sampled study of stellar spot evolution to date.  
+
+General comments:  
+
+I thank the authors for their replies to the queries I raised in the previous version of the paper. Although the authors have answered most of the queries I previously raised there are a few further points that I feel could be expanded upon more in the paper as I mention below.
+
+<--- Page Split --->
+
+As I remarked previously, overall the paper is well written and provides an intriguing look at the evolution of the spot features of XX Tri. This initial paper focuses on the astrometric impact of the spots features on the surface of XX Tri, which have been mapped by the authors using the technique of Doppler imaging. The authors find that the photocenter of the star changes by up to \(\sim 10\%\) of the stellar radius and is the main finding of the paper.  
+
+Main points:  
+
+- The photocenter calculations given by the authors are based on the reconstructed Doppler maps they create. The accuracy of the calculated photocenter offsets provided by the authors are reliant upon the accuracy of the reconstructed Doppler images in reproducing the stellar surface configuration. Given that the authors have access to contemporaneous photometry of XX Tri (Figure 7), is the change in flux measured from the Doppler maps matched by the changes in the photometry from the star? Can the author's comment on how well the Doppler maps are able to reproduce the photometric changes and thus the accuracy of the photocenter calculations?  
+
+- Given that the spots recovered in Doppler images are usually restricted to one hemisphere (normally nominated as the Northern hemisphere) of the star, can the authors comment on how the unrecovered spots from the southern hemisphere will impact the calculations of their photocenter? I assume that this will be mitigated somewhat by the projection effect and by the fact that latitudes below -60deg on the star are not visible, but some discussion on this would be useful.  
+
+Minor points:  
+
+- In the abstract the author's have included a mention about the impact the photocenter on the detection of exoplanets. I thank the authors for this, but where it has been included seems bit out of place and I would suggest some minor rewriting to fix this.  
+
+- On page 3 the author's mention that bright faculae are required to account for the observed brightness differences seen in the star's photometry. They then report on page 4, that they have found 5 such events in their recreated Doppler maps. Could the author's provide any indication if this would be enough to explain the brightness differences mentioned on page 3, at least for the time period of the reconstructed maps, or is this number likely to be an under (or over) estimate?  
+
+- In Section 2.3, the authors state that: "For a differentially rotating surface, like for XX Tri, and likely all cool stars...". It was my understanding that XX Tri was mapped assuming solid-body rotation,
+
+<--- Page Split --->
+
+correct? If so, are the authors simply assuming that XX Tri is differentially rotating? Some comment on the assumption of solid- body rotation in the mapping should be mentioned at least here.  
+
+- In Section 2.7, the author's mention that the image photocenter is a point on the apparent disk of the star. When calculating this do the authors take into account the stellar inclination as the star is visible to us? This should be mentioned.
+
+<--- Page Split --->
+
+## Reply to Referees  
+
+Reply to RefereesPls note that the current submission was reformatted according to the ncomms- formatting- instructions as requested by the editor. The paper is now split into the main text/figs/tables/refs and the supplementary material with its own figs/tables. There are no section numbers anymore in the Method's section but we quote them below for easier orientation.  
+
+## Reviewer #1 (Remarks to the Author):  
+
+- While the explanation about the north-south degeneracy is acceptable, more details on the star's inclination need to be given. The authors gave a thorough explanation in response to Referee #2's question of whether the inclination from previous study had been improved upon or not. The inclination is given in the text as though it was measured and not inferred through fits to the data. Please clarify this in the text. Additionally, on Page 3, inclination is included in a list of parameters that are considered "known with high precision," which is misleading given the inclination is not directly measured.  
+
+We added now a shortened version of our earlier reply to Referee #2 into the method's text in Sect. 2.3 ("General assumptions", p.23), and also added a sentence clarifying the indirect inclination "measurement" in the main text on p.3.  
+
+In response to my comment in Section 1.3 APT, the authors suggest that it would be damaging to the data and results to provide a qualitative analysis comparing their surfaces to the photometric data. I disagree. As the authors state, the non- cyclic, long- term trend seen in the photometry cannot yet be explained. While one approach taken in past studies was to eliminate this feature with a simple long- term trend and assess at the rotations individually, it is a stretch to consider this brutal when it is often largely a median subtraction from each observing period that occurs. It should be possible to just assess consistency of the Doppler images with the photometry- - even something as basic as the brightness in the Doppler surface goes up when photometry also indicate a brightness increase showing evidence that these are real structures. While the information to be garnered from photometry is limited, pointing out these consistencies makes the Doppler results more believable. Ignoring this in the text while including the photometry makes the reader suspicious that there are substantial disagreements that are being intentionally left out. Given the timespan of the data included in Figures 3c and 7 compressing the data so individual rotations cannot be seen, it is difficult for the reader to assess this on their own.  
+
+A new additional figure (now named Fig. 7 in the Supplementary material) shows a direct comparison of the predicted brightness variations from our Doppler images with the (existing) V- band APT photometry. As already stated, we cannot explain the observed long- term brightness trend with our (purely photospheric) Doppler images and must remove the V- band trend prior to a comparison on the scale of rotational modulation. This "de- trending" is not without concerns and is achieved with a simple normalization of the seasonal average brightness so that the seasonal averages are \(\Delta V = 0\) in the new figure. At this point we note that Fig. 3c already shows the predicted variations over time (relative flux in this case and normalized to a theoretical unspotted star). This figure also indicates the times when the variations were determined directly from the Doppler images or when they were determined from the interpolated images. Because the interpolated images are basically identical to the actual Doppler images for the time ranges when the Doppler images were obtained, we use only the interpolated images for this (long- term) light- curve comparison. The new figure shows that the reconstructed light curve follows the overall rotation changes relatively well, and that the differences are also due to the fact that in a number of cases the spectra used for DI cover more than one rotation cycle (listed in Supplementary Table 1 = previous Table 2), which essentially means time averaging of the rotational changes in the spectral lines (the figure additionally demonstrates the difficulty to include broad- band photometry in time- series Doppler imaging because the photometry is additionally influenced by chromospheric and circumstellar contributions as well). Therefore, and because already the seasonal de- trending is a hand- waiving procedure, our rotational- modulation comparison remains only qualitative. Nevertheless, the predicted light curves from the DIs match the photometric variations reasonably well. This info is added in Methods Sect. "Individual maps: fit quality and example result" (former Sect. 2.6) and also mentioned in the "Discussion".
+
+<--- Page Split --->
+
+- The mass of the star is mentioned multiple times, but only the mass function can be known, by the authors' own words. While this is explored in Section 2.3 General Assumptions, the paragraph that starts on Page 2 asserts the primary star's mass is "1.2 Msun, which in 2.3 is not given as the most probable mass. This may be resolved by including citations for the stellar parameters in that paragraph on Page 2 when the parameters are not going to be those determined by this work. For the value given in Table 1 and the description given in Section 2.3, it would be useful to understand what led to 1.1 Msun being the most probable value.  
+
+Oh, thanks. We indeed forgot to describe where the "most probable" mass (and age) in Table 1 came from. This is now added to the method's text in "General assumptions" (former Sect. 2.3). Both values and its errors were from a multiparameter track- and isochrone fit with the Teff, Fe/H, and luminosity from the DR3 parallax as listed in Table 1. The additional text required two new references (one for the isochrone- fitting code and one for the isochrones themselves). An H- R diagram with all isochrones considered exists (see below) but is not given in the paper because it is not a major result. Following the referee's suggestion the most- probable mass from this paper is now already mentioned in the first sentence of the main text on p.2 instead of our initial \(20\%\) orientation value from the average from "Künstler et al." (1.26 Msun) and "this paper" (1.1 Msun).  
+
+![PLACEHOLDER_6_0]
+  
+
+Working plot for the isochrone fit to the position (red dot with error bars) of XX Tri in the HRD. Isochrones are from PARSEC. XX Tri data from Table 1. Numerical result is listed on the top.
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+- The photocenter calculations given by the authors are based on the reconstructed Doppler maps they create. The accuracy of the calculated photocenter offsets provided by the authors are reliant upon the accuracy of the reconstructed Doppler images in reproducing the stellar surface configuration. Given that the authors have access to contemporaneous photometry of XX Tri (Figure 7), is the change in flux measured from the Doppler maps matched by the changes in the photometry from the star? Can the author's comment on how well the Doppler maps are able to reproduce the photometric changes and thus the accuracy of the photocenter calculations?  
+
+![PLACEHOLDER_7_0]
+  
+
+A new figure (above, now Supplementary Fig. 7) has been added showing a direct comparison of the predicted light curves from the interpolated DIs (line) with the pre- whitened APT data (red dots with its error bars). The grey areas denote the times of the Doppler images. Pls see also our reply to Referee #1 who addressed the same issue.
+
+<--- Page Split --->
+
+calculations of their photocenter? I assume that this will be mitigated somewhat by the projection effect and by the fact that latitudes below - 60deg on the star are not visible, but some discussion on this would be useful.  
+
+Correct. Spots at higher- to- intermediate latitudes in the adjacent "southern" hemisphere are reconstructed only if the fit to the data needs to be improved for a short phase duration (spot visibility at restricted phases) while not affecting the fit at other phases. Spots are not per se restricted to a "hemisphere". We frequently see spots between \(0^{\circ}\) and \(- 30^{\circ}\) , once or twice even down to \(- 50^{\circ}\) (e.g., for rotations #93 and 113). Its detections are strongly S/N- and phase- coverage dependent. Of course, even if perfect data exist, we are left with a north- south latitude- to- latitude cross talk that, however, gets smaller and smaller the lower the inclination up to the point when the lack of Doppler velocity per latitude bin takes over and starts to dominate the signal which is easily recognized (to the extreme of zero Doppler effect when the star is seen pole- on, independent of vsini). At the time we discovered large polar spots the question was raised whether the other (hidden) pole also has a large spot and whether we could see its rim in higher inclination targets. No such detection is known until today. We refer to the discussion in Piskunov & Rice (1993, PASP, 105, 1415). Whether such unrecovered spots, temperature or chemical ones, really exist remains undetermined (we unconsciously tend to assume hemispheric symmetry like on the Sun, but face a qualitative different dynamo in XX Tri) and, in any case, the impact on the photocenter determination would be always very small due to the much larger foreshortening angles as compared to the reconstructed spots. Because in our case \(i = 60^{\circ}\) , and because our line profiles are of high S/N, we do not think that there is such an unnoticed residual cross talk that could affect the photocenters. But even if, it moreover could affect only the y- component of the photocenter (i.e., in the direction parallel to the projected rotational axis), and not the full radial extent.  
+
+A discussion of this in the text appears overemphasizing, as we can draw no conclusions nor present quantitative limits or there- alike but still need the text space of a full paragraph. We would prefer keeping it out of the text and leave it just to the referee's discretion. Just a sentence with reference to the Piskunov & Rice (1993) discussion is added in the main text where we also refer to the north- south degeneracy.  
+
+## Minor points:  
+
+- In the abstract the author's have included a mention about the impact the photocenter on the detection of exoplanets. I thank the authors for this, but where it has been included seems bit out of place and I would suggest some minor rewriting to fix this.  
+
+The abstract had been rearranged due to the limitation to 150 words. The marked sentence was moved.  
+
+- On page 3 the author's mention that bright faculae are required to account for the observed brightness differences seen in the star's photometry. They then report on page 4, that they have found 5 such events in their recreated Doppler maps. Could the author's provide any indication if this would be enough to explain the brightness differences mentioned on page 3, at least for the time period of the reconstructed maps, or is this number likely to be an under (or over) estimate?  
+
+The statement on p.3 ("It requires bright faculae in order to explain the observed brightness differences by rotational modulation.") referred to the epochs of extreme V- band amplitudes due to rotational modulation, which could not be modelled anymore with just cool spots. No such extreme rotational- modulation amplitudes were observed in our current photometric data coverage during the Doppler imagery. The DIs recovered, independent of any photometry, just small (but numerically significant) bright/warm features with at most dT \(\approx 200\mathrm{K}\) wrt the photosphere. Regarding the long- term brightness change, which is of the order of another 0.6 magnitudes on top of the rotational modulation, an explanation with concrete warm/hot features is impossible because then we must
+
+<--- Page Split --->
+
+see larger rotational- modulation amplitudes when the star is overall brighter and necessarily also bluer (smaller B- V or V- I), but it appears that the opposite is the case (higher amplitudes when the star is fainter). When considering only the time period of the reconstructed maps, the amplitude is smaller but still large enough so that the effect is noticeable. And even there it the amplitudes are larger when fainter. So, the simple version of an answer to above question is - no-, it requires much larger and warmer features than those reconstructed on the five events in order to model just the 0.6- mag V- band amplitude due to rotational modulation in, e.g., 1998 or 2000 (Ref. 10 reconstructed a warm feature with dT=350±20K for 1998 and size about five times larger than the currently seen five features). The "number" is thus neither an under- nor an overestimate but just only part of the solution. We mention the situation now at the end of the Methods "Individual maps: fit quality and example result" (former Sect. 2.6).  
+
+- In Section 2.3, the authors state that: "For a differentially rotating surface, like for XX Tri, and likely all cool stars..." It was my understanding that XX Tri was mapped assuming solid-body rotation, correct? If so, are the authors simply assuming that XX Tri is differentially rotating? Some comment on the assumption of solid-body rotation in the mapping should be mentioned at least here.  
+
+Correct. The assumption of solid- body rotation for the inversion is justified even for a differentially- rotating star because we use typically only a single stellar rotation for the inversion, thus, no latitude- dependent phase smearing from one rotation to the next (=the "signal" for DR) is expected. Once we compare a particular DI to the next in the time sequence, e.g., with a 2D cross correlation, we can infer and quantify latitude- dependent DR. Weak, solar- like DR of XX Tri of shear \(\alpha = 0.016\) , was already detected in Ref. Künstler et al. (2015), so it is not just an assumption. We added a sentence in the Methods "General assumptions" section (former Sect. 2.3).  
+
+- In Section 2.7, the author's mention that the image photocenter is a point on the apparent disk of the star. When calculating this do the authors take into account the stellar inclination as the star is visible to us? This should be mentioned.  
+
+The photocenter is determined from the images, and the images are obtained with a fixed inclination of \(60^{\circ}\) . The apparent disk is just a projection of these images onto the plane of the sky (like the orthographic projection). Thus, yes, the inclination is implicitly included in the photocenter comp. We mention this now explicitly in Methods "Image photocenter" (former Sect. 2.7). Note that this is also noticable for the animated star in movie radial_offset_anim.mp4.
+
+<--- Page Split --->
+
+REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The revisions satisfy the issues I brought up in my previous reports. I have no further concerns.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Review of the manuscript entitled:  
+
+"A 16- yr movie of the spotted surface of the K- giant XX Triangular"  
+
+by Strassmeier et al.  
+
+The revised manuscript discusses the mapping of the surface spot features on the KOlII star XX Tri over a period of approximately 16 years using the technique of Doppler Imaging. The images produced by the authors are a significant contribution to our understanding of the evolution of spot features on such stars.  
+
+I once again thank the authors for their response to my queries and the changes implemented to the article. Given this, I am happy for the article in its current form to be accepted for publication.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42236cdce814b6f16452541b829274fa16205f7d595121945ec587de3c00c67c/supplementary_0_Peer Review File__42236cdce814b6f16452541b829274fa16205f7d595121945ec587de3c00c67c_det.mmd

@@ -0,0 +1,218 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 505, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[67, 111, 361, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>title<|/ref|><|det|>[[90, 152, 944, 220]]<|/det|>
+# Long-term Doppler imaging of the star XX Trianguli indicates chaotic nonperiodic dynamo  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 782]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 912, 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|>[[118, 84, 867, 138]]<|/det|>
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 212, 295, 227]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[118, 272, 403, 288]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 332, 872, 367]]<|/det|>
+The authors' responses to the comments posed were largely acceptable and addressed the concerns through clarifications and additional text. All of the attached movie files are also now accessible.  
+
+<|ref|>text<|/ref|><|det|>[[119, 411, 404, 426]]<|/det|>
+There are some concerns that remain.  
+
+<|ref|>text<|/ref|><|det|>[[117, 470, 857, 599]]<|/det|>
+- While the explanation about the north-south degeneracy is acceptable, more details on the star's inclination need to be given. The authors gave a thorough explanation in response to Referee #2's question of whether the inclination from previous study had been improved upon or not. The inclination is given in the text as though it was measured and not inferred through fits to the data. Please clarify this in the text. Additionally, on Page 3, inclination is included in a list of parameters that are considered "known with high precision," which is misleading given the inclination is not directly measured.  
+
+<|ref|>text<|/ref|><|det|>[[118, 642, 851, 695]]<|/det|>
+- In response to my comment in Section 1.3 APT, the authors suggest that it would be damaging to the data and results to provide a qualitative analysis comparing their surfaces to the photometric data. I disagree.  
+
+<|ref|>text<|/ref|><|det|>[[118, 709, 875, 780]]<|/det|>
+As the authors state, the non- cyclic, long- term trend seen in the photometry cannot yet be explained. While one approach taken in past studies was to eliminate this feature with a simple long- term trend and assess at the rotations individually, it is a stretch to consider this brutal when it is often largely a median subtraction from each observing period that occurs.  
+
+<|ref|>text<|/ref|><|det|>[[117, 794, 878, 902]]<|/det|>
+It should be possible to just assess consistency of the Doppler images with the photometry- - - even something as basic as the brightness in the Doppler surface goes up when photometry also indicate a brightness increase showing evidence that these are real structures. While the information to be garnered from photometry is limited, pointing out these consistencies makes the Doppler results more believable. Ignoring this in the text while including the photometry makes the reader suspicious that there are substantial disagreements that are being intentionally left out. Given the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 825, 119]]<|/det|>
+timespan of the data included in Figures 3c and 7 compressing the data so individual rotations cannot be seen, it is difficult for the reader to assess this on their own.  
+
+<|ref|>text<|/ref|><|det|>[[118, 163, 866, 290]]<|/det|>
+- The mass of the star is mentioned multiple times, but only the mass function can be known, by the authors' own words. While this is explored in Section 2.3 General Assumptions, the paragraph that starts on Page 2 asserts the primary star's mass is "1.2 Msun, which in 2.3 is not given as the most probable mass. This may be resolved by including citations for the stellar parameters in that paragraph on Page 2 when the parameters are not going to be those determined by this work. For the value given in Table 1 and the description given in Section 2.3, it would be useful to understand what led to 1.1 Msun being the most probable value.  
+
+<|ref|>text<|/ref|><|det|>[[119, 394, 403, 409]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 455, 379, 470]]<|/det|>
+Review of the manuscript entitled:  
+
+<|ref|>text<|/ref|><|det|>[[118, 515, 821, 549]]<|/det|>
+"A 16- yr movie of the spotted surface of the K- giant XX Trianguli: evidence for a chaotic stellar dynamo"  
+
+<|ref|>text<|/ref|><|det|>[[118, 594, 271, 609]]<|/det|>
+by Strassmeier et al.  
+
+<|ref|>text<|/ref|><|det|>[[118, 655, 864, 726]]<|/det|>
+This revised manuscript describes the mapping of the surface spot features on the K0III star XX Tri, covering "16 years of observations. The authors have used the technique of Doppler Imaging to recreate 99 images of the stellar surface over this time making it the most heavily sampled study of stellar spot evolution to date.  
+
+<|ref|>text<|/ref|><|det|>[[118, 771, 266, 785]]<|/det|>
+General comments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 830, 863, 883]]<|/det|>
+I thank the authors for their replies to the queries I raised in the previous version of the paper. Although the authors have answered most of the queries I previously raised there are a few further points that I feel could be expanded upon more in the paper as I mention below.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 872, 175]]<|/det|>
+As I remarked previously, overall the paper is well written and provides an intriguing look at the evolution of the spot features of XX Tri. This initial paper focuses on the astrometric impact of the spots features on the surface of XX Tri, which have been mapped by the authors using the technique of Doppler imaging. The authors find that the photocenter of the star changes by up to \(\sim 10\%\) of the stellar radius and is the main finding of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[118, 218, 214, 233]]<|/det|>
+Main points:  
+
+<|ref|>text<|/ref|><|det|>[[118, 278, 870, 404]]<|/det|>
+- The photocenter calculations given by the authors are based on the reconstructed Doppler maps they create. The accuracy of the calculated photocenter offsets provided by the authors are reliant upon the accuracy of the reconstructed Doppler images in reproducing the stellar surface configuration. Given that the authors have access to contemporaneous photometry of XX Tri (Figure 7), is the change in flux measured from the Doppler maps matched by the changes in the photometry from the star? Can the author's comment on how well the Doppler maps are able to reproduce the photometric changes and thus the accuracy of the photocenter calculations?  
+
+<|ref|>text<|/ref|><|det|>[[118, 448, 876, 540]]<|/det|>
+- Given that the spots recovered in Doppler images are usually restricted to one hemisphere (normally nominated as the Northern hemisphere) of the star, can the authors comment on how the unrecovered spots from the southern hemisphere will impact the calculations of their photocenter? I assume that this will be mitigated somewhat by the projection effect and by the fact that latitudes below -60deg on the star are not visible, but some discussion on this would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[118, 584, 222, 599]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[118, 643, 875, 696]]<|/det|>
+- In the abstract the author's have included a mention about the impact the photocenter on the detection of exoplanets. I thank the authors for this, but where it has been included seems bit out of place and I would suggest some minor rewriting to fix this.  
+
+<|ref|>text<|/ref|><|det|>[[118, 740, 880, 830]]<|/det|>
+- On page 3 the author's mention that bright faculae are required to account for the observed brightness differences seen in the star's photometry. They then report on page 4, that they have found 5 such events in their recreated Doppler maps. Could the author's provide any indication if this would be enough to explain the brightness differences mentioned on page 3, at least for the time period of the reconstructed maps, or is this number likely to be an under (or over) estimate?  
+
+<|ref|>text<|/ref|><|det|>[[118, 874, 866, 909]]<|/det|>
+- In Section 2.3, the authors state that: "For a differentially rotating surface, like for XX Tri, and likely all cool stars...". It was my understanding that XX Tri was mapped assuming solid-body rotation,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 861, 119]]<|/det|>
+correct? If so, are the authors simply assuming that XX Tri is differentially rotating? Some comment on the assumption of solid- body rotation in the mapping should be mentioned at least here.  
+
+<|ref|>text<|/ref|><|det|>[[118, 163, 864, 216]]<|/det|>
+- In Section 2.7, the author's mention that the image photocenter is a point on the apparent disk of the star. When calculating this do the authors take into account the stellar inclination as the star is visible to us? This should be mentioned.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[117, 84, 290, 102]]<|/det|>
+## Reply to Referees  
+
+<|ref|>text<|/ref|><|det|>[[117, 118, 878, 190]]<|/det|>
+Reply to RefereesPls note that the current submission was reformatted according to the ncomms- formatting- instructions as requested by the editor. The paper is now split into the main text/figs/tables/refs and the supplementary material with its own figs/tables. There are no section numbers anymore in the Method's section but we quote them below for easier orientation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 205, 408, 220]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 234, 880, 300]]<|/det|>
+- While the explanation about the north-south degeneracy is acceptable, more details on the star's inclination need to be given. The authors gave a thorough explanation in response to Referee #2's question of whether the inclination from previous study had been improved upon or not. The inclination is given in the text as though it was measured and not inferred through fits to the data. Please clarify this in the text. Additionally, on Page 3, inclination is included in a list of parameters that are considered "known with high precision," which is misleading given the inclination is not directly measured.  
+
+<|ref|>text<|/ref|><|det|>[[117, 313, 870, 365]]<|/det|>
+We added now a shortened version of our earlier reply to Referee #2 into the method's text in Sect. 2.3 ("General assumptions", p.23), and also added a sentence clarifying the indirect inclination "measurement" in the main text on p.3.  
+
+<|ref|>text<|/ref|><|det|>[[116, 378, 880, 512]]<|/det|>
+In response to my comment in Section 1.3 APT, the authors suggest that it would be damaging to the data and results to provide a qualitative analysis comparing their surfaces to the photometric data. I disagree. As the authors state, the non- cyclic, long- term trend seen in the photometry cannot yet be explained. While one approach taken in past studies was to eliminate this feature with a simple long- term trend and assess at the rotations individually, it is a stretch to consider this brutal when it is often largely a median subtraction from each observing period that occurs. It should be possible to just assess consistency of the Doppler images with the photometry- - even something as basic as the brightness in the Doppler surface goes up when photometry also indicate a brightness increase showing evidence that these are real structures. While the information to be garnered from photometry is limited, pointing out these consistencies makes the Doppler results more believable. Ignoring this in the text while including the photometry makes the reader suspicious that there are substantial disagreements that are being intentionally left out. Given the timespan of the data included in Figures 3c and 7 compressing the data so individual rotations cannot be seen, it is difficult for the reader to assess this on their own.  
+
+<|ref|>text<|/ref|><|det|>[[115, 525, 880, 928]]<|/det|>
+A new additional figure (now named Fig. 7 in the Supplementary material) shows a direct comparison of the predicted brightness variations from our Doppler images with the (existing) V- band APT photometry. As already stated, we cannot explain the observed long- term brightness trend with our (purely photospheric) Doppler images and must remove the V- band trend prior to a comparison on the scale of rotational modulation. This "de- trending" is not without concerns and is achieved with a simple normalization of the seasonal average brightness so that the seasonal averages are \(\Delta V = 0\) in the new figure. At this point we note that Fig. 3c already shows the predicted variations over time (relative flux in this case and normalized to a theoretical unspotted star). This figure also indicates the times when the variations were determined directly from the Doppler images or when they were determined from the interpolated images. Because the interpolated images are basically identical to the actual Doppler images for the time ranges when the Doppler images were obtained, we use only the interpolated images for this (long- term) light- curve comparison. The new figure shows that the reconstructed light curve follows the overall rotation changes relatively well, and that the differences are also due to the fact that in a number of cases the spectra used for DI cover more than one rotation cycle (listed in Supplementary Table 1 = previous Table 2), which essentially means time averaging of the rotational changes in the spectral lines (the figure additionally demonstrates the difficulty to include broad- band photometry in time- series Doppler imaging because the photometry is additionally influenced by chromospheric and circumstellar contributions as well). Therefore, and because already the seasonal de- trending is a hand- waiving procedure, our rotational- modulation comparison remains only qualitative. Nevertheless, the predicted light curves from the DIs match the photometric variations reasonably well. This info is added in Methods Sect. "Individual maps: fit quality and example result" (former Sect. 2.6) and also mentioned in the "Discussion".
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 880, 150]]<|/det|>
+- The mass of the star is mentioned multiple times, but only the mass function can be known, by the authors' own words. While this is explored in Section 2.3 General Assumptions, the paragraph that starts on Page 2 asserts the primary star's mass is "1.2 Msun, which in 2.3 is not given as the most probable mass. This may be resolved by including citations for the stellar parameters in that paragraph on Page 2 when the parameters are not going to be those determined by this work. For the value given in Table 1 and the description given in Section 2.3, it would be useful to understand what led to 1.1 Msun being the most probable value.  
+
+<|ref|>text<|/ref|><|det|>[[115, 161, 883, 326]]<|/det|>
+Oh, thanks. We indeed forgot to describe where the "most probable" mass (and age) in Table 1 came from. This is now added to the method's text in "General assumptions" (former Sect. 2.3). Both values and its errors were from a multiparameter track- and isochrone fit with the Teff, Fe/H, and luminosity from the DR3 parallax as listed in Table 1. The additional text required two new references (one for the isochrone- fitting code and one for the isochrones themselves). An H- R diagram with all isochrones considered exists (see below) but is not given in the paper because it is not a major result. Following the referee's suggestion the most- probable mass from this paper is now already mentioned in the first sentence of the main text on p.2 instead of our initial \(20\%\) orientation value from the average from "Künstler et al." (1.26 Msun) and "this paper" (1.1 Msun).  
+
+<|ref|>image<|/ref|><|det|>[[120, 340, 864, 858]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[115, 875, 829, 911]]<|/det|>
+Working plot for the isochrone fit to the position (red dot with error bars) of XX Tri in the HRD. Isochrones are from PARSEC. XX Tri data from Table 1. Numerical result is listed on the top.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 83, 408, 98]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 111, 872, 180]]<|/det|>
+- The photocenter calculations given by the authors are based on the reconstructed Doppler maps they create. The accuracy of the calculated photocenter offsets provided by the authors are reliant upon the accuracy of the reconstructed Doppler images in reproducing the stellar surface configuration. Given that the authors have access to contemporaneous photometry of XX Tri (Figure 7), is the change in flux measured from the Doppler maps matched by the changes in the photometry from the star? Can the author's comment on how well the Doppler maps are able to reproduce the photometric changes and thus the accuracy of the photocenter calculations?  
+
+<|ref|>image<|/ref|><|det|>[[118, 193, 864, 765]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[115, 776, 874, 847]]<|/det|>
+A new figure (above, now Supplementary Fig. 7) has been added showing a direct comparison of the predicted light curves from the interpolated DIs (line) with the pre- whitened APT data (red dots with its error bars). The grey areas denote the times of the Doppler images. Pls see also our reply to Referee #1 who addressed the same issue.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 851, 108]]<|/det|>
+calculations of their photocenter? I assume that this will be mitigated somewhat by the projection effect and by the fact that latitudes below - 60deg on the star are not visible, but some discussion on this would be useful.  
+
+<|ref|>text<|/ref|><|det|>[[115, 120, 876, 489]]<|/det|>
+Correct. Spots at higher- to- intermediate latitudes in the adjacent "southern" hemisphere are reconstructed only if the fit to the data needs to be improved for a short phase duration (spot visibility at restricted phases) while not affecting the fit at other phases. Spots are not per se restricted to a "hemisphere". We frequently see spots between \(0^{\circ}\) and \(- 30^{\circ}\) , once or twice even down to \(- 50^{\circ}\) (e.g., for rotations #93 and 113). Its detections are strongly S/N- and phase- coverage dependent. Of course, even if perfect data exist, we are left with a north- south latitude- to- latitude cross talk that, however, gets smaller and smaller the lower the inclination up to the point when the lack of Doppler velocity per latitude bin takes over and starts to dominate the signal which is easily recognized (to the extreme of zero Doppler effect when the star is seen pole- on, independent of vsini). At the time we discovered large polar spots the question was raised whether the other (hidden) pole also has a large spot and whether we could see its rim in higher inclination targets. No such detection is known until today. We refer to the discussion in Piskunov & Rice (1993, PASP, 105, 1415). Whether such unrecovered spots, temperature or chemical ones, really exist remains undetermined (we unconsciously tend to assume hemispheric symmetry like on the Sun, but face a qualitative different dynamo in XX Tri) and, in any case, the impact on the photocenter determination would be always very small due to the much larger foreshortening angles as compared to the reconstructed spots. Because in our case \(i = 60^{\circ}\) , and because our line profiles are of high S/N, we do not think that there is such an unnoticed residual cross talk that could affect the photocenters. But even if, it moreover could affect only the y- component of the photocenter (i.e., in the direction parallel to the projected rotational axis), and not the full radial extent.  
+
+<|ref|>text<|/ref|><|det|>[[116, 502, 872, 591]]<|/det|>
+A discussion of this in the text appears overemphasizing, as we can draw no conclusions nor present quantitative limits or there- alike but still need the text space of a full paragraph. We would prefer keeping it out of the text and leave it just to the referee's discretion. Just a sentence with reference to the Piskunov & Rice (1993) discussion is added in the main text where we also refer to the north- south degeneracy.  
+
+<|ref|>sub_title<|/ref|><|det|>[[117, 605, 192, 616]]<|/det|>
+## Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[116, 630, 840, 656]]<|/det|>
+- In the abstract the author's have included a mention about the impact the photocenter on the detection of exoplanets. I thank the authors for this, but where it has been included seems bit out of place and I would suggest some minor rewriting to fix this.  
+
+<|ref|>text<|/ref|><|det|>[[116, 669, 840, 702]]<|/det|>
+The abstract had been rearranged due to the limitation to 150 words. The marked sentence was moved.  
+
+<|ref|>text<|/ref|><|det|>[[116, 717, 870, 771]]<|/det|>
+- On page 3 the author's mention that bright faculae are required to account for the observed brightness differences seen in the star's photometry. They then report on page 4, that they have found 5 such events in their recreated Doppler maps. Could the author's provide any indication if this would be enough to explain the brightness differences mentioned on page 3, at least for the time period of the reconstructed maps, or is this number likely to be an under (or over) estimate?  
+
+<|ref|>text<|/ref|><|det|>[[115, 784, 871, 928]]<|/det|>
+The statement on p.3 ("It requires bright faculae in order to explain the observed brightness differences by rotational modulation.") referred to the epochs of extreme V- band amplitudes due to rotational modulation, which could not be modelled anymore with just cool spots. No such extreme rotational- modulation amplitudes were observed in our current photometric data coverage during the Doppler imagery. The DIs recovered, independent of any photometry, just small (but numerically significant) bright/warm features with at most dT \(\approx 200\mathrm{K}\) wrt the photosphere. Regarding the long- term brightness change, which is of the order of another 0.6 magnitudes on top of the rotational modulation, an explanation with concrete warm/hot features is impossible because then we must
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 82, 876, 281]]<|/det|>
+see larger rotational- modulation amplitudes when the star is overall brighter and necessarily also bluer (smaller B- V or V- I), but it appears that the opposite is the case (higher amplitudes when the star is fainter). When considering only the time period of the reconstructed maps, the amplitude is smaller but still large enough so that the effect is noticeable. And even there it the amplitudes are larger when fainter. So, the simple version of an answer to above question is - no-, it requires much larger and warmer features than those reconstructed on the five events in order to model just the 0.6- mag V- band amplitude due to rotational modulation in, e.g., 1998 or 2000 (Ref. 10 reconstructed a warm feature with dT=350±20K for 1998 and size about five times larger than the currently seen five features). The "number" is thus neither an under- nor an overestimate but just only part of the solution. We mention the situation now at the end of the Methods "Individual maps: fit quality and example result" (former Sect. 2.6).  
+
+<|ref|>text<|/ref|><|det|>[[117, 295, 833, 335]]<|/det|>
+- In Section 2.3, the authors state that: "For a differentially rotating surface, like for XX Tri, and likely all cool stars..." It was my understanding that XX Tri was mapped assuming solid-body rotation, correct? If so, are the authors simply assuming that XX Tri is differentially rotating? Some comment on the assumption of solid-body rotation in the mapping should be mentioned at least here.  
+
+<|ref|>text<|/ref|><|det|>[[115, 347, 883, 476]]<|/det|>
+Correct. The assumption of solid- body rotation for the inversion is justified even for a differentially- rotating star because we use typically only a single stellar rotation for the inversion, thus, no latitude- dependent phase smearing from one rotation to the next (=the "signal" for DR) is expected. Once we compare a particular DI to the next in the time sequence, e.g., with a 2D cross correlation, we can infer and quantify latitude- dependent DR. Weak, solar- like DR of XX Tri of shear \(\alpha = 0.016\) , was already detected in Ref. Künstler et al. (2015), so it is not just an assumption. We added a sentence in the Methods "General assumptions" section (former Sect. 2.3).  
+
+<|ref|>text<|/ref|><|det|>[[115, 488, 870, 515]]<|/det|>
+- In Section 2.7, the author's mention that the image photocenter is a point on the apparent disk of the star. When calculating this do the authors take into account the stellar inclination as the star is visible to us? This should be mentioned.  
+
+<|ref|>text<|/ref|><|det|>[[115, 528, 881, 618]]<|/det|>
+The photocenter is determined from the images, and the images are obtained with a fixed inclination of \(60^{\circ}\) . The apparent disk is just a projection of these images onto the plane of the sky (like the orthographic projection). Thus, yes, the inclination is implicitly included in the photocenter comp. We mention this now explicitly in Methods "Image photocenter" (former Sect. 2.7). Note that this is also noticable for the animated star in movie radial_offset_anim.mp4.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 85, 308, 100]]<|/det|>
+REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[119, 145, 404, 161]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 204, 828, 221]]<|/det|>
+The revisions satisfy the issues I brought up in my previous reports. I have no further concerns.  
+
+<|ref|>text<|/ref|><|det|>[[119, 296, 404, 312]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 356, 379, 372]]<|/det|>
+Review of the manuscript entitled:  
+
+<|ref|>text<|/ref|><|det|>[[118, 416, 605, 432]]<|/det|>
+"A 16- yr movie of the spotted surface of the K- giant XX Triangular"  
+
+<|ref|>text<|/ref|><|det|>[[118, 476, 271, 491]]<|/det|>
+by Strassmeier et al.  
+
+<|ref|>text<|/ref|><|det|>[[118, 536, 869, 608]]<|/det|>
+The revised manuscript discusses the mapping of the surface spot features on the KOlII star XX Tri over a period of approximately 16 years using the technique of Doppler Imaging. The images produced by the authors are a significant contribution to our understanding of the evolution of spot features on such stars.  
+
+<|ref|>text<|/ref|><|det|>[[118, 652, 875, 687]]<|/det|>
+I once again thank the authors for their response to my queries and the changes implemented to the article. Given this, I am happy for the article in its current form to be accepted for publication.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42d4699c4e65ed95775b859ea7b95f01d6cc639156a8bd344e894404ab1add48/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__42d4699c4e65ed95775b859ea7b95f01d6cc639156a8bd344e894404ab1add48/supplementary_0_Peer Review File__42d4699c4e65ed95775b859ea7b95f01d6cc639156a8bd344e894404ab1add48.mmd

@@ -0,0 +1,272 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Altered tRNA dynamics during translocation on slippery mRNA as determinant of spontaneous ribosome frameshifting  
+
+![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):  
+
+Reviewer #1 (Remarks to the Author):Manuscript by Poulis et al. presents additional smFRET look on the ribosome frameshifting associated with slippery sequences. Authors use EF- G mutants, different sets of slippery and non- slippery sequences, as well as slowly hydrolyzable GTPyS, fusidic acid and spectinomycin to test previously proposed model that slow frameshifting- prone translocation mode is responsible for reduced fidelity on slippery sequences. Manuscript lacks clear advancement over previous studies, multiple control experiments (Figure 3 and 6 with wt EF- G on slippery sequence), missing data (Supplementary Fig 3 data for Q507A and Q507N are missing) as well as clarity in data presentation (ie. what is TL in figures3 and Supplementary Fig 4?). The proposed model in Figure 7b is based on EF- G mutant Q507D data and increased population of CHI and P/P states which are almost identical in GTPyS conditions (Fig 5), and on contrary to authors comments not really similar to spectinomycin data (in Supplementary Fig 8), which blurs authors conclusions. For these reasons I suggest that manuscript should not be published in this form.  
+
+Reviewer #2 (Remarks to the Author):  
+
+This is a very good manuscript which continues a longstanding effort by the Gottingen MPG group to fully elucidate the mechanism of - 1 frameshifting. The major new results are the demonstration of two classes of ribosome during translation of slippery sequences, the correlation of the fraction of slow ribosome with frameshift efficiency, and the demonstration that tRNA dissociation from the E- site occurs more rapidly than the pep- tRNA fluctuation period. These results will be of interest to the community of researchers focused on the study of translation mechanisms. However, the points listed below need to be addressed before the MS would be acceptable for publication in Nature Communications.  
+
+## MAJOR POINTS REQUIRING FURTHER CONSIDERATION:  
+
+1. Correlation of the fraction of slow ribosomes due to slippery sequences with frameshift efficiency. The results backing this conclusion are presented in Fig. 2f, for wt-EF-G and three variants. But the data referred to (Fig. S3) only include results for wt and one variant (Q507D). The results for the other two variants should also be shown in Fig. S3.  
+
+2. E-site occupancy during pep-tRNA fluctuation between CHI and P/P sites. The results presented on p. 7 clearly show that tRNA dissociation from the E-site occurs more rapidly than the fluctuation period, providing a "time window for pept-tRNA to switch to the -1-frame." This conclusion appears to be in conflict with the statement in the Discussion (p.11) that "frameshifting can occur both with one or two tRNAs bound, provided pept-tRNA is trapped in fluctuations between CHI and P/P." This apparent discrepancy should be further discussed.  
+
+3. Insufficient documentation of results presented in Figs. 3b,h. The claimed transition from CR to AC is insufficiently documented in Fig. 3b. Contour plots, number of traces are needed. Similar comments for apply to transitions from CR in Fig. 3h. The lack of detail in Fig. 3h does not sufficiently support the two sentences beginning at the bottom of p.6 and continuing into p.7  
+
+" After some time, - 1-frame Val-tRNAVal-Cy5 is accommodated into POST2 and then fluctuates between A/A, A/P and A/P\* states. Accommodation of Val-tRNA provides a strong evidence that after the delayed translocation the ribosome moved into the - 1-frame exposing the Val codon in the A site (Fig. 3h-i)."  
+
+## OTHER SIGNIFICANT POINTS:  
+
+1. A fuller description is needed of the significant structural differences between the CHI, hybrid and classical states than that currently provided in the Introduction. Given the importance of the CHI
+
+<--- Page Split --->
+
+state, this will aid the understanding of the paper by all but the most knowledgeable readers.  
+
+2. It is unclear when the EF-G is added in the TIRF microscopy experiments monitoring translocation (Figs 3b,e,h; 4c,e; 5b; 6b). This point should be clearly addressed in either the Figure legends or the Experimental section.  
+
+3. Fig. 3b - The reason for the drift in Cy3 fluorescence in PRE1(0-47 s) should be explained 
+4. Fig. 4 - It is not clear from the data presented that the 0.9 - 0.6 transition represents a fluctuation, as opposed to an essentially unidirectional transition from 0.9 to 0.6. Contour plots should be provided for interconversions between the 0.9, 0.6 and 0.3 FRET states 
+5. The current arrangement of Figures and Supplementary Figures require the reader to make frequent excursions from the main text to the Supplementary section. Could more of the Supplementary Figures be included in the main text?  
+
+## MINOR EDITORIAL POINTS:  
+
+1. Last line of Discussion: Delete "the" "elucidated in the future work." 
+2. Fig. 3b legend: Replace "monitored" by "monitored"  
+
+Reviewer #3 (Remarks to the Author):  
+
+In this study the authors use single- molecule FRET (smFRET) to follow the translocation of tRNALys through the ribosome in the context of a slippery mRNA sequence. By altering the placement of Cy3 and Cy5 fluorophores, they are able to follow the dynamics of EF- G catalyzed translocation relative to translocation on a non- slippery mRNA as a control. They show that translocation on a slippery sequence can proceed through two pathways. In one pathway, both A and P site tRNAs rapidly move to the P and E sites with no change in reading frame. In the other pathway the deacylated P- site tRNA rapidly translocates while the A- site peptidyl- tRNA is delayed at a late chimeric stage of translocation, where it fluctuates between chimeric (ap/P) and post- translocation (P/P) states. During this period the authors confirm that the small subunit head is swiveled, conditions where codon- anticodon pairing is destabilized and alternative pairing frames can be explored. In the case of spontaneous - 1 frameshifting, rapid release of deacylated tRNA from the E- site following translocation may facilitate - 1 pairing of the stalled peptidyl- tRNA. These insights into translocation on a slippery sequence help to explain how spontaneous - 1 frameshifting occurs.  
+
+Central to this study is the demonstration of a positive correlation of - 1 frameshifting on a slippery mRNA sequence to the fraction of peptidyl- tRNA that is delayed in translocation. This was accomplished by carrying out translocation in the presence of several mutants of EF- G containing amino acid substitutions at position 507. In the wild- type factor, glutamine at this position helps to stabilize codon- anticodon pairing of the peptidyl- tRNA. Substitutions of this amino acid are known to promote - 1 frameshifting and the authors show that these same substitutions increase the fraction of translocation that is delayed due to fluctuation between ap/P and P/P states.  
+
+The smFRET studies appear to be carefully done. However, several minor revisions are needed. 1. As regards to the biochemical assay for - 1 frameshifting, it is not specified whether peptide formation was carried out in the presence of both tRNAPhe and tRNAVal or only in the presence of the latter. By conducting this type of assay in the presence of both the 0- frame and - 1- frame incoming tRNAs, any possibility of frameshifting in the P- site can be discounted. 2. The incorporation of 0- and - 1 frame aa- tRNAs presented in Figure 3 does not add a lot to the paper and could be included in the supplemental section. 3. It is not clear from the methods section (pg 15) how the POST fMAK complex is formed without going on to give a POST fMAKK complex. This needs to be more clearly described. 4. The findings of the work are consistent with recent work that elucidated the mechanism of spontaneous +1 frameshifting. In particular, the findings that - 1 frameshifting occurs at a later stage
+
+<--- Page Split --->
+
+of the translocation reaction, involving the head domain swiveling of the small subunit, are reminiscent of the findings for spontaneous \(+1\) frameshifting as reported in PMID: 33436566 and in PMID: 34330903. Both of these papers should be cited and discussed in the present manuscript. This will give readers a broader and more comprehensive view of the background of this work.
+
+<--- Page Split --->
+
+## Answer to the REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Manuscript by Poulis et al. presents additional smFRET look on the ribosome frameshifting associated with slippery sequences. Authors use EF- G mutants, different sets of slippery and non- slippery sequences, as well as slowly hydrolyzable GTPyS, fusidic acid and spectinomycin to test previously proposed model that slow frameshifting- prone translocation mode is responsible for reduced fidelity on slippery sequences.  
+
+Manuscript lacks clear advancement over previous studies, multiple control experiments (Figure 3 and 6 with wt EF- G on slippery sequence), missing data (Supplementary Fig 3 data for Q507A and Q507N are missing) as well as clarity in data presentation (ie. what is TL in figures3 and Supplementary Fig 4?).  
+
+Reply: We politely disagree with the referee concerning the novelty and originality of our study. While previous work (reviewed in detail in the Introduction) has identified the putative link between the translocation rate and the tendency for the ribosome to frameshift, in this work we reveal the molecular mechanism on the single molecule level and additionally dissect the timing and the exact coupling between ribosome motions and frameshifting. Notably, in the previous work on programmed ribosome frameshifting slow translocation was induced by the downstream secondary structure element on the mRNA (which is not very surprising, as the ribosome encounters a hurdle that it struggles to pass). In contrast, here we study spontaneous frameshifting without such hurdles and show that an mRNA slippery sequence by itself changes the choreography of translocation, leading to frameshifting.  
+
+One of the control experiments "Figure 6 with wt EF- G on slippery sequence" requested by the referee was already present in our original text (Fig. 6a,b). We now performed the second control experiment concerning "Figure 3 with wt EF- G on slippery sequence"; the data are presented in the revised Fig. 3 and Supplementary Fig. 6 and 7. We show that also with EF- G(wt) ribosomes show slow translocation of pept- tRNA on slippery mRNA before accommodation of \(- 1\) - frame Val- tRNAVal (Fig. 3e- h). However, since the majority of ribosomes stays in 0- frame with EF- G(wt) (Fig. 3a- d), \(- 1\) - frame Val- tRNAVal is predominantly rejected (Fig. 3a- d). We never observed accommodation of \(- 1\) - frame Val- tRNAVal after fast translocation of pept- tRNA, which makes prolonged fluctuations between CHI and P/P states an essential determinant of spontaneous frameshifting. We worked extensively on the data presentation in Fig. 3 and Supplementary Fig. 4, 5, 6, and 7. We simplified the cartoons, explained all abbreviations in the figure legends and replaced the term "TL" (translocation) for "CHI state", which was explained in the Introduction. The revised versions of Fig. 3 and Supplementary Fig.
+
+<--- Page Split --->
+
+4, 5, 6, and 7 show that slow translocation via CHI and P/P states is a major determinant of spontaneous ribosome frameshifting.  
+
+We also added the data on the frameshifting efficiency by EF- G(Q507A) and EF- G(Q507N) mutants in Supplementary Fig. 3.  
+
+The proposed model in Figure 7b is based on EF- G mutant Q507D data and increased population of CHI and P/P states which are almost identical in GTPyS conditions (Fig 5), and on contrary to authors comments not really similar to spectinomycin data (in Supplementary Fig 8), which blurs authors conclusions. For these reasons I suggest that manuscript should not be published in this form.  
+
+Reply: We are surprised by this comment of the referee, because our transition frequency analysis clearly shows differences in translocation in the presence of EF- G(Q507D) and EF- G—GTPyS (compare Fig. 2e and Fig. 5d). EF- G(Q507D) fluctuates predominantly between FRET 0.4 and 0.2, whereas EF- G—GTPyS fluctuates between FRET 0.6 and 0.4, i.e. EF- G is stalled at a different step of translocation. The frameshifting efficiency is high during translocation with EF- G(Q507D) and pept- tRNA fluctuates between CHI and P/P states (Fig. 2e and Supplemetary Fig. 3). In contrast, the frameshifting efficiency during translocation by EF- G—GTPyS is low and tRNAs fluctuate predominantly between CHI and A/P\* states (Fig. 5d). These results show that frameshifting occurs while ribosomes sample CHI and P/P states, whereas sampling of earlier intermediate states does not lead to frameshifting. Presumably, the referee is confused by the fact that contour plots with EF- G(Q507D) and EF- G—GTPyS look similar on first sight (Fig. 2d and Fig. 5c). This is because these plots are dominated by the end levels (the P/P state) and were therefore further analyzed with respect to the transition frequency (Fig. 2e). In this analysis, we counted the number of transitions between FRET 0.6 (A/A, A/P\*), FRET 0.4 (CHI) and 0.2 (P/P) states after EF- G binding to PRE complexes using idealized traces derived from the HMM fit of the FRET time course (see Methods). This information reveals in which step of the translocation pathway the movement of pept- tRNA is stalled (see also Adio et al. 2015). We added information in the Method section to better explain how the transition frequencies were derived.  
+
+The translocation pathway by EF- G—GTPyS is similar to that observed with EF- G(wt)—GTP in the presence of Spc (Fig. 5 and Supplementary Fig.11), as translocation is stalled at an early stage, consistent with other existing literature (Rundlet et al., Nature 2021, Belardinelli et al., RNA 2019 and Adio et al., Nat Commun 2015). In both cases, pept- tRNA is stalled in fluctuations between A/P, A/P\* and CHI states and transitions between CHI and P/P are practically absent. Translocation of deacylated tRNA is slow and occurs at the same rate as the translocation of pept- tRNA. We observed exactly the opposite behavior by EF- G(Q507D) (Fig. 2e, Fig.4e,f), where pept- tRNA is trapped in fluctuations between CHI and P/P, and translocation of deacylated tRNA is fast.
+
+<--- Page Split --->
+
+Our model (Fig. 7) shows the uncoupled movement of pept- and deacylated tRNA observed with the frameshifting- prone EF- G(Q507D) mutant. EF- G—GTPyS and Spc show only background levels of frameshifting (Fig. 5a), which is why these conditions serve as relevant controls but should not enter the main model.  
+
+Reviewer #2 (Remarks to the Author):  
+
+This is a very good manuscript which continues a longstanding effort by the Gottingen MPG group to fully elucidate the mechanism of - 1 frameshifting. The major new results are the demonstration of two classes of ribosome during translation of slippery sequences, the correlation of the fraction of slow ribosome with frameshift efficiency, and the demonstration that tRNA dissociation from the E- site occurs more rapidly than the pep- tRNA fluctuation period. These results will be of interest to the community of researchers focused on the study of translation mechanisms.  
+
+However, the points listed below need to be addressed before the MS would be acceptable for publication in Nature Communications.  
+
+## MAJOR POINTS REQUIRING FURTHER CONSIDERATION:  
+
+1. Correlation of the fraction of slow ribosomes due to slippery sequences with frameshift efficiency. The results backing this conclusion are presented in Fig. 2f, for wt-EF-G and three variants. But the data referred to (Fig. S3) only include results for wt and one variant (Q507D). The results for the other two variants should also be shown in Fig. S3.  
+
+Reply: We added the data on the frameshifting efficiency by EF- G(Q507A) and EF- G(Q507N) mutants in Supplementary Fig. 3.  
+
+2. E-site occupancy during pep-tRNA fluctuation between CHI and P/P sites. The results presented on p. 7 clearly show that tRNA dissociation from the E-site occurs more rapidly than the fluctuation period, providing a "time window for pept-tRNA to switch to the -1-frame." This conclusion appears to be in conflict with the statement in the Discussion (p.11) that "frameshifting can occur both with one or two tRNAs bound, provided pept-tRNA is trapped in fluctuations between CHI and P/P." This apparent discrepancy should be further discussed.  
+
+Reply: We show that upon spontaneous - 1- frameshifting, deacylated tRNA dissociates rapidly from the E site while pept- tRNA is still in the process of (slow) translocation (Fig. 1, 3 and 4). This indicates a one- tRNA slippage mechanism because the E- site codon is unoccupied while pept- tRNA has not reached the POST state. In contrast, during - 1PRF (described in the discussion on p.12) the translocation of deacylated and pept- tRNA is coupled and occurs at similar rate (Caliskan et al., Cell 2014). Hence, depending on whether frameshifting occurs
+
+<--- Page Split --->
+
+spontaneously or in a programmed manner, it can follow a one- or two- tRNA slippage mechanism. We modified the text (see top of p. 12) to point out that the two- tRNA slippage corresponds to \(- 1\mathrm{PRF}\) .  
+
+3. Insufficient documentation of results presented in Figs. 3b,h. The claimed transition from CR to AC is insufficiently documented in Fig. 3b. Contour plots, number of traces are needed. Similar comments for apply to transitions from CR in Fig. 3h. The lack of detail in Fig. 3h does not sufficiently support the two sentences beginning at the bottom of p.6 and continuing into p.7 "After some time, \(-1\) -frame Val-tRNAVal-Cy5 is accommodated into POST2 and then fluctuates between A/A, A/P and A/P\* states. Accommodation of Val-tRNA provides a strong evidence that after the delayed translocation the ribosome moved into the \(-1\) -frame exposing the Val codon in the A site (Fig. 3h-i)."  
+
+Reply: We revised Fig. 3 and the corresponding figure legend. We explain the meaning of the abbreviation "CR" (codon recognition) and show that upon accommodation in the A site, tRNAPhe and tRNAVal fluctuate between classic (A/A) and hybrid (A/P and A/P\*) conformations in a similar manner as pept-tRNALys (Supplementary Fig. 1). We included n=number of traces in the figure legend and compiled FRET signals reporting on the interaction of aa-tRNAs with POST2 complexes into contour plots (Fig. 3d,h,l and Supplementary Fig. 5d,h). The dwell time analysis of these FRET signals is now presented in the new Supplementary Fig. 6. The dissociation rates (koff) differ dramatically for the cognate aa-tRNA (which is accommodated) and the near-cognate tRNA (which is rapidly is rejected). Dissociation of \(-1\) -frame tRNAVal from PRE2 complexes formed after slow translocation on slippery mRNA is as slow as of the cognate O-frame tRNAPhe on PRE2 complexes formed on non-slippery mRNA. This provides a strong evidence in favor of our conclusion that slowly-translociating ribosomes have moved into the \(-1\) -frame exposing the Val codon in the A site.  
+
+Please note that we also unified the experiment schemes in Fig. 3a, e, i. In all experiments, PRE1 complexes are immobilized and EF- G (wt or mutant) is added together with EF- Tu- aa- tRNA- GTP complex to induce tRNA accommodation and subsequent translocation.  
+
+## OTHER SIGNIFICANT POINTS:  
+
+1. A fuller description is needed of the significant structural differences between the CHI, hybrid and classical states than that currently provided in the Introduction. Given the importance of the CHI state, this will aid the understanding of the paper by all but the most knowledgeable readers.  
+
+Reply: We have added the description of the dynamics of PRE translocation complexes including explanation of tRNA conformations in hybrid and classical states. We have further
+
+<--- Page Split --->
+
+explained that the CHI state is a transient intermediate state of translocation, which rapidly resolves into the POST state.  
+
+2. It is unclear when the EF-G is added in the TIRF microscopy experiments monitoring translocation (Figs 3b,e,h; 4c,e; 5b; 6b). This point should be clearly addressed in either the Figure legends or the Experimental section.  
+
+Reply: In the experiments described in Fig. 3, EF-G was added to immobilized PRE1 complexes together with EF-Tu-GTP-aa-tRNA complex. The components were added with the imaging buffer approximately 10 s prior to imaging. We added additional description in the figure legends and also in the Methods section.  
+
+In the experiments described in Fig. 4, 5, 6 EF-G was added to immobilized PRE complexes approximately 10 s before imaging. This is now clearly stated in the Methods section.  
+
+Please note that the vertical line in Fig. 6 represents the synchronization point of the smFRET traces. This is now explained in the figure legend.  
+
+3. Fig. 3b – The reason for the drift in Cy3 fluorescence in PRE1(0-47 s) should be explained.  
+
+Reply: "The drift" in the Cy3 fluorescence signal of the example trace shown in our previous Fig. 3b represented noise in the data and did not correspond to tRNA movement. We replaced the trace for a different example trace with stable Cy3 fluorescence.  
+
+4. Fig. 4 – It is not clear from the data presented that the 0.9 – 0.6 transition represents a fluctuation, as opposed to an essentially unidirectional transition from 0.9 to 0.6. Contour plots should be provided for interconversions between the 0.9, 0.6 and 0.3 FRET states  
+
+Reply: We changed the text (p. 7) to clarify that the transitions between 0.9 and 0.6 states are not unidirectional but much less frequent than fluctuations of pept-tRNA on PRE complexes. Additionally, we calculated the transition frequency of deacylated tRNA (i.e. the average number of transitions between 0.6 and 0.9 FRET per trace) and compared it with the transition frequency of pept-tRNA on PRE complexes. Values differ by about one order of magnitude (0.4 vs 5.4 transitions, Supplementary Fig. 1c and 8f), further supporting the point that fluctuations of deacylated tRNA on PRE complexes are relatively slow. Transitions into the POST (0.3 FRET) state are irreversible (Supplementary Fig. 9d) and occur from the PRE state with either FRET 0.9 or 0.6, which is shown in the contour plot (Fig. 4d,e).  
+
+5. The current arrangement of Figures and Supplementary Figures require the reader to make frequent excursions from the main text to the Supplementary section. Could more of the
+
+<--- Page Split --->
+
+Supplementary Figures be included in the main text?  
+
+Reply: In the current version of the manuscript, we describe all key experiments and the model in the main text and present the control experiments in the supplement. We absolutely see the point made by the referee, but the main Figs are already heavy on panels and information. We fear that including even more material into the main text will lead to overload and confusion. We are therefore reluctant to change the principle sequence of figures in the main text.  
+
+## MINOR EDITORIAL POINTS:  
+
+1. Last line of Discussion: Delete "the" "elucidated in the future work."  
+
+Corrected in the revised manuscript.  
+
+2. Fig. 3b legend: Replace "monitored" by "monitored" This is corrected in the revised manuscript.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In this study the authors use single- molecule FRET (smFRET) to follow the translocation of tRNALys through the ribosome in the context of a slippery mRNA sequence. By altering the placement of Cy3 and Cy5 fluorophores, they are able to follow the dynamics of EF- G catalyzed translocation relative to translocation on a non- slippery mRNA as a control. They show that translocation on a slippery sequence can proceed through two pathways. In one pathway, both A and P site tRNAs rapidly move to the P and E sites with no change in reading frame. In the other pathway the deacylated P- site tRNA rapidly translocates while the A- site peptidyl- tRNA is delayed at a late chimeric stage of translocation, where it fluctuates between chimeric (ap/P) and post- translocation (P/P) states. During this period the authors confirm that the small subunit head is swiveled, conditions where codon- anticodon pairing is destabilized and alternative pairing frames can be explored. In the case of spontaneous \(- 1\) frameshifting, rapid release of deacylated tRNA from the E- site following translocation may facilitate \(- 1\) pairing of the stalled peptidyl- tRNA. These insights into translocation on a slippery sequence help to explain how spontaneous \(- 1\) frameshifting occurs.  
+
+Central to this study is the demonstration of a positive correlation of \(- 1\) frameshifting on a slippery mRNA sequence to the fraction of peptidyl- tRNA that is delayed in translocation. This was accomplished by carrying out translocation in the presence of several mutants of EF- G containing amino acid substitutions at position 507. In the wild- type factor, glutamine at this position helps to stabilize codon- anticodon pairing of the peptidyl- tRNA. Substitutions of this
+
+<--- Page Split --->
+
+amino acid are known to promote \(- 1\) frameshifting and the authors show that these same substitutions increase the fraction of translocation that is delayed due to fluctuation between ap/P and P/P states.  
+
+The smFRET studies appear to be carefully done. However, several minor revisions are needed.  
+
+1. As regards to the biochemical assay for \(-1\) frameshifting, it is not specified whether peptide formation was carried out in the presence of both tRNAPhe and tRNAVal or only in the presence of the latter. By conducting this type of assay in the presence of both the O-frame and \(-1\) -frame incoming tRNAs, any possibility of frameshifting in the P-site can be discounted.  
+
+Reply: The biochemical assays to quantify frameshifting efficiency (Fig. 2f, 5a, and Supplementary Fig.3) were conducted in the presence of equal concentrations of the O-frame Phe- tRNAPhe and \(-1\) - frame Val- tRNAVal (5- fold excess over 70S IC). We clarified the description of the experimental procedure in the Methods section (p. 16).  
+
+2. The incorporation of O- and \(-1\) frame aa-tRNAs presented in Figure 3 does not add a lot to the paper and could be included in the supplemental section.  
+
+Reply: We feel that we cannot omit Fig. 3 from the main text figures because Fig 2 and Fig 3 represent two different aspects of pept- tRNA\(^{Lys}\) translocation. While Fig 2 reveals that pept- tRNA samples distinct intermediate states of translocation (CHI and POST) during spontaneous frameshifting, experiments presented in Fig. 3 verify that slow translocation via CHI states indeed precedes the accommodation of \(-1\) -frame-tRNA. Without this direct evidence, the correlation between slow translocation via distinct CHI states and spontaneous frameshifting would remain a conjecture.  
+
+3. It is not clear from the methods section (pg 15) how the POST fMAK complex is formed without going on to give a POST fMAKK complex. This needs to be more clearly described.  
+
+Reply: We revised the corresponding section in Methods (p. 17). Now it contains all the essential steps for the sample preparation.  
+
+4. The findings of the work are consistent with recent work that elucidated the mechanism of spontaneous \(+1\) frameshifting. In particular, the findings that \(-1\) frameshifting occurs at a later stage of the translocation reaction, involving the head domain swiveling of the small subunit, are reminiscent of the findings for spontaneous \(+1\) frameshifting as reported in PMID: 33436566 and in PMID: 34330903. Both of these papers should be cited and discussed in the
+
+<--- Page Split --->
+
+present manuscript. This will give readers a broader and more comprehensive view of the background of this work.  
+
+Reply: We agree that comparison of our results on spontaneous –1- frameshifting with spontaneous +1- frameshifting gives a broader view on the background of our study. We added a new paragraph on this subject into the Discussion section highlighting mechanistic similarities and differences between the two pathways.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The manuscript by Poulis et al., has improved in the representation of data and additional controls are included. The discussion of the data is also clarified and additional explanations are provided.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have responded well to my concerns. A few points still remain, as listed below.  
+
+Significant points:  
+
+1. Although a CR state is a plausible intermediate on the way to the PRE2 complex, the results interpreted as demonstrating its formation (Figs.3 h); Supplementary Fig. 5d) are not compelling, in part due to the small number of traces obtained. As a result, the current text (bottom, p. 6) is overstated, and should be reworded more conservatively.  
+
+2. In a similar vein, on p.7, line 3 from the bottom, the word "show" should be replaced by "suggest".  
+
+Minor points:  
+
+1. Supplementary Figure 3 legend needs to be edited to reflect the changes in part (a)  
+
+2. Three edits to p. 12, para2  
+
+- line 2: post-transcriptional  
+
+- line 8: replace "any type of frameshifting" with "-1 or +1 frameshifting"  
+
+-line 10: stalls  
+
+Reviewer #3 (Remarks to the Author):  
+
+The revisions by the authors have addressed all of my concerns. more comprehensive view of the background of this work.
+
+<--- Page Split --->
+
+## Point-by-point reply  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The manuscript by Poulis et al., has improved in the representation of data and additional controls are included. The discussion of the data is also clarified and additional explanations are provided.  
+
+Reply: We appreciate the positive feedback by Reviewer #1 and thank him/her for the helpful remarks on the initial submission.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have responded well to my concerns. A few points still remain, as listed below.  
+
+Significant points:  
+
+1. Although a CR state is a plausible intermediate on the way to the PRE2 complex, the results interpreted as demonstrating its formation (Figs.3 h); Supplementary Fig. 5d) are not compelling, in part due to the small number of traces obtained. As a result, the current text (bottom, p. 6) is overstated, and should be reworded more conservatively.  
+
+Reply: We replaced "CR" for codon recognition for "RB" for ribosome binding state, where Phe-tRNA<sup>Phe</sup>- Cy5 reads the codon in the A site according to the step assignment of the earlier work (Geggier et al., 2010).  
+
+2. In a similar vein, on p.7, line 3 from the bottom, the word "show" should be replaced by "suggest". Reply: Done  
+
+Minor points:  
+
+1. Supplementary Figure 3 legend needs to be edited to reflect the changes in part (a) Reply: We changed the legend of Supplementary Fig. 3.  
+
+2. Three edits to p. 12, para2  
+
+- line 2: post-transcriptional  
+
+- line 8: replace "any type of frameshifting" with "-1 or +1 frameshifting"  
+
+-line 10: stalls  
+
+Reply: We corrected the typos in the text.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The revisions by the authors have addressed all of my concerns.  
+
+Reply: We appreciate the positive response to the revised manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42d4699c4e65ed95775b859ea7b95f01d6cc639156a8bd344e894404ab1add48/supplementary_0_Peer Review File__42d4699c4e65ed95775b859ea7b95f01d6cc639156a8bd344e894404ab1add48_det.mmd

@@ -0,0 +1,376 @@
+<|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, 162, 890, 219]]<|/det|>
+Altered tRNA dynamics during translocation on slippery mRNA as determinant of spontaneous ribosome frameshifting  
+
+<|ref|>image<|/ref|><|det|>[[95, 732, 261, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[271, 732, 879, 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, 89, 305, 104]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 129, 404, 144]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 159, 880, 337]]<|/det|>
+Reviewer #1 (Remarks to the Author):Manuscript by Poulis et al. presents additional smFRET look on the ribosome frameshifting associated with slippery sequences. Authors use EF- G mutants, different sets of slippery and non- slippery sequences, as well as slowly hydrolyzable GTPyS, fusidic acid and spectinomycin to test previously proposed model that slow frameshifting- prone translocation mode is responsible for reduced fidelity on slippery sequences. Manuscript lacks clear advancement over previous studies, multiple control experiments (Figure 3 and 6 with wt EF- G on slippery sequence), missing data (Supplementary Fig 3 data for Q507A and Q507N are missing) as well as clarity in data presentation (ie. what is TL in figures3 and Supplementary Fig 4?). The proposed model in Figure 7b is based on EF- G mutant Q507D data and increased population of CHI and P/P states which are almost identical in GTPyS conditions (Fig 5), and on contrary to authors comments not really similar to spectinomycin data (in Supplementary Fig 8), which blurs authors conclusions. For these reasons I suggest that manuscript should not be published in this form.  
+
+<|ref|>text<|/ref|><|det|>[[116, 366, 404, 381]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 396, 880, 515]]<|/det|>
+This is a very good manuscript which continues a longstanding effort by the Gottingen MPG group to fully elucidate the mechanism of - 1 frameshifting. The major new results are the demonstration of two classes of ribosome during translation of slippery sequences, the correlation of the fraction of slow ribosome with frameshift efficiency, and the demonstration that tRNA dissociation from the E- site occurs more rapidly than the pep- tRNA fluctuation period. These results will be of interest to the community of researchers focused on the study of translation mechanisms. However, the points listed below need to be addressed before the MS would be acceptable for publication in Nature Communications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 530, 535, 545]]<|/det|>
+## MAJOR POINTS REQUIRING FURTHER CONSIDERATION:  
+
+<|ref|>text<|/ref|><|det|>[[116, 546, 860, 605]]<|/det|>
+1. Correlation of the fraction of slow ribosomes due to slippery sequences with frameshift efficiency. The results backing this conclusion are presented in Fig. 2f, for wt-EF-G and three variants. But the data referred to (Fig. S3) only include results for wt and one variant (Q507D). The results for the other two variants should also be shown in Fig. S3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 620, 877, 710]]<|/det|>
+2. E-site occupancy during pep-tRNA fluctuation between CHI and P/P sites. The results presented on p. 7 clearly show that tRNA dissociation from the E-site occurs more rapidly than the fluctuation period, providing a "time window for pept-tRNA to switch to the -1-frame." This conclusion appears to be in conflict with the statement in the Discussion (p.11) that "frameshifting can occur both with one or two tRNAs bound, provided pept-tRNA is trapped in fluctuations between CHI and P/P." This apparent discrepancy should be further discussed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 725, 880, 777]]<|/det|>
+3. Insufficient documentation of results presented in Figs. 3b,h. The claimed transition from CR to AC is insufficiently documented in Fig. 3b. Contour plots, number of traces are needed. Similar comments for apply to transitions from CR in Fig. 3h. The lack of detail in Fig. 3h does not sufficiently support the two sentences beginning at the bottom of p.6 and continuing into p.7  
+
+<|ref|>text<|/ref|><|det|>[[115, 785, 872, 845]]<|/det|>
+" After some time, - 1-frame Val-tRNAVal-Cy5 is accommodated into POST2 and then fluctuates between A/A, A/P and A/P\* states. Accommodation of Val-tRNA provides a strong evidence that after the delayed translocation the ribosome moved into the - 1-frame exposing the Val codon in the A site (Fig. 3h-i)."  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 859, 340, 873]]<|/det|>
+## OTHER SIGNIFICANT POINTS:  
+
+<|ref|>text<|/ref|><|det|>[[115, 875, 863, 904]]<|/det|>
+1. A fuller description is needed of the significant structural differences between the CHI, hybrid and classical states than that currently provided in the Introduction. Given the importance of the CHI
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 808, 105]]<|/det|>
+state, this will aid the understanding of the paper by all but the most knowledgeable readers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 106, 870, 150]]<|/det|>
+2. It is unclear when the EF-G is added in the TIRF microscopy experiments monitoring translocation (Figs 3b,e,h; 4c,e; 5b; 6b). This point should be clearly addressed in either the Figure legends or the Experimental section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 151, 881, 250]]<|/det|>
+3. Fig. 3b - The reason for the drift in Cy3 fluorescence in PRE1(0-47 s) should be explained 
+4. Fig. 4 - It is not clear from the data presented that the 0.9 - 0.6 transition represents a fluctuation, as opposed to an essentially unidirectional transition from 0.9 to 0.6. Contour plots should be provided for interconversions between the 0.9, 0.6 and 0.3 FRET states 
+5. The current arrangement of Figures and Supplementary Figures require the reader to make frequent excursions from the main text to the Supplementary section. Could more of the Supplementary Figures be included in the main text?  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 267, 323, 281]]<|/det|>
+## MINOR EDITORIAL POINTS:  
+
+<|ref|>text<|/ref|><|det|>[[115, 282, 641, 312]]<|/det|>
+1. Last line of Discussion: Delete "the" "elucidated in the future work." 
+2. Fig. 3b legend: Replace "monitored" by "monitored"  
+
+<|ref|>text<|/ref|><|det|>[[116, 373, 404, 388]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 401, 879, 596]]<|/det|>
+In this study the authors use single- molecule FRET (smFRET) to follow the translocation of tRNALys through the ribosome in the context of a slippery mRNA sequence. By altering the placement of Cy3 and Cy5 fluorophores, they are able to follow the dynamics of EF- G catalyzed translocation relative to translocation on a non- slippery mRNA as a control. They show that translocation on a slippery sequence can proceed through two pathways. In one pathway, both A and P site tRNAs rapidly move to the P and E sites with no change in reading frame. In the other pathway the deacylated P- site tRNA rapidly translocates while the A- site peptidyl- tRNA is delayed at a late chimeric stage of translocation, where it fluctuates between chimeric (ap/P) and post- translocation (P/P) states. During this period the authors confirm that the small subunit head is swiveled, conditions where codon- anticodon pairing is destabilized and alternative pairing frames can be explored. In the case of spontaneous - 1 frameshifting, rapid release of deacylated tRNA from the E- site following translocation may facilitate - 1 pairing of the stalled peptidyl- tRNA. These insights into translocation on a slippery sequence help to explain how spontaneous - 1 frameshifting occurs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 610, 875, 715]]<|/det|>
+Central to this study is the demonstration of a positive correlation of - 1 frameshifting on a slippery mRNA sequence to the fraction of peptidyl- tRNA that is delayed in translocation. This was accomplished by carrying out translocation in the presence of several mutants of EF- G containing amino acid substitutions at position 507. In the wild- type factor, glutamine at this position helps to stabilize codon- anticodon pairing of the peptidyl- tRNA. Substitutions of this amino acid are known to promote - 1 frameshifting and the authors show that these same substitutions increase the fraction of translocation that is delayed due to fluctuation between ap/P and P/P states.  
+
+<|ref|>text<|/ref|><|det|>[[115, 730, 880, 895]]<|/det|>
+The smFRET studies appear to be carefully done. However, several minor revisions are needed. 1. As regards to the biochemical assay for - 1 frameshifting, it is not specified whether peptide formation was carried out in the presence of both tRNAPhe and tRNAVal or only in the presence of the latter. By conducting this type of assay in the presence of both the 0- frame and - 1- frame incoming tRNAs, any possibility of frameshifting in the P- site can be discounted. 2. The incorporation of 0- and - 1 frame aa- tRNAs presented in Figure 3 does not add a lot to the paper and could be included in the supplemental section. 3. It is not clear from the methods section (pg 15) how the POST fMAK complex is formed without going on to give a POST fMAKK complex. This needs to be more clearly described. 4. The findings of the work are consistent with recent work that elucidated the mechanism of spontaneous +1 frameshifting. In particular, the findings that - 1 frameshifting occurs at a later stage
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 871, 150]]<|/det|>
+of the translocation reaction, involving the head domain swiveling of the small subunit, are reminiscent of the findings for spontaneous \(+1\) frameshifting as reported in PMID: 33436566 and in PMID: 34330903. Both of these papers should be cited and discussed in the present manuscript. This will give readers a broader and more comprehensive view of the background of this work.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 150, 426, 168]]<|/det|>
+## Answer to the REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 190, 420, 207]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 229, 880, 329]]<|/det|>
+Manuscript by Poulis et al. presents additional smFRET look on the ribosome frameshifting associated with slippery sequences. Authors use EF- G mutants, different sets of slippery and non- slippery sequences, as well as slowly hydrolyzable GTPyS, fusidic acid and spectinomycin to test previously proposed model that slow frameshifting- prone translocation mode is responsible for reduced fidelity on slippery sequences.  
+
+<|ref|>text<|/ref|><|det|>[[115, 339, 874, 418]]<|/det|>
+Manuscript lacks clear advancement over previous studies, multiple control experiments (Figure 3 and 6 with wt EF- G on slippery sequence), missing data (Supplementary Fig 3 data for Q507A and Q507N are missing) as well as clarity in data presentation (ie. what is TL in figures3 and Supplementary Fig 4?).  
+
+<|ref|>text<|/ref|><|det|>[[114, 428, 880, 627]]<|/det|>
+Reply: We politely disagree with the referee concerning the novelty and originality of our study. While previous work (reviewed in detail in the Introduction) has identified the putative link between the translocation rate and the tendency for the ribosome to frameshift, in this work we reveal the molecular mechanism on the single molecule level and additionally dissect the timing and the exact coupling between ribosome motions and frameshifting. Notably, in the previous work on programmed ribosome frameshifting slow translocation was induced by the downstream secondary structure element on the mRNA (which is not very surprising, as the ribosome encounters a hurdle that it struggles to pass). In contrast, here we study spontaneous frameshifting without such hurdles and show that an mRNA slippery sequence by itself changes the choreography of translocation, leading to frameshifting.  
+
+<|ref|>text<|/ref|><|det|>[[114, 638, 875, 898]]<|/det|>
+One of the control experiments "Figure 6 with wt EF- G on slippery sequence" requested by the referee was already present in our original text (Fig. 6a,b). We now performed the second control experiment concerning "Figure 3 with wt EF- G on slippery sequence"; the data are presented in the revised Fig. 3 and Supplementary Fig. 6 and 7. We show that also with EF- G(wt) ribosomes show slow translocation of pept- tRNA on slippery mRNA before accommodation of \(- 1\) - frame Val- tRNAVal (Fig. 3e- h). However, since the majority of ribosomes stays in 0- frame with EF- G(wt) (Fig. 3a- d), \(- 1\) - frame Val- tRNAVal is predominantly rejected (Fig. 3a- d). We never observed accommodation of \(- 1\) - frame Val- tRNAVal after fast translocation of pept- tRNA, which makes prolonged fluctuations between CHI and P/P states an essential determinant of spontaneous frameshifting. We worked extensively on the data presentation in Fig. 3 and Supplementary Fig. 4, 5, 6, and 7. We simplified the cartoons, explained all abbreviations in the figure legends and replaced the term "TL" (translocation) for "CHI state", which was explained in the Introduction. The revised versions of Fig. 3 and Supplementary Fig.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 836, 128]]<|/det|>
+4, 5, 6, and 7 show that slow translocation via CHI and P/P states is a major determinant of spontaneous ribosome frameshifting.  
+
+<|ref|>text<|/ref|><|det|>[[115, 140, 833, 178]]<|/det|>
+We also added the data on the frameshifting efficiency by EF- G(Q507A) and EF- G(Q507N) mutants in Supplementary Fig. 3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 189, 870, 288]]<|/det|>
+The proposed model in Figure 7b is based on EF- G mutant Q507D data and increased population of CHI and P/P states which are almost identical in GTPyS conditions (Fig 5), and on contrary to authors comments not really similar to spectinomycin data (in Supplementary Fig 8), which blurs authors conclusions. For these reasons I suggest that manuscript should not be published in this form.  
+
+<|ref|>text<|/ref|><|det|>[[113, 319, 880, 697]]<|/det|>
+Reply: We are surprised by this comment of the referee, because our transition frequency analysis clearly shows differences in translocation in the presence of EF- G(Q507D) and EF- G—GTPyS (compare Fig. 2e and Fig. 5d). EF- G(Q507D) fluctuates predominantly between FRET 0.4 and 0.2, whereas EF- G—GTPyS fluctuates between FRET 0.6 and 0.4, i.e. EF- G is stalled at a different step of translocation. The frameshifting efficiency is high during translocation with EF- G(Q507D) and pept- tRNA fluctuates between CHI and P/P states (Fig. 2e and Supplemetary Fig. 3). In contrast, the frameshifting efficiency during translocation by EF- G—GTPyS is low and tRNAs fluctuate predominantly between CHI and A/P\* states (Fig. 5d). These results show that frameshifting occurs while ribosomes sample CHI and P/P states, whereas sampling of earlier intermediate states does not lead to frameshifting. Presumably, the referee is confused by the fact that contour plots with EF- G(Q507D) and EF- G—GTPyS look similar on first sight (Fig. 2d and Fig. 5c). This is because these plots are dominated by the end levels (the P/P state) and were therefore further analyzed with respect to the transition frequency (Fig. 2e). In this analysis, we counted the number of transitions between FRET 0.6 (A/A, A/P\*), FRET 0.4 (CHI) and 0.2 (P/P) states after EF- G binding to PRE complexes using idealized traces derived from the HMM fit of the FRET time course (see Methods). This information reveals in which step of the translocation pathway the movement of pept- tRNA is stalled (see also Adio et al. 2015). We added information in the Method section to better explain how the transition frequencies were derived.  
+
+<|ref|>text<|/ref|><|det|>[[114, 708, 883, 886]]<|/det|>
+The translocation pathway by EF- G—GTPyS is similar to that observed with EF- G(wt)—GTP in the presence of Spc (Fig. 5 and Supplementary Fig.11), as translocation is stalled at an early stage, consistent with other existing literature (Rundlet et al., Nature 2021, Belardinelli et al., RNA 2019 and Adio et al., Nat Commun 2015). In both cases, pept- tRNA is stalled in fluctuations between A/P, A/P\* and CHI states and transitions between CHI and P/P are practically absent. Translocation of deacylated tRNA is slow and occurs at the same rate as the translocation of pept- tRNA. We observed exactly the opposite behavior by EF- G(Q507D) (Fig. 2e, Fig.4e,f), where pept- tRNA is trapped in fluctuations between CHI and P/P, and translocation of deacylated tRNA is fast.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 849, 168]]<|/det|>
+Our model (Fig. 7) shows the uncoupled movement of pept- and deacylated tRNA observed with the frameshifting- prone EF- G(Q507D) mutant. EF- G—GTPyS and Spc show only background levels of frameshifting (Fig. 5a), which is why these conditions serve as relevant controls but should not enter the main model.  
+
+<|ref|>text<|/ref|><|det|>[[115, 190, 419, 208]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 229, 870, 370]]<|/det|>
+This is a very good manuscript which continues a longstanding effort by the Gottingen MPG group to fully elucidate the mechanism of - 1 frameshifting. The major new results are the demonstration of two classes of ribosome during translation of slippery sequences, the correlation of the fraction of slow ribosome with frameshift efficiency, and the demonstration that tRNA dissociation from the E- site occurs more rapidly than the pep- tRNA fluctuation period. These results will be of interest to the community of researchers focused on the study of translation mechanisms.  
+
+<|ref|>text<|/ref|><|det|>[[115, 370, 872, 408]]<|/det|>
+However, the points listed below need to be addressed before the MS would be acceptable for publication in Nature Communications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 429, 558, 447]]<|/det|>
+## MAJOR POINTS REQUIRING FURTHER CONSIDERATION:  
+
+<|ref|>text<|/ref|><|det|>[[115, 450, 866, 529]]<|/det|>
+1. Correlation of the fraction of slow ribosomes due to slippery sequences with frameshift efficiency. The results backing this conclusion are presented in Fig. 2f, for wt-EF-G and three variants. But the data referred to (Fig. S3) only include results for wt and one variant (Q507D). The results for the other two variants should also be shown in Fig. S3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 559, 848, 597]]<|/det|>
+Reply: We added the data on the frameshifting efficiency by EF- G(Q507A) and EF- G(Q507N) mutants in Supplementary Fig. 3.  
+
+<|ref|>text<|/ref|><|det|>[[114, 628, 881, 747]]<|/det|>
+2. E-site occupancy during pep-tRNA fluctuation between CHI and P/P sites. The results presented on p. 7 clearly show that tRNA dissociation from the E-site occurs more rapidly than the fluctuation period, providing a "time window for pept-tRNA to switch to the -1-frame." This conclusion appears to be in conflict with the statement in the Discussion (p.11) that "frameshifting can occur both with one or two tRNAs bound, provided pept-tRNA is trapped in fluctuations between CHI and P/P." This apparent discrepancy should be further discussed.  
+
+<|ref|>text<|/ref|><|det|>[[114, 778, 866, 896]]<|/det|>
+Reply: We show that upon spontaneous - 1- frameshifting, deacylated tRNA dissociates rapidly from the E site while pept- tRNA is still in the process of (slow) translocation (Fig. 1, 3 and 4). This indicates a one- tRNA slippage mechanism because the E- site codon is unoccupied while pept- tRNA has not reached the POST state. In contrast, during - 1PRF (described in the discussion on p.12) the translocation of deacylated and pept- tRNA is coupled and occurs at similar rate (Caliskan et al., Cell 2014). Hence, depending on whether frameshifting occurs
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 850, 148]]<|/det|>
+spontaneously or in a programmed manner, it can follow a one- or two- tRNA slippage mechanism. We modified the text (see top of p. 12) to point out that the two- tRNA slippage corresponds to \(- 1\mathrm{PRF}\) .  
+
+<|ref|>text<|/ref|><|det|>[[113, 179, 875, 357]]<|/det|>
+3. Insufficient documentation of results presented in Figs. 3b,h. The claimed transition from CR to AC is insufficiently documented in Fig. 3b. Contour plots, number of traces are needed. Similar comments for apply to transitions from CR in Fig. 3h. The lack of detail in Fig. 3h does not sufficiently support the two sentences beginning at the bottom of p.6 and continuing into p.7 "After some time, \(-1\) -frame Val-tRNAVal-Cy5 is accommodated into POST2 and then fluctuates between A/A, A/P and A/P\* states. Accommodation of Val-tRNA provides a strong evidence that after the delayed translocation the ribosome moved into the \(-1\) -frame exposing the Val codon in the A site (Fig. 3h-i)."  
+
+<|ref|>text<|/ref|><|det|>[[113, 388, 880, 647]]<|/det|>
+Reply: We revised Fig. 3 and the corresponding figure legend. We explain the meaning of the abbreviation "CR" (codon recognition) and show that upon accommodation in the A site, tRNAPhe and tRNAVal fluctuate between classic (A/A) and hybrid (A/P and A/P\*) conformations in a similar manner as pept-tRNALys (Supplementary Fig. 1). We included n=number of traces in the figure legend and compiled FRET signals reporting on the interaction of aa-tRNAs with POST2 complexes into contour plots (Fig. 3d,h,l and Supplementary Fig. 5d,h). The dwell time analysis of these FRET signals is now presented in the new Supplementary Fig. 6. The dissociation rates (koff) differ dramatically for the cognate aa-tRNA (which is accommodated) and the near-cognate tRNA (which is rapidly is rejected). Dissociation of \(-1\) -frame tRNAVal from PRE2 complexes formed after slow translocation on slippery mRNA is as slow as of the cognate O-frame tRNAPhe on PRE2 complexes formed on non-slippery mRNA. This provides a strong evidence in favor of our conclusion that slowly-translociating ribosomes have moved into the \(-1\) -frame exposing the Val codon in the A site.  
+
+<|ref|>text<|/ref|><|det|>[[115, 658, 881, 716]]<|/det|>
+Please note that we also unified the experiment schemes in Fig. 3a, e, i. In all experiments, PRE1 complexes are immobilized and EF- G (wt or mutant) is added together with EF- Tu- aa- tRNA- GTP complex to induce tRNA accommodation and subsequent translocation.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 748, 348, 765]]<|/det|>
+## OTHER SIGNIFICANT POINTS:  
+
+<|ref|>text<|/ref|><|det|>[[115, 768, 881, 845]]<|/det|>
+1. A fuller description is needed of the significant structural differences between the CHI, hybrid and classical states than that currently provided in the Introduction. Given the importance of the CHI state, this will aid the understanding of the paper by all but the most knowledgeable readers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 858, 848, 896]]<|/det|>
+Reply: We have added the description of the dynamics of PRE translocation complexes including explanation of tRNA conformations in hybrid and classical states. We have further
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 847, 128]]<|/det|>
+explained that the CHI state is a transient intermediate state of translocation, which rapidly resolves into the POST state.  
+
+<|ref|>text<|/ref|><|det|>[[115, 160, 845, 219]]<|/det|>
+2. It is unclear when the EF-G is added in the TIRF microscopy experiments monitoring translocation (Figs 3b,e,h; 4c,e; 5b; 6b). This point should be clearly addressed in either the Figure legends or the Experimental section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 230, 864, 308]]<|/det|>
+Reply: In the experiments described in Fig. 3, EF-G was added to immobilized PRE1 complexes together with EF-Tu-GTP-aa-tRNA complex. The components were added with the imaging buffer approximately 10 s prior to imaging. We added additional description in the figure legends and also in the Methods section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 319, 840, 358]]<|/det|>
+In the experiments described in Fig. 4, 5, 6 EF-G was added to immobilized PRE complexes approximately 10 s before imaging. This is now clearly stated in the Methods section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 369, 860, 408]]<|/det|>
+Please note that the vertical line in Fig. 6 represents the synchronization point of the smFRET traces. This is now explained in the figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 438, 844, 458]]<|/det|>
+3. Fig. 3b – The reason for the drift in Cy3 fluorescence in PRE1(0-47 s) should be explained.  
+
+<|ref|>text<|/ref|><|det|>[[115, 489, 877, 548]]<|/det|>
+Reply: "The drift" in the Cy3 fluorescence signal of the example trace shown in our previous Fig. 3b represented noise in the data and did not correspond to tRNA movement. We replaced the trace for a different example trace with stable Cy3 fluorescence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 560, 877, 618]]<|/det|>
+4. Fig. 4 – It is not clear from the data presented that the 0.9 – 0.6 transition represents a fluctuation, as opposed to an essentially unidirectional transition from 0.9 to 0.6. Contour plots should be provided for interconversions between the 0.9, 0.6 and 0.3 FRET states  
+
+<|ref|>text<|/ref|><|det|>[[114, 629, 879, 809]]<|/det|>
+Reply: We changed the text (p. 7) to clarify that the transitions between 0.9 and 0.6 states are not unidirectional but much less frequent than fluctuations of pept-tRNA on PRE complexes. Additionally, we calculated the transition frequency of deacylated tRNA (i.e. the average number of transitions between 0.6 and 0.9 FRET per trace) and compared it with the transition frequency of pept-tRNA on PRE complexes. Values differ by about one order of magnitude (0.4 vs 5.4 transitions, Supplementary Fig. 1c and 8f), further supporting the point that fluctuations of deacylated tRNA on PRE complexes are relatively slow. Transitions into the POST (0.3 FRET) state are irreversible (Supplementary Fig. 9d) and occur from the PRE state with either FRET 0.9 or 0.6, which is shown in the contour plot (Fig. 4d,e).  
+
+<|ref|>text<|/ref|><|det|>[[115, 839, 866, 878]]<|/det|>
+5. The current arrangement of Figures and Supplementary Figures require the reader to make frequent excursions from the main text to the Supplementary section. Could more of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 537, 108]]<|/det|>
+Supplementary Figures be included in the main text?  
+
+<|ref|>text<|/ref|><|det|>[[115, 140, 877, 238]]<|/det|>
+Reply: In the current version of the manuscript, we describe all key experiments and the model in the main text and present the control experiments in the supplement. We absolutely see the point made by the referee, but the main Figs are already heavy on panels and information. We fear that including even more material into the main text will lead to overload and confusion. We are therefore reluctant to change the principle sequence of figures in the main text.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 270, 333, 288]]<|/det|>
+## MINOR EDITORIAL POINTS:  
+
+<|ref|>text<|/ref|><|det|>[[115, 290, 678, 308]]<|/det|>
+1. Last line of Discussion: Delete "the" "elucidated in the future work."  
+
+<|ref|>text<|/ref|><|det|>[[116, 321, 409, 338]]<|/det|>
+Corrected in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 370, 557, 408]]<|/det|>
+2. Fig. 3b legend: Replace "monitored" by "monitored" This is corrected in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 469, 419, 486]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 508, 878, 787]]<|/det|>
+In this study the authors use single- molecule FRET (smFRET) to follow the translocation of tRNALys through the ribosome in the context of a slippery mRNA sequence. By altering the placement of Cy3 and Cy5 fluorophores, they are able to follow the dynamics of EF- G catalyzed translocation relative to translocation on a non- slippery mRNA as a control. They show that translocation on a slippery sequence can proceed through two pathways. In one pathway, both A and P site tRNAs rapidly move to the P and E sites with no change in reading frame. In the other pathway the deacylated P- site tRNA rapidly translocates while the A- site peptidyl- tRNA is delayed at a late chimeric stage of translocation, where it fluctuates between chimeric (ap/P) and post- translocation (P/P) states. During this period the authors confirm that the small subunit head is swiveled, conditions where codon- anticodon pairing is destabilized and alternative pairing frames can be explored. In the case of spontaneous \(- 1\) frameshifting, rapid release of deacylated tRNA from the E- site following translocation may facilitate \(- 1\) pairing of the stalled peptidyl- tRNA. These insights into translocation on a slippery sequence help to explain how spontaneous \(- 1\) frameshifting occurs.  
+
+<|ref|>text<|/ref|><|det|>[[115, 809, 865, 907]]<|/det|>
+Central to this study is the demonstration of a positive correlation of \(- 1\) frameshifting on a slippery mRNA sequence to the fraction of peptidyl- tRNA that is delayed in translocation. This was accomplished by carrying out translocation in the presence of several mutants of EF- G containing amino acid substitutions at position 507. In the wild- type factor, glutamine at this position helps to stabilize codon- anticodon pairing of the peptidyl- tRNA. Substitutions of this
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 866, 148]]<|/det|>
+amino acid are known to promote \(- 1\) frameshifting and the authors show that these same substitutions increase the fraction of translocation that is delayed due to fluctuation between ap/P and P/P states.  
+
+<|ref|>text<|/ref|><|det|>[[115, 169, 872, 188]]<|/det|>
+The smFRET studies appear to be carefully done. However, several minor revisions are needed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 200, 878, 279]]<|/det|>
+1. As regards to the biochemical assay for \(-1\) frameshifting, it is not specified whether peptide formation was carried out in the presence of both tRNAPhe and tRNAVal or only in the presence of the latter. By conducting this type of assay in the presence of both the O-frame and \(-1\) -frame incoming tRNAs, any possibility of frameshifting in the P-site can be discounted.  
+
+<|ref|>text<|/ref|><|det|>[[115, 290, 866, 368]]<|/det|>
+Reply: The biochemical assays to quantify frameshifting efficiency (Fig. 2f, 5a, and Supplementary Fig.3) were conducted in the presence of equal concentrations of the O-frame Phe- tRNAPhe and \(-1\) - frame Val- tRNAVal (5- fold excess over 70S IC). We clarified the description of the experimental procedure in the Methods section (p. 16).  
+
+<|ref|>text<|/ref|><|det|>[[115, 399, 857, 437]]<|/det|>
+2. The incorporation of O- and \(-1\) frame aa-tRNAs presented in Figure 3 does not add a lot to the paper and could be included in the supplemental section.  
+
+<|ref|>text<|/ref|><|det|>[[114, 479, 870, 617]]<|/det|>
+Reply: We feel that we cannot omit Fig. 3 from the main text figures because Fig 2 and Fig 3 represent two different aspects of pept- tRNA\(^{Lys}\) translocation. While Fig 2 reveals that pept- tRNA samples distinct intermediate states of translocation (CHI and POST) during spontaneous frameshifting, experiments presented in Fig. 3 verify that slow translocation via CHI states indeed precedes the accommodation of \(-1\) -frame-tRNA. Without this direct evidence, the correlation between slow translocation via distinct CHI states and spontaneous frameshifting would remain a conjecture.  
+
+<|ref|>text<|/ref|><|det|>[[115, 648, 833, 687]]<|/det|>
+3. It is not clear from the methods section (pg 15) how the POST fMAK complex is formed without going on to give a POST fMAKK complex. This needs to be more clearly described.  
+
+<|ref|>text<|/ref|><|det|>[[115, 729, 812, 767]]<|/det|>
+Reply: We revised the corresponding section in Methods (p. 17). Now it contains all the essential steps for the sample preparation.  
+
+<|ref|>text<|/ref|><|det|>[[114, 799, 870, 896]]<|/det|>
+4. The findings of the work are consistent with recent work that elucidated the mechanism of spontaneous \(+1\) frameshifting. In particular, the findings that \(-1\) frameshifting occurs at a later stage of the translocation reaction, involving the head domain swiveling of the small subunit, are reminiscent of the findings for spontaneous \(+1\) frameshifting as reported in PMID: 33436566 and in PMID: 34330903. Both of these papers should be cited and discussed in the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 833, 128]]<|/det|>
+present manuscript. This will give readers a broader and more comprehensive view of the background of this work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 140, 878, 218]]<|/det|>
+Reply: We agree that comparison of our results on spontaneous –1- frameshifting with spontaneous +1- frameshifting gives a broader view on the background of our study. We added a new paragraph on this subject into the Discussion section highlighting mechanistic similarities and differences between the two pathways.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 88, 321, 105]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 119, 404, 135]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 149, 877, 179]]<|/det|>
+The manuscript by Poulis et al., has improved in the representation of data and additional controls are included. The discussion of the data is also clarified and additional explanations are provided.  
+
+<|ref|>text<|/ref|><|det|>[[116, 223, 404, 239]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 253, 795, 269]]<|/det|>
+The authors have responded well to my concerns. A few points still remain, as listed below.  
+
+<|ref|>text<|/ref|><|det|>[[116, 283, 250, 298]]<|/det|>
+Significant points:  
+
+<|ref|>text<|/ref|><|det|>[[116, 299, 868, 357]]<|/det|>
+1. Although a CR state is a plausible intermediate on the way to the PRE2 complex, the results interpreted as demonstrating its formation (Figs.3 h); Supplementary Fig. 5d) are not compelling, in part due to the small number of traces obtained. As a result, the current text (bottom, p. 6) is overstated, and should be reworded more conservatively.  
+
+<|ref|>text<|/ref|><|det|>[[113, 371, 872, 388]]<|/det|>
+2. In a similar vein, on p.7, line 3 from the bottom, the word "show" should be replaced by "suggest".  
+
+<|ref|>text<|/ref|><|det|>[[116, 402, 213, 416]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[116, 417, 760, 433]]<|/det|>
+1. Supplementary Figure 3 legend needs to be edited to reflect the changes in part (a)  
+
+<|ref|>text<|/ref|><|det|>[[116, 447, 338, 461]]<|/det|>
+2. Three edits to p. 12, para2  
+
+<|ref|>text<|/ref|><|det|>[[116, 462, 336, 475]]<|/det|>
+- line 2: post-transcriptional  
+
+<|ref|>text<|/ref|><|det|>[[116, 476, 661, 491]]<|/det|>
+- line 8: replace "any type of frameshifting" with "-1 or +1 frameshifting"  
+
+<|ref|>text<|/ref|><|det|>[[116, 492, 225, 506]]<|/det|>
+-line 10: stalls  
+
+<|ref|>text<|/ref|><|det|>[[116, 550, 404, 566]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 580, 844, 610]]<|/det|>
+The revisions by the authors have addressed all of my concerns. more comprehensive view of the background of this work.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 91, 281, 108]]<|/det|>
+## Point-by-point reply  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 120, 400, 137]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 139, 866, 174]]<|/det|>
+The manuscript by Poulis et al., has improved in the representation of data and additional controls are included. The discussion of the data is also clarified and additional explanations are provided.  
+
+<|ref|>text<|/ref|><|det|>[[115, 185, 866, 220]]<|/det|>
+Reply: We appreciate the positive feedback by Reviewer #1 and thank him/her for the helpful remarks on the initial submission.  
+
+<|ref|>text<|/ref|><|det|>[[115, 250, 395, 266]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 269, 784, 286]]<|/det|>
+The authors have responded well to my concerns. A few points still remain, as listed below.  
+
+<|ref|>text<|/ref|><|det|>[[115, 305, 247, 321]]<|/det|>
+Significant points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 323, 880, 394]]<|/det|>
+1. Although a CR state is a plausible intermediate on the way to the PRE2 complex, the results interpreted as demonstrating its formation (Figs.3 h); Supplementary Fig. 5d) are not compelling, in part due to the small number of traces obtained. As a result, the current text (bottom, p. 6) is overstated, and should be reworded more conservatively.  
+
+<|ref|>text<|/ref|><|det|>[[115, 396, 876, 450]]<|/det|>
+Reply: We replaced "CR" for codon recognition for "RB" for ribosome binding state, where Phe-tRNA<sup>Phe</sup>- Cy5 reads the codon in the A site according to the step assignment of the earlier work (Geggier et al., 2010).  
+
+<|ref|>text<|/ref|><|det|>[[115, 479, 857, 515]]<|/det|>
+2. In a similar vein, on p.7, line 3 from the bottom, the word "show" should be replaced by "suggest". Reply: Done  
+
+<|ref|>text<|/ref|><|det|>[[115, 544, 215, 560]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 562, 743, 597]]<|/det|>
+1. Supplementary Figure 3 legend needs to be edited to reflect the changes in part (a) Reply: We changed the legend of Supplementary Fig. 3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 627, 328, 643]]<|/det|>
+2. Three edits to p. 12, para2  
+
+<|ref|>text<|/ref|><|det|>[[115, 647, 328, 662]]<|/det|>
+- line 2: post-transcriptional  
+
+<|ref|>text<|/ref|><|det|>[[115, 664, 645, 681]]<|/det|>
+- line 8: replace "any type of frameshifting" with "-1 or +1 frameshifting"  
+
+<|ref|>text<|/ref|><|det|>[[115, 684, 219, 699]]<|/det|>
+-line 10: stalls  
+
+<|ref|>text<|/ref|><|det|>[[115, 702, 422, 718]]<|/det|>
+Reply: We corrected the typos in the text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 738, 393, 754]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 774, 582, 790]]<|/det|>
+The revisions by the authors have addressed all of my concerns.  
+
+<|ref|>text<|/ref|><|det|>[[115, 803, 629, 819]]<|/det|>
+Reply: We appreciate the positive response to the revised manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42e00d20e9d549cfc52e3f47786978426d2bd90cf868b9b2017627c51f0f960a/images_list.json

@@ -0,0 +1,108 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Updated Figure 1. (A) Schematic of the continuous flow setup and chemical structure of the polymer (PEO44-b-PS86) used in this work. (B) DLS particle size distributions obtained at different asymmetric flow rates \\((Q_{\\mathrm{organic}} / Q_{\\mathrm{total}})\\) . (C) Intensity-averaged hydrodynamic diameters \\((D_{\\mathrm{h, intensity}})\\) and polydispersity indices (PDI) derived from the data shown in B. The different shades of black in C depict a pseudo-phase diagram. TEM images of (D) micelles obtained at \\(Q_{\\mathrm{organic}} / Q_{\\mathrm{total}} = 0.2\\) , (E) a mixture of micelles and polymersomes at \\(Q_{\\mathrm{organic}} / Q_{\\mathrm{water}} = 0.4\\) and (F) polymersomes obtained at \\(Q_{\\mathrm{organic}} / Q_{\\mathrm{total}} = 0.6\\) . All samples were analyzed in their respective organic solvent/water mixtures.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Updated Figure 4. (A) Schematic of continuous flow setup used for polymersome self-assembly and downstream annealing to manipulate polymersome size. BPR, backpressure regulator. (B) DLS particle size distributions of aqueous polymersomes prepared at different annealing temperatures ( \\(T_{\\text{annealing}}\\) ). (C) Intensity-averaged hydrodynamic diameters ( \\(D_{\\text{h, intensity}}\\) ) and polydispersity indices (PDI) derived from the data shown in B. TEM images of polymersomes annealed at (D) \\(20^{\\circ}\\mathrm{C}\\) , (E) \\(50^{\\circ}\\mathrm{C}\\) and (F) \\(70^{\\circ}\\mathrm{C}\\) for a residence time under heating ( \\(t_{\\text{residence, annealing}}\\) ) of \\(30\\mathrm{~s}\\) . Flow conditions used for polymersome formation: \\(Q_{\\text{total}} = 4 \\mathrm{~mL / min}\\) , \\(Q_{\\text{organic}} / Q_{\\text{total}} = 0.7\\) and \\(C_{\\text{polymer}} = 1 \\mathrm{mg / mL}\\) . All samples in B-F were dialyzed against water prior to analysis.",
+    "footnote": [],
+    "bbox": [
+      [
+        221,
+        60,
+        770,
+        480
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6. Large liposomes protruding out of the vitrified film gives an image that is darkest in the central, thickest part. Compare the drawing in Fig. 2.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure R3. A simplified schematic illustrating why polymersomes (or vesicles in general) sometimes have \"dark cores\" when visualized under cryo-TEM.",
+    "footnote": [],
+    "bbox": [
+      [
+        277,
+        460,
+        757,
+        592
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2a.jpg",
+    "caption": "Figure 2a-f from Rikken et al. Nat. Commun. 2016, 7, 12606",
+    "footnote": [],
+    "bbox": [
+      [
+        243,
+        68,
+        625,
+        275
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2A.jpg",
+    "caption": "Figure 2A-C from Ridolfo et al. Polym. Chem. 2020, 11, 2775-2780",
+    "footnote": [],
+    "bbox": [
+      [
+        175,
+        325,
+        694,
+        464
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2B.jpg",
+    "caption": "Figure 2B-C from Rijpkema et al. Biomacromolecules 2020, 21, 1853-1864",
+    "footnote": [],
+    "bbox": [
+      [
+        96,
+        525,
+        446,
+        644
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2A.jpg",
+    "caption": "Figure 2A from Abdelmohsen et al. JACS 2016, 138, 9353-9356",
+    "footnote": [],
+    "bbox": [
+      [
+        537,
+        522,
+        714,
+        644
+      ]
+    ],
+    "page_idx": 19
+  }
+]

peer_reviews/supplementary_0_Peer Review File__42e00d20e9d549cfc52e3f47786978426d2bd90cf868b9b2017627c51f0f960a/supplementary_0_Peer Review File__42e00d20e9d549cfc52e3f47786978426d2bd90cf868b9b2017627c51f0f960a.mmd

@@ -0,0 +1,530 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Dynamic Metastable Polymersomes Enable Continuous Flow Manufacturing  
+
+![](images/Figure_1.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 --->
+
+Editorial Note: Parts of this Peer Review File have been redacted as indicated to maintain the confidentiality of unpublished data.  
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have provided a very unique process that can create polymersomes with controllable sizes at much higher throughput than most, if not all, available formulation methods. This work has the potential to be highly impactful in aiding the translation of polymerome technology into clinical trials and beyond. I am very enthusiastic about this paper and really enjoyed reading it. The flow was very logical and experimental evidence is presented for the great majority of claims.  
+
+I recommend very minor revisions in accordance with the following remarks:  
+
+Robert Prud'homme's group in Princeton has developed a method called Flash Inverse Nanoprecipitation that is capable of high throughput monodisperse nanoparticle production. I think it could be important to compare and contrast benefits associated with the system developed here and this unique system. Furthermore, inverse nanoprecipitation (solvent injection) is capable of forming monodisperse polymersomes, albeit at low concentrations. It could aid in the discussion to include information about this.  
+
+The statement made on Page 4 line 14- 15 regarding the syringe size, etc having no effect on the selfassembly process should be supported by citations or experiments.  
+
+In Figure 1 B it is difficult to distinguish which lines correspond with which concentrations. A color may help here (Similar to Figure 4B).  
+
+Figure 2G makes it appear as if osmotic pressure is felt only at a point, when Jan Van Hest's group suggests it is more of an elongation force that ultimately leads to internal collapse of stomatocytes. If it is believed that this force is unidirectional, I think it would be helpful to explain why this is believed and support with citations.  
+
+Based on figures and data alone it is difficult to understand precisely where the "sub- 40 nm precision" conclusion is coming from. Is this meant to be supported by TEM images? It may be helpful to explain this in more detail on page 10 line 6.  
+
+Claim made at the end of the paragraph on page 12 line 13 needs to be supported with citations.  
+
+Again, I thoroughly enjoyed this paper and applaud the authors for their very interesting approach to an important translational problem. However, there appears to be a limitation associated which each polymerome having a polystyrene hydrophobic block. PS is not always used in clinical applications, which appear to be dominated by polyesters and other biodegradable blocks. I think it could really increase the impact of the conclusions to discuss potential translation to less hydrophobic or stimuli- responsive blocks.
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+Wong et al. describe in this article a continuous flow methodology for production of polymersomes at a relatively large scale ( \(\geq 3 \mathrm{~g} / \mathrm{h}\) ). While the topic of the article is important due to the need of efficient production of nanocarriers for various applications, there are critical issues, which prevent this manuscript for acceptance in Nature Communications. After solving the issues, a revised version will be appropriate to be submitted to a more specialized journal.  
+
+## General comments:  
+
+1. The method presented in the manuscript is based on the combination of a static mixing tee (Y-junction) as a small-scale mixing chamber and a continuous flow setup which reduces the mixing time, while the equilibration loop allows for a good control over the size and shape of polymersomes due to the polymersomes' metastability in the organic solvent/water mixture. However, compared to the current progress of science in the self-assembly process of vesicles formation (polymersomes and giant unilamellar vesicles), the method brings an elegant optimization in one of the polymersome preparation methods however without being a breakthrough in the field.  
+
+2. The Introduction does not contain the real state-of-art in the field regarding the polymersomes production because the well known film rehydration method for polymersome formation and loading with molecules is not presented with its advantages both in terms of polymersomes high yield production and encapsulation efficiency. The Introduction should be improved to present all relevant methods for polymersomes formation and their advantages or still open questions in the field.  
+
+3. There is a confusion the authors include in the Introduction by considering synthetic giant unilamellar vesicles (GUVs) as polymersomes. In the field there is a clear distinction between the vesicles with sizes in the nanometer range (polymersomes) and the vesicles with micrometer sizes (GUVs), similarly to the notions of liposomes and lipid GUVs. This distinction is based both on the methods of production and the properties of the vesicles membrane (stability, curvature, etc). The focus of this manuscript is in preparation of polymersomes. Therefore, the Introduction should be corrected accordingly to avoid misunderstanding and to clearly indicate the relevance of the method the authors propose.  
+
+4. A serious critical aspect is related to the low mechanic stability of polymersomes obtained by the present method, which downgrade them for any application, as they are less stable even than PEGylated-liposomes. Usually polymersomes stability is of several months, depending on the type of amphiphilic copolymer. Therefore, this method should be significantly improved to allow formation of stable polymersomes fort longer periods of time than one week, as this is a real bottle neck for further applications.  
+
+5. The authors indicate that the method they propose can be used for any type of amphiphilic copolymer. However, they selected as example to test their method for polymersomes formation, PEO
+
+<--- Page Split --->
+
+b- PS, which has specific properties that cannot be extrapolated for other types of amphiphilic copolymers. In addition, both the chemistry of synthesized block copolymers has been well established and the theoretical background of polymer self-assembly via the solvent switch approach has been previously explored in detail, e.g. by the groups of A. Eisenberg, T.P. Lodge or J.C.M. van Hest. To validate this method for a variety of copolymers, the authors should prove it at least for two different types of copolymers in terms of molecular properties.  
+
+6. The characterization of polymersomes only by DLS (Pag.4 line 25) is not enough to prove the hollow sphere architecture. Static light scattering experiments should be performed and combined to distinguish whether the spherical nanoobjects are polymersomes.  
+
+7. Cryo-TEM will be important to give the necessary details of the polymersome membrane that are not clearly visible from the presented TEM micrographs (Fig. 1). Besides, the resolution of all TEM micrographs should be significantly improved. The change in TEM micrographs as stated, "The three accessible morphological phases are highlighted in Figure 1C using different shades of black" is confusing. It is essential to present the raw data, not highlighted images, to avoid biases.  
+
+8. Figures 2 D and E regarding TEM micrographs of polymersomes are different from what we expect to have when hollow sphere architecture is present; the spherical nanoobjects have a darker core that is not specific for polymersomes. Therefore it is not clear that the nanoobjects in these figures are polymersomes. Cryo-TEM will elucidate this issue.  
+
+9. Figures 4 D-F indicate aggregation of the spherical nanoobjects, which is a severe limitation for further applications. How the aggregation can be avoided?  
+
+Reviewer #3 (Remarks to the Author):  
+
+In this manuscript, the authors reported a continuous flow methodology capable of producing near- monodisperse polymersomes at scale \((\geq 3 \text{g / h})\) . They also demonstrated downstream processes (thermal annealing and/or secondary micro- mixing) to manipulate polymersome size (with sub- 40 nm precision) and/or polymersome shape. This work is meaningful for the production of near- monodisperse polymersomes at scale, as well as the control of polymersomes properties under flow conditions. I would recommend its publication on NC after major revision.  
+
+1. The authors claimed that they specifically adjusted the salinity of the polymersome solution to \(50 \text{mM NaCl}\) , followed by dialysis to remove the organic solvents. Why choose \(50 \text{mM NaCl}\) ?  
+
+2. At the highest total flow rate (Qtotal= 8 mL/min), a low PDI of \(0.045 \pm 0.015\) was obtained, indicating near-monodisperse polymersomes. The monodispersity is very interesting, and the authors should investigate the formation mechanism of the monodispersity.  
+
+3. A production rate of \(\geq 3 \text{g}\) of polymersomes/hour was achieved. Further increasing Copolymer and Qtotal, the production rate can be improved. Why the authors did not pursue this?
+
+<--- Page Split --->
+
+4. The authors also manipulated polymersome shape by expanding the flow setup to include a cooling loop and a secondary micromixer connected to another syringe pump. It was crucial to introduce only a small amount of concentrated NaCl solution, as this ensures minimal deviations in solvent quality after micromixing, thus preventing any morphological deviations. How to precisely control a small amount of concentrated NaCl solution? What is the exact value and exact quantity? Any preliminary data/experiment?
+
+<--- Page Split --->
+
+Black = comments from the reviewers  
+
+Blue = our response to the reviewers  
+
+Green = reproduced texts from the original/revised manuscript  
+
+All additions/changes made to the manuscript are highlighted in yellow to enable tracked changes.  
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The authors have provided a very unique process that can create polymersomes with controllable sizes at much higher throughput than most, if not all, available formulation methods. This work has the potential to be highly impactful in aiding the translation of polymersome technology into clinical trials and beyond. I am very enthusiastic about this paper and really enjoyed reading it. The flow was very logical and experimental evidence is presented for the great majority of claims.  
+
+I recommend very minor revisions in accordance with the following remarks:  
+
+Comment #1: Robert Prud'homme's group in Princeton has developed a method called Flash Inverse Nanoprecipitation that is capable of high throughput monodisperse nanoparticle production. I think it could be important to compare and contrast benefits associated with the system developed here and this unique system. Furthermore, inverse nanoprecipitation (solvent injection) is capable of forming monodisperse polymersomes, albeit at low concentrations. It could aid in the discussion to include information about this.  
+
+## Response:  
+
+We thank the Reviewer for their suggestion.  
+
+We have added a reference to Prof. Robert Prud'homme's first paper where he first introduced the term "flash nanoprecipitation". We have also modified our Introduction to clarify that the use of micromixers to conduct nanoprecipitation is sometimes referred to as flash nanoprecipitation in the literature:  
+
+"An alternative flow- based approach, sometimes referred to as flash nanoprecipitation, \(^{38}\) relies on the use of miniaturized mixing chambers (micromixers) that reduce the mixing timescale between two incoming solution streams down to the millisecond regime. By employing a micromixer for nanoprecipitation, as opposed to simply conducting nanoprecipitation under batch conditions, one can effectively enhance the uniformity of an overall block copolymer self- assembly process to generate polymersomes in a highly reproducible manner. \(^{39 - }\) \(^{42}\) Although proven effective, most reports on this approach employ the use of micromixers with complex internal geometries that are difficult and expensive to manufacture."  
+
+Note, however, that we have chosen not to elaborate on inverse flash nanoprecipitation specifically because we view the process as somewhat akin to templated layer- by- layer (LbL) self- assembly, as opposed to a true bottom- up self- assembly/nanoprecipitation process such as that seen in traditional flash nanoprecipitation. In fact, we support our view by quoting a statement made in a recent paper \(^{1}\) by Prof. Prud'homme's group where they distinguished nanoparticles made from inverse flash nanoprecipitation from polymersomes:  
+
+"Called "inverse Flash NanoPrecipitation (iFNP)," the technique achieves biologic loadings (wt% of total formulation) that are 5- 15x higher than typical values (9- 27% versus < 2%). In contrast to liposomes and polymersomes, we sequentially assemble the polymer layers to form the final nanocarrier".  
+
+Comment #2: The statement made on Page 4 line 14- 15 regarding the syringe size, etc having no effect on the self- assembly process should be supported by citations or experiments.
+
+<--- Page Split --->
+
+## Response:  
+
+We appreciate the suggestion provided by the Reviewer. However, we do not believe that it is necessary to provide citations or conduct experiments to support our statement made on previously on Page 4, Line 14- 15 (Page 4, Line 17 in the revised manuscript) for the following reasons:  
+
+Firstly, both syringe size and solution volume are simply device- related parameters that a user inputs into a pump program. They should not be considered as experimental parameters because they do not bear any impact on the self- assembly process. To illustrate this, let's consider two scenarios where a \(10~\mathrm{mL}\) and 20 mL syringe are used in separate self- assembly experiments. Irrespective of which syringe is used, the self- assembly would remain identical because in both experiments the pumps would have been set to the dispense at identical flow rates. The only notably distinction between the two experiments would be the volume of nanoparticles generated, as the \(10~\mathrm{mL}\) syringe has half the capacity of the \(20~\mathrm{mL}\) syringe.  
+
+Secondly, the length of the equilibration loop likewise has no impact on the self- assembly process. To elaborate, let's again consider two hypothetical experiments: one using a \(1~\mathrm{mL}\) equilibration loop and the other using a \(2~\mathrm{mL}\) equilibration loop. In both scenarios, the self- assembly process would be identical because the equilibration loop is placed downstream to the micromixer, where the self- assembly process truly occurs. The only real difference between the two experiments is the duration of time the nanoparticles resides in the equilibration loop immediately after self- assembly \((t_{\mathrm{residence}})\) . Assuming a total flow rate of \(1~\mathrm{mL / min}\) is used, the \(1~\mathrm{mL}\) equilibration loop would provide a \(t_{\mathrm{residence}} = 1\) min, while the \(2~\mathrm{mL}\) equilibration loop would provide \(t_{\mathrm{residence}} = 2\) min.  
+
+Now, although our polymersomes are metastable in nature, we know from our aging studies (refer back to Figure 2C) that the lifetime of this metastable state \((t_{\mathrm{metastable}})\) is ca. 7 days. Considering the significant difference in timescales between \(t_{\mathrm{residence}}\) (min/sec) and \(t_{\mathrm{metastable}}\) ( \(\sim 7\) days), it is reasonable to claim that the length of the equilibration loop has no impact on the self- assembly process. Of course, one may argue that the equilibration loop can be extended to cover a \(t_{\mathrm{residence}}\) of several days, but such an experiment would not be very practical.  
+
+Comment #3: In Figure 1 B it is difficult to distinguish which lines correspond with which concentrations. A color may help here (Similar to Figure 4B).  
+
+## Response:  
+
+We have updated Figure 1B and Figure 4B to incorporate the Reviewer's suggestions. Both updated figures are reproduced below for clarity.
+
+<--- Page Split --->
+![](images/Figure_4.jpg)
+
+<center>Updated Figure 1. (A) Schematic of the continuous flow setup and chemical structure of the polymer (PEO44-b-PS86) used in this work. (B) DLS particle size distributions obtained at different asymmetric flow rates \((Q_{\mathrm{organic}} / Q_{\mathrm{total}})\) . (C) Intensity-averaged hydrodynamic diameters \((D_{\mathrm{h, intensity}})\) and polydispersity indices (PDI) derived from the data shown in B. The different shades of black in C depict a pseudo-phase diagram. TEM images of (D) micelles obtained at \(Q_{\mathrm{organic}} / Q_{\mathrm{total}} = 0.2\) , (E) a mixture of micelles and polymersomes at \(Q_{\mathrm{organic}} / Q_{\mathrm{water}} = 0.4\) and (F) polymersomes obtained at \(Q_{\mathrm{organic}} / Q_{\mathrm{total}} = 0.6\) . All samples were analyzed in their respective organic solvent/water mixtures. </center>
+
+<--- Page Split --->
+![](images/Figure_6.jpg)
+
+<center>Updated Figure 4. (A) Schematic of continuous flow setup used for polymersome self-assembly and downstream annealing to manipulate polymersome size. BPR, backpressure regulator. (B) DLS particle size distributions of aqueous polymersomes prepared at different annealing temperatures ( \(T_{\text{annealing}}\) ). (C) Intensity-averaged hydrodynamic diameters ( \(D_{\text{h, intensity}}\) ) and polydispersity indices (PDI) derived from the data shown in B. TEM images of polymersomes annealed at (D) \(20^{\circ}\mathrm{C}\) , (E) \(50^{\circ}\mathrm{C}\) and (F) \(70^{\circ}\mathrm{C}\) for a residence time under heating ( \(t_{\text{residence, annealing}}\) ) of \(30\mathrm{~s}\) . Flow conditions used for polymersome formation: \(Q_{\text{total}} = 4 \mathrm{~mL / min}\) , \(Q_{\text{organic}} / Q_{\text{total}} = 0.7\) and \(C_{\text{polymer}} = 1 \mathrm{mg / mL}\) . All samples in B-F were dialyzed against water prior to analysis. </center>  
+
+Comment #4: Figure 2G makes it appear as if osmotic pressure is felt only at a point, when Jan Van Hes't's group suggests it is more of an elongation force that ultimately leads to internal collapse of stomatocytes. If it is believed that this force is unidirectional, I think it would be helpful to explain why this is believed and support with citations.  
+
+## Response:  
+
+We thank the Reviewer for raising this issue.  
+
+The Reviewer is correct in that the osmotic pressure that is applied onto a polymersome structure during shape transformation is not unidirectional. We had initially thought that the addition of the red arrow in Figure 2G would help guide a non- expert reader to envisage how an originally spherical polymersomes can be deformed into a (bowl- like) stomatocyte structure. In hindsight, we admit that this is misleading since the deformation process is not caused by the application of an external force at a single point as we have indicated with the red arrow in Figure 2G. We clarify that we are not suggesting a shape transformation pathway that is any different to what was proposed by the van Hest group in their seminal work.2 The deformation process is in fact driven by a reduction in internal volume, which is in turn caused by the rapid efflux of organic solvents from the polymersome core due to osmotic imbalance. Despite the misleading schematic, we had already described the shape transformation mechanism on Page 11, Lines 7- 11:
+
+<--- Page Split --->
+
+"This change in salinity generates an osmotic imbalance between the polymersomes' inner compartment and their surrounding solution, causing a net efflux of solvent molecules out of the polymersomes. This in effect drives a reduction in the polymersomes' internal volume and causes the (initially spherical) polymersomes to deform into indented polymersomes known as stomatocytes (TEM and cryo- TEM images in Figure 2H)."  
+
+To prevent further confusion regarding this matter, we have proceeded to remove the red arrow from Figure 2G.  
+
+Comment #5: Based on figures and data alone it is difficult to understand precisely where the "sub- 40 nm precision" conclusion is coming from. Is this meant to be supported by TEM images? It may be helpful to explain this in more detail on page 10 line 6.  
+
+## Response:  
+
+We apologise for the lack of clarity.  
+
+We have added a reference to our DLS data in Figure 2C and Table S2 to back our claim made previously on Page 10, Line 6 (Page 10, Line 9 in the revised manuscript). We have also amended the sentence slightly to enhance clarity:  
+
+"Our ability to control polymersome- size- distribution mean polymersome size with sub- 40 nm precision (Figure 2C and Table S2) is a feat inconceivable with conventional polymersomes formation methods."  
+
+Comment #6: Claim made at the end of the paragraph on page 12 line 13 needs to be supported with citations.  
+
+## Response:  
+
+To support our claim made previously on Page 12, Line 13 (Page 12, Line 11 in the revised manuscript), we have added 4 references to papers published between 2011- 2021, where PEO- b- PS polymersomes have been prepared by batch nanoprecipitation. The polymersome production rate in each reference is provided in [bolded brackets] below.  
+
+50. Meeuwissen, S. A., Kim, K. T., Chen, Y., Pochan, D. J. & van Hest, J. C. M. Controlled shape transformation of polymersome stomatocytes. Angew. Chem. Int. Ed. 50, 7070-7073 (2011). [Polymersome production rate = 3.33 milligrams/hour]  
+
+51. Nijemeisland, M., Abdelmohsen, L. K. E. A., Huck, W. T. S., Wilson, D. A. & van Hest, J. C. M. A compartmentalized out-of-equilibrium enzymatic reaction network for sustained autonomous movement. ACS Cent. Sci. 2, 843-849 (2016). [Polymersome production rate = 6.67 milligrams/hour]  
+
+[Polymersome production rate = 6.67 milligrams/hour]  
+
+52. Kim, J. & Kim, K. T. Polymersome-Based Modular Nanoreactors with Size-Selective Transmembrane Permeability. ACS Appl. Mater. Interfaces 12, 23502-23513 (2020). [Polymersome production rate = 20 milligrams/hour]  
+
+53. Sun, J., Rijpkema, S. J., Luan, J., Zhang, S. & Wilson, D. A. Generating biomembrane-like local curvature in polymersomes via dynamic polymer insertion. Nat. Commun. 12, 2235 (2021). [Polymersome production rate = 3.33 milligrams/hour]  
+
+For comparison, the highest polymersome production rate we have achieved is 3.02 grams/hour (data shown in Figure 3D in our manuscript).
+
+<--- Page Split --->
+
+Comment #7: Again, I thoroughly enjoyed this paper and applaud the authors for their very interesting approach to an important translational problem. However, there appears to be a limitation associated which each polymerome having a polystyrene hydrophobic block. PS is not always used in clinical applications, which appear to be dominated by polyesters and other biodegradable blocks. I think it could really increase the impact of the conclusions to discuss potential translation to less hydrophobic or stimuli- responsive blocks.  
+
+## Response:  
+
+We acknowledge the concerns raised by the Reviewer regarding the lack of biological relevance of polystyrene (PS) in clinical applications. Although we share the same sentiment as the Reviewer, it is worthwhile pointing out that, despite being non- biodegradable, PEO- b- PS polymersomes still possess value in the medical realms. To elaborate, the Leroux group from ETH Zurich, have for example, recently reported<sup>3</sup> on the use of PEO- b- PS polymersomes for the oral treatment and diagnosis of hyperammonia, a metabolic condition characterized by an abnormally high level of ammonia in blood, which can, at times, be lifethreatening. Prior to publishing this work, the authors of this work had already filed for patent applications worldwide (see e.g., Patent No.: WO2019053578A1, US20200283583A1, EP3668927A1). According to the authors in their Conflict of Interest disclosure statement, the patents have been licensed to Versantis AG, a clinical- stage pharmaceutical company that focuses on the development of new generation orphan medications.  
+
+Having said the above, we do not deter the fact that clinical applications would ultimately benefit from the use of clinically relevant polymersomes. Biodegradable polyesters or biocompatible stimuli- responsive polymers, as suggested by the Reviewer, certainly hold great promise in this regard. We have taken the advice of the Reviewer and expanded our conclusion to emphasize the need for more clinically relevant polymersomes:  
+
+"Finally, in order to accelerate the clinical translation of polymersomes, further advancements in this area should prioritize the development of more clinically relevant polymersomes (e.g., biodegradable/stimuli- responsive polymersomes) to ensure optimal efficacy and safety for patients."  
+
+Finally, we extend our gratitude to the Reviewer for their compliment and valuable feedback aimed at improving our manuscript. After nearly dedicating a decade on research on polymersomes, we genuinely believe that this work stands as one of our most significant contributions to the field. We hope that forthcoming readers will likewise recognize and appreciate the impact of our work.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+Wong et al. describe in this article a continuous flow methodology for production of polymersomes at a relatively large scale (≥ 3 g/h). While the topic of the article is important due to the need of efficient production of nanocarriers for various applications, there are critical issues, which prevent this manuscript for acceptance in Nature Communications. After solving the issues, a revised version will be appropriate to be submitted to a more specialized journal.  
+
+## Response:  
+
+We appreciate the Reviewer's acknowledgement of our work in addressing the pressing need for "efficient production of nanocarriers for various applications" despite the presence of certain "critical issues". We have responded to these concerns in detail below.  
+
+## General comments:  
+
+Comment #1: The method presented in the manuscript is based on the combination of a static mixing tee (Y- junction) as a small- scale mixing chamber and a continuous flow setup which reduces the mixing time, while the equilibration loop allows for a good control over the size and shape of polymersomes due to the polymersomes' metastability in the organic solvent/water mixture. However, compared to the current progress of science in the self- assembly process of vesicles formation (polymersomes and giant unilamellar vesicles), the method brings an elegant optimization in one of the polymersome preparation methods however without being a breakthrough in the field.
+
+<--- Page Split --->
+
+## Response:  
+
+We thank the Reviewer for acknowledging our methodology as an "elegant optimization" of nanoprecipitation.  
+
+Contrary to the Reviewer's remark that "the equilibration loop allows for a good control over the size and shape of polymersomes", the equilibration loop plays no actual role in size or shape control. In our system, size control was achieved through the use of an annealing loop, which provides heat energy to "nudge" the polymersomes out of their metastable state into lower free energy states. Shape control, on the other hand, was enabled by a secondary downstream micromixer—this allowed us to continuously introduce an additive (NaCl) to osmotically deform the polymersomes into their non- spherical, stomatocyte shape. Neither of the two processes have been demonstrated in a continuous downstream fashion as we have reported.  
+
+Although we regret that the Reviewer identifies our work as not "being a breakthrough in the field", we will nonetheless attempt to clarify both the novelty and significance of our work through all remaining responses to Reviewer #2 below.  
+
+Comment #2: The Introduction does not contain the real state- of- art in the field regarding the polymersomes production because the well known film rehydration method for polymersome formation and loading with molecules is not presented with its advantages both in terms of polymersomes high yield production and encapsulation efficiency. The Introduction should be improved to present all relevant methods for polymersomes formation and their advantages or still open questions in the field.  
+
+## Response:  
+
+We believe that we have provided a fairly comprehensive overview of the current state- of- the- art in the Introduction, in particular within the context of nanoprecipitation—the most common bottom- up self- assembly approach in the polymersome field. We acknowledge the existence and importance of the film rehydration method raised by the Reviewer; however, the film rehydration method is a top- down self- assembly approach which bears significant difference to nanoprecipitation (a bottom- up approach) and thus does not fit within the context of our Introduction. The Reviewer also mentions "encapsulation efficiency"; however, since we have not performed any encapsulation experiments using our methodology, we do not see any relevance in elaborating on that topic in the Introduction, especially considering its complexity. Below, we outline how our Introduction has been carefully structured to cover a large breadth of information pertaining to polymersome formation via nanoprecipitation, both in batch and in flow, as well as their associated advantages and limitations, and how our methodology opens up the possibility of performing downstream manipulations—processes that would not have been possible with traditional, kinetically trapped polymersomes:  
+
+## Paragraph 1  
+
+What polymersomes are and how they structurally resemble liposomes What physicochemical properties of polymersomes can be modified How the above, combined with their ability to load both hydrophilic and hydrophobic materials, has led to widespread applications in drug delivery, synthetic biology and nanoreactor science.  
+
+## Paragraph 2  
+
+Step- by- step explanation outlining how nanoprecipitation is performed in batch to produce polymersomes Explanation of logic behind individual steps in a typical nanoprecipitation process including other important factors to consider  
+
+## Paragraph 3  
+
+Limitations of batch nanoprecipitation Why nanoprecipitation leads to polydisperse polymersomes due to poor mixing efficiency Why nanoprecipitation is poorly scalable  
+
+## Paragraph 4  
+
+How researchers have turned to flow- based systems to negate the effects of mixing in batch
+
+<--- Page Split --->
+
+- Examples and advantages/disadvantages of current microfluidic chip-based polymerase formation methods, including:  
+
+- Double emulsion templating using flow-focusing chips- Laminar and plugged flow mixing using flow-focusing chips- Polymerization-induced self-assembly (PISA) under flow  
+
+## Paragraph 5  
+
+- How miniaturized mixing chambers (micromixers) are superior to flow-focusing chips in both scalability and reproducibility- How micromixers work by minimizing the timescale of mixing between two incoming streams down to the millisecond regime- Drawbacks of current polymerase formation methods with micromixers, in particular their ability to only produce kinetically trapped polymersomes- Explanation as to why downstream processing/manipulation unlocks the full potential of a continuous flow process  
+
+## Paragraph 6  
+
+- Summary, novelty, and rationale behind our work  
+
+Comment #3: There is a confusion the authors include in the Introduction by considering synthetic giant unilamellar vesicles (GUVs) as polymersomes. In the field there is a clear distinction between the vesicles with sizes in the nanometer range (polymersomes) and the vesicles with micrometer sizes (GUVs), similarly to the notions of liposomes and lipid GUVs. This distinction is based both on the methods of production and the properties of the vesicles membrane (stability, curvature, etc). The focus of this manuscript is in preparation of polymersomes. Therefore, the Introduction should be corrected accordingly to avoid misunderstanding and to clearly indicate the relevance of the method the authors propose.  
+
+## Response:  
+
+The Reviewer's concerns stem from what was written in the Introduction starting Page 2, Line 19. We note that this is the only section in our manuscript where we have discussed micrometre- sized polymersomes that could qualify as so- called giant unilamellar vesicles (GUVs). For clarity, we reproduce the text with the phrase in question underlined below:  
+
+"To negate the effects of batch mixing, researchers have turned to flow- based systems such as microfluidics. A reliable microfluidics approach is the double emulsion templating method,28- 30 which relies on the use of flow- focusing chips to confine and self- assemble block copolymers in the oil phase of water/oil/water (w/o/w) double emulsion droplets. Although the approach generates monodisperse polymersomes with high reproducibility, it is somewhat limited in terms of accessible polymersome size (tens to hundreds of \(\mu \mathrm{m}\) ), and production scalability because the devices used typically only operate at flow rates of only several \(\mu \mathrm{L} / \mathrm{min}\) ."  
+
+We hope it is clear from the above that we have only briefly mentioned micrometre- sized polymersomes from a microfluidic/flow self- assembly context. Given this information, we do not believe it is necessary to explicitly classify or describe polymersomes that are micrometre- sized as giant unilamellar vesicles (GUVs), especially if we also consider the fact that the pioneers of the technique (Prof. David Weitz and his colleagues) themselves referred to such structures simply as "polymersomes" in their publications.4- 6 Furthermore, from a morphological perspective, there really is no real distinction between sub- micron polymersomes and micrometre- sized polymersomes (e.g., GUVs) other than their sizes, as both "classes" of polymersomes share the same morphology—a bilayer membrane structure and a hollow core. We appreciate that the nomenclature "giant unilamellar vesicles (GUVs)" originates from the liposome field; however, we believe the use of this nomenclature should be context dependent. For instance, it may be appropriate when describing a polymersome system that consists of a mixture of giant unilamellar vesicles (GUVs) and multilamellar vesicles (MLVs).
+
+<--- Page Split --->
+
+Comment #4: A serious critical aspect is related to the low mechanic stability of polymersomes obtained by the present method, which downgrade them for any application, as they are less stable even than PEGylated- liposomes. Usually polymersomes stability is of several months, depending on the type of amphiphilic copolymer. Therefore, this method should be significantly improved to allow formation of stable polymersomes for longer periods of time than one week, as this is a real bottle neck for further applications.  
+
+## Response:  
+
+The Reviewer is certainly correct that our polymersomes are inherently unstable- this is exactly the novelty of our methodology. In the absence of this metastable state, one would not have been able to accomplish downstream manipulation with the level of control that we have demonstrated throughout our manuscript. We would like to further clarify the potential misconception about (i) our polymersomes' metastability in organic solvent/water mixture and (ii) their inherent stability in water below:  
+
+Regarding (i) polymersome metastability: Briefly, the term metastability is used to describe a system that exists in an apparent state of equilibrium, when in fact it can transition into a more stable (equilibrium) state if energy is provided to the system. The amount of energy needed for this transition to occur can be quantified as an activation energy barrier \((E_{\mathrm{A}})\) , which in the case of our system, is in the order of \(k_{\mathrm{B}}T\) at room temperature (see Figure 2A in main text for proposed free energy diagram). Under ambient conditions, and in organic solvent/water mixture, our polymersomes grow as a result of metastability for ca. 7 days. After this 7- day period, this growth process ceases entirely as the system has transitioned out of its initial metastable state and into an equilibrium state. As we have described in our manuscript, this transition from metastable to equilibrium state is key to the implementation of the downstream annealing setup in Figure 4, which we used to demonstrate size control immediately after self- assembly (and perhaps more importantly, in the very same continuous stream).  
+
+Regarding (ii) polymersome stability: As we have discussed on page 9 (line 13 onwards), the metastable state observed in our system can be quenched at any point in time (e.g., during the 7- day growth process or even after the growth has ceased) to trap the system in different kinetically arrested states. This can be done simply by removing the organic solvent from the system (e.g., by extensively dialyzing the polymersomes against water). This is possible owing to high glass transition of polystyrene, PS \((T_{\mathrm{g,PS}})\) , which is \(\sim 100^{\circ}\mathrm{C}\) . Upon removal of the plasticizing organic solvent, PS chains which constitute the polymersome membrane structure transitions from a dynamic plasticized state into a glassy quenched state. Once this quenched state has been reached, no further chain rearrangements (and thus no further morphological changes) are possible. In the quenched state, the polymersomes are indefinitely stable unless, of course, some organic solvent is reintroduced into the system to plasticize the PS membrane or if the block copolymer undergoes chemical degradation (which is unlikely in the case of PEO- b- PS). Throughout the undertaking of this project, we have observed minimal macroscopic precipitation or sedimentation in all our quenched polymersome samples, some of which have been stored at room temperature for as long as 2 years (although we state here that this stability is not intrinsic to our system and is likely common even for PEO- b- PS polymersomes prepared by batch nanoprecipitation). Even if precipitates were present, the samples can simply be filtered through a \(0.45 \mu \mathrm{m}\) polyethersulfone (PES) membrane filter without any deterioration in sample quality as all our polymersome samples are \(< 0.45 \mu \mathrm{m}\) in diameter.  
+
+With all the above, we dispute the Reviewer's assertion that our polymersomes have "low mechanic(al) stability", "are less stable even than PEGylated- liposomes", and that there "is a real bottle neck for further applications".  
+
+Comment #5: The authors indicate that the method they propose can be used for any type of amphiphilic copolymer. However, they selected as example to test their method for polymersomes formation, PEO- b- PS, which has specific properties that cannot be extrapolated for other types of amphiphilic copolymers. In addition, both the chemistry of synthesized block copolymers has been well established and the theoretical background of polymer self- assembly via the solvent switch approach has been previously explored in detail, e.g. by the groups of A. Eisenberg, T.P. Lodge or J.C.M. van Hest. To validate this method for a variety of copolymers, the authors should prove it at least for two different types of copolymers in terms of molecular properties.
+
+<--- Page Split --->
+
+Response:
+
+<--- Page Split --->
+![](images/Figure_unknown_0.jpg)
+
+
+<--- Page Split --->
+
+Comment #6: The characterization of polymersomes only by DLS (Pag.4 line 25) is not enough to prove the hollow sphere architecture. Static light scattering experiments should be performed and combined to distinguish whether the spherical nanoobjects are polymersomes.  
+
+## Response:  
+
+For the purpose of this response, we reproduce the statement in question made previously on Page 4, Line 25 underlined and italicized below (n.b., this statement is now on Page 4, Line 24 in the revised manuscript), along with the two sentences that preceded it. Also reproduced below is the subsequent paragraph (starting Page 5, Line 3 in the revised manuscript; italicized) to help contextualize the statement in question:  
+
+"We performed the self- assembly process at 7 different asymmetric flow rates ranging from \(Q_{organic} / Q_{total} =\) 0.1- 0.7 (in 0.1 increments). In every case, the product was collected directly into a quartz cuvette and immediately analyzed by dynamic light scattering (DLS). The resulting particle size distributions are shown in Figure 1B. Each sample's intensity- averaged hydrodynamic diameter \((D_{h,intensity})\) and polydispersity index (PDI) are 25 plotted in Figure 1C. A summary of the DLS data is further provided in Table S1.  
+
+All 7 asymmetric flow rates resulted in monomodal particle size distributions with relatively low PDIs of \(< 0.16\) (Figure 1B- C and Table S1). At \(Q_{organic} / Q_{total} \leq 0.2\) , minimal changes in particle size were observed. Increments above this value, however, resulted in a linear increase in particle size (see \(D_{h,intensity}\) datapoints for \(Q_{organic} / Q_{total} = 0.3 - 0.7\) in Figure 1C). We note here that asymmetric flow rates of \(Q_{organic} / Q_{total} > 0.7\) were also tested, but these flow conditions did not result in any particle formation because \(PEO_{44} - b - PS_{86}\) remains molecularly dissolved when the organic solvent content exceeds 70 vol%.  
+
+As can be seen from the above, we did not make any claims regarding particle morphology using our DLS data. In fact, all our discussions around DLS (data provided in Figure 1B- C and Table S1) was on the effect of asymmetric flow rates \((Q_{organic} / Q_{total})\) on particle size \((D_{h,intensity})\) . We specifically noted that (i) particle size does not change when the asymmetric flow rate, \(Q_{organic} / Q_{total} \leq 0.2\) , (ii) particle size increases linearly when the asymmetric flow rate is increased from \(Q_{organic} / Q_{total} = 0.3 - 0.7\) , and (iii) no particles form beyond \(Q_{organic} / Q_{total} > 0.7\) because \(PEO_{44} - b - PS_{86}\) is molecularly soluble under those self- assembly conditions. Notice how we were careful in using the term "particle size" as opposed to "micelle size" or "polymersome size" to discuss our DLS data.  
+
+We only began making claims on particle morphology starting Page 6, Line 3. For the sake of clarity, we reproduce the entire paragraph below in italics, with the claims underlined:  
+
+"Next, we used transmission electron microscopy (TEM) to probe particle morphology. Shown in Figure 1D- F are three TEM images of particles produced at \(Q_{organic} / Q_{total} = 0.2\) , 0.4 and 0.6, respectively. The gradual increase in \(Q_{organic} / Q_{total}\) generated a morphological transition from micelles (Figure 1D) to a mixed phase of micelles/polymersomes (Figure 1E), and finally to polymersomes (Figure 1F). For clarity, the three accessible morphological phases are highlighted in Figure 1C using different shades of black."  
+
+One can see from the above that our claims about particle morphology are solely based on the supporting evidence from our TEM data. We claimed the existence of micelles, a mixed phase of micelles/polymersomes, and polymersomes based on the TEM images provided in Figure 1D, Figure 1E and Figure 1F, respectively.  
+
+To summarize, we did not use DLS as a means to prove the hollow structure of our polymersomes. Instead, we relied on TEM (and cryo- TEM throughout many parts of our manuscript) to visualize and confirm our polymersome morphology.  
+
+In closing Comment #6, the Reviewer suggested that we perform static light scattering (SLS); however, we do not see any value in doing so since SLS can only (within this context) provide indirect evidence on particle shape based on the shape factor \(\rho = R_{g} / R_{h}\) , and not particle morphology as the Reviewer suggested.7 We are confident based on our years of research contributions in the polymersome field (Chem. Soc. Rev. 2019,8 Nat. Commun. 2017,9 Chem. Sci. 2019,10 JACS 2020,11 ACS Nano 2020,12 etc) that TEM and cryo- TEM, both of which we have used to provide direct visual evidence of particle morphology, is a reliable method for confirming polymersome morphology.
+
+<--- Page Split --->
+
+Comment #7: Cryo- TEM will be important to give the necessary details of the polymersome membrane that are not clearly visible from the presented TEM micrographs (Fig. 1). Besides, the resolution of all TEM micrographs should be significantly improved. The change in TEM micrographs as stated, "The three accessible morphological phases are highlighted in Figure 1C using different shades of black" is confusing. It is essential to present the raw data, not highlighted images, to avoid biases.  
+
+## Response:  
+
+First of all, we presume that the TEM image in question is Figure 1F specifically (and not Figure 1 as a whole as noted by the Reviewer in Comment #7). We made this presumption because we only provided one TEM image of polymersomes in Figure 1. For the sake of clarity, we reproduce Figure 1F below:  
+
+![](images/Figure_2a.jpg)
+  
+
+Reproduced Figure 1F. TEM image of polymersomes obtained at \(Q_{\mathrm{organic}} / Q_{\mathrm{total}} = 0.6\) . This sample was analyzed in their respective organic solvent/water mixtures. Shown on the right (and highlighted red) is a magnified region of the same TEM image where the morphology of individual polymersomes can clearly be seen.  
+
+We struggle to understand why the Reviewer isn't convinced that the particles shown above in Reproduced Figure 1F are polymersomes considering how well- resolved the membrane structure of individual particles are in the image. In their comment, the Reviewer further suggested that "cryo- TEM will be important to give the necessary details of the polymersome membrane".  
+
+In response, we point out that there was in fact a reason why we could not measure cryo- TEM for the sample in Figure 1F. To explain this, we refer to Page 6, Lines 1- 2, where we stated that the sample in Figure 1F was "analyzed in their respective organic solvent/water mixtures". For clarity, the "organic solvent/water mixture" in this sample consists of 60 vol% of organic solvent (20% THF/dioxane) and 40 vol% of water. Due to the large amount of organic solvent present in this sample (and its miscibility with liquid ethane, which we use for sample vitrification), we are unable to obtain a vitrified sample that was good enough for cryo- TEM imaging. That said, it is well established in the literature that PEO- b- PS can form glassy polymersomes that retain their structure in the dry state—well enough to be imaged by dry state TEM. \(^{2,13,14}\) TEM imaging of PEO- b- PS polymersomes generally becomes challenging (i) if the polymersomes studied are e.g., \(\geq 500 \mathrm{nm}\) in diameter, because at such sizes, they tend to buckle or collapse when dried because their ( \(\sim 20 \mathrm{nm}\) - thin) membrane structure can no longer support the overall diameter of the structure, or (ii) if the polymersomes have non- spherical shapes that are difficult to properly characterize in the dry- state. In such cases, cryo- TEM becomes a necessary tool to confirm polymersome morphology in a pristine, frozen- hydrated state. \(^{15}\)  
+
+We further point out that we have in fact provided a fair amount of cryo- TEM data throughout the manuscript (wherever feasible and necessary) to confirm our polymersomes' morphology and/or non- spherical shapes. These cryo- TEM images can be found in Figure 2D, Figure 2E, Figure 2H(ii), Figure 2H(iii), and Figure 5C.  
+
+In closing Comment #7, the Reviewer suggests that (i) the "resolution of all TEM micrographs should be significantly improved", (ii) our statement that "The three accessible morphological phases are highlighted in Figure 1C using different shades of black" is confusing, and that "it is essential to present the raw data, not highlighted images, to avoid biases.
+
+<--- Page Split --->
+
+In response to (i) issue with TEM image resolution: All TEM images presented throughout our manuscript were acquired at the highest possible resolution ( \(\sim 4112 \times 3008\) pixels) and saved at a file size between 35- 40 megabytes (MB). Although we disagree that the resolution of our images needs to be improved, we point out that further improvements to image resolution are not possible due to the limitations of the camera (EMSIS Phurona) on our microscope.  
+
+In response to (ii) issue with our statement: We begin by clarifying that the statement in question was made on Page 5, Line 13. It was provided to supplement the figure caption of Figure 1C, which we reproduce as follows:  
+
+"(C) Intensity-averaged hydrodynamic diameters \((D_{h, \text{intensity}})\) and polydispersity indices (PDI) derived from the data shown in B. The different shades of black in C depict a pseudo-phase diagram."  
+
+As can be seen from the above, the figure caption advises readers that the DLS data in Figure 1C has different shades of black to depict a pseudo- phase diagram and to help readers visually identify the different morphologies that can be accessed. The Reviewer appears to have misunderstood our statement since they somehow suggested that "it is essential to present the raw data, not highlighted images, to avoid biases". We clarify that none of our TEM images have been "highlighted" to potentially generate bias—the only thing that has been highlighted was the DLS data in Figure 1C to depict a pseudo- phase diagram.  
+
+We have nevertheless amended the phrases "different shades of black" with "different shades of gray" in our manuscript to hopefully avoid future confusion.  
+
+Comment #8: Figures 2 D and E regarding TEM micrographs of polymersomes are different from what we expect to have when hollow sphere architecture is present; the spherical nanoobjects have a darker core that is not specific for polymersomes. Therefore it is not clear that the nanoobjects in these figures are polymersomes. Cryo- TEM will elucidate this issue.  
+
+## Response:  
+
+We begin our response by reproducing Figures 2D and 2E below:  
+
+![](images/Figure_2A.jpg)
+  
+
+Reproduced Figure 2. TEM images of (D) pristine polymersomes quenched on day 0 (immediately after continuous flow self- assembly) and (E) aged polymersomes quenched after 14 days of aging. Corresponding cryo- TEM images are shown inset in D and E.  
+
+We presume that the Reviewer is referring to the inset cryo- TEM images in Figure 2D and 2E (and not Figure 2D and 2E as a whole) as the inset images are the only images where our polymersomes "have a darker core" that is, according to the Reviewer, "not specific for polymersomes".  
+
+We argue that this "dark core" that the Reviewer raises as an issue is in fact a cryo- TEM imaging artifact that has long been known to exist in the vesicle literature. This imaging artifact has been reported as early as in 2000 by Almgren et al.16 For clarity, we reproduce a cryo- TEM image of liposomes with "dark cores" reported in the cited work, along with the figure caption that was published alongside the cryo- TEM image.
+
+<--- Page Split --->
+![](images/Figure_2B.jpg)
+
+<center>Fig. 6. Large liposomes protruding out of the vitrified film gives an image that is darkest in the central, thickest part. Compare the drawing in Fig. 2. </center>  
+
+Figure R2. A cryo- TEM image of liposomes reported by Almgren et al. \(^{16}\) with "dark cores" similar to what we have presented in Figures 2D and 2E of our manuscript. Note that the original figure caption has been reproduced below the cryo- TEM image for clarity.  
+
+As pointed out by Almgren et al. \(^{16}\) in their figure caption, "the large liposomes protruding out of the vitrified film gives an image that is darkest in the central, thickest part". To help the Reviewer understand this statement, we provide a schematic below in Figure R3 to illustrate the phenomenon:  
+
+![](images/Figure_2A.jpg)
+
+<center>Figure R3. A simplified schematic illustrating why polymersomes (or vesicles in general) sometimes have "dark cores" when visualized under cryo-TEM. </center>  
+
+We further provide four literature examples \(^{15,17 - 19}\) below in Figure R4 where polymersomes have been reported with "dark cores" under cryo- TEM.
+
+<--- Page Split --->
+![PLACEHOLDER_20_0]
+
+<center>Figure 2a-f from Rikken et al. Nat. Commun. 2016, 7, 12606 </center>  
+
+![PLACEHOLDER_20_1]
+
+<center>Figure 2A-C from Ridolfo et al. Polym. Chem. 2020, 11, 2775-2780 </center>  
+
+![PLACEHOLDER_20_2]
+
+<center>Figure 2B-C from Rijpkema et al. Biomacromolecules 2020, 21, 1853-1864 </center>  
+
+![PLACEHOLDER_20_3]
+
+<center>Figure 2A from Abdelmohsen et al. JACS 2016, 138, 9353-9356 </center>  
+
+Figure R4. Cryo- TEM images of polymersomes from 4 different references. In each example provided, every polymersome can be seen to exhibit the same "dark core", which the Reviewer has claimed to be non- specific to polymersomes.  
+
+Finally, in closing Comment #8, the Reviewer suggested that the use of cryo- TEM could potentially "elucidate this issue" (n.b., "this issue" implies the observation of the dark cores). We clarify here that the images shown inset in Figures 2E and 2D, which the Reviewer raised issues with, are in fact cryo- TEM images.  
+
+Comment #9: Figures 4 D- F indicate aggregation of the spherical nanoobjects, which is a severe limitation for further applications. How the aggregation can be avoided?  
+
+## Response:  
+
+We understand that the density of particles in these TEM images gives the impression that our polymersomes are aggregated. However, our DLS data presented in Figures 4B- C very clearly indicates the absence of any aggregation phenomena. Every polymersome sample that was annealed between 20- 70 °C (including those
+
+<--- Page Split --->
+
+whose TEM images were provided in Figure 4D- F) gave monomodal size distributions on DLS with PDIs \(\leq\) 0.10, indicating sample uniformity without the presence of any aggregates. See Table S7 for exact \(D_{\mathrm{h,intensity}}\) and PDI values for each sample. We prefer to retain the same TEM images in Figure 4D- F as they provide an overview of \(>30\) polymersomes per image as opposed to just a select few. Furthermore, if we were to dilute these samples and replace the TEM images to show only a few scattered polymersomes per image, the size differences may not as immediately clear to readers.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+In this manuscript, the authors reported a continuous flow methodology capable of producing near- monodisperse polymersomes at scale ( \(\geq 3\) g/h). They also demonstrated downstream processes (thermal annealing and/or secondary micro- mixing) to manipulate polymersome size (with sub- 40 nm precision) and/or polymersome shape. This work is meaningful for the production of near- monodisperse polymersomes at scale, as well as the control of polymersomes properties under flow conditions. I would recommend its publication on NC after major revision.  
+
+Comment #1: The authors claimed that they specifically adjusted the salinity of the polymersome solution to \(50 \text{mM NaCl}\) , followed by dialysis to remove the organic solvents. Why choose \(50 \text{mM NaCl}\) ?  
+
+## Response:  
+
+We apologize for the lack of clarity.  
+
+We specifically chose \(50 \text{mM NaCl}\) as the osmotic additive because this particular salt and concentration is commonly used to induce osmotic pressures that are sufficiently strong to cause polymersomes to deform into non- spherical shapes. Previous systematic studies \(^{20 - 22}\) have demonstrated that NaCl concentrations below \(50 \text{mM}\) generally result in only partial deformation, while NaCl concentrations exceeding \(50 \text{mM}\) do no produce any noticeable effects beyond complete shape transformation.  
+
+We have amended the text on Page 11, Line 6 and added 3 references to clarify this:  
+
+"We specifically adjusted the salinity of the polymersome solution to \(50 \text{mM NaCl}\) (a concentration regularly used to deform polymersomes by osmotic pressure), \(^{47 - 49}\) followed by dialysis to remove the organic solvents."  
+
+Comment #2: At the highest total flow rate (Qtotal= 8 mL/min), a low PDI of \(0.045 \pm 0.015\) was obtained, indicating near- monodisperse polymersomes. The monodispersity is very interesting, and the authors should investigate the formation mechanism of the monodispersity.  
+
+## Response:  
+
+We begin our response by clarifying that the Reviewer is referring to our data in Figure 3B, which demonstrates that an increase in total flow rate (Qtotal) leads to a significant reduction in polydispersity (PDI). At the highest total flow rate that was tested (Qtotal = 8 mL/min), we observed a particularly low PDI of 0.045 \(\pm 0.015\) , indicating the presence of near- monodisperse polymersomes.  
+
+One plausible explanation for the observed decrease in PDI with increasing total flow rate (Qtotal) is the transition from transient to turbulent flow regime. In the lower range of Qtotal (0.5- 4 mL/min), we observed noticeable improvements in PDI as Qtotal increased. Given that the extent of improvement in PDI was very pronounced in this range, these improvements are likely attributed to micromixing occurring in the transient flow regime. \(^{23}\)  
+
+At the highest Qtotal tested (8 mL/min), the remarkably low PDI observed (0.045 \(\pm 0.015\) ) suggests that the micromixing process is most efficient in this flow regime, thereby enhancing the uniformity of the self- assembly process. Considering how there were minimal improvements in PDI beyond this, we postulate that a complete transition from transient to turbulent flow regime occurs at Qtotal = 8 mL/min.
+
+<--- Page Split --->
+
+While these observations offer valuable insights, it is important to acknowledge that our understanding of the exact flow regimes involved is currently limited, and a comprehensive understanding would require extensive computational studies. This is particularly the case because the micromixer employed in our study is a commercially available product, and thus we lack detailed knowledge of its internal geometry and specifications to properly calculate Reynolds number \((Re)\) , a dimensionless parameter that is used to characterize a fluid flow profile. We would have to engineer a custom micromixer in order to conduct such computational investigations. While we acknowledge the importance of this, we want to clarify that such endeavors are beyond the scope of our current work, which revolves around polymersome metastability and its implications in continuous flow manufacturing. We will consider this in our future research directions, but at this stage, we wish to avoid speculating on flow regimes/profiles in the main text, and prefer to retain our original general explanation attributing the improvements in PDI to an increase in flow turbulence in the micromixer (which we believe is reasonable since total flow rate \((Q_{\text{total}})\) is the only parameter changed in the experiment):  
+
+"We attribute the decreasing size and PDI trends at higher \(Q_{\text{total}}\) to an increase in flow turbulence during micromixing (Figure S5). The effect of \(Q_{\text{total}}\) , however, diminishes beyond \(Q_{\text{total}} \geq 8 \text{mL/min}\) as flow turbulence can no longer be improved beyond the limitations imposed by the geometry of the micromixer."  
+
+Comment #3: A production rate of \(\geq 3 \text{g}\) of polymersomes/hour was achieved. Further increasing Copolymer and Qtotal, the production rate can be improved. Why the authors did not pursue this?  
+
+## Response:  
+
+The Reviewer poses an excellent question here.  
+
+Indeed, the production can be improved by simultaneously increasingly polymer concentration \((c_{\text{polymer}})\) and the total flow rate \((Q_{\text{total}})\) . By augmenting both of these variables in parallel, one can effectively enhance the production rate beyond \(3 \text{g/h}\) . However, we refrained from exploring conditions surpassing \(c_{\text{polymer}} = 9 \text{mg/mL}\) and \(Q_{\text{total}} = 8 \text{mL/min}\) (data shown in Figure 3D) due to the need of more than \(72 \text{mg}\) of \(\text{PEO}_{44} - b - \text{PS}_{86}\) per minute to conduct such experiments.  
+
+To put things into a perspective, let's consider an experimental scenario employing the following self- assembly conditions: \(c_{\text{polymer}} = 20 \text{mg/mL}\) , \(Q_{\text{total}} = 20 \text{mL/min}\) (and \(Q_{\text{organic}} / Q_{\text{total}} = 0.7\) to target polymersomes). Under these conditions, the production rate could reach as high as \(16.8 \text{g/h}\) . However, one needs to appreciate that a mere 5- minute experiment under such conditions would necessitate \(1.4 \text{g}\) of \(\text{PEO}_{44} - b - \text{PS}_{86}\) , which accounts for approximately three- fourths of the total amount of \(\text{PEO}_{44} - b - \text{PS}_{86}\) we have synthesized for the entire project (see experimental section in SI).  
+
+We clarify here that a discussion on this had already been provided in the main text starting Page 13, Line 9:  
+
+"What is important to recognize here is that these flow conditions equate to a production rate of \(\geq 3 \text{g}\) of polymersomes/hour, far exceeding the capabilities of typical batch self- assembly processes. The production rate demonstrated herein can undoubtedly be improved by further increasing \(c_{\text{polymer}}\) and \(Q_{\text{total}}\) , but we did not pursue this as such experiments would require \(>72 \text{mg}\) of \(\text{PEO}_{44} - b - \text{PS}_{86}\) /minute to conduct."  
+
+Comment #4: The authors also manipulated polymersome shape by expanding the flow setup to include a cooling loop and a secondary micromixer connected to another syringe pump. It was crucial to introduce only a small amount of concentrated NaCl solution, as this ensures minimal deviations in solvent quality after micromixing, thus preventing any morphological deviations. How to precisely control a small amount of concentrated NaCl solution? What is the exact value and exact quantity? Any preliminary data/experiment?  
+
+## Response:  
+
+The Reviewer accurately notes that the flow setup had to be expanded to include a cooling loop and a secondary micromixer (connected to a third syringe pump) to manipulate polymersome shape.  
+
+To answer the Reviewer's query, the small amount of concentrated NaCl solution \((c_{\text{NaCl}} = 5.05 \text{M})\) was introduced into the system through the inlet of the secondary micromixer. A photograph of the entire setup
+
+<--- Page Split --->
+
+can be found in Figure S9 in the SI. The amount of 5.0 M NaCl solution added was precisely controlled by the syringe pump connected to the secondary micromixer.  
+
+We clarify that the information mentioned above has been provided in the main text starting Page 17, Line 11:  
+
+"The secondary micromixer, which is placed downstream of the cooling loop (Figure 5A), serves as a junction for the introduction of an additive (NaCl solution) needed to osmotically deform the annealed/grown polymersomes. In a typical experiment, we would generate a salinity change of 50 mM NaCl by micromixing the annealed/grown polymersome solution with a concentrated NaCl solution (5.05 M) at a flow rate of 4 mL/min and 0.04 mL/min, respectively. We found it crucial to introduce only a small amount of concentrated NaCl solution (as opposed to larger volumes of diluted NaCl solution) as this ensures minimal deviations in solvent quality after micromixing, thus preventing any morphological deviations beyond the intended shape transformation process."  
+
+As shown above, we specified that the 5.05 M NaCl solution was introduced into the system at a flow rate of 0.04 mL/min (n.b., these experimental conditions were also provided in the caption of Figure 5C and in the Experimental Section on Page 19 of the SI). In other words, our statement implies that, in every 1 minute, the syringe pump dispenses a total of \(40 \mu \mathrm{L}\) of 5.05 M NaCl solution into the incoming stream of annealed/grown polymersomes, which is conversely flowed at 4 mL/min. Accounting mutual dilution when these two solutions are mixed, the final salinity is thus \(\mathrm{C_{NaCl, final}} = [(\mathrm{C_{NaCl}} \times \mathrm{V_{NaCl}}) / (\mathrm{V_{NaCl}} + \mathrm{V_{polymersome}})] = [(5.05 \mathrm{M} \times 0.040 \mathrm{mL}) / (0.040 \mathrm{mL} + 4 \mathrm{mL})] = 50 \mathrm{mM}\) . Based on the same information, we can also calculate the extent of dilution of the incoming stream using the following formula: \(\% \mathrm{dilution} = [1 - (\mathrm{Q_{total, original}} / \mathrm{Q_{total, original + NaCl}})] \times 100\% ] = [1 - (4 / 4.04) \mathrm{mL / min} \times 100\% ] = 1\%\) , and hence our claims of "minimal deviations in solvent quality after micromixing".  
+
+## Rebuttal References  
+
+1. Markwalter, C. E. et al. Polymeric Nanocarrier Formulations of Biologics Using Inverse Flash NanoPrecipitation. AAPS J. 22, 1–16 (2020).  
+2. Kim, K. T. et al. Polymersome stomatocytes: controlled shape transformation in polymer vesicles. J. Am. Chem. Soc. 132, 12522–12524 (2010).  
+3. Matoori, S. et al. An Investigation of PS-b-PEO Polymersomes for the Oral Treatment and Diagnosis of Hyperammonemia. Small 15, 1–13 (2019).  
+4. Ho, C. S., Kim, J. W. & Weitz, D. A. Microfluidic fabrication of monodisperse biocompatible and biodegradable polymersomes with controlled permeability. J. Am. Chem. Soc. 130, 9543–9549 (2008).  
+5. Shum, H. C., Zhao, Y. J., Kim, S. H. & Weitz, D. A. Multicompartment polymersomes from double emulsions. Angew. Chem. Int. Ed. 50, 1648–1651 (2011).  
+6. Amstad, E., Kim, S. H. & Weitz, D. A. Photo- and thermoresponsive polymersomes for triggered release. Angew. Chem. Int. Ed. 51, 12499–12503 (2012).  
+7. Abdelmohsen, L. K. E. A., Rikken, R. S. M., Christianen, P. C. M., van Hest, J. C. M. & Wilson, D. A. Shape characterization of polymersome morphologies via light scattering techniques. Polymer 107, 445–449 (2016).  
+8. Wong, C. K., Stenzel, M. H. & Thordarson, P. Non-spherical polymersomes: Formation and characterization. Chem. Soc. Rev. 48, 4019–4035 (2019).  
+9. Wong, C. K., Mason, A. F., Stenzel, M. H. & Thordarson, P. Formation of non-spherical polymersomes driven by hydrophobic directional aromatic perylene interactions. Nat. Commun. 8, 1240 (2017).  
+10. Wong, C. K. et al. Faceted polymersomes: a sphere-to-polyhedron shape transformation. Chem. Sci. 10, 2725–2731 (2019).  
+11. Wong, C. K. et al. Vesicular Polymer Hexosomes Exhibit Topological Defects. J. Am. Chem. Soc. 142, 10989–10995 (2020).  
+12. Gröschel, T. I., Wong, C. K., Haataja, J. S., Dias, M. A. & Gröschel, A. H. Direct observation of
+
+<--- Page Split --->
+
+topological defects in striped block copolymer discs and polymersomes. ACS Nano 14, 4829- 4838 (2020).13. Wilson, D. A., Nolte, R. J. M. & van Hest, J. C. M. Autonomous movement of platinum-loaded stomatocytes. Nat. Chem. 4, 268- 74 (2012).14. Van Rhee, P. G. et al. Polymersome magneto- valves for reversible capture and release of nanoparticles. Nat. Commun. 5, 1- 8 (2014).15. Rikken, R. S. M. et al. Shaping polymersomes into predictable morphologies via out- of- equilibrium self- assembly. Nat. Commun. 7, 12606 (2016).16. Almgren, M., Edwards, K. & Karlsson, G. Cryo transmission electron microscopy of liposomes and related structures. Colloids Surfaces A Physicochem. Eng. Asp. 174, 3- 21 (2000).17. Abdelmohsen, L. K. E. A. et al. Formation of well- defined, functional nanotubes via osmotically induced shape transformation of biodegradable polymersomes. J. Am. Chem. Soc. 138, 9353- 9356 (2016).18. Ridolfo, R., Williams, D. S. & Van Hest, J. C. M. Influence of surface charge on the formulation of elongated PEG- : B- PDLLA nanoparticles. Polym. Chem. 11, 2775- 2780 (2020).19. Rijpkema, S. J. et al. Modular Approach to the Functionalization of Polymersomes. Biomacromolecules 21, 1853- 1864 (2020).20. Pijpers, I. A. B., Abdelmohsen, L. K. E. A., Williams, D. S. & Van Hest, J. C. M. Morphology under Control: Engineering Biodegradable Stomatocytes. ACS Macro Lett. 6, 1217- 1222 (2017).21. Wauters, A. C. et al. Development of Morphologically Discrete PEG- PDLLA Nanotubes for Precision Nanomedicine. Biomacromolecules 20, 177- 183 (2019).22. Men, Y., Li, W., Lebleu, C., Sun, J. & Wilson, D. A. Tailoring Polymersome Shape Using the Hofmeister Effect. Biomacromolecules 21, 89- 94 (2020).23. Plutschack, M. B., Pieber, B., Gilmore, K. & Seeberger, P. H. The Hitchhiker's Guide to Flow Chemistry. Chem. Rev. 117, 11796- 11893 (2017).
+
+<--- Page Split --->
+
+Editorial note: Reviewer 2 was unable to look over the responses to the comments, and therefore Reviewer 1 assessed the responses to these comments.  
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+My comments have been thoroughly addressed. I support acceptance of this publication and applaud the authors on their hard work.  
+
+I also believe that the comments of Reviewer 2 were thoroughly addressed.  
+
+Reviewer #3 (Remarks to the Author):  
+
+After revision, the authors did quite a few works to improve the quality of the paper. I would recommend this manuscript for this format to be accepted and published.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+My comments have been thoroughly addressed. I support acceptance of this publication and applaud the authors on their hard work. I also believe that the comments of Reviewer 2 were thoroughly addressed.  
+
+## Response:  
+
+We appreciate the time and support that the Reviewer has dedicated to the reviewing process.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+After revision, the authors did quite a few works to improve the quality of the paper. I would recommend this manuscript for this format to be accepted and published.  
+
+## Response:  
+
+We thank the Reviewer for their time and for endorsing publication of our manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__42e00d20e9d549cfc52e3f47786978426d2bd90cf868b9b2017627c51f0f960a/supplementary_0_Peer Review File__42e00d20e9d549cfc52e3f47786978426d2bd90cf868b9b2017627c51f0f960a_det.mmd

@@ -0,0 +1,717 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[108, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 162, 870, 220]]<|/det|>
+Dynamic Metastable Polymersomes Enable Continuous Flow Manufacturing  
+
+<|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|>text<|/ref|><|det|>[[115, 41, 802, 75]]<|/det|>
+Editorial Note: Parts of this Peer Review File have been redacted as indicated to maintain the confidentiality of unpublished data.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 109, 291, 125]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 130, 394, 146]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 167, 870, 264]]<|/det|>
+The authors have provided a very unique process that can create polymersomes with controllable sizes at much higher throughput than most, if not all, available formulation methods. This work has the potential to be highly impactful in aiding the translation of polymerome technology into clinical trials and beyond. I am very enthusiastic about this paper and really enjoyed reading it. The flow was very logical and experimental evidence is presented for the great majority of claims.  
+
+<|ref|>text<|/ref|><|det|>[[116, 285, 677, 302]]<|/det|>
+I recommend very minor revisions in accordance with the following remarks:  
+
+<|ref|>text<|/ref|><|det|>[[115, 323, 864, 438]]<|/det|>
+Robert Prud'homme's group in Princeton has developed a method called Flash Inverse Nanoprecipitation that is capable of high throughput monodisperse nanoparticle production. I think it could be important to compare and contrast benefits associated with the system developed here and this unique system. Furthermore, inverse nanoprecipitation (solvent injection) is capable of forming monodisperse polymersomes, albeit at low concentrations. It could aid in the discussion to include information about this.  
+
+<|ref|>text<|/ref|><|det|>[[115, 459, 855, 497]]<|/det|>
+The statement made on Page 4 line 14- 15 regarding the syringe size, etc having no effect on the selfassembly process should be supported by citations or experiments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 517, 866, 555]]<|/det|>
+In Figure 1 B it is difficult to distinguish which lines correspond with which concentrations. A color may help here (Similar to Figure 4B).  
+
+<|ref|>text<|/ref|><|det|>[[115, 575, 872, 652]]<|/det|>
+Figure 2G makes it appear as if osmotic pressure is felt only at a point, when Jan Van Hest's group suggests it is more of an elongation force that ultimately leads to internal collapse of stomatocytes. If it is believed that this force is unidirectional, I think it would be helpful to explain why this is believed and support with citations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 673, 864, 730]]<|/det|>
+Based on figures and data alone it is difficult to understand precisely where the "sub- 40 nm precision" conclusion is coming from. Is this meant to be supported by TEM images? It may be helpful to explain this in more detail on page 10 line 6.  
+
+<|ref|>text<|/ref|><|det|>[[115, 751, 823, 769]]<|/det|>
+Claim made at the end of the paragraph on page 12 line 13 needs to be supported with citations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 790, 877, 906]]<|/det|>
+Again, I thoroughly enjoyed this paper and applaud the authors for their very interesting approach to an important translational problem. However, there appears to be a limitation associated which each polymerome having a polystyrene hydrophobic block. PS is not always used in clinical applications, which appear to be dominated by polyesters and other biodegradable blocks. I think it could really increase the impact of the conclusions to discuss potential translation to less hydrophobic or stimuli- responsive blocks.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 130, 393, 145]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 167, 875, 264]]<|/det|>
+Wong et al. describe in this article a continuous flow methodology for production of polymersomes at a relatively large scale ( \(\geq 3 \mathrm{~g} / \mathrm{h}\) ). While the topic of the article is important due to the need of efficient production of nanocarriers for various applications, there are critical issues, which prevent this manuscript for acceptance in Nature Communications. After solving the issues, a revised version will be appropriate to be submitted to a more specialized journal.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 286, 261, 301]]<|/det|>
+## General comments:  
+
+<|ref|>text<|/ref|><|det|>[[114, 305, 879, 440]]<|/det|>
+1. The method presented in the manuscript is based on the combination of a static mixing tee (Y-junction) as a small-scale mixing chamber and a continuous flow setup which reduces the mixing time, while the equilibration loop allows for a good control over the size and shape of polymersomes due to the polymersomes' metastability in the organic solvent/water mixture. However, compared to the current progress of science in the self-assembly process of vesicles formation (polymersomes and giant unilamellar vesicles), the method brings an elegant optimization in one of the polymersome preparation methods however without being a breakthrough in the field.  
+
+<|ref|>text<|/ref|><|det|>[[114, 460, 863, 556]]<|/det|>
+2. The Introduction does not contain the real state-of-art in the field regarding the polymersomes production because the well known film rehydration method for polymersome formation and loading with molecules is not presented with its advantages both in terms of polymersomes high yield production and encapsulation efficiency. The Introduction should be improved to present all relevant methods for polymersomes formation and their advantages or still open questions in the field.  
+
+<|ref|>text<|/ref|><|det|>[[114, 577, 878, 711]]<|/det|>
+3. There is a confusion the authors include in the Introduction by considering synthetic giant unilamellar vesicles (GUVs) as polymersomes. In the field there is a clear distinction between the vesicles with sizes in the nanometer range (polymersomes) and the vesicles with micrometer sizes (GUVs), similarly to the notions of liposomes and lipid GUVs. This distinction is based both on the methods of production and the properties of the vesicles membrane (stability, curvature, etc). The focus of this manuscript is in preparation of polymersomes. Therefore, the Introduction should be corrected accordingly to avoid misunderstanding and to clearly indicate the relevance of the method the authors propose.  
+
+<|ref|>text<|/ref|><|det|>[[114, 732, 866, 847]]<|/det|>
+4. A serious critical aspect is related to the low mechanic stability of polymersomes obtained by the present method, which downgrade them for any application, as they are less stable even than PEGylated-liposomes. Usually polymersomes stability is of several months, depending on the type of amphiphilic copolymer. Therefore, this method should be significantly improved to allow formation of stable polymersomes fort longer periods of time than one week, as this is a real bottle neck for further applications.  
+
+<|ref|>text<|/ref|><|det|>[[114, 869, 872, 906]]<|/det|>
+5. The authors indicate that the method they propose can be used for any type of amphiphilic copolymer. However, they selected as example to test their method for polymersomes formation, PEO
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 871, 205]]<|/det|>
+b- PS, which has specific properties that cannot be extrapolated for other types of amphiphilic copolymers. In addition, both the chemistry of synthesized block copolymers has been well established and the theoretical background of polymer self-assembly via the solvent switch approach has been previously explored in detail, e.g. by the groups of A. Eisenberg, T.P. Lodge or J.C.M. van Hest. To validate this method for a variety of copolymers, the authors should prove it at least for two different types of copolymers in terms of molecular properties.  
+
+<|ref|>text<|/ref|><|det|>[[114, 226, 866, 282]]<|/det|>
+6. The characterization of polymersomes only by DLS (Pag.4 line 25) is not enough to prove the hollow sphere architecture. Static light scattering experiments should be performed and combined to distinguish whether the spherical nanoobjects are polymersomes.  
+
+<|ref|>text<|/ref|><|det|>[[114, 303, 875, 399]]<|/det|>
+7. Cryo-TEM will be important to give the necessary details of the polymersome membrane that are not clearly visible from the presented TEM micrographs (Fig. 1). Besides, the resolution of all TEM micrographs should be significantly improved. The change in TEM micrographs as stated, "The three accessible morphological phases are highlighted in Figure 1C using different shades of black" is confusing. It is essential to present the raw data, not highlighted images, to avoid biases.  
+
+<|ref|>text<|/ref|><|det|>[[114, 420, 875, 496]]<|/det|>
+8. Figures 2 D and E regarding TEM micrographs of polymersomes are different from what we expect to have when hollow sphere architecture is present; the spherical nanoobjects have a darker core that is not specific for polymersomes. Therefore it is not clear that the nanoobjects in these figures are polymersomes. Cryo-TEM will elucidate this issue.  
+
+<|ref|>text<|/ref|><|det|>[[114, 517, 882, 555]]<|/det|>
+9. Figures 4 D-F indicate aggregation of the spherical nanoobjects, which is a severe limitation for further applications. How the aggregation can be avoided?  
+
+<|ref|>text<|/ref|><|det|>[[116, 616, 393, 632]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 654, 881, 770]]<|/det|>
+In this manuscript, the authors reported a continuous flow methodology capable of producing near- monodisperse polymersomes at scale \((\geq 3 \text{g / h})\) . They also demonstrated downstream processes (thermal annealing and/or secondary micro- mixing) to manipulate polymersome size (with sub- 40 nm precision) and/or polymersome shape. This work is meaningful for the production of near- monodisperse polymersomes at scale, as well as the control of polymersomes properties under flow conditions. I would recommend its publication on NC after major revision.  
+
+<|ref|>text<|/ref|><|det|>[[113, 772, 880, 808]]<|/det|>
+1. The authors claimed that they specifically adjusted the salinity of the polymersome solution to \(50 \text{mM NaCl}\) , followed by dialysis to remove the organic solvents. Why choose \(50 \text{mM NaCl}\) ?  
+
+<|ref|>text<|/ref|><|det|>[[113, 810, 875, 866]]<|/det|>
+2. At the highest total flow rate (Qtotal= 8 mL/min), a low PDI of \(0.045 \pm 0.015\) was obtained, indicating near-monodisperse polymersomes. The monodispersity is very interesting, and the authors should investigate the formation mechanism of the monodispersity.  
+
+<|ref|>text<|/ref|><|det|>[[113, 869, 842, 906]]<|/det|>
+3. A production rate of \(\geq 3 \text{g}\) of polymersomes/hour was achieved. Further increasing Copolymer and Qtotal, the production rate can be improved. Why the authors did not pursue this?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 89, 876, 204]]<|/det|>
+4. The authors also manipulated polymersome shape by expanding the flow setup to include a cooling loop and a secondary micromixer connected to another syringe pump. It was crucial to introduce only a small amount of concentrated NaCl solution, as this ensures minimal deviations in solvent quality after micromixing, thus preventing any morphological deviations. How to precisely control a small amount of concentrated NaCl solution? What is the exact value and exact quantity? Any preliminary data/experiment?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[344, 42, 655, 58]]<|/det|>
+Black = comments from the reviewers  
+
+<|ref|>text<|/ref|><|det|>[[346, 79, 650, 94]]<|/det|>
+Blue = our response to the reviewers  
+
+<|ref|>text<|/ref|><|det|>[[241, 114, 755, 131]]<|/det|>
+Green = reproduced texts from the original/revised manuscript  
+
+<|ref|>text<|/ref|><|det|>[[92, 150, 900, 169]]<|/det|>
+All additions/changes made to the manuscript are highlighted in yellow to enable tracked changes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 189, 271, 206]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 241, 388, 258]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[58, 267, 941, 349]]<|/det|>
+The authors have provided a very unique process that can create polymersomes with controllable sizes at much higher throughput than most, if not all, available formulation methods. This work has the potential to be highly impactful in aiding the translation of polymersome technology into clinical trials and beyond. I am very enthusiastic about this paper and really enjoyed reading it. The flow was very logical and experimental evidence is presented for the great majority of claims.  
+
+<|ref|>text<|/ref|><|det|>[[58, 357, 680, 373]]<|/det|>
+I recommend very minor revisions in accordance with the following remarks:  
+
+<|ref|>text<|/ref|><|det|>[[58, 408, 941, 491]]<|/det|>
+Comment #1: Robert Prud'homme's group in Princeton has developed a method called Flash Inverse Nanoprecipitation that is capable of high throughput monodisperse nanoparticle production. I think it could be important to compare and contrast benefits associated with the system developed here and this unique system. Furthermore, inverse nanoprecipitation (solvent injection) is capable of forming monodisperse polymersomes, albeit at low concentrations. It could aid in the discussion to include information about this.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 500, 154, 516]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 525, 417, 542]]<|/det|>
+We thank the Reviewer for their suggestion.  
+
+<|ref|>text<|/ref|><|det|>[[58, 551, 941, 600]]<|/det|>
+We have added a reference to Prof. Robert Prud'homme's first paper where he first introduced the term "flash nanoprecipitation". We have also modified our Introduction to clarify that the use of micromixers to conduct nanoprecipitation is sometimes referred to as flash nanoprecipitation in the literature:  
+
+<|ref|>text<|/ref|><|det|>[[58, 609, 941, 722]]<|/det|>
+"An alternative flow- based approach, sometimes referred to as flash nanoprecipitation, \(^{38}\) relies on the use of miniaturized mixing chambers (micromixers) that reduce the mixing timescale between two incoming solution streams down to the millisecond regime. By employing a micromixer for nanoprecipitation, as opposed to simply conducting nanoprecipitation under batch conditions, one can effectively enhance the uniformity of an overall block copolymer self- assembly process to generate polymersomes in a highly reproducible manner. \(^{39 - }\) \(^{42}\) Although proven effective, most reports on this approach employ the use of micromixers with complex internal geometries that are difficult and expensive to manufacture."  
+
+<|ref|>text<|/ref|><|det|>[[58, 732, 941, 813]]<|/det|>
+Note, however, that we have chosen not to elaborate on inverse flash nanoprecipitation specifically because we view the process as somewhat akin to templated layer- by- layer (LbL) self- assembly, as opposed to a true bottom- up self- assembly/nanoprecipitation process such as that seen in traditional flash nanoprecipitation. In fact, we support our view by quoting a statement made in a recent paper \(^{1}\) by Prof. Prud'homme's group where they distinguished nanoparticles made from inverse flash nanoprecipitation from polymersomes:  
+
+<|ref|>text<|/ref|><|det|>[[58, 821, 941, 871]]<|/det|>
+"Called "inverse Flash NanoPrecipitation (iFNP)," the technique achieves biologic loadings (wt% of total formulation) that are 5- 15x higher than typical values (9- 27% versus < 2%). In contrast to liposomes and polymersomes, we sequentially assemble the polymer layers to form the final nanocarrier".  
+
+<|ref|>text<|/ref|><|det|>[[58, 905, 940, 939]]<|/det|>
+Comment #2: The statement made on Page 4 line 14- 15 regarding the syringe size, etc having no effect on the self- assembly process should be supported by citations or experiments.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[58, 42, 153, 57]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 66, 940, 116]]<|/det|>
+We appreciate the suggestion provided by the Reviewer. However, we do not believe that it is necessary to provide citations or conduct experiments to support our statement made on previously on Page 4, Line 14- 15 (Page 4, Line 17 in the revised manuscript) for the following reasons:  
+
+<|ref|>text<|/ref|><|det|>[[58, 125, 940, 239]]<|/det|>
+Firstly, both syringe size and solution volume are simply device- related parameters that a user inputs into a pump program. They should not be considered as experimental parameters because they do not bear any impact on the self- assembly process. To illustrate this, let's consider two scenarios where a \(10~\mathrm{mL}\) and 20 mL syringe are used in separate self- assembly experiments. Irrespective of which syringe is used, the self- assembly would remain identical because in both experiments the pumps would have been set to the dispense at identical flow rates. The only notably distinction between the two experiments would be the volume of nanoparticles generated, as the \(10~\mathrm{mL}\) syringe has half the capacity of the \(20~\mathrm{mL}\) syringe.  
+
+<|ref|>text<|/ref|><|det|>[[57, 247, 940, 378]]<|/det|>
+Secondly, the length of the equilibration loop likewise has no impact on the self- assembly process. To elaborate, let's again consider two hypothetical experiments: one using a \(1~\mathrm{mL}\) equilibration loop and the other using a \(2~\mathrm{mL}\) equilibration loop. In both scenarios, the self- assembly process would be identical because the equilibration loop is placed downstream to the micromixer, where the self- assembly process truly occurs. The only real difference between the two experiments is the duration of time the nanoparticles resides in the equilibration loop immediately after self- assembly \((t_{\mathrm{residence}})\) . Assuming a total flow rate of \(1~\mathrm{mL / min}\) is used, the \(1~\mathrm{mL}\) equilibration loop would provide a \(t_{\mathrm{residence}} = 1\) min, while the \(2~\mathrm{mL}\) equilibration loop would provide \(t_{\mathrm{residence}} = 2\) min.  
+
+<|ref|>text<|/ref|><|det|>[[57, 387, 940, 485]]<|/det|>
+Now, although our polymersomes are metastable in nature, we know from our aging studies (refer back to Figure 2C) that the lifetime of this metastable state \((t_{\mathrm{metastable}})\) is ca. 7 days. Considering the significant difference in timescales between \(t_{\mathrm{residence}}\) (min/sec) and \(t_{\mathrm{metastable}}\) ( \(\sim 7\) days), it is reasonable to claim that the length of the equilibration loop has no impact on the self- assembly process. Of course, one may argue that the equilibration loop can be extended to cover a \(t_{\mathrm{residence}}\) of several days, but such an experiment would not be very practical.  
+
+<|ref|>text<|/ref|><|det|>[[57, 519, 940, 553]]<|/det|>
+Comment #3: In Figure 1 B it is difficult to distinguish which lines correspond with which concentrations. A color may help here (Similar to Figure 4B).  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 562, 153, 577]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 587, 940, 620]]<|/det|>
+We have updated Figure 1B and Figure 4B to incorporate the Reviewer's suggestions. Both updated figures are reproduced below for clarity.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[108, 48, 850, 404]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[56, 417, 941, 533]]<|/det|>
+<center>Updated Figure 1. (A) Schematic of the continuous flow setup and chemical structure of the polymer (PEO44-b-PS86) used in this work. (B) DLS particle size distributions obtained at different asymmetric flow rates \((Q_{\mathrm{organic}} / Q_{\mathrm{total}})\) . (C) Intensity-averaged hydrodynamic diameters \((D_{\mathrm{h, intensity}})\) and polydispersity indices (PDI) derived from the data shown in B. The different shades of black in C depict a pseudo-phase diagram. TEM images of (D) micelles obtained at \(Q_{\mathrm{organic}} / Q_{\mathrm{total}} = 0.2\) , (E) a mixture of micelles and polymersomes at \(Q_{\mathrm{organic}} / Q_{\mathrm{water}} = 0.4\) and (F) polymersomes obtained at \(Q_{\mathrm{organic}} / Q_{\mathrm{total}} = 0.6\) . All samples were analyzed in their respective organic solvent/water mixtures. </center>
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[221, 60, 770, 480]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[57, 494, 941, 608]]<|/det|>
+<center>Updated Figure 4. (A) Schematic of continuous flow setup used for polymersome self-assembly and downstream annealing to manipulate polymersome size. BPR, backpressure regulator. (B) DLS particle size distributions of aqueous polymersomes prepared at different annealing temperatures ( \(T_{\text{annealing}}\) ). (C) Intensity-averaged hydrodynamic diameters ( \(D_{\text{h, intensity}}\) ) and polydispersity indices (PDI) derived from the data shown in B. TEM images of polymersomes annealed at (D) \(20^{\circ}\mathrm{C}\) , (E) \(50^{\circ}\mathrm{C}\) and (F) \(70^{\circ}\mathrm{C}\) for a residence time under heating ( \(t_{\text{residence, annealing}}\) ) of \(30\mathrm{~s}\) . Flow conditions used for polymersome formation: \(Q_{\text{total}} = 4 \mathrm{~mL / min}\) , \(Q_{\text{organic}} / Q_{\text{total}} = 0.7\) and \(C_{\text{polymer}} = 1 \mathrm{mg / mL}\) . All samples in B-F were dialyzed against water prior to analysis. </center>  
+
+<|ref|>text<|/ref|><|det|>[[57, 642, 941, 708]]<|/det|>
+Comment #4: Figure 2G makes it appear as if osmotic pressure is felt only at a point, when Jan Van Hes't's group suggests it is more of an elongation force that ultimately leads to internal collapse of stomatocytes. If it is believed that this force is unidirectional, I think it would be helpful to explain why this is believed and support with citations.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 718, 153, 733]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 743, 423, 758]]<|/det|>
+We thank the Reviewer for raising this issue.  
+
+<|ref|>text<|/ref|><|det|>[[57, 768, 941, 931]]<|/det|>
+The Reviewer is correct in that the osmotic pressure that is applied onto a polymersome structure during shape transformation is not unidirectional. We had initially thought that the addition of the red arrow in Figure 2G would help guide a non- expert reader to envisage how an originally spherical polymersomes can be deformed into a (bowl- like) stomatocyte structure. In hindsight, we admit that this is misleading since the deformation process is not caused by the application of an external force at a single point as we have indicated with the red arrow in Figure 2G. We clarify that we are not suggesting a shape transformation pathway that is any different to what was proposed by the van Hest group in their seminal work.2 The deformation process is in fact driven by a reduction in internal volume, which is in turn caused by the rapid efflux of organic solvents from the polymersome core due to osmotic imbalance. Despite the misleading schematic, we had already described the shape transformation mechanism on Page 11, Lines 7- 11:
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 41, 941, 108]]<|/det|>
+"This change in salinity generates an osmotic imbalance between the polymersomes' inner compartment and their surrounding solution, causing a net efflux of solvent molecules out of the polymersomes. This in effect drives a reduction in the polymersomes' internal volume and causes the (initially spherical) polymersomes to deform into indented polymersomes known as stomatocytes (TEM and cryo- TEM images in Figure 2H)."  
+
+<|ref|>text<|/ref|><|det|>[[57, 115, 940, 149]]<|/det|>
+To prevent further confusion regarding this matter, we have proceeded to remove the red arrow from Figure 2G.  
+
+<|ref|>text<|/ref|><|det|>[[57, 182, 940, 232]]<|/det|>
+Comment #5: Based on figures and data alone it is difficult to understand precisely where the "sub- 40 nm precision" conclusion is coming from. Is this meant to be supported by TEM images? It may be helpful to explain this in more detail on page 10 line 6.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 242, 154, 257]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 267, 344, 283]]<|/det|>
+We apologise for the lack of clarity.  
+
+<|ref|>text<|/ref|><|det|>[[57, 292, 940, 341]]<|/det|>
+We have added a reference to our DLS data in Figure 2C and Table S2 to back our claim made previously on Page 10, Line 6 (Page 10, Line 9 in the revised manuscript). We have also amended the sentence slightly to enhance clarity:  
+
+<|ref|>text<|/ref|><|det|>[[57, 350, 940, 383]]<|/det|>
+"Our ability to control polymersome- size- distribution mean polymersome size with sub- 40 nm precision (Figure 2C and Table S2) is a feat inconceivable with conventional polymersomes formation methods."  
+
+<|ref|>text<|/ref|><|det|>[[57, 417, 940, 450]]<|/det|>
+Comment #6: Claim made at the end of the paragraph on page 12 line 13 needs to be supported with citations.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 460, 154, 476]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 485, 940, 550]]<|/det|>
+To support our claim made previously on Page 12, Line 13 (Page 12, Line 11 in the revised manuscript), we have added 4 references to papers published between 2011- 2021, where PEO- b- PS polymersomes have been prepared by batch nanoprecipitation. The polymersome production rate in each reference is provided in [bolded brackets] below.  
+
+<|ref|>text<|/ref|><|det|>[[57, 559, 880, 608]]<|/det|>
+50. Meeuwissen, S. A., Kim, K. T., Chen, Y., Pochan, D. J. & van Hest, J. C. M. Controlled shape transformation of polymersome stomatocytes. Angew. Chem. Int. Ed. 50, 7070-7073 (2011). [Polymersome production rate = 3.33 milligrams/hour]  
+
+<|ref|>text<|/ref|><|det|>[[57, 617, 901, 666]]<|/det|>
+51. Nijemeisland, M., Abdelmohsen, L. K. E. A., Huck, W. T. S., Wilson, D. A. & van Hest, J. C. M. A compartmentalized out-of-equilibrium enzymatic reaction network for sustained autonomous movement. ACS Cent. Sci. 2, 843-849 (2016). [Polymersome production rate = 6.67 milligrams/hour]  
+
+<|ref|>text<|/ref|><|det|>[[57, 666, 590, 682]]<|/det|>
+[Polymersome production rate = 6.67 milligrams/hour]  
+
+<|ref|>text<|/ref|><|det|>[[57, 691, 808, 740]]<|/det|>
+52. Kim, J. & Kim, K. T. Polymersome-Based Modular Nanoreactors with Size-Selective Transmembrane Permeability. ACS Appl. Mater. Interfaces 12, 23502-23513 (2020). [Polymersome production rate = 20 milligrams/hour]  
+
+<|ref|>text<|/ref|><|det|>[[57, 750, 896, 799]]<|/det|>
+53. Sun, J., Rijpkema, S. J., Luan, J., Zhang, S. & Wilson, D. A. Generating biomembrane-like local curvature in polymersomes via dynamic polymer insertion. Nat. Commun. 12, 2235 (2021). [Polymersome production rate = 3.33 milligrams/hour]  
+
+<|ref|>text<|/ref|><|det|>[[57, 808, 940, 841]]<|/det|>
+For comparison, the highest polymersome production rate we have achieved is 3.02 grams/hour (data shown in Figure 3D in our manuscript).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[58, 40, 940, 122]]<|/det|>
+Comment #7: Again, I thoroughly enjoyed this paper and applaud the authors for their very interesting approach to an important translational problem. However, there appears to be a limitation associated which each polymerome having a polystyrene hydrophobic block. PS is not always used in clinical applications, which appear to be dominated by polyesters and other biodegradable blocks. I think it could really increase the impact of the conclusions to discuss potential translation to less hydrophobic or stimuli- responsive blocks.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 132, 154, 147]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 157, 941, 336]]<|/det|>
+We acknowledge the concerns raised by the Reviewer regarding the lack of biological relevance of polystyrene (PS) in clinical applications. Although we share the same sentiment as the Reviewer, it is worthwhile pointing out that, despite being non- biodegradable, PEO- b- PS polymersomes still possess value in the medical realms. To elaborate, the Leroux group from ETH Zurich, have for example, recently reported<sup>3</sup> on the use of PEO- b- PS polymersomes for the oral treatment and diagnosis of hyperammonia, a metabolic condition characterized by an abnormally high level of ammonia in blood, which can, at times, be lifethreatening. Prior to publishing this work, the authors of this work had already filed for patent applications worldwide (see e.g., Patent No.: WO2019053578A1, US20200283583A1, EP3668927A1). According to the authors in their Conflict of Interest disclosure statement, the patents have been licensed to Versantis AG, a clinical- stage pharmaceutical company that focuses on the development of new generation orphan medications.  
+
+<|ref|>text<|/ref|><|det|>[[58, 345, 940, 411]]<|/det|>
+Having said the above, we do not deter the fact that clinical applications would ultimately benefit from the use of clinically relevant polymersomes. Biodegradable polyesters or biocompatible stimuli- responsive polymers, as suggested by the Reviewer, certainly hold great promise in this regard. We have taken the advice of the Reviewer and expanded our conclusion to emphasize the need for more clinically relevant polymersomes:  
+
+<|ref|>text<|/ref|><|det|>[[58, 418, 940, 468]]<|/det|>
+"Finally, in order to accelerate the clinical translation of polymersomes, further advancements in this area should prioritize the development of more clinically relevant polymersomes (e.g., biodegradable/stimuli- responsive polymersomes) to ensure optimal efficacy and safety for patients."  
+
+<|ref|>text<|/ref|><|det|>[[58, 477, 940, 544]]<|/det|>
+Finally, we extend our gratitude to the Reviewer for their compliment and valuable feedback aimed at improving our manuscript. After nearly dedicating a decade on research on polymersomes, we genuinely believe that this work stands as one of our most significant contributions to the field. We hope that forthcoming readers will likewise recognize and appreciate the impact of our work.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 577, 388, 593]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[58, 602, 940, 685]]<|/det|>
+Wong et al. describe in this article a continuous flow methodology for production of polymersomes at a relatively large scale (≥ 3 g/h). While the topic of the article is important due to the need of efficient production of nanocarriers for various applications, there are critical issues, which prevent this manuscript for acceptance in Nature Communications. After solving the issues, a revised version will be appropriate to be submitted to a more specialized journal.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 694, 154, 710]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 720, 940, 768]]<|/det|>
+We appreciate the Reviewer's acknowledgement of our work in addressing the pressing need for "efficient production of nanocarriers for various applications" despite the presence of certain "critical issues". We have responded to these concerns in detail below.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 804, 220, 819]]<|/det|>
+## General comments:  
+
+<|ref|>text<|/ref|><|det|>[[58, 829, 940, 944]]<|/det|>
+Comment #1: The method presented in the manuscript is based on the combination of a static mixing tee (Y- junction) as a small- scale mixing chamber and a continuous flow setup which reduces the mixing time, while the equilibration loop allows for a good control over the size and shape of polymersomes due to the polymersomes' metastability in the organic solvent/water mixture. However, compared to the current progress of science in the self- assembly process of vesicles formation (polymersomes and giant unilamellar vesicles), the method brings an elegant optimization in one of the polymersome preparation methods however without being a breakthrough in the field.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[58, 43, 153, 58]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[60, 67, 936, 85]]<|/det|>
+We thank the Reviewer for acknowledging our methodology as an "elegant optimization" of nanoprecipitation.  
+
+<|ref|>text<|/ref|><|det|>[[57, 93, 940, 206]]<|/det|>
+Contrary to the Reviewer's remark that "the equilibration loop allows for a good control over the size and shape of polymersomes", the equilibration loop plays no actual role in size or shape control. In our system, size control was achieved through the use of an annealing loop, which provides heat energy to "nudge" the polymersomes out of their metastable state into lower free energy states. Shape control, on the other hand, was enabled by a secondary downstream micromixer—this allowed us to continuously introduce an additive (NaCl) to osmotically deform the polymersomes into their non- spherical, stomatocyte shape. Neither of the two processes have been demonstrated in a continuous downstream fashion as we have reported.  
+
+<|ref|>text<|/ref|><|det|>[[58, 216, 940, 264]]<|/det|>
+Although we regret that the Reviewer identifies our work as not "being a breakthrough in the field", we will nonetheless attempt to clarify both the novelty and significance of our work through all remaining responses to Reviewer #2 below.  
+
+<|ref|>text<|/ref|><|det|>[[58, 299, 940, 381]]<|/det|>
+Comment #2: The Introduction does not contain the real state- of- art in the field regarding the polymersomes production because the well known film rehydration method for polymersome formation and loading with molecules is not presented with its advantages both in terms of polymersomes high yield production and encapsulation efficiency. The Introduction should be improved to present all relevant methods for polymersomes formation and their advantages or still open questions in the field.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 391, 153, 406]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 415, 940, 610]]<|/det|>
+We believe that we have provided a fairly comprehensive overview of the current state- of- the- art in the Introduction, in particular within the context of nanoprecipitation—the most common bottom- up self- assembly approach in the polymersome field. We acknowledge the existence and importance of the film rehydration method raised by the Reviewer; however, the film rehydration method is a top- down self- assembly approach which bears significant difference to nanoprecipitation (a bottom- up approach) and thus does not fit within the context of our Introduction. The Reviewer also mentions "encapsulation efficiency"; however, since we have not performed any encapsulation experiments using our methodology, we do not see any relevance in elaborating on that topic in the Introduction, especially considering its complexity. Below, we outline how our Introduction has been carefully structured to cover a large breadth of information pertaining to polymersome formation via nanoprecipitation, both in batch and in flow, as well as their associated advantages and limitations, and how our methodology opens up the possibility of performing downstream manipulations—processes that would not have been possible with traditional, kinetically trapped polymersomes:  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 620, 161, 635]]<|/det|>
+## Paragraph 1  
+
+<|ref|>text<|/ref|><|det|>[[87, 645, 940, 713]]<|/det|>
+What polymersomes are and how they structurally resemble liposomes What physicochemical properties of polymersomes can be modified How the above, combined with their ability to load both hydrophilic and hydrophobic materials, has led to widespread applications in drug delivery, synthetic biology and nanoreactor science.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 723, 161, 738]]<|/det|>
+## Paragraph 2  
+
+<|ref|>text<|/ref|><|det|>[[87, 749, 940, 814]]<|/det|>
+Step- by- step explanation outlining how nanoprecipitation is performed in batch to produce polymersomes Explanation of logic behind individual steps in a typical nanoprecipitation process including other important factors to consider  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 825, 161, 840]]<|/det|>
+## Paragraph 3  
+
+<|ref|>text<|/ref|><|det|>[[87, 851, 849, 901]]<|/det|>
+Limitations of batch nanoprecipitation Why nanoprecipitation leads to polydisperse polymersomes due to poor mixing efficiency Why nanoprecipitation is poorly scalable  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 911, 161, 926]]<|/det|>
+## Paragraph 4  
+
+<|ref|>text<|/ref|><|det|>[[87, 937, 872, 953]]<|/det|>
+How researchers have turned to flow- based systems to negate the effects of mixing in batch
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[88, 41, 940, 75]]<|/det|>
+- Examples and advantages/disadvantages of current microfluidic chip-based polymerase formation methods, including:  
+
+<|ref|>text<|/ref|><|det|>[[148, 75, 663, 124]]<|/det|>
+- Double emulsion templating using flow-focusing chips- Laminar and plugged flow mixing using flow-focusing chips- Polymerization-induced self-assembly (PISA) under flow  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 134, 162, 150]]<|/det|>
+## Paragraph 5  
+
+<|ref|>text<|/ref|><|det|>[[87, 160, 941, 291]]<|/det|>
+- How miniaturized mixing chambers (micromixers) are superior to flow-focusing chips in both scalability and reproducibility- How micromixers work by minimizing the timescale of mixing between two incoming streams down to the millisecond regime- Drawbacks of current polymerase formation methods with micromixers, in particular their ability to only produce kinetically trapped polymersomes- Explanation as to why downstream processing/manipulation unlocks the full potential of a continuous flow process  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 301, 162, 317]]<|/det|>
+## Paragraph 6  
+
+<|ref|>text<|/ref|><|det|>[[87, 327, 520, 344]]<|/det|>
+- Summary, novelty, and rationale behind our work  
+
+<|ref|>text<|/ref|><|det|>[[57, 377, 941, 493]]<|/det|>
+Comment #3: There is a confusion the authors include in the Introduction by considering synthetic giant unilamellar vesicles (GUVs) as polymersomes. In the field there is a clear distinction between the vesicles with sizes in the nanometer range (polymersomes) and the vesicles with micrometer sizes (GUVs), similarly to the notions of liposomes and lipid GUVs. This distinction is based both on the methods of production and the properties of the vesicles membrane (stability, curvature, etc). The focus of this manuscript is in preparation of polymersomes. Therefore, the Introduction should be corrected accordingly to avoid misunderstanding and to clearly indicate the relevance of the method the authors propose.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 503, 153, 518]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 527, 941, 592]]<|/det|>
+The Reviewer's concerns stem from what was written in the Introduction starting Page 2, Line 19. We note that this is the only section in our manuscript where we have discussed micrometre- sized polymersomes that could qualify as so- called giant unilamellar vesicles (GUVs). For clarity, we reproduce the text with the phrase in question underlined below:  
+
+<|ref|>text<|/ref|><|det|>[[57, 600, 941, 700]]<|/det|>
+"To negate the effects of batch mixing, researchers have turned to flow- based systems such as microfluidics. A reliable microfluidics approach is the double emulsion templating method,28- 30 which relies on the use of flow- focusing chips to confine and self- assemble block copolymers in the oil phase of water/oil/water (w/o/w) double emulsion droplets. Although the approach generates monodisperse polymersomes with high reproducibility, it is somewhat limited in terms of accessible polymersome size (tens to hundreds of \(\mu \mathrm{m}\) ), and production scalability because the devices used typically only operate at flow rates of only several \(\mu \mathrm{L} / \mathrm{min}\) ."  
+
+<|ref|>text<|/ref|><|det|>[[57, 708, 941, 902]]<|/det|>
+We hope it is clear from the above that we have only briefly mentioned micrometre- sized polymersomes from a microfluidic/flow self- assembly context. Given this information, we do not believe it is necessary to explicitly classify or describe polymersomes that are micrometre- sized as giant unilamellar vesicles (GUVs), especially if we also consider the fact that the pioneers of the technique (Prof. David Weitz and his colleagues) themselves referred to such structures simply as "polymersomes" in their publications.4- 6 Furthermore, from a morphological perspective, there really is no real distinction between sub- micron polymersomes and micrometre- sized polymersomes (e.g., GUVs) other than their sizes, as both "classes" of polymersomes share the same morphology—a bilayer membrane structure and a hollow core. We appreciate that the nomenclature "giant unilamellar vesicles (GUVs)" originates from the liposome field; however, we believe the use of this nomenclature should be context dependent. For instance, it may be appropriate when describing a polymersome system that consists of a mixture of giant unilamellar vesicles (GUVs) and multilamellar vesicles (MLVs).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[58, 40, 940, 121]]<|/det|>
+Comment #4: A serious critical aspect is related to the low mechanic stability of polymersomes obtained by the present method, which downgrade them for any application, as they are less stable even than PEGylated- liposomes. Usually polymersomes stability is of several months, depending on the type of amphiphilic copolymer. Therefore, this method should be significantly improved to allow formation of stable polymersomes for longer periods of time than one week, as this is a real bottle neck for further applications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 131, 154, 147]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 157, 940, 239]]<|/det|>
+The Reviewer is certainly correct that our polymersomes are inherently unstable- this is exactly the novelty of our methodology. In the absence of this metastable state, one would not have been able to accomplish downstream manipulation with the level of control that we have demonstrated throughout our manuscript. We would like to further clarify the potential misconception about (i) our polymersomes' metastability in organic solvent/water mixture and (ii) their inherent stability in water below:  
+
+<|ref|>text<|/ref|><|det|>[[57, 248, 940, 426]]<|/det|>
+Regarding (i) polymersome metastability: Briefly, the term metastability is used to describe a system that exists in an apparent state of equilibrium, when in fact it can transition into a more stable (equilibrium) state if energy is provided to the system. The amount of energy needed for this transition to occur can be quantified as an activation energy barrier \((E_{\mathrm{A}})\) , which in the case of our system, is in the order of \(k_{\mathrm{B}}T\) at room temperature (see Figure 2A in main text for proposed free energy diagram). Under ambient conditions, and in organic solvent/water mixture, our polymersomes grow as a result of metastability for ca. 7 days. After this 7- day period, this growth process ceases entirely as the system has transitioned out of its initial metastable state and into an equilibrium state. As we have described in our manuscript, this transition from metastable to equilibrium state is key to the implementation of the downstream annealing setup in Figure 4, which we used to demonstrate size control immediately after self- assembly (and perhaps more importantly, in the very same continuous stream).  
+
+<|ref|>text<|/ref|><|det|>[[57, 435, 940, 712]]<|/det|>
+Regarding (ii) polymersome stability: As we have discussed on page 9 (line 13 onwards), the metastable state observed in our system can be quenched at any point in time (e.g., during the 7- day growth process or even after the growth has ceased) to trap the system in different kinetically arrested states. This can be done simply by removing the organic solvent from the system (e.g., by extensively dialyzing the polymersomes against water). This is possible owing to high glass transition of polystyrene, PS \((T_{\mathrm{g,PS}})\) , which is \(\sim 100^{\circ}\mathrm{C}\) . Upon removal of the plasticizing organic solvent, PS chains which constitute the polymersome membrane structure transitions from a dynamic plasticized state into a glassy quenched state. Once this quenched state has been reached, no further chain rearrangements (and thus no further morphological changes) are possible. In the quenched state, the polymersomes are indefinitely stable unless, of course, some organic solvent is reintroduced into the system to plasticize the PS membrane or if the block copolymer undergoes chemical degradation (which is unlikely in the case of PEO- b- PS). Throughout the undertaking of this project, we have observed minimal macroscopic precipitation or sedimentation in all our quenched polymersome samples, some of which have been stored at room temperature for as long as 2 years (although we state here that this stability is not intrinsic to our system and is likely common even for PEO- b- PS polymersomes prepared by batch nanoprecipitation). Even if precipitates were present, the samples can simply be filtered through a \(0.45 \mu \mathrm{m}\) polyethersulfone (PES) membrane filter without any deterioration in sample quality as all our polymersome samples are \(< 0.45 \mu \mathrm{m}\) in diameter.  
+
+<|ref|>text<|/ref|><|det|>[[58, 721, 940, 770]]<|/det|>
+With all the above, we dispute the Reviewer's assertion that our polymersomes have "low mechanic(al) stability", "are less stable even than PEGylated- liposomes", and that there "is a real bottle neck for further applications".  
+
+<|ref|>text<|/ref|><|det|>[[57, 805, 940, 936]]<|/det|>
+Comment #5: The authors indicate that the method they propose can be used for any type of amphiphilic copolymer. However, they selected as example to test their method for polymersomes formation, PEO- b- PS, which has specific properties that cannot be extrapolated for other types of amphiphilic copolymers. In addition, both the chemistry of synthesized block copolymers has been well established and the theoretical background of polymer self- assembly via the solvent switch approach has been previously explored in detail, e.g. by the groups of A. Eisenberg, T.P. Lodge or J.C.M. van Hest. To validate this method for a variety of copolymers, the authors should prove it at least for two different types of copolymers in terms of molecular properties.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 42, 155, 56]]<|/det|>
+Response:
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[112, 32, 870, 968]]<|/det|>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 66, 940, 116]]<|/det|>
+Comment #6: The characterization of polymersomes only by DLS (Pag.4 line 25) is not enough to prove the hollow sphere architecture. Static light scattering experiments should be performed and combined to distinguish whether the spherical nanoobjects are polymersomes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 125, 154, 141]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 150, 940, 216]]<|/det|>
+For the purpose of this response, we reproduce the statement in question made previously on Page 4, Line 25 underlined and italicized below (n.b., this statement is now on Page 4, Line 24 in the revised manuscript), along with the two sentences that preceded it. Also reproduced below is the subsequent paragraph (starting Page 5, Line 3 in the revised manuscript; italicized) to help contextualize the statement in question:  
+
+<|ref|>text<|/ref|><|det|>[[58, 225, 940, 306]]<|/det|>
+"We performed the self- assembly process at 7 different asymmetric flow rates ranging from \(Q_{organic} / Q_{total} =\) 0.1- 0.7 (in 0.1 increments). In every case, the product was collected directly into a quartz cuvette and immediately analyzed by dynamic light scattering (DLS). The resulting particle size distributions are shown in Figure 1B. Each sample's intensity- averaged hydrodynamic diameter \((D_{h,intensity})\) and polydispersity index (PDI) are 25 plotted in Figure 1C. A summary of the DLS data is further provided in Table S1.  
+
+<|ref|>text<|/ref|><|det|>[[58, 315, 940, 413]]<|/det|>
+All 7 asymmetric flow rates resulted in monomodal particle size distributions with relatively low PDIs of \(< 0.16\) (Figure 1B- C and Table S1). At \(Q_{organic} / Q_{total} \leq 0.2\) , minimal changes in particle size were observed. Increments above this value, however, resulted in a linear increase in particle size (see \(D_{h,intensity}\) datapoints for \(Q_{organic} / Q_{total} = 0.3 - 0.7\) in Figure 1C). We note here that asymmetric flow rates of \(Q_{organic} / Q_{total} > 0.7\) were also tested, but these flow conditions did not result in any particle formation because \(PEO_{44} - b - PS_{86}\) remains molecularly dissolved when the organic solvent content exceeds 70 vol%.  
+
+<|ref|>text<|/ref|><|det|>[[58, 421, 940, 552]]<|/det|>
+As can be seen from the above, we did not make any claims regarding particle morphology using our DLS data. In fact, all our discussions around DLS (data provided in Figure 1B- C and Table S1) was on the effect of asymmetric flow rates \((Q_{organic} / Q_{total})\) on particle size \((D_{h,intensity})\) . We specifically noted that (i) particle size does not change when the asymmetric flow rate, \(Q_{organic} / Q_{total} \leq 0.2\) , (ii) particle size increases linearly when the asymmetric flow rate is increased from \(Q_{organic} / Q_{total} = 0.3 - 0.7\) , and (iii) no particles form beyond \(Q_{organic} / Q_{total} > 0.7\) because \(PEO_{44} - b - PS_{86}\) is molecularly soluble under those self- assembly conditions. Notice how we were careful in using the term "particle size" as opposed to "micelle size" or "polymersome size" to discuss our DLS data.  
+
+<|ref|>text<|/ref|><|det|>[[58, 560, 940, 594]]<|/det|>
+We only began making claims on particle morphology starting Page 6, Line 3. For the sake of clarity, we reproduce the entire paragraph below in italics, with the claims underlined:  
+
+<|ref|>text<|/ref|><|det|>[[58, 602, 940, 685]]<|/det|>
+"Next, we used transmission electron microscopy (TEM) to probe particle morphology. Shown in Figure 1D- F are three TEM images of particles produced at \(Q_{organic} / Q_{total} = 0.2\) , 0.4 and 0.6, respectively. The gradual increase in \(Q_{organic} / Q_{total}\) generated a morphological transition from micelles (Figure 1D) to a mixed phase of micelles/polymersomes (Figure 1E), and finally to polymersomes (Figure 1F). For clarity, the three accessible morphological phases are highlighted in Figure 1C using different shades of black."  
+
+<|ref|>text<|/ref|><|det|>[[58, 693, 940, 759]]<|/det|>
+One can see from the above that our claims about particle morphology are solely based on the supporting evidence from our TEM data. We claimed the existence of micelles, a mixed phase of micelles/polymersomes, and polymersomes based on the TEM images provided in Figure 1D, Figure 1E and Figure 1F, respectively.  
+
+<|ref|>text<|/ref|><|det|>[[58, 767, 940, 817]]<|/det|>
+To summarize, we did not use DLS as a means to prove the hollow structure of our polymersomes. Instead, we relied on TEM (and cryo- TEM throughout many parts of our manuscript) to visualize and confirm our polymersome morphology.  
+
+<|ref|>text<|/ref|><|det|>[[58, 825, 940, 940]]<|/det|>
+In closing Comment #6, the Reviewer suggested that we perform static light scattering (SLS); however, we do not see any value in doing so since SLS can only (within this context) provide indirect evidence on particle shape based on the shape factor \(\rho = R_{g} / R_{h}\) , and not particle morphology as the Reviewer suggested.7 We are confident based on our years of research contributions in the polymersome field (Chem. Soc. Rev. 2019,8 Nat. Commun. 2017,9 Chem. Sci. 2019,10 JACS 2020,11 ACS Nano 2020,12 etc) that TEM and cryo- TEM, both of which we have used to provide direct visual evidence of particle morphology, is a reliable method for confirming polymersome morphology.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[58, 66, 941, 148]]<|/det|>
+Comment #7: Cryo- TEM will be important to give the necessary details of the polymersome membrane that are not clearly visible from the presented TEM micrographs (Fig. 1). Besides, the resolution of all TEM micrographs should be significantly improved. The change in TEM micrographs as stated, "The three accessible morphological phases are highlighted in Figure 1C using different shades of black" is confusing. It is essential to present the raw data, not highlighted images, to avoid biases.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 158, 154, 174]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 183, 941, 231]]<|/det|>
+First of all, we presume that the TEM image in question is Figure 1F specifically (and not Figure 1 as a whole as noted by the Reviewer in Comment #7). We made this presumption because we only provided one TEM image of polymersomes in Figure 1. For the sake of clarity, we reproduce Figure 1F below:  
+
+<|ref|>image<|/ref|><|det|>[[252, 250, 744, 430]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[58, 444, 941, 510]]<|/det|>
+Reproduced Figure 1F. TEM image of polymersomes obtained at \(Q_{\mathrm{organic}} / Q_{\mathrm{total}} = 0.6\) . This sample was analyzed in their respective organic solvent/water mixtures. Shown on the right (and highlighted red) is a magnified region of the same TEM image where the morphology of individual polymersomes can clearly be seen.  
+
+<|ref|>text<|/ref|><|det|>[[58, 520, 940, 585]]<|/det|>
+We struggle to understand why the Reviewer isn't convinced that the particles shown above in Reproduced Figure 1F are polymersomes considering how well- resolved the membrane structure of individual particles are in the image. In their comment, the Reviewer further suggested that "cryo- TEM will be important to give the necessary details of the polymersome membrane".  
+
+<|ref|>text<|/ref|><|det|>[[57, 594, 941, 805]]<|/det|>
+In response, we point out that there was in fact a reason why we could not measure cryo- TEM for the sample in Figure 1F. To explain this, we refer to Page 6, Lines 1- 2, where we stated that the sample in Figure 1F was "analyzed in their respective organic solvent/water mixtures". For clarity, the "organic solvent/water mixture" in this sample consists of 60 vol% of organic solvent (20% THF/dioxane) and 40 vol% of water. Due to the large amount of organic solvent present in this sample (and its miscibility with liquid ethane, which we use for sample vitrification), we are unable to obtain a vitrified sample that was good enough for cryo- TEM imaging. That said, it is well established in the literature that PEO- b- PS can form glassy polymersomes that retain their structure in the dry state—well enough to be imaged by dry state TEM. \(^{2,13,14}\) TEM imaging of PEO- b- PS polymersomes generally becomes challenging (i) if the polymersomes studied are e.g., \(\geq 500 \mathrm{nm}\) in diameter, because at such sizes, they tend to buckle or collapse when dried because their ( \(\sim 20 \mathrm{nm}\) - thin) membrane structure can no longer support the overall diameter of the structure, or (ii) if the polymersomes have non- spherical shapes that are difficult to properly characterize in the dry- state. In such cases, cryo- TEM becomes a necessary tool to confirm polymersome morphology in a pristine, frozen- hydrated state. \(^{15}\)  
+
+<|ref|>text<|/ref|><|det|>[[58, 815, 940, 863]]<|/det|>
+We further point out that we have in fact provided a fair amount of cryo- TEM data throughout the manuscript (wherever feasible and necessary) to confirm our polymersomes' morphology and/or non- spherical shapes. These cryo- TEM images can be found in Figure 2D, Figure 2E, Figure 2H(ii), Figure 2H(iii), and Figure 5C.  
+
+<|ref|>text<|/ref|><|det|>[[58, 872, 940, 937]]<|/det|>
+In closing Comment #7, the Reviewer suggests that (i) the "resolution of all TEM micrographs should be significantly improved", (ii) our statement that "The three accessible morphological phases are highlighted in Figure 1C using different shades of black" is confusing, and that "it is essential to present the raw data, not highlighted images, to avoid biases.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 40, 941, 123]]<|/det|>
+In response to (i) issue with TEM image resolution: All TEM images presented throughout our manuscript were acquired at the highest possible resolution ( \(\sim 4112 \times 3008\) pixels) and saved at a file size between 35- 40 megabytes (MB). Although we disagree that the resolution of our images needs to be improved, we point out that further improvements to image resolution are not possible due to the limitations of the camera (EMSIS Phurona) on our microscope.  
+
+<|ref|>text<|/ref|><|det|>[[57, 133, 941, 181]]<|/det|>
+In response to (ii) issue with our statement: We begin by clarifying that the statement in question was made on Page 5, Line 13. It was provided to supplement the figure caption of Figure 1C, which we reproduce as follows:  
+
+<|ref|>text<|/ref|><|det|>[[57, 191, 940, 224]]<|/det|>
+"(C) Intensity-averaged hydrodynamic diameters \((D_{h, \text{intensity}})\) and polydispersity indices (PDI) derived from the data shown in B. The different shades of black in C depict a pseudo-phase diagram."  
+
+<|ref|>text<|/ref|><|det|>[[57, 234, 941, 331]]<|/det|>
+As can be seen from the above, the figure caption advises readers that the DLS data in Figure 1C has different shades of black to depict a pseudo- phase diagram and to help readers visually identify the different morphologies that can be accessed. The Reviewer appears to have misunderstood our statement since they somehow suggested that "it is essential to present the raw data, not highlighted images, to avoid biases". We clarify that none of our TEM images have been "highlighted" to potentially generate bias—the only thing that has been highlighted was the DLS data in Figure 1C to depict a pseudo- phase diagram.  
+
+<|ref|>text<|/ref|><|det|>[[57, 340, 940, 373]]<|/det|>
+We have nevertheless amended the phrases "different shades of black" with "different shades of gray" in our manuscript to hopefully avoid future confusion.  
+
+<|ref|>text<|/ref|><|det|>[[57, 407, 941, 472]]<|/det|>
+Comment #8: Figures 2 D and E regarding TEM micrographs of polymersomes are different from what we expect to have when hollow sphere architecture is present; the spherical nanoobjects have a darker core that is not specific for polymersomes. Therefore it is not clear that the nanoobjects in these figures are polymersomes. Cryo- TEM will elucidate this issue.  
+
+<|ref|>sub_title<|/ref|><|det|>[[57, 482, 154, 497]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 508, 597, 524]]<|/det|>
+We begin our response by reproducing Figures 2D and 2E below:  
+
+<|ref|>image<|/ref|><|det|>[[273, 533, 720, 693]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[57, 706, 941, 755]]<|/det|>
+Reproduced Figure 2. TEM images of (D) pristine polymersomes quenched on day 0 (immediately after continuous flow self- assembly) and (E) aged polymersomes quenched after 14 days of aging. Corresponding cryo- TEM images are shown inset in D and E.  
+
+<|ref|>text<|/ref|><|det|>[[57, 765, 941, 813]]<|/det|>
+We presume that the Reviewer is referring to the inset cryo- TEM images in Figure 2D and 2E (and not Figure 2D and 2E as a whole) as the inset images are the only images where our polymersomes "have a darker core" that is, according to the Reviewer, "not specific for polymersomes".  
+
+<|ref|>text<|/ref|><|det|>[[57, 823, 941, 888]]<|/det|>
+We argue that this "dark core" that the Reviewer raises as an issue is in fact a cryo- TEM imaging artifact that has long been known to exist in the vesicle literature. This imaging artifact has been reported as early as in 2000 by Almgren et al.16 For clarity, we reproduce a cryo- TEM image of liposomes with "dark cores" reported in the cited work, along with the figure caption that was published alongside the cryo- TEM image.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[339, 48, 656, 223]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[331, 231, 660, 264]]<|/det|>
+<center>Fig. 6. Large liposomes protruding out of the vitrified film gives an image that is darkest in the central, thickest part. Compare the drawing in Fig. 2. </center>  
+
+<|ref|>text<|/ref|><|det|>[[57, 279, 940, 328]]<|/det|>
+Figure R2. A cryo- TEM image of liposomes reported by Almgren et al. \(^{16}\) with "dark cores" similar to what we have presented in Figures 2D and 2E of our manuscript. Note that the original figure caption has been reproduced below the cryo- TEM image for clarity.  
+
+<|ref|>text<|/ref|><|det|>[[57, 362, 940, 411]]<|/det|>
+As pointed out by Almgren et al. \(^{16}\) in their figure caption, "the large liposomes protruding out of the vitrified film gives an image that is darkest in the central, thickest part". To help the Reviewer understand this statement, we provide a schematic below in Figure R3 to illustrate the phenomenon:  
+
+<|ref|>image<|/ref|><|det|>[[277, 460, 757, 592]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[57, 610, 940, 642]]<|/det|>
+<center>Figure R3. A simplified schematic illustrating why polymersomes (or vesicles in general) sometimes have "dark cores" when visualized under cryo-TEM. </center>  
+
+<|ref|>text<|/ref|><|det|>[[57, 677, 940, 710]]<|/det|>
+We further provide four literature examples \(^{15,17 - 19}\) below in Figure R4 where polymersomes have been reported with "dark cores" under cryo- TEM.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[243, 68, 625, 275]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[196, 46, 675, 62]]<|/det|>
+<center>Figure 2a-f from Rikken et al. Nat. Commun. 2016, 7, 12606 </center>  
+
+<|ref|>image<|/ref|><|det|>[[175, 325, 694, 464]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[180, 300, 711, 316]]<|/det|>
+<center>Figure 2A-C from Ridolfo et al. Polym. Chem. 2020, 11, 2775-2780 </center>  
+
+<|ref|>image<|/ref|><|det|>[[96, 525, 446, 644]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[108, 488, 436, 519]]<|/det|>
+<center>Figure 2B-C from Rijpkema et al. Biomacromolecules 2020, 21, 1853-1864 </center>  
+
+<|ref|>image<|/ref|><|det|>[[537, 522, 714, 644]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[488, 488, 768, 519]]<|/det|>
+<center>Figure 2A from Abdelmohsen et al. JACS 2016, 138, 9353-9356 </center>  
+
+<|ref|>text<|/ref|><|det|>[[57, 666, 940, 715]]<|/det|>
+Figure R4. Cryo- TEM images of polymersomes from 4 different references. In each example provided, every polymersome can be seen to exhibit the same "dark core", which the Reviewer has claimed to be non- specific to polymersomes.  
+
+<|ref|>text<|/ref|><|det|>[[57, 748, 940, 798]]<|/det|>
+Finally, in closing Comment #8, the Reviewer suggested that the use of cryo- TEM could potentially "elucidate this issue" (n.b., "this issue" implies the observation of the dark cores). We clarify here that the images shown inset in Figures 2E and 2D, which the Reviewer raised issues with, are in fact cryo- TEM images.  
+
+<|ref|>text<|/ref|><|det|>[[57, 832, 940, 865]]<|/det|>
+Comment #9: Figures 4 D- F indicate aggregation of the spherical nanoobjects, which is a severe limitation for further applications. How the aggregation can be avoided?  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 875, 153, 890]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 899, 940, 949]]<|/det|>
+We understand that the density of particles in these TEM images gives the impression that our polymersomes are aggregated. However, our DLS data presented in Figures 4B- C very clearly indicates the absence of any aggregation phenomena. Every polymersome sample that was annealed between 20- 70 °C (including those
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 40, 940, 138]]<|/det|>
+whose TEM images were provided in Figure 4D- F) gave monomodal size distributions on DLS with PDIs \(\leq\) 0.10, indicating sample uniformity without the presence of any aggregates. See Table S7 for exact \(D_{\mathrm{h,intensity}}\) and PDI values for each sample. We prefer to retain the same TEM images in Figure 4D- F as they provide an overview of \(>30\) polymersomes per image as opposed to just a select few. Furthermore, if we were to dilute these samples and replace the TEM images to show only a few scattered polymersomes per image, the size differences may not as immediately clear to readers.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 173, 388, 190]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[57, 199, 940, 297]]<|/det|>
+In this manuscript, the authors reported a continuous flow methodology capable of producing near- monodisperse polymersomes at scale ( \(\geq 3\) g/h). They also demonstrated downstream processes (thermal annealing and/or secondary micro- mixing) to manipulate polymersome size (with sub- 40 nm precision) and/or polymersome shape. This work is meaningful for the production of near- monodisperse polymersomes at scale, as well as the control of polymersomes properties under flow conditions. I would recommend its publication on NC after major revision.  
+
+<|ref|>text<|/ref|><|det|>[[57, 332, 940, 364]]<|/det|>
+Comment #1: The authors claimed that they specifically adjusted the salinity of the polymersome solution to \(50 \text{mM NaCl}\) , followed by dialysis to remove the organic solvents. Why choose \(50 \text{mM NaCl}\) ?  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 375, 154, 390]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 400, 344, 416]]<|/det|>
+We apologize for the lack of clarity.  
+
+<|ref|>text<|/ref|><|det|>[[57, 425, 940, 506]]<|/det|>
+We specifically chose \(50 \text{mM NaCl}\) as the osmotic additive because this particular salt and concentration is commonly used to induce osmotic pressures that are sufficiently strong to cause polymersomes to deform into non- spherical shapes. Previous systematic studies \(^{20 - 22}\) have demonstrated that NaCl concentrations below \(50 \text{mM}\) generally result in only partial deformation, while NaCl concentrations exceeding \(50 \text{mM}\) do no produce any noticeable effects beyond complete shape transformation.  
+
+<|ref|>text<|/ref|><|det|>[[58, 515, 752, 531]]<|/det|>
+We have amended the text on Page 11, Line 6 and added 3 references to clarify this:  
+
+<|ref|>text<|/ref|><|det|>[[57, 541, 940, 574]]<|/det|>
+"We specifically adjusted the salinity of the polymersome solution to \(50 \text{mM NaCl}\) (a concentration regularly used to deform polymersomes by osmotic pressure), \(^{47 - 49}\) followed by dialysis to remove the organic solvents."  
+
+<|ref|>text<|/ref|><|det|>[[57, 608, 940, 658]]<|/det|>
+Comment #2: At the highest total flow rate (Qtotal= 8 mL/min), a low PDI of \(0.045 \pm 0.015\) was obtained, indicating near- monodisperse polymersomes. The monodispersity is very interesting, and the authors should investigate the formation mechanism of the monodispersity.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 668, 154, 683]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[57, 693, 940, 757]]<|/det|>
+We begin our response by clarifying that the Reviewer is referring to our data in Figure 3B, which demonstrates that an increase in total flow rate (Qtotal) leads to a significant reduction in polydispersity (PDI). At the highest total flow rate that was tested (Qtotal = 8 mL/min), we observed a particularly low PDI of 0.045 \(\pm 0.015\) , indicating the presence of near- monodisperse polymersomes.  
+
+<|ref|>text<|/ref|><|det|>[[57, 767, 940, 847]]<|/det|>
+One plausible explanation for the observed decrease in PDI with increasing total flow rate (Qtotal) is the transition from transient to turbulent flow regime. In the lower range of Qtotal (0.5- 4 mL/min), we observed noticeable improvements in PDI as Qtotal increased. Given that the extent of improvement in PDI was very pronounced in this range, these improvements are likely attributed to micromixing occurring in the transient flow regime. \(^{23}\)  
+
+<|ref|>text<|/ref|><|det|>[[57, 857, 940, 922]]<|/det|>
+At the highest Qtotal tested (8 mL/min), the remarkably low PDI observed (0.045 \(\pm 0.015\) ) suggests that the micromixing process is most efficient in this flow regime, thereby enhancing the uniformity of the self- assembly process. Considering how there were minimal improvements in PDI beyond this, we postulate that a complete transition from transient to turbulent flow regime occurs at Qtotal = 8 mL/min.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 40, 940, 252]]<|/det|>
+While these observations offer valuable insights, it is important to acknowledge that our understanding of the exact flow regimes involved is currently limited, and a comprehensive understanding would require extensive computational studies. This is particularly the case because the micromixer employed in our study is a commercially available product, and thus we lack detailed knowledge of its internal geometry and specifications to properly calculate Reynolds number \((Re)\) , a dimensionless parameter that is used to characterize a fluid flow profile. We would have to engineer a custom micromixer in order to conduct such computational investigations. While we acknowledge the importance of this, we want to clarify that such endeavors are beyond the scope of our current work, which revolves around polymersome metastability and its implications in continuous flow manufacturing. We will consider this in our future research directions, but at this stage, we wish to avoid speculating on flow regimes/profiles in the main text, and prefer to retain our original general explanation attributing the improvements in PDI to an increase in flow turbulence in the micromixer (which we believe is reasonable since total flow rate \((Q_{\text{total}})\) is the only parameter changed in the experiment):  
+
+<|ref|>text<|/ref|><|det|>[[58, 260, 940, 310]]<|/det|>
+"We attribute the decreasing size and PDI trends at higher \(Q_{\text{total}}\) to an increase in flow turbulence during micromixing (Figure S5). The effect of \(Q_{\text{total}}\) , however, diminishes beyond \(Q_{\text{total}} \geq 8 \text{mL/min}\) as flow turbulence can no longer be improved beyond the limitations imposed by the geometry of the micromixer."  
+
+<|ref|>text<|/ref|><|det|>[[58, 344, 940, 378]]<|/det|>
+Comment #3: A production rate of \(\geq 3 \text{g}\) of polymersomes/hour was achieved. Further increasing Copolymer and Qtotal, the production rate can be improved. Why the authors did not pursue this?  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 387, 153, 402]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 412, 454, 428]]<|/det|>
+The Reviewer poses an excellent question here.  
+
+<|ref|>text<|/ref|><|det|>[[58, 437, 940, 518]]<|/det|>
+Indeed, the production can be improved by simultaneously increasingly polymer concentration \((c_{\text{polymer}})\) and the total flow rate \((Q_{\text{total}})\) . By augmenting both of these variables in parallel, one can effectively enhance the production rate beyond \(3 \text{g/h}\) . However, we refrained from exploring conditions surpassing \(c_{\text{polymer}} = 9 \text{mg/mL}\) and \(Q_{\text{total}} = 8 \text{mL/min}\) (data shown in Figure 3D) due to the need of more than \(72 \text{mg}\) of \(\text{PEO}_{44} - b - \text{PS}_{86}\) per minute to conduct such experiments.  
+
+<|ref|>text<|/ref|><|det|>[[58, 527, 940, 626]]<|/det|>
+To put things into a perspective, let's consider an experimental scenario employing the following self- assembly conditions: \(c_{\text{polymer}} = 20 \text{mg/mL}\) , \(Q_{\text{total}} = 20 \text{mL/min}\) (and \(Q_{\text{organic}} / Q_{\text{total}} = 0.7\) to target polymersomes). Under these conditions, the production rate could reach as high as \(16.8 \text{g/h}\) . However, one needs to appreciate that a mere 5- minute experiment under such conditions would necessitate \(1.4 \text{g}\) of \(\text{PEO}_{44} - b - \text{PS}_{86}\) , which accounts for approximately three- fourths of the total amount of \(\text{PEO}_{44} - b - \text{PS}_{86}\) we have synthesized for the entire project (see experimental section in SI).  
+
+<|ref|>text<|/ref|><|det|>[[60, 634, 940, 652]]<|/det|>
+We clarify here that a discussion on this had already been provided in the main text starting Page 13, Line 9:  
+
+<|ref|>text<|/ref|><|det|>[[58, 660, 940, 727]]<|/det|>
+"What is important to recognize here is that these flow conditions equate to a production rate of \(\geq 3 \text{g}\) of polymersomes/hour, far exceeding the capabilities of typical batch self- assembly processes. The production rate demonstrated herein can undoubtedly be improved by further increasing \(c_{\text{polymer}}\) and \(Q_{\text{total}}\) , but we did not pursue this as such experiments would require \(>72 \text{mg}\) of \(\text{PEO}_{44} - b - \text{PS}_{86}\) /minute to conduct."  
+
+<|ref|>text<|/ref|><|det|>[[58, 761, 940, 843]]<|/det|>
+Comment #4: The authors also manipulated polymersome shape by expanding the flow setup to include a cooling loop and a secondary micromixer connected to another syringe pump. It was crucial to introduce only a small amount of concentrated NaCl solution, as this ensures minimal deviations in solvent quality after micromixing, thus preventing any morphological deviations. How to precisely control a small amount of concentrated NaCl solution? What is the exact value and exact quantity? Any preliminary data/experiment?  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 853, 153, 868]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 877, 940, 910]]<|/det|>
+The Reviewer accurately notes that the flow setup had to be expanded to include a cooling loop and a secondary micromixer (connected to a third syringe pump) to manipulate polymersome shape.  
+
+<|ref|>text<|/ref|><|det|>[[58, 918, 940, 952]]<|/det|>
+To answer the Reviewer's query, the small amount of concentrated NaCl solution \((c_{\text{NaCl}} = 5.05 \text{M})\) was introduced into the system through the inlet of the secondary micromixer. A photograph of the entire setup
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 41, 939, 74]]<|/det|>
+can be found in Figure S9 in the SI. The amount of 5.0 M NaCl solution added was precisely controlled by the syringe pump connected to the secondary micromixer.  
+
+<|ref|>text<|/ref|><|det|>[[57, 83, 940, 115]]<|/det|>
+We clarify that the information mentioned above has been provided in the main text starting Page 17, Line 11:  
+
+<|ref|>text<|/ref|><|det|>[[57, 125, 940, 255]]<|/det|>
+"The secondary micromixer, which is placed downstream of the cooling loop (Figure 5A), serves as a junction for the introduction of an additive (NaCl solution) needed to osmotically deform the annealed/grown polymersomes. In a typical experiment, we would generate a salinity change of 50 mM NaCl by micromixing the annealed/grown polymersome solution with a concentrated NaCl solution (5.05 M) at a flow rate of 4 mL/min and 0.04 mL/min, respectively. We found it crucial to introduce only a small amount of concentrated NaCl solution (as opposed to larger volumes of diluted NaCl solution) as this ensures minimal deviations in solvent quality after micromixing, thus preventing any morphological deviations beyond the intended shape transformation process."  
+
+<|ref|>text<|/ref|><|det|>[[57, 265, 940, 426]]<|/det|>
+As shown above, we specified that the 5.05 M NaCl solution was introduced into the system at a flow rate of 0.04 mL/min (n.b., these experimental conditions were also provided in the caption of Figure 5C and in the Experimental Section on Page 19 of the SI). In other words, our statement implies that, in every 1 minute, the syringe pump dispenses a total of \(40 \mu \mathrm{L}\) of 5.05 M NaCl solution into the incoming stream of annealed/grown polymersomes, which is conversely flowed at 4 mL/min. Accounting mutual dilution when these two solutions are mixed, the final salinity is thus \(\mathrm{C_{NaCl, final}} = [(\mathrm{C_{NaCl}} \times \mathrm{V_{NaCl}}) / (\mathrm{V_{NaCl}} + \mathrm{V_{polymersome}})] = [(5.05 \mathrm{M} \times 0.040 \mathrm{mL}) / (0.040 \mathrm{mL} + 4 \mathrm{mL})] = 50 \mathrm{mM}\) . Based on the same information, we can also calculate the extent of dilution of the incoming stream using the following formula: \(\% \mathrm{dilution} = [1 - (\mathrm{Q_{total, original}} / \mathrm{Q_{total, original + NaCl}})] \times 100\% ] = [1 - (4 / 4.04) \mathrm{mL / min} \times 100\% ] = 1\%\) , and hence our claims of "minimal deviations in solvent quality after micromixing".  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 461, 238, 477]]<|/det|>
+## Rebuttal References  
+
+<|ref|>text<|/ref|><|det|>[[55, 486, 940, 952]]<|/det|>
+1. Markwalter, C. E. et al. Polymeric Nanocarrier Formulations of Biologics Using Inverse Flash NanoPrecipitation. AAPS J. 22, 1–16 (2020).  
+2. Kim, K. T. et al. Polymersome stomatocytes: controlled shape transformation in polymer vesicles. J. Am. Chem. Soc. 132, 12522–12524 (2010).  
+3. Matoori, S. et al. An Investigation of PS-b-PEO Polymersomes for the Oral Treatment and Diagnosis of Hyperammonemia. Small 15, 1–13 (2019).  
+4. Ho, C. S., Kim, J. W. & Weitz, D. A. Microfluidic fabrication of monodisperse biocompatible and biodegradable polymersomes with controlled permeability. J. Am. Chem. Soc. 130, 9543–9549 (2008).  
+5. Shum, H. C., Zhao, Y. J., Kim, S. H. & Weitz, D. A. Multicompartment polymersomes from double emulsions. Angew. Chem. Int. Ed. 50, 1648–1651 (2011).  
+6. Amstad, E., Kim, S. H. & Weitz, D. A. Photo- and thermoresponsive polymersomes for triggered release. Angew. Chem. Int. Ed. 51, 12499–12503 (2012).  
+7. Abdelmohsen, L. K. E. A., Rikken, R. S. M., Christianen, P. C. M., van Hest, J. C. M. & Wilson, D. A. Shape characterization of polymersome morphologies via light scattering techniques. Polymer 107, 445–449 (2016).  
+8. Wong, C. K., Stenzel, M. H. & Thordarson, P. Non-spherical polymersomes: Formation and characterization. Chem. Soc. Rev. 48, 4019–4035 (2019).  
+9. Wong, C. K., Mason, A. F., Stenzel, M. H. & Thordarson, P. Formation of non-spherical polymersomes driven by hydrophobic directional aromatic perylene interactions. Nat. Commun. 8, 1240 (2017).  
+10. Wong, C. K. et al. Faceted polymersomes: a sphere-to-polyhedron shape transformation. Chem. Sci. 10, 2725–2731 (2019).  
+11. Wong, C. K. et al. Vesicular Polymer Hexosomes Exhibit Topological Defects. J. Am. Chem. Soc. 142, 10989–10995 (2020).  
+12. Gröschel, T. I., Wong, C. K., Haataja, J. S., Dias, M. A. & Gröschel, A. H. Direct observation of
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 40, 941, 510]]<|/det|>
+topological defects in striped block copolymer discs and polymersomes. ACS Nano 14, 4829- 4838 (2020).13. Wilson, D. A., Nolte, R. J. M. & van Hest, J. C. M. Autonomous movement of platinum-loaded stomatocytes. Nat. Chem. 4, 268- 74 (2012).14. Van Rhee, P. G. et al. Polymersome magneto- valves for reversible capture and release of nanoparticles. Nat. Commun. 5, 1- 8 (2014).15. Rikken, R. S. M. et al. Shaping polymersomes into predictable morphologies via out- of- equilibrium self- assembly. Nat. Commun. 7, 12606 (2016).16. Almgren, M., Edwards, K. & Karlsson, G. Cryo transmission electron microscopy of liposomes and related structures. Colloids Surfaces A Physicochem. Eng. Asp. 174, 3- 21 (2000).17. Abdelmohsen, L. K. E. A. et al. Formation of well- defined, functional nanotubes via osmotically induced shape transformation of biodegradable polymersomes. J. Am. Chem. Soc. 138, 9353- 9356 (2016).18. Ridolfo, R., Williams, D. S. & Van Hest, J. C. M. Influence of surface charge on the formulation of elongated PEG- : B- PDLLA nanoparticles. Polym. Chem. 11, 2775- 2780 (2020).19. Rijpkema, S. J. et al. Modular Approach to the Functionalization of Polymersomes. Biomacromolecules 21, 1853- 1864 (2020).20. Pijpers, I. A. B., Abdelmohsen, L. K. E. A., Williams, D. S. & Van Hest, J. C. M. Morphology under Control: Engineering Biodegradable Stomatocytes. ACS Macro Lett. 6, 1217- 1222 (2017).21. Wauters, A. C. et al. Development of Morphologically Discrete PEG- PDLLA Nanotubes for Precision Nanomedicine. Biomacromolecules 20, 177- 183 (2019).22. Men, Y., Li, W., Lebleu, C., Sun, J. & Wilson, D. A. Tailoring Polymersome Shape Using the Hofmeister Effect. Biomacromolecules 21, 89- 94 (2020).23. Plutschack, M. B., Pieber, B., Gilmore, K. & Seeberger, P. H. The Hitchhiker's Guide to Flow Chemistry. Chem. Rev. 117, 11796- 11893 (2017).
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[108, 74, 901, 108]]<|/det|>
+Editorial note: Reviewer 2 was unable to look over the responses to the comments, and therefore Reviewer 1 assessed the responses to these comments.  
+
+<|ref|>sub_title<|/ref|><|det|>[[110, 125, 298, 142]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[110, 159, 389, 175]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[108, 184, 852, 220]]<|/det|>
+My comments have been thoroughly addressed. I support acceptance of this publication and applaud the authors on their hard work.  
+
+<|ref|>text<|/ref|><|det|>[[108, 225, 661, 242]]<|/det|>
+I also believe that the comments of Reviewer 2 were thoroughly addressed.  
+
+<|ref|>text<|/ref|><|det|>[[108, 274, 387, 291]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[108, 298, 787, 335]]<|/det|>
+After revision, the authors did quite a few works to improve the quality of the paper. I would recommend this manuscript for this format to be accepted and published.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[58, 40, 272, 58]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 92, 389, 109]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[58, 118, 940, 152]]<|/det|>
+My comments have been thoroughly addressed. I support acceptance of this publication and applaud the authors on their hard work. I also believe that the comments of Reviewer 2 were thoroughly addressed.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 161, 155, 176]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 185, 830, 202]]<|/det|>
+We appreciate the time and support that the Reviewer has dedicated to the reviewing process.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 237, 389, 254]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[58, 263, 940, 297]]<|/det|>
+After revision, the authors did quite a few works to improve the quality of the paper. I would recommend this manuscript for this format to be accepted and published.  
+
+<|ref|>sub_title<|/ref|><|det|>[[58, 305, 155, 321]]<|/det|>
+## Response:  
+
+<|ref|>text<|/ref|><|det|>[[58, 330, 754, 347]]<|/det|>
+We thank the Reviewer for their time and for endorsing publication of our manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__43470755cca0263d20ba27a2e315ebda8bd3f51b485e75fb05673e70392cb22a/images_list.json

@@ -0,0 +1,10 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure for reviewer. Comparison of Simoa assays with Luminex in TruCulture stimulations; Null, Poly:IC, LPS, R848. Spearman correlations were performed per stimuli on each assay comparison.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  }
+]

peer_reviews/supplementary_0_Peer Review File__43470755cca0263d20ba27a2e315ebda8bd3f51b485e75fb05673e70392cb22a/supplementary_0_Peer Review File__43470755cca0263d20ba27a2e315ebda8bd3f51b485e75fb05673e70392cb22a.mmd

@@ -0,0 +1,218 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Detection of defective type I interferon immunity is associated with increasing COVID- 19 severity  
+
+![](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 this manuscript, Smith, Possemé and Bondet et al studied type I interferon responses in samples from several COVID- 19 patient cohorts in depth. First, the authors confirmed their previous findings (PMID: 32661059) showing that IFN responses were impaired in critical patients, where the concentration of IFN- alpha found in plasma samples inversely correlated with disease severity. For this, they highlighted the need to use digital ELISA, a highly sensitive assay to measure interferon concentration, as Luminex couldn't pick differences between patient groups. Moreover, and contrary to the Luminex data, data from digital ELISA correlated well with the "ISG score" (based on the relative quantification of 6 prototype ISG expression) and interferon activity (as measured by protection against virus- induced cytopathic effects). They further showed that blood cells from critical patients globally responded less well to various stimuli (poly(I:C), LPS, R848) and confirmed that their pDC count was significantly lower, whereas monocyte counts were increased. Using the nanostring technology on 800 immunology and host response related genes on patient cells exposed or not to IFN- alpha, poly(I:C) and R848, they showed a perturbation at baseline of immune responses and a lack of ISG response following stimulation in COVID- 19 patients. Moreover, they could detect increased inflammatory responses in hospitalized patients.  
+
+This is a highly interesting area of research and the manuscript presents an impressive amount of data. However, in order to improve the manuscript and facilitate its reading, some clarifications in the text are required. Moreover, some of the data sets used to generate some of the main figures should be shown in the supplemental figures. Please find these recommendations below.  
+
+For clarity, in the Results section and in the figure legends, when mentioning the cohorts, the authors should always refer to the corresponding supplementary tables (this was done only for Table S3 and S5).  
+
+The authors took into account the sex and age of patients in their statistical analyses. However, the sampling time (number of days post symptoms), when available, did not seem to have been taken into account in the analyses for Cohort 1 (Fig. 1, S1). Importantly, for this cohort (Table S2) the average of days post symptoms were different between the 3 categories (moderate, severe, critical) (see review figure - Days post symptom Cohort 1 and 3), and although the differences appeared not significant (using a Mann- Whitney test), it might somehow impact the findings. The authors should acknowledge this. Could the authors plot the IFN concentrations according to the number of days post symptoms for each sample (and using the disease severity color code)? This might help discard the possibility that the time of sampling parameter impacted the data.  
+
+In the third cohort (Table S4), according to date of sample and date of onset of symptoms, some samples were harvested \(>60\) days post- symptom, in particular for patients with moderate disease (see review figure - Days post symptom Cohort 1 and 3). One might wonder why the authors kept such outliers for moderate disease and whether they impacted the results.  
+
+The discrepancy between Luminex and digital ELISA data is worrying and, as such, an important finding as it could explain some of the discrepancies found in the literature. In Fig 1c (Luminex analysis) compared to 1a - b (digital ELISA) (data obtained with Cohort 1), the
+
+<--- Page Split --->
+
+concentrations measured were completely different; between \(10^{\wedge}0\) to \(10^{\wedge}3 \mathrm{pg / mL}\) for instance in the healthy group with Luminex and between 10- 1 and 10- 2 pg/mL with digital ELISA. How could this be explained? As the values are quite high with the Luminex analyses, a problem with detection limit is not what first comes to mind. What was the limit of detection for the Luminex assay?  
+
+The experiments performed in Fig. 1 need to be better explained and the data provided. Notably, the ISG score and functional cytopathic experiments were not explained. For this, an article (PMID: 32661059) was quoted, but one has to download the supplemental methods from that paper to get the information. The authors should define in detail how the ISG score was obtained and provide the actual data (in a supplemental figure) for the expression measure of the 6 chosen ISGs (data corresponding to Fig. 1d- f). Moreover, it was unclear what the calibrator sample (used for 2- DDCT calculation) was. The authors should also define how the functional cytopathic effect experiment was performed (for this, reference 2, PMID: 32661059, actually referred to a paper from 1979 without further details) and the data should be provided (i.e. data corresponding to Fig. 1g- i).  
+
+The ISG score measured using the nanostring technology (Fig. 2b- c) should also be explained, and the data provided. Moreover, could the authors define the paired blood samples?  
+
+Fig. 2c- d, S2f- g: there are no units on the y- axis and y- axis, respectively  
+
+In the following paragraph, the figures are mislabeled, S2b should be replaced by S2e "... This showed low transcriptional levels of all measured IFNA subtypes as expected (Fig. S2b). Despite these low baseline levels, among the 7 IFNA subtypes examined we did observe some subtype differences, with notably higher levels of IFNA6 in both COVID- 19 patient groups and IFNA1/13 and IFNA5 only in the hospitalized group (Fig. S2b). IFNA2 was notably no different between all three groups (Fig. S2e)."  
+
+Fig 2c- d description in the main text: (>95 considered positive): please specify the units  
+
+Please correct the figure numbers in the following paragraph: "Additional correlation analysis between cytometry measured intracellular IFNa, and digital ELISA measured plasma IFNa, showed in the absence of stimulation an association between monocytes and plasma IFNa levels (Fig. 4d). Following R848 stimulation, both pDCs and monocytes showed an association with secreted IFNa (Fig 4h), although the percentage of IFNa+ cells was lower in critical patients compared to severe and healthy controls (Fig. 4i)." (Fig. 4d actually shows the % of IFNa+ pDCs, Fig. 4h shows only data for pDC: R848...)  
+
+"For this, we performed the same standardized whole blood ex vivo stimulation with recombinant IFNa2 and measured gene expression by Nanostring as previously described." \(\rightarrow\) please state where previously described  
+
+Fig S3 legend: CXCL10 (c), IL- 10 (c) \(\rightarrow\) CXCL10 (c), IL- 10 (d)  
+
+Fig. 4c: No increase in the % of pIRF3+ monocytes was observed with the healthy control cells when treated with R848. Could the authors comment on this?  
+
+Table S2: "Débit 02 max": please translate to English
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+The role of type I IFNs in protecting against or aggravating disease outcomes during SARS- CoV- 2 infection remains controversial, and likely depends on the kinetics of IFN treatment vs. disease progression. This study adds important insights into the biological role of IFNs during different severities of COVID- 19. The authors demonstrate that patients with critical COVID- 19 have a more inflammatory gene expression profile. This has been demonstrated by other studies previously. However, interestingly, blood cells from hospitalized patients also mounted an inflammatory response on stimulation with TLR ligands, in contrast to cells from healthy individuals that mount an IFN- dependent ISG response. Overall, this study adds more information to the slowly growing body of literature that supports the likely role of a perturbed IFN response during severe COVID- 19. Some comments for the authors are mentioned below:  
+
+1. What was the clinical criteria/score used to define patients as mild, moderate, or severe? Please elaborate on this in the methodology. While the authors have included this in the methods, it is important to note how these vary in comparison to other studies. How is moderate different from 'mild' as defined by other publications? Putting this in context is important for the readers and for reproducibility. 
+2. Were the patient cohorts scanned for comorbidities or ongoing treatments? 
+3. In Fig 1, the authors demonstrate that digital ELISA is more sensitive than other methods, such as Luminex. However, are such low levels, as detected by digital ELISA have any physiological relevance? Do these low IFNa2 concentrations perform well (induce gene expression) when used to stimulate human cells experimentally? 
+4. Are the differences in Fig. 3a statistically significant? 
+5. Although TLR4 and TLR8 gene counts are up in hospitalized patients (Fig. 3d), why don't their cells respond to stimulation by LPS and R848 (Figs. 3a-c)? Do TLR gene counts correlate with protein levels of TLRs in the cells? 
+6. Label for Fig. 6f is missing 
+7. Heatmap legend is missing for Fig. 6b 
+8. MDA5 plays a major role in recognizing SARS-CoV-2 RNA. Is there a reason why MDA5 stimulation wasn't included/discussed?
+
+<--- Page Split --->
+
+Dear Nature Communications  
+
+We are grateful for the careful reviews of our manuscript, which we have fully addressed as described below in a point by point response. We believe that the modifications we have made to our revised manuscript have furthered strengthened it and made our message clearer.  
+
+We look forward to your responses.  
+
+Regards Darragh Duffy
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+Reviewer #1 (Remarks to the Author):In this manuscript, Smith, Possemé and Bondet et al studied type I interferon responses in samples from several COVID- 19 patient cohorts in depth. First, the authors confirmed their previous findings (PMID: 32661059) showing that IFN responses were impaired in critical patients, where the concentration of IFN- alpha found in plasma samples inversely correlated with disease severity. For this, they highlighted the need to use digital ELISA, a highly sensitive assay to measure interferon concentration, as Luminex couldn't pick differences between patient groups. Moreover, and contrary to the Luminex data, data from digital ELISA correlated well with the "ISG score" (based on the relative quantification of 6 prototype ISG expression) and interferon activity (as measured by protection against virus- induced cytopathic effects). They further showed that blood cells from critical patients globally responded less well to various stimuli (poly(I:C), LPS, R848) and confirmed that their pDC count was significantly lower, whereas monocyte counts were increased. Using the nanostring technology on 800 immunology and host response related genes on patient cells exposed or not to IFN- alpha, poly(I:C) and R848, they showed a perturbation at baseline of immune responses and a lack of ISG response following stimulation in COVID- 19 patients. Moreover, they could detect increased inflammatory responses in hospitalized patients. This is a highly interesting area of research and the manuscript presents an impressive amount of data. However, in order to improve the manuscript and facilitate its reading, some clarifications in the text are required. Moreover, some of the data sets used to generate some of the main figures should be shown in the supplemental figures. Please find these recommendations below.  
+
+- We thank the reviewer for the concise summary of our study and noting its high interest and impressive amount of data. We have responded to each specific point below and made corresponding modifications to the revised manuscript.  
+
+For clarity, in the Results section and in the figure legends, when mentioning the cohorts, the authors should always refer to the corresponding supplementary tables (this was done only for Table S3 and S5).  
+
+## We have made this correction throughout the manuscript.  
+
+The authors took into account the sex and age of patients in their statistical analyses. However, the sampling time (number of days post symptoms), when available, did not seem to have been taken into account in the analyses for Cohort 1 (Fig. 1, S1). Importantly, for this cohort (Table S2) the average of days post symptoms were different between the 3 categories (moderate, severe, critical) (see review figure - Days post symptom Cohort 1 and 3), and although the differences appeared not significant (using a Mann- Whitney test), it might somehow impact the findings. The authors should acknowledge this. Could the authors plot the IFN concentrations according to the number of days post symptoms for each sample (and using the disease severity color code)? This might help discard the possibility that the time of sampling parameter impacted the data.
+
+<--- Page Split --->
+
+- We thank the reviewer for highlighting this important point. We have acknowledged this in the description of the results and plotted the data as suggested, shown below for cohort 1 (Table S2), and also include in the revised supplemental Fig S1d, e & f. Analysis of which does not suggest any significant impact of days since symptoms on IFNα levels in this cohort. We had previously performed this analysis for cohort 3 (Table S4) which is included in Fig 2e and 2f.  
+
+![PLACEHOLDER_6_0]
+  
+
+In the third cohort (Table S4), according to date of sample and date of onset of symptoms, some samples were harvested \(>60\) days post- symptom, in particular for patients with moderate disease (see review figure - Days post symptom Cohort 1 and 3). One might wonder why the authors kept such outliers for moderate disease and whether they impacted the results.  
+
+- We apologize for an error which caused this confusion. Within this original cohort there were some convalescent samples of the same patients after they had cleared the virus. In the final submitted manuscript we did not include analysis of these samples in order to keep a focused message on the IFN response during acute infection. However, some of these samples were listed by mistake in the supplemental table S4 which explains the observation of the reviewer that some samples were harvested \(>60\) days post-symptom. This has now been corrected and we thank the reviewer for spotting this error.  
+
+The discrepancy between Luminex and digital ELISA data is worrying and, as such, an important finding as it could explain some of the discrepancies found in the literature. In Fig 1c (Luminex analysis) compared to 1a - b (digital ELISA) (data obtained with Cohort 1), the concentrations measured were completely different; between \(10^4 0\) to \(10^4 3 \mathrm{pg / mL}\) for instance in the healthy group with Luminex and between 10- 1 and 10- 2 pg/mL with digital ELISA. How could this be explained? As the values are quite high with the Luminex analyses, a problem with detection limit is not what first comes to mind. What was the limit of detection for the Luminex assay?  
+
+- We completely agree with the reviewer that the discrepancy between these technologies is worrying. We believe that it could be due to a combination of low affinity antibodies, a relatively high LOD, and a highly multiplexed assay (44 analytes) that may report non-specific background fluorescence as a low but positive signal. The LOD from the commercial supplier (Biotechne) is reported
+
+<--- Page Split --->
+
+to be \(0.29\mathrm{pg / mL}\) which with a dilution factor of 7 used in our analysis corresponds to \(2 - 3\mathrm{pg / mL}\) . However, the lowest point on the standard curve is 9.8 pg/mL indicating that these low values are extrapolated from the curve. This detection limit was determined by adding two standard deviations to the mean response of twenty zero standard replicates and calculating the corresponding concentration. For the digital ELISA IFNa2 and IFNα multi-subtype assays, the detection limits are \(2\mathrm{fg / mL}\) and \(0.6\mathrm{fg / mL}\) respectively, plus integration of the sample dilution factor. These detection limits were determined from the background level of each assay + 2SD ( \(95\%\) confidence) for the IFNa2 assay and 3SD ( \(99\%\) confidence) for the IFNα multi-subtype assay. In Figures 1a and 1b, the concentrations we measured using the digital ELISA assays are between \(10^{- }\) - 2 and \(10^{- }2\mathrm{pg / mL}\) and in Fig 1c between \(10^{- }0\) and \(10^{- }3\mathrm{pg / mL}\) using the Luminex assay. All the concentrations that are below the Luminex and above the digital ELISA detection limits are correctly quantified (they correlate with ISG score and IFN activity) using the digital ELISA assays, but quantified close to the detection limit by this Luminex assay, thus generating the results obtained with this test: all data is above \(2 - 3\mathrm{pg / mL}\) , thus abolishing the differences between healthy and COVID- 19 patients and between the severity classes, and correlations with ISG score and IFN activity are lost.  
+
+We have reanalyzed some of our whole blood stimulation results where there are high concentrations of IFNα after stimulation. Correlation analysis with the Simoa assays also suggests that there is a problem with calibration standards of the Luminex kit for IFNα, as there is almost 2 log differences between the reported concentrations (Figure shown below for reviewer). Despite this discrepancy, the Luminex results only correlate with the Simoa values after the strongest stimulation (R848) where the concentrations are above \(10\mathrm{pg / mL}\) (according to the Simoa measures). This result strongly suggests that Luminex assays should only be used to quantify high concentrations of IFNα.
+
+<--- Page Split --->
+![PLACEHOLDER_8_0]
+
+<center>Figure for reviewer. Comparison of Simoa assays with Luminex in TruCulture stimulations; Null, Poly:IC, LPS, R848. Spearman correlations were performed per stimuli on each assay comparison. </center>  
+
+The experiments performed in Fig. 1 need to be better explained and the data provided. Notably, the ISG score and functional cytopathic experiments were not explained. For this, an article (PMID: 32661059) was quoted, but one has to download the supplemental methods from that paper to get the information. The authors should define in detail how the ISG score was obtained and provide the actual data (in a supplemental figure) for the expression measure of the 6 chosen ISGs (data corresponding to Fig. 1d- f). Moreover, it was unclear what the calibrator sample (used for 2- DDCT calculation) was. The authors should also define how the functional cytopathic effect experiment was performed (for this, reference 2, PMID: 32661059, actually referred to a paper from 1979
+
+<--- Page Split --->
+
+without further details) and the data should be provided (i.e. data corresponding to Fig. 1g-i).  
+
+- We have included these specific details as requested in the revised manuscript, and the data has been included in the supplemental tables.  
+
+The ISG score measured using the nanostring technology (Fig. 2b- c) should also be explained, and the data provided. Moreover, could the authors define the paired blood samples?  
+
+- We have included additional descriptions in the Methods of the revised manuscript and included the data in the relevant supplemental tables.  
+
+Fig. 2c- d, S2f- g: there are no units on the y- axis and y- axis, respectively  
+
+- This has been corrected.  
+
+In the following paragraph, the figures are mislabeled, S2b should be replaced by S2e "... This showed low transcriptional levels of all measured IFNA subtypes as expected (Fig. S2b). Despite these low baseline levels, among the 7 IFNA subtypes examined we did observe some subtype differences, with notably higher levels of IFNA6 in both COVID- 19 patient groups and IFNA1/13 and IFNA5 only in the hospitalized group (Fig. S2b). IFNA2 was notably no different between all three groups (Fig. S2e)."  
+
+- This has been corrected.  
+
+Fig 2c- d description in the main text: (>95 considered positive): please specify the units  
+
+- This has been included.  
+
+Please correct the figure numbers in the following paragraph: "Additional correlation analysis between cytometry measured intracellular IFNa, and digital ELISA measured plasma IFNa, showed in the absence of stimulation an association between monocytes and plasma IFNa levels (Fig. 4d). Following R848 stimulation, both pDCs and monocytes showed an association with secreted IFNa (Fig 4h), although the percentage of IFNa+ cells was lower in critical patients compared to severe and healthy controls (Fig. 4i)." (Fig. 4d actually shows the % of IFNa+ pDCs, Fig. 4h shows only data for pDC: R848...)  
+
+(Fig. 4d actually shows the % of IFNa+ pDCs, Fig. 4h shows only data for pDC: R848...)  
+
+- This has been corrected.  
+
+"For this, we performed the same standardized whole blood ex vivo stimulation with recombinant IFNa2 and measured gene expression by Nanostring as previously described." - > please state where previously described  
+
+- This has been corrected.  
+
+Fig S3 legend: CXCL10 (c), IL- 10 (c) \(\rightarrow\) CXCL10 (c), IL- 10 (d) - This has been corrected.
+
+<--- Page Split --->
+
+Fig. 4c: No increase in the \(\%\) of pIRF3+ monocytes was observed with the healthy control cells when treated with R848. Could the authors comment on this?  
+
+- We thank the reviewer for highlighting this statement. While we did not see an increase in the \(\%\) of pIRF3+ monocytes we did see an increase in MFI, we have modified the text to reflect this.  
+
+Table S2: "Débit 02 max": please translate to English  
+
+- This has been corrected.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The role of type I IFNs in protecting against or aggravating disease outcomes during SARS- CoV- 2 infection remains controversial, and likely depends on the kinetics of IFN treatment vs. disease progression. This study adds important insights into the biological role of IFNs during different severities of COVID- 19. The authors demonstrate that patients with critical COVID- 19 have a more inflammatory gene expression profile. This has been demonstrated by other studies previously. However, interestingly, blood cells from hospitalized patients also mounted an inflammatory response on stimulation with TLR ligands, in contrast to cells from healthy individuals that mount an IFN- dependent ISG response. Overall, this study adds more information to the slowly growing body of literature that supports the likely role of a perturbed IFN response during severe COVID- 19. Some comments for the authors are mentioned below:  
+
+- We thank the reviewer for recognizing how our study adds important insights into the biological role of IFNs during different severities of COVID-19. We have responded to each specific point below and made corresponding modifications to the revised manuscript.  
+
+1. What was the clinical criteria/score used to define patients as mild, moderate, or severe? Please elaborate on this in the methodology. While the authors have included this in the methods, it is important to note how these vary in comparison to other studies. How is moderate different from 'mild' as defined by other publications? Putting this in context is important for the readers and for reproducibility.  
+
+- We apologize for any confusion here. We defined the patients (as described in the methods) based on internationally accepted criteria which is the requirements for supplemental oxygen at the time of sampling. Moderate patients did not require hospitalization at any timepoint. Hospitalized patients requiring supplemental oxygen via nasal cannula (maximal supplemental oxygen flow of up to 6L/min) were considered severe, with critical disease classified as requiring more than 6L of oxygen per minute, either delivered via high-flow nasal oxygen (Airvo) or a venturi mask, a clinical definition previously defined<sup>3,4</sup>. In the final version of the manuscript we decided not to distinguish mild from moderate patients as this is often based on the opinion of the clinician and is thus challenging to compare across studies. We have corrected this term in the manuscript and now all patients that did not require hospitalization or oxygen supplementation are defined as moderate.
+
+<--- Page Split --->
+
+2. Were the patient cohorts scanned for comorbidities or ongoing treatments?  
+
+- In some of the cohorts this was possible, but due to the incomplete nature of the data sets we were not able to integrate these factors into the analysis which is why they were not included.  
+
+3. In Fig 1, the authors demonstrate that digital ELISA is more sensitive than other methods, such as Luminex. However, are such low levels, as detected by digital ELISA have any physiological relevance? Do these low IFNa2 concentrations perform well (induce gene expression) when used to stimulate human cells experimentally?  
+
+- This is a very interesting point, and one that we often discuss internally. The main set of evidence that these levels are physiological come from our multitude of published studies with these assays, in particular in cases of autoimmune disease where IFNa has been clinically implicated in the pathology and where levels of the plasma protein are within these ranges (>100fg/mL). Whether these low IFNa2 concentrations induce gene expression is dependent on the sensitivity of the assay measuring the gene expression, which do not always match the sensitivity of these digital ELISA. Recombinant IFN is often described in units, with 1 unit/mL of interferon being the quantity necessary to produce a cytopathic effect of 50%. 1 unit/mL is estimated to be between 200-300fg/mL depending on the viral cytopathic assay utilized, which also have their limitations in terms of sensitivity. However, in previous studies we have observed that an ISG score begins to correlate with Simoa measures between 1 and 10 fg/mL IFNa in SLE and JDM patients, and that the IFN activity begins to correlate between 10 and 100 fg/mL IFNa in JDM patients (Rodero et al, JEM 2016).  
+
+4. Are the differences in Fig. 3a statistically significant?  
+
+- No the differences between the different patient groups per stimulation are not statistically significant.  
+
+5. Although TLR4 and TLR8 gene counts are up in hospitalized patients (Fig. 3d), why don't their cells respond to stimulation by LPS and R848 (Figs. 3a-c)? Do TLR gene counts correlate with protein levels of TLRs in the cells?  
+
+- This is an interesting observation for which we do not currently have an evidence based explanation, but we may hypothesis that there are intracellular perturbations downstream of TLR signaling. It is challenging to measure TLR proteins in cells and were not able to perform this particular analysis.  
+
+6. Label for Fig. 6f is missing  
+
+- This has been corrected.  
+
+7. Heatmap legend is missing for Fig. 6b  
+
+- This has been corrected.
+
+<--- Page Split --->
+
+8. MDA5 plays a major role in recognizing SARS-CoV-2 RNA. Is there a reason why MDA5 stimulation wasn't included/discussed?  
+
+- This initial study was started during the first wave of the pandemic when little was known about the receptors that recognize SARS-CoV-2 RNA, which is the main reason why it was not included as a stimulation. We did include analysis of the levels of IFIH1 (gene encoding MDA5 in humans) in Fig 3d.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+## Reviewer #1 (Remarks to the Author):  
+
+The authors have addressed my concerns.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__43470755cca0263d20ba27a2e315ebda8bd3f51b485e75fb05673e70392cb22a/supplementary_0_Peer Review File__43470755cca0263d20ba27a2e315ebda8bd3f51b485e75fb05673e70392cb22a_det.mmd

@@ -0,0 +1,290 @@
+<|ref|>title<|/ref|><|det|>[[100, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 154, 893, 210]]<|/det|>
+Detection of defective type I interferon immunity is associated with increasing COVID- 19 severity  
+
+<|ref|>image<|/ref|><|det|>[[93, 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|>[[119, 83, 331, 99]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 115, 450, 132]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 148, 874, 425]]<|/det|>
+In this manuscript, Smith, Possemé and Bondet et al studied type I interferon responses in samples from several COVID- 19 patient cohorts in depth. First, the authors confirmed their previous findings (PMID: 32661059) showing that IFN responses were impaired in critical patients, where the concentration of IFN- alpha found in plasma samples inversely correlated with disease severity. For this, they highlighted the need to use digital ELISA, a highly sensitive assay to measure interferon concentration, as Luminex couldn't pick differences between patient groups. Moreover, and contrary to the Luminex data, data from digital ELISA correlated well with the "ISG score" (based on the relative quantification of 6 prototype ISG expression) and interferon activity (as measured by protection against virus- induced cytopathic effects). They further showed that blood cells from critical patients globally responded less well to various stimuli (poly(I:C), LPS, R848) and confirmed that their pDC count was significantly lower, whereas monocyte counts were increased. Using the nanostring technology on 800 immunology and host response related genes on patient cells exposed or not to IFN- alpha, poly(I:C) and R848, they showed a perturbation at baseline of immune responses and a lack of ISG response following stimulation in COVID- 19 patients. Moreover, they could detect increased inflammatory responses in hospitalized patients.  
+
+<|ref|>text<|/ref|><|det|>[[118, 425, 875, 505]]<|/det|>
+This is a highly interesting area of research and the manuscript presents an impressive amount of data. However, in order to improve the manuscript and facilitate its reading, some clarifications in the text are required. Moreover, some of the data sets used to generate some of the main figures should be shown in the supplemental figures. Please find these recommendations below.  
+
+<|ref|>text<|/ref|><|det|>[[118, 521, 878, 570]]<|/det|>
+For clarity, in the Results section and in the figure legends, when mentioning the cohorts, the authors should always refer to the corresponding supplementary tables (this was done only for Table S3 and S5).  
+
+<|ref|>text<|/ref|><|det|>[[117, 586, 880, 749]]<|/det|>
+The authors took into account the sex and age of patients in their statistical analyses. However, the sampling time (number of days post symptoms), when available, did not seem to have been taken into account in the analyses for Cohort 1 (Fig. 1, S1). Importantly, for this cohort (Table S2) the average of days post symptoms were different between the 3 categories (moderate, severe, critical) (see review figure - Days post symptom Cohort 1 and 3), and although the differences appeared not significant (using a Mann- Whitney test), it might somehow impact the findings. The authors should acknowledge this. Could the authors plot the IFN concentrations according to the number of days post symptoms for each sample (and using the disease severity color code)? This might help discard the possibility that the time of sampling parameter impacted the data.  
+
+<|ref|>text<|/ref|><|det|>[[118, 764, 874, 845]]<|/det|>
+In the third cohort (Table S4), according to date of sample and date of onset of symptoms, some samples were harvested \(>60\) days post- symptom, in particular for patients with moderate disease (see review figure - Days post symptom Cohort 1 and 3). One might wonder why the authors kept such outliers for moderate disease and whether they impacted the results.  
+
+<|ref|>text<|/ref|><|det|>[[118, 862, 868, 911]]<|/det|>
+The discrepancy between Luminex and digital ELISA data is worrying and, as such, an important finding as it could explain some of the discrepancies found in the literature. In Fig 1c (Luminex analysis) compared to 1a - b (digital ELISA) (data obtained with Cohort 1), the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 877, 165]]<|/det|>
+concentrations measured were completely different; between \(10^{\wedge}0\) to \(10^{\wedge}3 \mathrm{pg / mL}\) for instance in the healthy group with Luminex and between 10- 1 and 10- 2 pg/mL with digital ELISA. How could this be explained? As the values are quite high with the Luminex analyses, a problem with detection limit is not what first comes to mind. What was the limit of detection for the Luminex assay?  
+
+<|ref|>text<|/ref|><|det|>[[117, 181, 870, 343]]<|/det|>
+The experiments performed in Fig. 1 need to be better explained and the data provided. Notably, the ISG score and functional cytopathic experiments were not explained. For this, an article (PMID: 32661059) was quoted, but one has to download the supplemental methods from that paper to get the information. The authors should define in detail how the ISG score was obtained and provide the actual data (in a supplemental figure) for the expression measure of the 6 chosen ISGs (data corresponding to Fig. 1d- f). Moreover, it was unclear what the calibrator sample (used for 2- DDCT calculation) was. The authors should also define how the functional cytopathic effect experiment was performed (for this, reference 2, PMID: 32661059, actually referred to a paper from 1979 without further details) and the data should be provided (i.e. data corresponding to Fig. 1g- i).  
+
+<|ref|>text<|/ref|><|det|>[[119, 358, 825, 407]]<|/det|>
+The ISG score measured using the nanostring technology (Fig. 2b- c) should also be explained, and the data provided. Moreover, could the authors define the paired blood samples?  
+
+<|ref|>text<|/ref|><|det|>[[118, 423, 708, 440]]<|/det|>
+Fig. 2c- d, S2f- g: there are no units on the y- axis and y- axis, respectively  
+
+<|ref|>text<|/ref|><|det|>[[118, 455, 851, 553]]<|/det|>
+In the following paragraph, the figures are mislabeled, S2b should be replaced by S2e "... This showed low transcriptional levels of all measured IFNA subtypes as expected (Fig. S2b). Despite these low baseline levels, among the 7 IFNA subtypes examined we did observe some subtype differences, with notably higher levels of IFNA6 in both COVID- 19 patient groups and IFNA1/13 and IFNA5 only in the hospitalized group (Fig. S2b). IFNA2 was notably no different between all three groups (Fig. S2e)."  
+
+<|ref|>text<|/ref|><|det|>[[118, 569, 832, 586]]<|/det|>
+Fig 2c- d description in the main text: (>95 considered positive): please specify the units  
+
+<|ref|>text<|/ref|><|det|>[[118, 601, 877, 715]]<|/det|>
+Please correct the figure numbers in the following paragraph: "Additional correlation analysis between cytometry measured intracellular IFNa, and digital ELISA measured plasma IFNa, showed in the absence of stimulation an association between monocytes and plasma IFNa levels (Fig. 4d). Following R848 stimulation, both pDCs and monocytes showed an association with secreted IFNa (Fig 4h), although the percentage of IFNa+ cells was lower in critical patients compared to severe and healthy controls (Fig. 4i)." (Fig. 4d actually shows the % of IFNa+ pDCs, Fig. 4h shows only data for pDC: R848...)  
+
+<|ref|>text<|/ref|><|det|>[[118, 731, 870, 780]]<|/det|>
+"For this, we performed the same standardized whole blood ex vivo stimulation with recombinant IFNa2 and measured gene expression by Nanostring as previously described." \(\rightarrow\) please state where previously described  
+
+<|ref|>text<|/ref|><|det|>[[118, 796, 628, 813]]<|/det|>
+Fig S3 legend: CXCL10 (c), IL- 10 (c) \(\rightarrow\) CXCL10 (c), IL- 10 (d)  
+
+<|ref|>text<|/ref|><|det|>[[118, 829, 852, 861]]<|/det|>
+Fig. 4c: No increase in the % of pIRF3+ monocytes was observed with the healthy control cells when treated with R848. Could the authors comment on this?  
+
+<|ref|>text<|/ref|><|det|>[[118, 877, 553, 893]]<|/det|>
+Table S2: "Débit 02 max": please translate to English
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 100, 449, 117]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 132, 872, 310]]<|/det|>
+The role of type I IFNs in protecting against or aggravating disease outcomes during SARS- CoV- 2 infection remains controversial, and likely depends on the kinetics of IFN treatment vs. disease progression. This study adds important insights into the biological role of IFNs during different severities of COVID- 19. The authors demonstrate that patients with critical COVID- 19 have a more inflammatory gene expression profile. This has been demonstrated by other studies previously. However, interestingly, blood cells from hospitalized patients also mounted an inflammatory response on stimulation with TLR ligands, in contrast to cells from healthy individuals that mount an IFN- dependent ISG response. Overall, this study adds more information to the slowly growing body of literature that supports the likely role of a perturbed IFN response during severe COVID- 19. Some comments for the authors are mentioned below:  
+
+<|ref|>text<|/ref|><|det|>[[115, 325, 872, 618]]<|/det|>
+1. What was the clinical criteria/score used to define patients as mild, moderate, or severe? Please elaborate on this in the methodology. While the authors have included this in the methods, it is important to note how these vary in comparison to other studies. How is moderate different from 'mild' as defined by other publications? Putting this in context is important for the readers and for reproducibility. 
+2. Were the patient cohorts scanned for comorbidities or ongoing treatments? 
+3. In Fig 1, the authors demonstrate that digital ELISA is more sensitive than other methods, such as Luminex. However, are such low levels, as detected by digital ELISA have any physiological relevance? Do these low IFNa2 concentrations perform well (induce gene expression) when used to stimulate human cells experimentally? 
+4. Are the differences in Fig. 3a statistically significant? 
+5. Although TLR4 and TLR8 gene counts are up in hospitalized patients (Fig. 3d), why don't their cells respond to stimulation by LPS and R848 (Figs. 3a-c)? Do TLR gene counts correlate with protein levels of TLRs in the cells? 
+6. Label for Fig. 6f is missing 
+7. Heatmap legend is missing for Fig. 6b 
+8. MDA5 plays a major role in recognizing SARS-CoV-2 RNA. Is there a reason why MDA5 stimulation wasn't included/discussed?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 166, 362, 182]]<|/det|>
+Dear Nature Communications  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 872, 248]]<|/det|>
+We are grateful for the careful reviews of our manuscript, which we have fully addressed as described below in a point by point response. We believe that the modifications we have made to our revised manuscript have furthered strengthened it and made our message clearer.  
+
+<|ref|>text<|/ref|><|det|>[[118, 264, 408, 280]]<|/det|>
+We look forward to your responses.  
+
+<|ref|>text<|/ref|><|det|>[[117, 298, 240, 330]]<|/det|>
+Regards Darragh Duffy
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 100, 358, 118]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[119, 133, 430, 150]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 165, 875, 512]]<|/det|>
+Reviewer #1 (Remarks to the Author):In this manuscript, Smith, Possemé and Bondet et al studied type I interferon responses in samples from several COVID- 19 patient cohorts in depth. First, the authors confirmed their previous findings (PMID: 32661059) showing that IFN responses were impaired in critical patients, where the concentration of IFN- alpha found in plasma samples inversely correlated with disease severity. For this, they highlighted the need to use digital ELISA, a highly sensitive assay to measure interferon concentration, as Luminex couldn't pick differences between patient groups. Moreover, and contrary to the Luminex data, data from digital ELISA correlated well with the "ISG score" (based on the relative quantification of 6 prototype ISG expression) and interferon activity (as measured by protection against virus- induced cytopathic effects). They further showed that blood cells from critical patients globally responded less well to various stimuli (poly(I:C), LPS, R848) and confirmed that their pDC count was significantly lower, whereas monocyte counts were increased. Using the nanostring technology on 800 immunology and host response related genes on patient cells exposed or not to IFN- alpha, poly(I:C) and R848, they showed a perturbation at baseline of immune responses and a lack of ISG response following stimulation in COVID- 19 patients. Moreover, they could detect increased inflammatory responses in hospitalized patients. This is a highly interesting area of research and the manuscript presents an impressive amount of data. However, in order to improve the manuscript and facilitate its reading, some clarifications in the text are required. Moreover, some of the data sets used to generate some of the main figures should be shown in the supplemental figures. Please find these recommendations below.  
+
+<|ref|>text<|/ref|><|det|>[[147, 544, 868, 595]]<|/det|>
+- We thank the reviewer for the concise summary of our study and noting its high interest and impressive amount of data. We have responded to each specific point below and made corresponding modifications to the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 609, 870, 660]]<|/det|>
+For clarity, in the Results section and in the figure legends, when mentioning the cohorts, the authors should always refer to the corresponding supplementary tables (this was done only for Table S3 and S5).  
+
+<|ref|>sub_title<|/ref|><|det|>[[147, 676, 678, 694]]<|/det|>
+## We have made this correction throughout the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[117, 709, 875, 874]]<|/det|>
+The authors took into account the sex and age of patients in their statistical analyses. However, the sampling time (number of days post symptoms), when available, did not seem to have been taken into account in the analyses for Cohort 1 (Fig. 1, S1). Importantly, for this cohort (Table S2) the average of days post symptoms were different between the 3 categories (moderate, severe, critical) (see review figure - Days post symptom Cohort 1 and 3), and although the differences appeared not significant (using a Mann- Whitney test), it might somehow impact the findings. The authors should acknowledge this. Could the authors plot the IFN concentrations according to the number of days post symptoms for each sample (and using the disease severity color code)? This might help discard the possibility that the time of sampling parameter impacted the data.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[147, 85, 866, 199]]<|/det|>
+- We thank the reviewer for highlighting this important point. We have acknowledged this in the description of the results and plotted the data as suggested, shown below for cohort 1 (Table S2), and also include in the revised supplemental Fig S1d, e & f. Analysis of which does not suggest any significant impact of days since symptoms on IFNα levels in this cohort. We had previously performed this analysis for cohort 3 (Table S4) which is included in Fig 2e and 2f.  
+
+<|ref|>image<|/ref|><|det|>[[137, 223, 850, 396]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[147, 422, 877, 504]]<|/det|>
+In the third cohort (Table S4), according to date of sample and date of onset of symptoms, some samples were harvested \(>60\) days post- symptom, in particular for patients with moderate disease (see review figure - Days post symptom Cohort 1 and 3). One might wonder why the authors kept such outliers for moderate disease and whether they impacted the results.  
+
+<|ref|>text<|/ref|><|det|>[[147, 522, 872, 654]]<|/det|>
+- We apologize for an error which caused this confusion. Within this original cohort there were some convalescent samples of the same patients after they had cleared the virus. In the final submitted manuscript we did not include analysis of these samples in order to keep a focused message on the IFN response during acute infection. However, some of these samples were listed by mistake in the supplemental table S4 which explains the observation of the reviewer that some samples were harvested \(>60\) days post-symptom. This has now been corrected and we thank the reviewer for spotting this error.  
+
+<|ref|>text<|/ref|><|det|>[[147, 669, 877, 800]]<|/det|>
+The discrepancy between Luminex and digital ELISA data is worrying and, as such, an important finding as it could explain some of the discrepancies found in the literature. In Fig 1c (Luminex analysis) compared to 1a - b (digital ELISA) (data obtained with Cohort 1), the concentrations measured were completely different; between \(10^4 0\) to \(10^4 3 \mathrm{pg / mL}\) for instance in the healthy group with Luminex and between 10- 1 and 10- 2 pg/mL with digital ELISA. How could this be explained? As the values are quite high with the Luminex analyses, a problem with detection limit is not what first comes to mind. What was the limit of detection for the Luminex assay?  
+
+<|ref|>text<|/ref|><|det|>[[147, 817, 870, 900]]<|/det|>
+- We completely agree with the reviewer that the discrepancy between these technologies is worrying. We believe that it could be due to a combination of low affinity antibodies, a relatively high LOD, and a highly multiplexed assay (44 analytes) that may report non-specific background fluorescence as a low but positive signal. The LOD from the commercial supplier (Biotechne) is reported
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[177, 83, 878, 396]]<|/det|>
+to be \(0.29\mathrm{pg / mL}\) which with a dilution factor of 7 used in our analysis corresponds to \(2 - 3\mathrm{pg / mL}\) . However, the lowest point on the standard curve is 9.8 pg/mL indicating that these low values are extrapolated from the curve. This detection limit was determined by adding two standard deviations to the mean response of twenty zero standard replicates and calculating the corresponding concentration. For the digital ELISA IFNa2 and IFNα multi-subtype assays, the detection limits are \(2\mathrm{fg / mL}\) and \(0.6\mathrm{fg / mL}\) respectively, plus integration of the sample dilution factor. These detection limits were determined from the background level of each assay + 2SD ( \(95\%\) confidence) for the IFNa2 assay and 3SD ( \(99\%\) confidence) for the IFNα multi-subtype assay. In Figures 1a and 1b, the concentrations we measured using the digital ELISA assays are between \(10^{- }\) - 2 and \(10^{- }2\mathrm{pg / mL}\) and in Fig 1c between \(10^{- }0\) and \(10^{- }3\mathrm{pg / mL}\) using the Luminex assay. All the concentrations that are below the Luminex and above the digital ELISA detection limits are correctly quantified (they correlate with ISG score and IFN activity) using the digital ELISA assays, but quantified close to the detection limit by this Luminex assay, thus generating the results obtained with this test: all data is above \(2 - 3\mathrm{pg / mL}\) , thus abolishing the differences between healthy and COVID- 19 patients and between the severity classes, and correlations with ISG score and IFN activity are lost.  
+
+<|ref|>text<|/ref|><|det|>[[178, 411, 872, 560]]<|/det|>
+We have reanalyzed some of our whole blood stimulation results where there are high concentrations of IFNα after stimulation. Correlation analysis with the Simoa assays also suggests that there is a problem with calibration standards of the Luminex kit for IFNα, as there is almost 2 log differences between the reported concentrations (Figure shown below for reviewer). Despite this discrepancy, the Luminex results only correlate with the Simoa values after the strongest stimulation (R848) where the concentrations are above \(10\mathrm{pg / mL}\) (according to the Simoa measures). This result strongly suggests that Luminex assays should only be used to quantify high concentrations of IFNα.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[192, 95, 930, 696]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[117, 714, 875, 742]]<|/det|>
+<center>Figure for reviewer. Comparison of Simoa assays with Luminex in TruCulture stimulations; Null, Poly:IC, LPS, R848. Spearman correlations were performed per stimuli on each assay comparison. </center>  
+
+<|ref|>text<|/ref|><|det|>[[117, 757, 878, 904]]<|/det|>
+The experiments performed in Fig. 1 need to be better explained and the data provided. Notably, the ISG score and functional cytopathic experiments were not explained. For this, an article (PMID: 32661059) was quoted, but one has to download the supplemental methods from that paper to get the information. The authors should define in detail how the ISG score was obtained and provide the actual data (in a supplemental figure) for the expression measure of the 6 chosen ISGs (data corresponding to Fig. 1d- f). Moreover, it was unclear what the calibrator sample (used for 2- DDCT calculation) was. The authors should also define how the functional cytopathic effect experiment was performed (for this, reference 2, PMID: 32661059, actually referred to a paper from 1979
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 870, 101]]<|/det|>
+without further details) and the data should be provided (i.e. data corresponding to Fig. 1g-i).  
+
+<|ref|>text<|/ref|><|det|>[[147, 118, 857, 152]]<|/det|>
+- We have included these specific details as requested in the revised manuscript, and the data has been included in the supplemental tables.  
+
+<|ref|>text<|/ref|><|det|>[[118, 168, 815, 217]]<|/det|>
+The ISG score measured using the nanostring technology (Fig. 2b- c) should also be explained, and the data provided. Moreover, could the authors define the paired blood samples?  
+
+<|ref|>text<|/ref|><|det|>[[147, 234, 792, 267]]<|/det|>
+- We have included additional descriptions in the Methods of the revised manuscript and included the data in the relevant supplemental tables.  
+
+<|ref|>text<|/ref|><|det|>[[118, 283, 704, 300]]<|/det|>
+Fig. 2c- d, S2f- g: there are no units on the y- axis and y- axis, respectively  
+
+<|ref|>text<|/ref|><|det|>[[149, 317, 390, 333]]<|/det|>
+- This has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 349, 856, 448]]<|/det|>
+In the following paragraph, the figures are mislabeled, S2b should be replaced by S2e "... This showed low transcriptional levels of all measured IFNA subtypes as expected (Fig. S2b). Despite these low baseline levels, among the 7 IFNA subtypes examined we did observe some subtype differences, with notably higher levels of IFNA6 in both COVID- 19 patient groups and IFNA1/13 and IFNA5 only in the hospitalized group (Fig. S2b). IFNA2 was notably no different between all three groups (Fig. S2e)."  
+
+<|ref|>text<|/ref|><|det|>[[149, 466, 390, 482]]<|/det|>
+- This has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 499, 830, 515]]<|/det|>
+Fig 2c- d description in the main text: (>95 considered positive): please specify the units  
+
+<|ref|>text<|/ref|><|det|>[[149, 532, 382, 548]]<|/det|>
+- This has been included.  
+
+<|ref|>text<|/ref|><|det|>[[118, 565, 878, 666]]<|/det|>
+Please correct the figure numbers in the following paragraph: "Additional correlation analysis between cytometry measured intracellular IFNa, and digital ELISA measured plasma IFNa, showed in the absence of stimulation an association between monocytes and plasma IFNa levels (Fig. 4d). Following R848 stimulation, both pDCs and monocytes showed an association with secreted IFNa (Fig 4h), although the percentage of IFNa+ cells was lower in critical patients compared to severe and healthy controls (Fig. 4i)." (Fig. 4d actually shows the % of IFNa+ pDCs, Fig. 4h shows only data for pDC: R848...)  
+
+<|ref|>text<|/ref|><|det|>[[118, 666, 848, 682]]<|/det|>
+(Fig. 4d actually shows the % of IFNa+ pDCs, Fig. 4h shows only data for pDC: R848...)  
+
+<|ref|>text<|/ref|><|det|>[[149, 699, 390, 714]]<|/det|>
+- This has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 746, 878, 795]]<|/det|>
+"For this, we performed the same standardized whole blood ex vivo stimulation with recombinant IFNa2 and measured gene expression by Nanostring as previously described." - > please state where previously described  
+
+<|ref|>text<|/ref|><|det|>[[149, 813, 390, 828]]<|/det|>
+- This has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 845, 645, 899]]<|/det|>
+Fig S3 legend: CXCL10 (c), IL- 10 (c) \(\rightarrow\) CXCL10 (c), IL- 10 (d) - This has been corrected.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 850, 117]]<|/det|>
+Fig. 4c: No increase in the \(\%\) of pIRF3+ monocytes was observed with the healthy control cells when treated with R848. Could the authors comment on this?  
+
+<|ref|>text<|/ref|><|det|>[[148, 134, 860, 185]]<|/det|>
+- We thank the reviewer for highlighting this statement. While we did not see an increase in the \(\%\) of pIRF3+ monocytes we did see an increase in MFI, we have modified the text to reflect this.  
+
+<|ref|>text<|/ref|><|det|>[[178, 200, 613, 216]]<|/det|>
+Table S2: "Débit 02 max": please translate to English  
+
+<|ref|>text<|/ref|><|det|>[[149, 234, 390, 250]]<|/det|>
+- This has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[119, 267, 430, 283]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 299, 876, 479]]<|/det|>
+The role of type I IFNs in protecting against or aggravating disease outcomes during SARS- CoV- 2 infection remains controversial, and likely depends on the kinetics of IFN treatment vs. disease progression. This study adds important insights into the biological role of IFNs during different severities of COVID- 19. The authors demonstrate that patients with critical COVID- 19 have a more inflammatory gene expression profile. This has been demonstrated by other studies previously. However, interestingly, blood cells from hospitalized patients also mounted an inflammatory response on stimulation with TLR ligands, in contrast to cells from healthy individuals that mount an IFN- dependent ISG response. Overall, this study adds more information to the slowly growing body of literature that supports the likely role of a perturbed IFN response during severe COVID- 19. Some comments for the authors are mentioned below:  
+
+<|ref|>text<|/ref|><|det|>[[148, 496, 876, 562]]<|/det|>
+- We thank the reviewer for recognizing how our study adds important insights into the biological role of IFNs during different severities of COVID-19. We have responded to each specific point below and made corresponding modifications to the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[119, 578, 854, 662]]<|/det|>
+1. What was the clinical criteria/score used to define patients as mild, moderate, or severe? Please elaborate on this in the methodology. While the authors have included this in the methods, it is important to note how these vary in comparison to other studies. How is moderate different from 'mild' as defined by other publications? Putting this in context is important for the readers and for reproducibility.  
+
+<|ref|>text<|/ref|><|det|>[[148, 679, 877, 891]]<|/det|>
+- We apologize for any confusion here. We defined the patients (as described in the methods) based on internationally accepted criteria which is the requirements for supplemental oxygen at the time of sampling. Moderate patients did not require hospitalization at any timepoint. Hospitalized patients requiring supplemental oxygen via nasal cannula (maximal supplemental oxygen flow of up to 6L/min) were considered severe, with critical disease classified as requiring more than 6L of oxygen per minute, either delivered via high-flow nasal oxygen (Airvo) or a venturi mask, a clinical definition previously defined<sup>3,4</sup>. In the final version of the manuscript we decided not to distinguish mild from moderate patients as this is often based on the opinion of the clinician and is thus challenging to compare across studies. We have corrected this term in the manuscript and now all patients that did not require hospitalization or oxygen supplementation are defined as moderate.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 744, 117]]<|/det|>
+2. Were the patient cohorts scanned for comorbidities or ongoing treatments?  
+
+<|ref|>text<|/ref|><|det|>[[148, 135, 864, 184]]<|/det|>
+- In some of the cohorts this was possible, but due to the incomplete nature of the data sets we were not able to integrate these factors into the analysis which is why they were not included.  
+
+<|ref|>text<|/ref|><|det|>[[118, 200, 866, 266]]<|/det|>
+3. In Fig 1, the authors demonstrate that digital ELISA is more sensitive than other methods, such as Luminex. However, are such low levels, as detected by digital ELISA have any physiological relevance? Do these low IFNa2 concentrations perform well (induce gene expression) when used to stimulate human cells experimentally?  
+
+<|ref|>text<|/ref|><|det|>[[148, 283, 878, 530]]<|/det|>
+- This is a very interesting point, and one that we often discuss internally. The main set of evidence that these levels are physiological come from our multitude of published studies with these assays, in particular in cases of autoimmune disease where IFNa has been clinically implicated in the pathology and where levels of the plasma protein are within these ranges (>100fg/mL). Whether these low IFNa2 concentrations induce gene expression is dependent on the sensitivity of the assay measuring the gene expression, which do not always match the sensitivity of these digital ELISA. Recombinant IFN is often described in units, with 1 unit/mL of interferon being the quantity necessary to produce a cytopathic effect of 50%. 1 unit/mL is estimated to be between 200-300fg/mL depending on the viral cytopathic assay utilized, which also have their limitations in terms of sensitivity. However, in previous studies we have observed that an ISG score begins to correlate with Simoa measures between 1 and 10 fg/mL IFNa in SLE and JDM patients, and that the IFN activity begins to correlate between 10 and 100 fg/mL IFNa in JDM patients (Rodero et al, JEM 2016).  
+
+<|ref|>text<|/ref|><|det|>[[118, 545, 573, 562]]<|/det|>
+4. Are the differences in Fig. 3a statistically significant?  
+
+<|ref|>text<|/ref|><|det|>[[147, 579, 858, 613]]<|/det|>
+- No the differences between the different patient groups per stimulation are not statistically significant.  
+
+<|ref|>text<|/ref|><|det|>[[118, 629, 873, 678]]<|/det|>
+5. Although TLR4 and TLR8 gene counts are up in hospitalized patients (Fig. 3d), why don't their cells respond to stimulation by LPS and R848 (Figs. 3a-c)? Do TLR gene counts correlate with protein levels of TLRs in the cells?  
+
+<|ref|>text<|/ref|><|det|>[[148, 695, 878, 761]]<|/det|>
+- This is an interesting observation for which we do not currently have an evidence based explanation, but we may hypothesis that there are intracellular perturbations downstream of TLR signaling. It is challenging to measure TLR proteins in cells and were not able to perform this particular analysis.  
+
+<|ref|>text<|/ref|><|det|>[[118, 777, 361, 794]]<|/det|>
+6. Label for Fig. 6f is missing  
+
+<|ref|>text<|/ref|><|det|>[[148, 811, 390, 827]]<|/det|>
+- This has been corrected.  
+
+<|ref|>text<|/ref|><|det|>[[118, 844, 450, 860]]<|/det|>
+7. Heatmap legend is missing for Fig. 6b  
+
+<|ref|>text<|/ref|><|det|>[[148, 878, 390, 894]]<|/det|>
+- This has been corrected.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 100, 875, 134]]<|/det|>
+8. MDA5 plays a major role in recognizing SARS-CoV-2 RNA. Is there a reason why MDA5 stimulation wasn't included/discussed?  
+
+<|ref|>text<|/ref|><|det|>[[148, 150, 873, 218]]<|/det|>
+- This initial study was started during the first wave of the pandemic when little was known about the receptors that recognize SARS-CoV-2 RNA, which is the main reason why it was not included as a stimulation. We did include analysis of the levels of IFIH1 (gene encoding MDA5 in humans) in Fig 3d.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 83, 349, 99]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 116, 450, 133]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[119, 149, 470, 165]]<|/det|>
+The authors have addressed my concerns.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__43865603d5a26c4d69a5cc2b37b48b2437fdf8a2a31357deb4840ab90c8a6924/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__43865603d5a26c4d69a5cc2b37b48b2437fdf8a2a31357deb4840ab90c8a6924/supplementary_0_Peer Review File__43865603d5a26c4d69a5cc2b37b48b2437fdf8a2a31357deb4840ab90c8a6924.mmd

@@ -0,0 +1,650 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Conserved effector families render Phytophthora species vulnerable to recognition by NLR receptors in nonhost plants  
+
+![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):  
+
+Oh et al manuscript shows that multiple RxLR effectors are broadly conserved across Phytophthora species and most of them are recognized by corresponding Solanum NLRs. They also show that expression three of these NLRs conferred broad- spectrum resistance against multiple Phytophthora species. Based on these results, they suggest that nonhost resistance (NHR) can be mediated by NLRs. The manuscript is well written. The topic of this manuscript would be of broad interest and the results are potentially novel. Even though the experiments are well done, the data is not presented well and that made me less convincing as described below.  
+
+## Major Comments:  
+
+Line 35: Author suggests that "they have developed a homology- based approach to identify functional NLR. However, it is not clear whether this method is used for the first time by the author or the method is adopted from previously published approaches. In addition, in the abstract it will be useful to state which scientific questions were addressed and how they were addressed. This could perhaps be done in a sentence or two.  
+
+Line 46- 59: Although author has tried to explain NHR, however, it is not clear through this paragraph how the non- host resistance works. Author can briefly explain NHR mechanism before getting into unaddressed questions in NHR research domain.  
+
+Line 68- 69: This is a convoluted sentence- authors are suggested to modify the sentence to convey the message- Functional homologs of solanum NLRs recognizing effectors of Potato late blight pathogen- Phytophthora infestans were found in non- host plant chilli pepper.  
+
+Line 76: Acronym RxLR appears without defining it for the first time. Although, this might be an obvious terminology for a specialist in the field perhaps non- specialist reader would be left puzzled.  
+
+Line 137: Author should clarify what is the rationale for cloning 69 out of 89 effectors chosen. Was the additional 20 effectors could not be cloned?  
+
+Line 138: Although, author mixed effectors and putative solanum NLR in a 1:1 fashion, however, simply mixing does not ensure that two Agrobacterium strains separately carrying an effector and NLR would deliver two of these molecules to a single plant cell, even though this approach is commonly used. The variation observed in cell death could be due to this issue. Perhaps this could have been avoided by cloning the effector and NLR into a single T- DNA vector for expression? I do acknowledge this increases the work significantly. Perhaps they can test few examples to see if they observe less variation in cell death.  
+
+Line 167- 169: These sentences are miss leading. Author should amend the sentences to express clearly that Avr1, Avr8, or Avramr1 were transiently expressed in T0 plants expressing R1, R8, or Rpi- amr1. It is also not clear promoter and terminator used for the expression of R1, R8, or Rpi- amr1, and T- DNA backbone used for the cloning these genes. It will be useful if the author can list the sequences of the constructs used in supplementary info. In addition, it will be useful to provide representative map of construct used for the expression of R1, R8, or Rpi- amr1 and others in figure 3.  
+
+Line 229- 335: Discussion section could be shortened a bit and include more references Line 423: plasmid sequence of pICH31160 should be provided in the supplementary material. In addition, sequence of all 69 effectors cloned should be provided along with primers pairs used for their
+
+<--- Page Split --->
+
+amplification.  
+
+Fig. 2a is not well explained. What does the white blank space mean? No HR? but "- " also means no HR. The scale for the heat map doesn't make sense to me. There is no dark green in the scale. I presume dark green is strong HR.  
+
+Fig. 2b is also not well explained. There are single asterisks without connecting two box plots! What does this mean? It is statistically different than GFP control? The same applies to Supplemental figures 8, 11, and 12. There are single asterisks all over the place in some cases below the box plot! No explanation in the legend regarding what these asterisks are for! In Fig 2b, the lesion size of P. palmivora infection upon expression of R8 significantly increased instead of decreasing. The same is observed with P. capsici infection after expression of Rpi- amr1 and Rpi- blb2. The authors did not explain the reason behind this. How the R gene expression can make it more susceptible to disease?  
+
+There are serious issues with supplemental figures 7b, 9b, and 10b. I am not sure if this is data manipulation or inadvertent error. In 7b the images of PiAvr2::R2 and PiAvramr1::Rpi- amr1 interactions are identical! The same is observed with PiAvrblb1::Rpi- blb1 and PiAvrnt1::Rpi- vnt1 interactions where the images are identical! In Fig 9b PiAvr8::R8 and PccaAvramr1::Rpi- amr1 are identical; PiAvrnt1- Rpi- vnt1 and PccaAvrnt1::Rpi- vnt1 are identical. In Fig 10b PiAvramr1::Rpi- amr1 is identical to Fig 9b. It is not clear if the experiment was repeated. I did not compare all the images across all the figures. There may be more like this.  
+
+Fig 3 is not well prepared and difficult to understand. In Fig 3 c- f is cited in the text before Fig 3a. In Fig 3c- f, I am not able to see red and green dotted lines in the graphs as mentioned in the legend! However, I do see a very light green or red shading. Maybe give different colors for R1, R8 and Rpi- amr1 since the shades given can be confusing especially between R8 and Rpi- amr1. The percentages shown in the graph is very tiny and hard to see unless zoomed. The reader will not be able to read this in a printed copy. It is not clear what the difference between Supplemental figures 14 and 15 is.  
+
+Minor Comments:  
+
+Please expand gene names when appropriate during first mention. For example, Rpi- amr3.  
+
+Line 46: "now NHR" should be changed to "currently NHR" or "at present NHR".  
+
+Line 49: Need a reference.  
+
+Line 109, Avramr3 is missing in the Supplementary Fig 3 legend.  
+
+Line 165: agrobacterium is Agrobacterium  
+
+Line 198: Bit Score Score?  
+
+Line 221: In conclusion, among our homologous effector candidates- Authors are suggested to remove "our" from the sentence. Perhaps, this can be applied through the manuscript.  
+
+Line 237: needs a reference.  
+
+Line 241- 251: Need a reference.  
+
+Line 337: Plant materials and growth conditions- Author should provide details about the nutrition of the plants or if not refer to an appropriate prior study.  
+
+Line 349: "Incubated" should be changed to "incubating".  
+
+Line 406, it should be P. infestans.  
+
+Line 451: Washed out thrice with what?  
+
+Line 452: growth for four weeks in MS media, whether it is light or dark?
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+"I co- reviewed this manuscript with one of the reviewers who provided the listed reports". Please recognize my contributions accordingly.  
+
+Reviewer #3 (Remarks to the Author):  
+
+In the reviewed paper Oh et all. explored the hypothesis that NLRs conferring resistance to P. infestans can also recognise effectors and thus provide resistance against other Phytophthora species. The findings in this paper are overall interesting and provide new knowledge. I have some comments for the authors to consider:  
+
+Major points:  
+
+- The major hypothesis is that Rpi genes from Solanum species may confer non-host resistance to other Phytophthora species in addition to host resistance to P. infestans. However, care should be taken to provide evidence that the source of these genes (particularly S. americanum and S. demissum) actually are non-hosts to these other Phytophthora. Currently no references or experiments are provided to support this. Similarly, S. americanum is a reported non-host of P. infestans, so the suggestion that P. infestans is adapted to S. americanum are misleading. 
+- The finding that Rpi-amr1 confers recognition of P. cactorum and P. parasitica is already published (Witek et al 2021) while the study expands upon this and demonstrates additional recognition as well as resistance conferred by Rpi-amr1, care should be taken to make this clear. 
+- The quality of writing makes it difficult to understand at times  
+
+Minor points:  
+
+26 - The term 'the corresponding Solanum NLR' is misleading, there may be multiple NLRs capable of recognising these effectors (e.g. AVR2 is recognised by the unrelated NLRs Rpi-mcq1 and R2), not all may recognise the same orthologues from other Phytophthora species - Similar in Line- 80  
+
+57- 59 - This statement directly contradicts the following paragraph and the general hypothesis of the paper.  
+
+121 - Functionally conserved is a big assumption, conserved predicted structure does not mean that function is conserved.  
+
+163/164 - 'More natural condition' means non- transient expression? Is transgenic N. benthamiana natural?
+
+<--- Page Split --->
+
+190- 227 section and figure 4 - Isn't the finding obvious?  
+
+- 201/202 - "Overall, more closely related effectors were more possibly being recognized by corresponding Solanum NLRs."  
+
+- Similar effectors are more likely to be recognised than distantly related proteins  
+
+246- 251 - What does this mean?  
+
+270- 273 - This hypothesis is a repeat of the previous referenced statement  
+
+275- 294  
+
+- Do NLRs that mediate indirect recognition have lower sequence conservation than NLRs which directly recognise effectors? Are there references for this?  
+
+- 280-282 - Both convergently and divergently evolved NLRs could mediate NHR? The implication is that any NLR could mediate NHR  
+
+- 293-294 - Is there any evidence that convergently evolved NLRs are more likely to have an indirect recognition mode? Or that NLRs cannot convergently evolve direct recognition of an effector?  
+
+- How is this section relevant to the findings in the paper? Direct and indirect recognition is not distinguished elsewhere in the manuscript.  
+
+304/305  
+
+- Do you mean suppression of the NRC helper NLR? Could you correlate NRC dependency of the tested sensors with resistance observed? E.g. does Rpi-amr1-mediated recognition of P. parasitica evade suppression due to signalling through additional NRCs compared to R8 or other tested NLRs? 
+- There may be other reasons that recognition does not translate into resistance like: the sensor NLRs could also be suppressed, there are differences in the expression levels of the effectors between pathogens or due to the difference in the interaction strength between effectors from each pathogen, the effectors are overexpressed in the HR assay.  
+
+338 - N. benthamiana is not tobacco  
+
+341- 350 - Include details for P. capsici  
+
+406 - infestans rather than infestation  
+
+Figure 1  
+
+- Minor comment - It would be much easier to interpret if the species in (a) were not abbreviated, or if abbreviations were indicated in the legend  
+
+- "The presence or absence of parentheses indicates whether certain motif or domain is required or optional to be classified as each category" - It is not clear what is meant by this. Also, it appears that categories are distinguished by the presence/absence of Sig/WY, rather than the motif or domain in
+
+<--- Page Split --->
+
+parentheses.  
+
+Figure 2 (a): Green is not indicated on the scale, is this a pdf formatting issue? Positions without either - /HR show where there are no orthologues identified? Was only one orthologue from each species cloned? There are 40 HR assay results shown, but 77 effectors are mentioned in the legend - in the text - "60.87% (42/69) of the tested effectors induced cell death upon co- expressed with their putative corresponding Solanum NLR". It is not clear how many effectors are represented in this figure, if multiple effectors fit into each category, it would be more informative to indicate how many induce HR with the tested NLR (b) How many lesions were measured for each set? It would be helpful to represent this in a similar way to (c). Why were some combinations not tested - e.g. Rpi- vnt1 and P. cactorum which is shown as HR in panel (a)?  
+
+Figure 3 (a) Include representative images of all lines indicated in panel (b); (b) The pathogen (P. capsici?) is not indicated in the figure or legend, the second R8 transgenic line is not resistant? Consider changing the colours or datapoint style as R8 #11 and Rpi- amr1#2 are indistinguishable. How many plants of each line were tested? (c/d/e/f) WT lesion size plots are coloured the same as R1/R8/Rpi- amr1 lines, please use a different colour to indicate the WT control. It is strange that number of inoculation sites varies between the pictures, it should be consistent between WT and transgenic lines  
+
+Figure 4: Similar to line 121, functionally conserved is an assumption.  
+
+Figure 5d: Consider rephrasing 'number of functional NLRs' to something closer to 'compatible R- gene and avr- gene pairs'. Also evolutionary distance could be changed to 'coevolving to non- adapted"  
+
+Reviewer #4 (Remarks to the Author):  
+
+The manuscript by Oh et al investigates whether the conserved avirulence effectors across four Phytophthora spp. infecting different plant species were cross- recognised by immune receptors from different Solanaceous plants. They cloned 69 effectors closely related to 12 avirulence effectors into a binary vector and tested for their recognition by nine cognate immune receptors by agroinfiltration using both transient and stable transgenic Nicotiana benthamiana plants expressing these receptors. They found that some immune receptors recognised closely related effectors present in Phytophthora spp. that do not normally infect the plant species harboring those receptors. Additionally, they discovered that three of the immune receptors, tested using transgenic N. benthamiana plants, conferred broad resistance against "non- infecting" Phytophthora species. The authors concluded that these NLRs found in Solanaceous plants contribute to the non- host resistance (NHR) in this plant family.  
+
+The manuscript is mostly well written and logically laid out. Experiments were performed to excellent
+
+<--- Page Split --->
+
+standards and statistical analyses were included where applicable. The figures are of high quality (although there were some issues) with enough details for the readers to scrutinize. There are sufficient details in the materials and methods that other researchers can use to reproduce the results if need be. The authors are highly commendable for their thoroughness and robust experiments with meticulous presentation of the results.  
+
+I am convinced that the NLRs the authors have tested have the potential and capability of recognising the effectors from "non- infecting" pathogens and some do confer resistance against these pathogens when tested on a surrogate host such as N. benthamiana. However, my main concern is that the authors have not provided explicit enough answer to the question: Are these NLRs actually involved in NHR? In my opinion, the authors have not provided enough evidence to overcome the burden of proof that these NLRs are involved in NHR (e.g., Would NLR knock out in a host plant be successfully infected by a "non- infecting" pathogen in this experimental set up? Or would the "non- infecting" pathogen become infectious if the effectors that are recognised by non- host plant NLRs were knocked out? Are the NLRs and effectors expressed in the right places at the right time to the right amount? Would there be any NLRs that are involved in NHR against slightly more distant Albugo or Pythium spp?). The authors would be less controversial and more productive if they re- frame their findings along the evolutionary lines, such as evolutionary transitions, regressive evolution, diversification and differential loss of pathogenicity to host ranges etc, instead of tying them to NHR (although they can speculate about this phenomena). The fact that pathogen species as well as the host species used in this manuscript are "closely" related makes defining and resolving NHR difficult. As far as I am concerned, the infection assays conducted in N. benthamiana to show the context of NHR is confusing and not conclusive enough for authors' claims. N. benthamiana itself can be host or non- host to the pathogens they tested depending on circumstances as P. capsici and N. benthamiana might have never encountered each other in their evolutionary time until they were put together by humans. NHR could get more complicated if one considers P. capsici and P. palmivora to have wider host range than other Phytophthora species. Furthermore, the authors still cannot eliminate the possibility that other factors and genes, in addition to the NLRs, might be responsible for NHR in these interactions.  
+
+## Some minor issues:  
+
+- The authors would benefit from reconciling some concepts with Panstruga and Moscou (2020) https://doi.org/10.1094/MPMI-06-20-0161-CR in their introduction and discussions.- Perhaps discussions could be streamlined. It almost reads like a review article as it was written.  
+
+## Some corrections:  
+
+Line 32: "Solanaceae palnts" should be either "Solanaceous plants" or "Solanaceae family of plants" Line 325: "Similar with" should be "Similar to" Line 333: "more identification of NLRs" should be "identification of more NLRs" Line 338: Please delete "Tobacco" and parenthesis around N. benthamiana as it is not Tabacco plant Line 342: "rye agar plate" should be "rye sucrose agar plate"
+
+<--- Page Split --->
+
+Line 346: Please define "TDW"  
+
+Line 349: "after incubated" should be "after being incubated"  
+
+Line 415: Please state the cutoff values for bit- score and pTM score  
+
+Line 429: GV3101- it should be precisely written as GV3101 (pMP90) if this is indeed used (refer to http://www.bio.net/bionet/mm/arab- gen/2016- January/013588. html). Otherwise, GV3101 alone cannot transform plants as it is cured of Ti plasmid.  
+
+Line 432: "after adjusted" should be "after being adjusted"  
+
+Line 433: "were used for test" should be either "were used for testing" or "were used for agroinfiltration"  
+
+Line 435: "1\~4 scale indexing method" - please provide a reference, or describe the scale in details  
+
+Line 456: "For the root infection assay" - please provide the soil type used, and whether it was sterilized  
+
+Line 459: "Phenotypes were scored" - please describe specific phenotypes being scored  
+
+Figure 2a - Color scale bar next to the heat map - gradient application is wrong, e.g. 300 and 0 will have the same color according to the scale.  
+
+Figure 2b - White bar on the leaves - what does this represent? A scale bar? This should be described in the legend.  
+
+Figure 5a - does not capture the concept described in the legend very well. Perhaps it needs to be re- considered, for example, effectors and NLRs are missing. Figure 1 in their reference #1 seems to better represent the concept.  
+
+Figure 5b - what does each tick on the X- axis represent? Number of effectors? Different effector groups? This should be included in the graph. "Effectors" is not sufficient to understand the graph.
+
+<--- Page Split --->
+
+## Reply to Reviewer's Comments  
+
+## Reviewer #1 (Remarks to the Author):  
+
+Oh et al manuscript shows that multiple RxLR effectors are broadly conserved across Phytophthora species and most of them are recognized by corresponding Solanum NLRs. They also show that expression three of these NLRs conferred broad- spectrum resistance against multiple Phytophthora species. Based on these results, they suggest that nonhost resistance (NHR) can be mediated by NLRs.  
+
+The manuscript is well written. The topic of this manuscript would be of broad interest and the results are potentially novel. Even though the experiments are well done, the data is not presented well and that made me less convincing as described below.  
+
+## Major Comments:  
+
+Q1. Line 35: Author suggests that "they have developed a homology- based approach to identify functional NLR. However, it is not clear whether this method is used for the first time by the author or the method is adopted from previously published approaches. In addition, in the abstract it will be useful to state which scientific questions were addressed and how they were addressed. This could perhaps be done in a sentence or two.  
+
+Reply: As the reviewer's concern, we revised the abstract as below:  
+
+'Moreover, considering that resistance genes against most Phytophthora species, except for P. infestans, have never been identified, a homology- based approach could provide an alternative strategy of genetic mapping for identifying functional NLRs against multiple pathogens threatening crop production.' In Line 35- 38  
+
+Additionally, we referenced several papers performed similar approaches such as (Lin et al., Mol. Plant 2022 & Witek et al., Nat. Plant 2021; Laflamme et al., Science 2020) and revised related parts (added statement about the previously reported result in Line 150- 151 / legend in Figure 2a) to tone down previous statements. We also removed 'our' words from the most part of manuscript to emphasize the approach/methods used in the study is modified/adopted from the previous methods.  
+
+We added sentence and revised abstracts to clarify our question and how it was addressed as below.  
+
+[Question in Line 21- 22] However, the evolutionary process of how plants develop receptors for recognizing wide range of non- adapted pathogens is still elusive.  
+
+[Finding and meaning in Line 32- 35] Combined results suggest that conserved effector families of Phytophthora species allow Solanaceae family of plants to recognize a wide range of pathogens via NLRs that originally reported to recognize P. infestans. Thus, NLR- mediated recognition would contribute to NHR against pathogens that possess similar repertoires of effectors.  
+
+Q2. Line 46- 59: Although author has tried to explain NHR, however, it is not clear through this paragraph how the non- host resistance works. Author can briefly explain NHR mechanism before getting into unaddressed questions in NHR research domain.  
+
+Reply: As the reviewer's concerns, we additionally describe about 'NHR' in Line 47- 50, and
+
+<--- Page Split --->
+
+newly added reference (Panstruga and Moscou MPMI., 2020) to link the what is NHR and receptor- mediated NHR (especially NLRs) in introduction part.  
+
+Q3. Line 68- 69: This is a convoluted sentence- authors are suggested to modify the sentence to convey the message- Functional homologs of solanum NLRs recognizing effectors of Potato late blight pathogen- Phytophthora infestans were found in non- host plant chilli pepper.  
+
+Reply: As the reviewer's concerns, related part is removed from the revised manuscript.  
+
+Q4. Line 76: Acronym RxLR appears without defining it for the first time. Although, this might be an obvious terminology for a specialist in the field perhaps non- specialist reader would be left puzzled.  
+
+Reply: We added description as ' (conserved N- terminal Arg- Xaa- Leu- Arg motif)' as suggested in Line 23, and 82- 83.  
+
+Q5. Line 137: Author should clarify what is the rationale for cloning 69 out of 89 effectors chosen. Was the additional 20 effectors could not be cloned?  
+
+Reply: There were disparities between the genome information of reference strains (of Phytophthora species) and experimental (domestic/Korean) strains that we used in this study (as shown in Supplementary Table 4). We assume that several candidates were not existed (or possess SNPs in primer site) in our strain (were not amplified from PCR).  
+
+To clarify procedure and rationale, we revised method section (Line 423 - 424), and provided how we select (cut off, or detailed information about cloned effectors) in Line 415- 416, and Supplementary figure 3, 6, and supporting information about initial cloning targets (with raw data, excel file named as 'initial_sets_effector_numbering').  
+
+Q6. Line 138: Although, author mixed effectors and putative solanum NLR in a 1:1 fashion, however, simply mixing does not ensure that two Agrobacterium strains separately carrying an effector and NLR would deliver two of these molecules to a single plant cell, even though this approach is commonly used. The variation observed in cell death could be due to this issue. Perhaps this could have been avoided by cloning the effector and NLR into a single T-DNA vector for expression? I do acknowledge this increases the work significantly. Perhaps they can test few examples to see if they observe less variation in cell death.  
+
+Reply: We agree with the reviewer's suggestion and we could identify additional combination by stabilizing co- expression of effector and NLRs by cloning them into single vector. However, we are concerned about the benefits that could be gained by further elaborating screening method would not be significant because we could already obtain a plenty of candidates which exhibited consistent (at least 3 replications) and intensive cell death phenotype with commonly used 1:1 co- expression screening. We beg reviewer's generous accept about this part.  
+
+Q7. Line 167- 169: These sentences are miss leading. Author should amend the sentences to express clearly that Avr1, Avr8, or Avramr1 were transiently expressed in T0 plants expressing R1, R8, or Rpi- amr1. It is also not clear promoter and terminator used for the expression of R1, R8, or Rpi- amr1, and T- DNA backbone used for the cloning these genes. It will be useful if the author can list the sequences of the constructs used in supplementary info. In addition, it will be useful to provide representative map of construct used for the
+
+<--- Page Split --->
+
+expression of R1, R8, or Rpi- amr1 and others in figure 3.  
+
+Reply: As the reviewer's concerns, we revised sentences in Line 171- 175. And, we added whole plasmid sequencing data (in Supplementary table 7) of vectors used in this study, and we also submitted vector map files (as a supporting information).  
+
+Q8. Line 229- 335: Discussion section could be shortened a bit and include more references.  
+
+Reply: We shortened the discussion section from \(1,219 \Rightarrow 996\) words by removing 'Convergent and divergent evolution of NLRs and NHR' section. We also revised first section 'NLRs recognizing effectors of broad- spectrum pathogens have a potential to be exploited to confer durable resistance in crops' in Line 261- 287 for the clarification, and added some references including (Schulze- Lefert et al., Trends in Plant Sci. 2011; Witek et al., Nat. Plants 2021).  
+
+Q9. Line 423: plasmid sequence of pICH31160 should be provided in the supplementary material. In addition, sequence of all 69 effectors cloned should be provided along with primers pairs used for their amplification.  
+
+Reply: As the reviewer's concern, sequences of all vectors used in this study is added in Supplementary table 7. And also, coding sequences of 69 cloned effectors are added with the primer information in Supplementary table 6.  
+
+Q10. Fig. 2a is not well explained. What does the white blank space mean? No HR? but "- " also means no HR. The scale for the heat map doesn't make sense to me. There is no dark green in the scale. I presume dark green is strong HR.  
+
+Reply: We corrected figure 2, absence of dark green part was a file converting error and we fixed it. We revised the figure legend and design of figures, described detailed information of scales, and added description about white blank spaces that mean no conserved homologs or not cloned, thus not tested.  
+
+Q11. Fig. 2b is also not well explained. There are single asterisks without connecting two box plots! What does this mean? It is statistically different than GFP control? The same applies to Supplemental figures 8, 11, and 12. There are single asterisks all over the place in some cases below the box plot! No explanation in the legend regarding what these asterisks are for! In Fig 2b, the lesion size of P. palmivora infection upon expression of R8 significantly increased instead of decreasing. The same is observed with P. capsici infection after expression of Rpi- amr1 and Rpi- blb2. The authors did not explain the reason behind this. How the R gene expression can make it more susceptible to disease?  
+
+Reply: It was not a single asterix without connecting boxes but data point which were excluded drawing boxes because they exceed threshold from average. However, we notice this type of presentation could cause misunderstanding, thus we replaced all the graph format in this manuscript (also in supplementary figures).  
+
+About the lesion size increasing: for now, we have no clear answer about this phenomenon, though it was consistently observed for several cases (Rpi- blb2 against \(P\) . capsici and R8 against \(P\) . palmivora) as reviewer's concern. At least, as shown in the Fig 2a, Rpi- blb2 and R8 cannot recognized effectors of \(P\) . capsici (Avrblb2 is not conserved) or Avr8 of \(P\) . palmivora. Thus, we could just assume both NLRs cannot function as resistance gene (because they are not activated) against each pathogen but the physiological changes occurred by over- expressing those NLRs make more infectious environment in \(N\) . benthamiana leaves to the each Phytophthora pathogen. But still, we have no clear answer about this phenomenon.
+
+<--- Page Split --->
+
+Q12. There are serious issues with supplemental figures 7b, 9b, and 10b. I am not sure if this is data manipulation or inadvertent error. In 7b the images of PiAvr2::R2 and PiAvramr1::Rpi-amr1 interactions are identical! The same is observed with PiAvrblb1::Rpi-blb1 and PiAvrvt1::Rpi-vnt1 interactions where the images are identical! In Fig 9b PiAvr8::R8 and PcacAvramr1::Rpi-amr1 are identical; PiAvrvt1-Rpi-vnt1 and PcacAvrvt1::Rpi-vnt1 are identical. In Fig 10b PiAvramr1::Rpi-amr1 is identical to Fig 9b. It is not clear if the experiment was repeated. I did not compare all the images across all the figures. There may be more like this.  
+
+Reply: First of all, we thanks to the reviewer for pointing out this serious flaws and give us an opportunity for revising it. We thoroughly inspect our images (even for used in other figures) and replaced duplicated pictures with proper pictures, and also provided related raw data and information when each picture was taken with the picture of research note for each corresponding date.  
+
+As shown in the submitted raw data and research scheme (supporting information file named as 'HR_test_raw_data_scheme'), although the experiments were performed properly, we regret that there was an image modification issue on editing Supplementary figures, and once again, thanks for the reviewer and editor for the concerns and devotion on revising our manuscript.  
+
+Q13. Fig 3 is not well prepared and difficult to understand. In Fig 3 c-f is cited in the text before Fig 3a. In Fig 3c-f, I am not able to see red and green dotted lines in the graphs as mentioned in the legend! However, I do see a very light green or red shading. Maybe give different colors for R1, R8 and Rpi-amr1 since the shades given can be confusing especially between R8 and Rpi-amr1. The percentages shown in the graph is very tiny and hard to see unless zoomed. The reader will not be able to read this in a printed copy.  
+
+Reply: Figure legend are revised [dotted lines \(\Rightarrow\) shades], and all the graph in the Figure 3 are re- designed and re- colored for the clarification as the reviewer's suggestion (font size of percentage are also increased).  
+
+Q14. It is not clear what the difference between Supplemental figures 14 and 15 is.  
+
+Reply: We presented combined data on the main Figure 3, and Supplementary figure 14 and 15 were the replication of same experiment (1st and 2nd trials). To clarifying it, we described as 1st / 2nd trials in figure legends and revised design of both Supplementary Figures, and we also marked as (3rd trial) for the newly added Supplementary figure 16.  
+
+## Minor Comments:  
+
+Q1. Please expand gene names when appropriate during first mention. For example, Rpi-amr3. Reply: we added 'resistance genes against P. infestans' in Line 65-66  
+
+Q2. Line 46: "now NHR" should be changed to "currently NHR" or "at present NHR". Reply: revised as 'currently NHR' in Line 45  
+
+Q3. Line 49: Need a reference. Reply: reference is added in Line 51 (Oh et al., EBC. 2022).  
+
+Q4. Line 109, Avramr3 is missing in the Supplementary Fig 3 legend. Reply: We added Avramr3 in figure legend.
+
+<--- Page Split --->
+
+Q5. Line 165: agrobacterium is Agrobacterium Reply: revised in the whole manuscript.  
+
+Q6. Line 198: Bit Score Score? Reply: revised in Line 205  
+
+Q7. Line 221: In conclusion, among our homologous effector candidates- Authors are suggested to remove "our" from the sentence. Perhaps, this can be applied through the manuscript.  
+
+Reply: revised in the whole part of manuscript (removed 'our' word in related parts)  
+
+Q8. Line 237: needs a reference.  
+
+Reply: We added reference (Haverkort et al., Potato Res. 2016) in Line 255.  
+
+Q9. Line 241- 251: Need a reference.  
+
+Reply: Added reference in Line 266 (P. schulze- Lefert and R. panstruga et al., 2011), and revised related part.  
+
+Q10. Line 337: Plant materials and growth conditions- Author should provide details about the nutrition of the plants or if not refer to an appropriate prior study.  
+
+Reply: we added descriptions about detailed conditions for plant materials in Line 335- 338.  
+
+Q11. Line 349: "Incubated" should be changed to "incubating". Reply: revised as 'being incubated' in Line 349  
+
+Q12. Line 406, it should be P. infestans. Reply: revised as concerned, in Line 406  
+
+Q13. Line 451: Washed out thrice with what? Reply: we added detailed information about 'washing media' in Line 453.  
+
+Q14. Line 452: growth for four weeks in MS media, whether it is light or dark? Reply: under the continuous light, revised in Line 455.  
+
+Above all, we sincerely thanks to the Reviewer 1's devotion for polishing this manuscript.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+"I co- reviewed this manuscript with one of the reviewers who provided the listed reports". Please recognize my contributions accordingly.  
+
+Reply: We sincerely thanks to the Reviewer 2's contribution for revising this manuscript.
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+In the reviewed paper Oh et all. explored the hypothesis that NLRs conferring resistance to \(P\) infestans can also recognise effectors and thus provide resistance against other Phytophthora species. The findings in this paper are overall interesting and provide new knowledge. I have some comments for the authors to consider:  
+
+## Major points:  
+
+Q1. The major hypothesis is that Rpi genes from Solanum species may confer non-host resistance to other Phytophthora species in addition to host resistance to \(P\) infestans. However, care should be taken to provide evidence that the source of these genes (particularly S. americanum and S. demissum) actually are non-hosts to these other Phytophthora species. Currently no references or experiments are provided to support this. Similarly, S. americanum is a reported non-host of \(P\) infestans, so the suggestion that \(P\) infestans is adapted to S. americanum are misleading.  
+
+Reply: We completely agree with the reviewer's concern that there is no actual evidence in this study whether the sources of the tested NLRs are nonhost or not against tested Phytophthora species (Though the pathogens used in this study could be regarded as 'normally not infecting the Solanum species harboring the tested NLRs' because it has never been reported). However, we are not able to obtain all the wild Solanum species described in this study.  
+
+Thus, we tried to tone down and reduced statements about nonhost resistance (started from the title of the manuscript, in total, the number of 'NHR' in the manuscript decreased from \(22 \Rightarrow 11\) ; and 'non- adapted' decreased from \(14 \Rightarrow 5\) ), and most of those parts remain only in discussion & very first of introduction section. As presented in the replaced title, we tried to state our result as the broad- spectrum resistance by recognizing conserved effectors but stated as these mechanisms 'would' contribute to the nonhost resistance.  
+
+We revised the related part for clarification, in Line: 277- 280 as below: Considering that S. americanum is evolutionarily more distant from potato compared to the other wild Solanum species such as S. demissum or S. bulbocastanum, we could assume that \(P\) . parasitica is relatively well adapted to potato and its closely related species but not to S. americanum.  
+
+Q2. The finding that Rpi- amr1 confers recognition of \(P\) . cactorum and \(P\) . parasitica is already published (Witek et al 2021) while the study expands upon this and demonstrates additional recognition as well as resistance conferred by Rpi- amr1, care should be taken to make this clear.  
+
+Reply: As the reviewer's concern, we added statement about the previously reported result in Line 151, 275 and referenced (Witek et al 2021; though we already cited this paper but we added statement about the result), and tried to emphasize the previous results in Figure 2a using asterix (\*) mark with Rpi- amr1/3- related results and described it in figure legends.  
+
+Q3. The quality of writing makes it difficult to understand at times Reply: We revised the whole manuscript according to the three reviewers' major/minor concerns. And corrected grammars. We sincerely thanks to the reviewers' contribution for revising the manuscript.
+
+<--- Page Split --->
+
+## Minor points:  
+
+Q1. 26 - The term 'the corresponding Solanum NLR' is misleading, there may be multiple NLRs capable of recognising these effectors (e.g. AVR2 is recognised by the unrelated NLRs Rpi-mcq1 and R2), not all may recognise the same orthologues from other Phytophthora species. Similar in Line-80  
+
+Reply: we removed 'corresponding' from the Line 26 (abstract) and 86 (introduction)  
+
+Q2. 57- 59 - This statement directly contradicts the following paragraph and the general hypothesis of the paper.  
+
+Reply: Our intension was to state effectors are 'relatively variable' compared to PAMPs. Indeed, though PAMPs are conserved across the kingdom level, effectors are less broadly conserved (Phytophthora genus level in our results). We revised related part (in Line 57- 60) to clarify our intension.  
+
+Q3. 121 - Functionally conserved is a big assumption, conserved predicted structure does not mean that function is conserved.  
+
+Reply: We removed 'functionally conserved' from the related part (also in the whole manuscript), and revised as 'Combined results suggest that multiple effector families are conserved among Phytophthora species.' In Line 127- 128  
+
+Q4. 163/164 - 'More natural condition' means non- transient expression? Is transgenic N. benthamiana natural?  
+
+Reply: Revised as 'To validate Solanum NLR- mediated resistance against multiple Phytophthora species using transgenic plants' in Line 169- 170.  
+
+Q5. 190- 227 section and figure 4 - Isn't the finding obvious? 201/202 - "Overall, more closely related effectors were more possibly being recognized by corresponding Solanum NLRs." Similar effectors are more likely to be recognised than distantly related proteins  
+
+Reply: As a response to the reviewer's concern, we removed the concerned sentence and re- described the related section for the clarification (Line 198- 215 in revised manuscript).  
+
+Q6. 246- 251 - What does this mean?  
+
+Reply: We revised the related part for the clarification, in Line 261- 272, and also revised related part of Figure 5 and its legend.  
+
+Q7. 270- 273 - This hypothesis is a repeat of the previous referenced statement  
+
+Reply: We removed the repetitive statement and replace it with our own perspective about the durable resistance in Line 283- 288.  
+
+Q8. 275- 294 - Do NLRs that mediate indirect recognition have lower sequence conservation than NLRs which directly recognise effectors? Are there references for this?  
+
+Reply: As the reviewer's concerns, we realized the part named 'Convergent and divergent evolution of NLRs and NHR' was not well matched with this study. Thus, the related part is removed from the revised manuscript (in the discussion section).  
+
+Q9. 280- 282 - Both convergently and divergently evolved NLRs could mediate NHR? The implication is that any NLR could mediate NHR  
+
+Reply: As the reviewer's concerns, we realized the part named 'Convergent and divergent
+
+<--- Page Split --->
+
+evolution of NLRs and NHR' was not well matched with this study. Thus, the related part is removed from the revised manuscript (in the discussion section).  
+
+Q10. 293- 294 - Is there any evidence that convergently evolved NLRs are more likely to have an indirect recognition mode? Or that NLRs cannot convergently evolve direct recognition of an effector? How is this section relevant to the findings in the paper? Direct and indirect recognition is not distinguished elsewhere in the manuscript.  
+
+Reply: As the reviewer's concerns, we realized the part named 'Convergent and divergent evolution of NLRs and NHR' was not well matched with this study. Thus, the related part is removed from the revised manuscript (in the discussion section) for the uniformity.  
+
+Q11. 304/305 - Do you mean suppression of the NRC helper NLR? Could you correlate NRC dependency of the tested sensors with resistance observed? E.g. does Rpi-amr1- mediated recognition of P. parasitica evade suppression due to signaling through additional NRCs compared to R8 or other tested NLRs?  
+
+Reply: As the reviewer's comment, NRC is one of the possible candidate to be suppressed by pathogen, and similar with the reviewer's concern, we also assumed that different NRC- dependency of tested sensor NLRs could be correlated with the disparities between HR cell death against effectors and resistance against Phytophthora pathogens that we observed in Figure 2. Indeed, Rpi- amr1 (NRC2/3- dependent) cannot conferred resistance against P. capsici while R1 (NRC4- dependent) and R8 (NRC2/3/4- dependent) conferred significant resistance against P. capsici, even though all these three Solanum NLRs were able to recognize and induced cell death against effectors of P. capsici, Therefore, we hypothesized that P. capsici would suppress NRC2/3 but not NRC4 of N. benthamiana. To test this hypothesis, we used nrc4 knockout N. benthamiana. However, the result was negative (expression of R8 were still able to confer resistance against P. capsici even in nrc4 knockout N. benthamiana as shown in the submitted raw data of DLA experiment performed in '220415'using P. capsici). This result indirectly indicates that NbNRC2/3 still properly work during the P. capsici infection. As suggested by the reviewer, if we conduct similar experiments with a more variety of NRC knockout plants, we may be able to find examples that explain the correlation of NRC- dependency and the discrepancies between resistance and HR cell death phenotypes, but unfortunately, with the materials we currently have, it was not achieved.  
+
+Q12. There may be other reasons that recognition does not translate into resistance like: the sensor NLRs could also be suppressed, there are differences in the expression levels of the effectors between pathogens or due to the difference in the interaction strength between effectors from each pathogen, the effectors are overexpressed in the HR assay.  
+
+Reply: Agree, we added the several more possible reasons in the discussion part including the reviewer3's concerns in Line 300- 303.  
+
+Q13. 338 - N. benthamiana is not tobacco Reply: revised in Line 335.  
+
+Q14. 341- 350 - Include details for P. capsici Reply: added details for P. capsici in Line 343 and 346.  
+
+Q15. 406 - infestans rather than infestance Reply: revised in Line 406
+
+<--- Page Split --->
+
+Q16. Figure 1 (Minor comment) - It would be much easier to interpret if the species in (a) were not abbreviated, or if abbreviations were indicated in the legend  
+
+- "The presence or absence of parentheses indicates whether certain motif or domain is required or optional to be classified as each category" - It is not clear what is meant by this. Also, it appears that categories are distinguished by the presence/absence of Sig/WY, rather than the motif or domain in parentheses.  
+
+Reply: We have revised Figure 1 based on the reviewers' suggestions. Species names were assigned in Figure 1a rather than using abbreviations. The categories in Figure 1b were classified according to the presence or absence of a signal peptide (Sig) and WY domain (WY). Additionally, we have added a note to Figure 1b indicating that detailed domain and motif information regarding homologous effectors is provided in Supplementary Table 3.  
+
+Q17. Figure 2 (a): Green is not indicated on the scale, is this a pdf formatting issue? Positions without either - /HR show where there are no orthologues identified?  
+
+Reply: The absence of green part was a formatting issue, corrected. Avr2 homologs of \(P\) capsici were missed from dataset for drawing heat-map. It's also revised (added into the heatmap). Also, we revised the figure legend and design of figures, described detailed information of scales, and added description about white blank spaces that mean no conserved homologs or not cloned, thus not tested.  
+
+Q18. Was only one orthologue from each species cloned? There are 40 HR assay results shown, but 77 effectors are mentioned in the legend - in the text - "60.87% (42/69) of the tested effectors induced cell death upon co-expressed with their putative corresponding Solanum NLR".  
+
+Reply: We revised figure and its legend to clarify how to draw this figure, and 69 was correct number, we revised it.  
+
+Q19. It is not clear how many effectors are represented in this figure, if multiple effectors fit into each category, it would be more informative to indicate how many induce HR with the tested NLR  
+
+Reply: The number of HR positive / test effectors are presented as fraction for each cases in Figure 2a as the reviewer's concern.  
+
+Q20. (b) How many lesions were measured for each set? It would be helpful to represent this in a similar way to (c). Why were some combinations not tested - e.g. Rpi-vnt1 and \(P\) . cactorum which is shown as HR in panel (a)?  
+
+Reply: All dots presented as the reviewer's suggestion (similar with Figure 2c, for the all presented graph in main figures). However, the number of replications are different for each case and if it presented, the figures would become too complicated. Thus, detailed information (numbers) are presented in Supplementary figures 7- 12.  
+
+Not tested cases: For the cases of \(P\) . palmirova and \(P\) . cactorum, we did not test for several case (NLRs) when the corresponding effectors are not conserved in each pathogen (only R3a was tested even though \(P\) . cactorum did not possess Avr3a homologs in our criteria).  
+
+However, as the reviewer's concern, for the case of Rpi- vnt1, We have auto- activated cell death issue with p35s:Rpi- vnt1 construct. It induces moderate level of cell death on infiltrated region (weak cell death observed with naked eyes but merely detected using
+
+<--- Page Split --->
+
+FoBI machine – red leaves/black dead tissue images) at 2 dpi, but it kills most of infiltrated leaves at 2.5 dpi of agroinfiltration. While \(P\) . parasitica and \(P\) . capsici fully expand their lesion in \(N\) . benthamiana leaves within 2.0 days after inoculation, \(P\) . palmivora and \(P\) . cactorum take more than 3 days for the measureable lesion size. Thus, we could not test \(Rpi\) - vnt1- mediated resistance against \(P\) . cactorum in \(N\) . benthamiana (also, we could obtain other promising candidates \(R1\) , \(R8\) , and \(Rpi\) - amr1, it did not stimulate us to test \(Rpi\) - vnt1 with different/optimized experimental conditions). For the clarification, we added auto- active phenotypes of \(Rpi\) - vnt1 expressed leaves in Supplementary Figure 7a (weak cell death at 2 dpi) and 9a (severe cell death at 3 dpi), and described about these issues in legends. In addition, for the cell death test using \(Rpi\) - vnt1, we also provided the raw data (file named as 'HR_test_raw_data_scheme') with a proper control (photo taken before \(Rpi\) - vnt1 induced cell death itself, as shown in experiments performed at 230214_P. cactorum, 230309_P. parasitica, and 230406_P. sojae, respectively), and presented control cases of \(Rpi\) - vnt1 in Supplementary Figure 7a and 9a.  
+
+Q21. Figure 3 (a) Include representative images of all lines indicated in panel (b);  
+
+Reply: We could not take the pictures (as good as could be presented in the main figure) of representative images of R1 #3 and R8 #11, however we provided whole plant images in Supplementary figure 14 / 15, and newly added Supplementary figure 16, and all the pictures taken using transgenic plants (file named as 'transgenic_plant_raw_data')  
+
+Q22. Figure 3 (b) The pathogen ( \(P\) . capsici?) is not indicated in the figure or legend, the second R8 transgenic line is not resistant? Consider changing the colours or datapoint style as R8 #11 and Rpi- amr1#2 are indistinguishable. How many plants of each line were tested?  
+
+Reply: It was \(P\) . capsici, the figure and legend are revised to include \(P\) . capsici.  
+
+For the concern about the second line (#11) expressing R8, we observed characteristics of 'heterozygous line' of R8- expressing plants (especially #11) because we used T1 plants which could exhibit genetic segregation. The similar patterns were also observed in R1- expressing lines (2\~30% of plants were infected by \(P\) . capsici as wild type plants). And unfortunately, it was worse in the 1st trial (Supplementary figure 14). Thus we performed additional experiments (triplicate) as presented in Supplementary figure 15 (2nd trial) and Supplementary Figure 16 (3rd, newly performed) and observed that R8- expressing lines (#11) able to confer relatively weak resistance (compared to R1- expressing lines) against \(P\) . capsici. Indeed, this phenomenon was similarly observed in transient- expression- based experiments as shown in Figure 2c (R1 > R8, in terms of intensity of resistance).  
+
+In addition, we changed color of graph in Figure 3b, and the numbers of tested plants are presented in Figure 3a and Supplementary figure 14\~16.  
+
+Q23. (c/d/e/f) WT lesion size plots are coloured the same as R1/R8/Rpi- amr1 lines, please use a different colour to indicate the WT control. It is strange that number of inoculation sites varies between the pictures, it should be consistent between WT and transgenic lines  
+
+Reply: We changed colors, re- design, and revised legend of graph for the clear distinguish between R1 / R8 / Rpi- amr1 lines as the reviewer's concern. We also replaced images and synchronized all the numbers of inoculation sites of photos in Figure 3c\~f).  
+
+Q24. Figure 4: Similar to line 121, functionally conserved is an assumption.  
+
+Reply: We removed 'functionally conserved' as the reviewer's concern. In Line 127- 128, and also from the whole manuscript.
+
+<--- Page Split --->
+
+Q25. Figure 5d: Consider rephrasing 'number of functional NLRs' to something closer to 'compatible R-gene and avr-gene pairs'. Also evolutionary distance could be changed to 'coevolving to non-adapted"  
+
+Reply: Revised as reviewer's concerns. As [Number of functional NLRs => Compatible NLR/effector paris] [Evolutionary distance => Divergent time from adapted pathogen (A) of given plant (a)]  
+
+Above all, we sincerely thanks to the Reviewer 3's suggestions and devotion for revising this manuscript
+
+<--- Page Split --->
+
+## Reviewer #4 (Remarks to the Author):  
+
+The manuscript by Oh et al investigates whether the conserved avirulence effectors across four Phytophthora spp. infecting different plant species were cross- recognised by immune receptors from different Solanaceous plants. They cloned 69 effectors closely related to 12 avirulence effectors into a binary vector and tested for their recognition by nine cognate immune receptors by agroinfiltration using both transient and stable transgenic Nicotiana benthamiana plants expressing these receptors. They found that some immune receptors recognised closely related effectors present in Phytophthora spp. that do not normally infect the plant species harboring those receptors. Additionally, they discovered that three of the immune receptors, tested using transgenic N. benthamiana plants, conferred broad resistance against "non- infecting" Phytophthora species. The authors concluded that these NLRs found in Solanaceous plants contribute to the non- host resistance (NHR) in this plant family.  
+
+The manuscript is mostly well written and logically laid out. Experiments were performed to excellent standards and statistical analyses were included where applicable. The figures are of high quality (although there were some issues) with enough details for the readers to scrutinize. There are sufficient details in the materials and methods that other researchers can use to reproduce the results if need be. The authors are highly commendable for their thoroughness and robust experiments with meticulous presentation of the results.  
+
+## Major issues:  
+
+Q1. I am convinced that the NLRs the authors have tested have the potential and capability of recognising the effectors from "non-infecting" pathogens and some do confer resistance against these pathogens when tested on a surrogate host such as N. benthamiana. However, my main concern is that the authors have not provided explicit enough answer to the question: Are these NLRs actually involved in NHR? In my opinion, the authors have not provided enough evidence to overcome the burden of proof that these NLRs are involved in NHR (e.g., Would NLR knock out in a host plant be successfully infected by a "non- infecting" pathogen in this experimental set up? Or would the "non- infecting" pathogen become infectious if the effectors that are recognised by non- host plant NLRs were knocked out? Are the NLRs and effectors expressed in the right places at the right time to the right amount? Would there be any NLRs that are involved in NHR against slightly more distant Albugo or Pythium spp?). The authors would be less controversial and more productive if they re- frame their findings along the evolutionary lines, such as evolutionary transitions, regressive evolution, diversification and differential loss of pathogenicity to host ranges etc, instead of tying them to NHR (although they can speculate about this phenomena). The fact that pathogen species as well as the host species used in this manuscript are "closely" related makes defining and resolving NHR difficult. As far as I am concerned, the infection assays conducted in N. benthamiana to show the context of NHR is confusing and not conclusive enough for authors' claims. N. benthamiana itself can be host or non- host to the pathogens they tested depending on circumstances as P. capsici and N. benthamiana might have never encountered each other in their evolutionary time until they were put together by humans. NHR could get more complicated if one considers P. capsici and P. palmivora to have wider host range than other Phytophthora species. Furthermore, the authors still cannot eliminate the possibility that other factors and genes, in addition to the NLRs, might be responsible for NHR in these interactions.  
+
+Reply: We sincerely thanks to the reviewer's suggestion for the better presentation of our work
+
+<--- Page Split --->
+
+and pointing out the weak point of the manuscript. We completely agree with the reviewer's suggestion that 'to re- frame the findings along the evolutionary lines instead of tying them to NHR' would be less controversial and more productive.  
+
+Therefore, we tried to tone down and reduced statements about nonhost resistance (started from the title of the manuscript, in total, the number of 'NHR' in the manuscript decreased from \(22 \Rightarrow 11\) ; and 'non- adapted' decreased from \(14 \Rightarrow 5\) ), and those parts remain only in discussion & very first part of introduction section. As shown in the presented in the replaced title, we tried to state our result as the broad- spectrum resistance by recognizing conserved effectors of Phytophthora species but stated this kind of mechanism 'would' have potential to contribute to the nonhost resistance.  
+
+## Some minor issues:  
+
+Q1. The authors would benefit from reconciling some concepts with Panstruga and Moscou (2020) https://doi.org/10.1094/MPMI- 06- 20- 0161- CR in their introduction and discussions.  
+
+Reply: We newly cited (Panstruga and Moscou et al., MPMI. 2020) and tried to adopt their concept and description into this manuscript (in Line 47- 60 of introduction part), and we also revised related part in discussion section (in Line 261- 272 of discussion) for the clarification/reframing of our previous description.  
+
+Q2. Perhaps discussions could be streamlined. It almost reads like a review article as it was written.  
+
+Reply: We shortened the discussion section from \(1,219 \Rightarrow 996\) words by removing 'Convergent and divergent evolution of NLRs and NHR' part and revised most other parts to make it streamlined and reduce description about NHR or non- adapted pathogens.  
+
+## Some corrections:  
+
+Q1. Line 32: "Solanaceae palnts" should be either "Solanaceous plants" or "Solanaceae family of plants"  
+
+Reply: revised in Line 33, and also in the whole manuscript.  
+
+Q2. Line 325: "Similar with" should be "Similar to" Reply: revised in Line 323.  
+
+Q3. Line 333: "more identification of NLRs" should be "identification of more NLRs" Reply: revised in Line 330.  
+
+Q4. Line 338: Please delete "Tobacco" and parenthesis around \(N\) . benthamiana as it is not Tabacco plant  
+
+Reply: revised in Line 335.  
+
+Q5. Line 342: "rye agar plate" should be "rye sucrose agar plate" Reply: revised in Line 341.  
+
+Q6. Line 346: Please define "TDW" Reply: revised in Line 345.
+
+<--- Page Split --->
+
+Q7. Line 349: "after incubated" should be "after being incubated" Reply: revised in Line 349.  
+
+Q8. Line 415: Please state the cutoff values for bit-score and pTM score  
+
+Reply: We provided cut- off in Line 415- 416 as the reviewer's concerns and we also provided initially selected effectors for cloning in submitted raw data excel file named 'initial_sets_effector_numbering'  
+
+Q9. Line 429: GV3101- it should be precisely written as GV3101 (pMP90) if this is indeed used (refer to http://www.bio.net/bionet/mm/arab- gen/2016- January/013588. html). Otherwise, GV3101 alone cannot transform plants as it is cured of Ti plasmid.  
+
+Reply: revised as commented, thanks for the notification.  
+
+Q10. Line 432: "after adjusted" should be "after being adjusted" Reply: revised in Line 433.  
+
+Q11. Line 433: "were used for test" should be either "were used for testing" or "were used for agroinfiltration"  
+
+Reply: revised in Line 435.  
+
+Q12. Line 435: "1\~4 scale indexing method" - please provide a reference, or describe the scale in details  
+
+Reply: it was our mistake, we only scored HR cell death phenotypes with positive/negative in this study, it revised as presented in Supplementary table 6, thus related part is removed.  
+
+Q13. Line 456: "For the root infection assay" - please provide the soil type used, and whether it was sterilized  
+
+Reply: details about soil is added in Line 464- 466.  
+
+Q14. Line 459: "Phenotypes were scored" - please describe specific phenotypes being scored Reply: we add descriptions about how we determined wilt/dead plants in material method in Line 462- 463.  
+
+Q15. Figure 2a - Color scale bar next to the heat map - gradient application is wrong, e.g. 300 and 0 will have the same color according to the scale.  
+
+Reply: It was an error occurred during file converting in submission procedure, we corrected.  
+
+Q16. Figure 2b - White bar on the leaves - what does this represent? A scale bar? This should be described in the legend.  
+
+Reply: It's removed.  
+
+Q17. Figure 5a - does not capture the concept described in the legend very well. Perhaps it needs to be re- considered, for example, effectors and NLRs are missing. Figure 1 in their reference #1 seems to better represent the concept.  
+
+Reply: We added NLR and effectors in the picture, and revised figure legend and related part of the manuscript in Line 261- 272.  
+
+Q18. Figure 5b - what does each tick on the X-axis represent? Number of effectors? Different effector groups? This should be included in the graph. "Effectors" is not sufficient to
+
+<--- Page Split --->
+
+understand the graph.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+In the revised manuscript, the authors have addressed most of my concerns but needs a bit more improvement as mentioned below.  
+
+Include NHR in keywords  
+
+In Fig 2a legend, please mention blank space means not tested.  
+
+In Fig 2b, regarding the author's response to my concern about the lesion size of P. palmirova infection upon expression of R8 significantly increased instead of decreasing, the authors need to discuss this in the discussion section.  
+
+Also in fig 2b, please explain in the legend that the dots are data points. The dots look more like asterisk. It will be good to distinguish this form the asterisk. Can actual dots be used instead of asterisk? Maybe even a different colored dots.  
+
+Regarding washing media in line 453, It will be useful to clarify what is BA in liquid MS media with cefotaxime.  
+
+Reviewer #4 (Remarks to the Author):  
+
+The manuscript by Oh et al is a resubmission by the authors with revisions. The authors have addressed majority of my concerns and I do not have any other issues except the title which could be re- worded so that it conveys the message more clearly, perhaps along the line of "Conserved effector families renders Phytophthora species vulnerable to recognition by Nucleotide- binding leucine- rich repeat receptors in nonhost plants" or something similar. I appreciate the authors' openness and willingness to converse with the reviewers.
+
+<--- Page Split --->
+
+## REPLY to REVIEWERS' COMMENTS  
+
+Above all, thanks to the anonymous reviewers for their invaluable comments on our manuscript.  
+
+The new version of manuscript has been revised according to the reviewer's comments and replies to the reviewer's comments are followed.  
+
+Reviewer #1 (Remarks to the Author):  
+
+In the revised manuscript, the authors have addressed most of my concerns but needs a bit more improvement as mentioned below.  
+
+S1: Include NHR in keywords  
+
+Done  
+
+S2: In Fig 2a legend, please mention blank space means not tested.  
+
+Adjusted as suggested.  
+
+S3: In Fig 2b, regarding the author's response to my concern about the lesion size of P. palmiyora infection upon expression of R8 significantly increased instead of decreasing, the authors need to discuss this in the discussion section.  
+
+Added description in Line 292- 297 of discussion session.  
+
+S4: Also in fig 2b, please explain in the legend that the dots are data points. The dots look more like asterisk. It will be good to distinguish this form the asterisk. Can actual dots be used instead of asterisk? Maybe even a different colored dot.  
+
+We increased font size of asterisk for the clear distinguishment from data points.  
+
+S5: Regarding washing media in line 453, It will be useful to clarify what is BA in liquid MS media with cefotaxime.  
+
+We added full description as (BA \(= >\) benzyladenine)  
+
+Reviewer #4 (Remarks to the Author):  
+
+The manuscript by Oh et al is a resubmission by the authors with revisions.  
+
+The authors have addressed majority of my concerns and I do not have any other issues except the title which could be re- worded so that it conveys the message more clearly, perhaps along the line of "Conserved effector families renders Phytophthora species vulnerable to recognition by Nucleotide- binding leucine- rich repeat receptors in nonhost plants" or something similar. I appreciate the authors' openness and willingness to converse with the reviewers.  
+
+We decided below sentence as a New Title according to the reviewer's suggestion and editorial guide lines (less than 16 words)  
+
+'Conserved effector families render Phytophthora species vulnerable to recognition by NLR receptors in nonhost plants
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__43865603d5a26c4d69a5cc2b37b48b2437fdf8a2a31357deb4840ab90c8a6924/supplementary_0_Peer Review File__43865603d5a26c4d69a5cc2b37b48b2437fdf8a2a31357deb4840ab90c8a6924_det.mmd

@@ -0,0 +1,921 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 507, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[70, 110, 362, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 154, 907, 211]]<|/det|>
+Conserved effector families render Phytophthora species vulnerable to recognition by NLR receptors in nonhost plants  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 780]]<|/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, 91, 290, 107]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 127, 392, 144]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 163, 864, 291]]<|/det|>
+Oh et al manuscript shows that multiple RxLR effectors are broadly conserved across Phytophthora species and most of them are recognized by corresponding Solanum NLRs. They also show that expression three of these NLRs conferred broad- spectrum resistance against multiple Phytophthora species. Based on these results, they suggest that nonhost resistance (NHR) can be mediated by NLRs. The manuscript is well written. The topic of this manuscript would be of broad interest and the results are potentially novel. Even though the experiments are well done, the data is not presented well and that made me less convincing as described below.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 311, 250, 326]]<|/det|>
+## Major Comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 328, 874, 418]]<|/det|>
+Line 35: Author suggests that "they have developed a homology- based approach to identify functional NLR. However, it is not clear whether this method is used for the first time by the author or the method is adopted from previously published approaches. In addition, in the abstract it will be useful to state which scientific questions were addressed and how they were addressed. This could perhaps be done in a sentence or two.  
+
+<|ref|>text<|/ref|><|det|>[[115, 419, 855, 473]]<|/det|>
+Line 46- 59: Although author has tried to explain NHR, however, it is not clear through this paragraph how the non- host resistance works. Author can briefly explain NHR mechanism before getting into unaddressed questions in NHR research domain.  
+
+<|ref|>text<|/ref|><|det|>[[115, 474, 861, 529]]<|/det|>
+Line 68- 69: This is a convoluted sentence- authors are suggested to modify the sentence to convey the message- Functional homologs of solanum NLRs recognizing effectors of Potato late blight pathogen- Phytophthora infestans were found in non- host plant chilli pepper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 529, 867, 565]]<|/det|>
+Line 76: Acronym RxLR appears without defining it for the first time. Although, this might be an obvious terminology for a specialist in the field perhaps non- specialist reader would be left puzzled.  
+
+<|ref|>text<|/ref|><|det|>[[115, 566, 865, 601]]<|/det|>
+Line 137: Author should clarify what is the rationale for cloning 69 out of 89 effectors chosen. Was the additional 20 effectors could not be cloned?  
+
+<|ref|>text<|/ref|><|det|>[[115, 602, 868, 729]]<|/det|>
+Line 138: Although, author mixed effectors and putative solanum NLR in a 1:1 fashion, however, simply mixing does not ensure that two Agrobacterium strains separately carrying an effector and NLR would deliver two of these molecules to a single plant cell, even though this approach is commonly used. The variation observed in cell death could be due to this issue. Perhaps this could have been avoided by cloning the effector and NLR into a single T- DNA vector for expression? I do acknowledge this increases the work significantly. Perhaps they can test few examples to see if they observe less variation in cell death.  
+
+<|ref|>text<|/ref|><|det|>[[115, 730, 870, 840]]<|/det|>
+Line 167- 169: These sentences are miss leading. Author should amend the sentences to express clearly that Avr1, Avr8, or Avramr1 were transiently expressed in T0 plants expressing R1, R8, or Rpi- amr1. It is also not clear promoter and terminator used for the expression of R1, R8, or Rpi- amr1, and T- DNA backbone used for the cloning these genes. It will be useful if the author can list the sequences of the constructs used in supplementary info. In addition, it will be useful to provide representative map of construct used for the expression of R1, R8, or Rpi- amr1 and others in figure 3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 840, 870, 894]]<|/det|>
+Line 229- 335: Discussion section could be shortened a bit and include more references Line 423: plasmid sequence of pICH31160 should be provided in the supplementary material. In addition, sequence of all 69 effectors cloned should be provided along with primers pairs used for their
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 216, 105]]<|/det|>
+amplification.  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 883, 162]]<|/det|>
+Fig. 2a is not well explained. What does the white blank space mean? No HR? but "- " also means no HR. The scale for the heat map doesn't make sense to me. There is no dark green in the scale. I presume dark green is strong HR.  
+
+<|ref|>text<|/ref|><|det|>[[113, 163, 880, 291]]<|/det|>
+Fig. 2b is also not well explained. There are single asterisks without connecting two box plots! What does this mean? It is statistically different than GFP control? The same applies to Supplemental figures 8, 11, and 12. There are single asterisks all over the place in some cases below the box plot! No explanation in the legend regarding what these asterisks are for! In Fig 2b, the lesion size of P. palmivora infection upon expression of R8 significantly increased instead of decreasing. The same is observed with P. capsici infection after expression of Rpi- amr1 and Rpi- blb2. The authors did not explain the reason behind this. How the R gene expression can make it more susceptible to disease?  
+
+<|ref|>text<|/ref|><|det|>[[113, 291, 875, 417]]<|/det|>
+There are serious issues with supplemental figures 7b, 9b, and 10b. I am not sure if this is data manipulation or inadvertent error. In 7b the images of PiAvr2::R2 and PiAvramr1::Rpi- amr1 interactions are identical! The same is observed with PiAvrblb1::Rpi- blb1 and PiAvrnt1::Rpi- vnt1 interactions where the images are identical! In Fig 9b PiAvr8::R8 and PccaAvramr1::Rpi- amr1 are identical; PiAvrnt1- Rpi- vnt1 and PccaAvrnt1::Rpi- vnt1 are identical. In Fig 10b PiAvramr1::Rpi- amr1 is identical to Fig 9b. It is not clear if the experiment was repeated. I did not compare all the images across all the figures. There may be more like this.  
+
+<|ref|>text<|/ref|><|det|>[[113, 418, 884, 528]]<|/det|>
+Fig 3 is not well prepared and difficult to understand. In Fig 3 c- f is cited in the text before Fig 3a. In Fig 3c- f, I am not able to see red and green dotted lines in the graphs as mentioned in the legend! However, I do see a very light green or red shading. Maybe give different colors for R1, R8 and Rpi- amr1 since the shades given can be confusing especially between R8 and Rpi- amr1. The percentages shown in the graph is very tiny and hard to see unless zoomed. The reader will not be able to read this in a printed copy. It is not clear what the difference between Supplemental figures 14 and 15 is.  
+
+<|ref|>text<|/ref|><|det|>[[115, 549, 250, 563]]<|/det|>
+Minor Comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 566, 775, 583]]<|/det|>
+Please expand gene names when appropriate during first mention. For example, Rpi- amr3.  
+
+<|ref|>text<|/ref|><|det|>[[115, 584, 696, 600]]<|/det|>
+Line 46: "now NHR" should be changed to "currently NHR" or "at present NHR".  
+
+<|ref|>text<|/ref|><|det|>[[115, 603, 308, 618]]<|/det|>
+Line 49: Need a reference.  
+
+<|ref|>text<|/ref|><|det|>[[115, 621, 580, 638]]<|/det|>
+Line 109, Avramr3 is missing in the Supplementary Fig 3 legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 424, 656]]<|/det|>
+Line 165: agrobacterium is Agrobacterium  
+
+<|ref|>text<|/ref|><|det|>[[115, 658, 303, 673]]<|/det|>
+Line 198: Bit Score Score?  
+
+<|ref|>text<|/ref|><|det|>[[115, 676, 860, 711]]<|/det|>
+Line 221: In conclusion, among our homologous effector candidates- Authors are suggested to remove "our" from the sentence. Perhaps, this can be applied through the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 713, 321, 728]]<|/det|>
+Line 237: needs a reference.  
+
+<|ref|>text<|/ref|><|det|>[[115, 731, 350, 746]]<|/det|>
+Line 241- 251: Need a reference.  
+
+<|ref|>text<|/ref|><|det|>[[115, 749, 875, 784]]<|/det|>
+Line 337: Plant materials and growth conditions- Author should provide details about the nutrition of the plants or if not refer to an appropriate prior study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 785, 535, 801]]<|/det|>
+Line 349: "Incubated" should be changed to "incubating".  
+
+<|ref|>text<|/ref|><|det|>[[115, 804, 362, 819]]<|/det|>
+Line 406, it should be P. infestans.  
+
+<|ref|>text<|/ref|><|det|>[[115, 822, 408, 837]]<|/det|>
+Line 451: Washed out thrice with what?  
+
+<|ref|>text<|/ref|><|det|>[[115, 840, 645, 857]]<|/det|>
+Line 452: growth for four weeks in MS media, whether it is light or dark?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 108, 392, 125]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 144, 829, 181]]<|/det|>
+"I co- reviewed this manuscript with one of the reviewers who provided the listed reports". Please recognize my contributions accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[116, 255, 392, 272]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 290, 880, 364]]<|/det|>
+In the reviewed paper Oh et all. explored the hypothesis that NLRs conferring resistance to P. infestans can also recognise effectors and thus provide resistance against other Phytophthora species. The findings in this paper are overall interesting and provide new knowledge. I have some comments for the authors to consider:  
+
+<|ref|>text<|/ref|><|det|>[[115, 401, 215, 417]]<|/det|>
+Major points:  
+
+<|ref|>text<|/ref|><|det|>[[113, 419, 875, 600]]<|/det|>
+- The major hypothesis is that Rpi genes from Solanum species may confer non-host resistance to other Phytophthora species in addition to host resistance to P. infestans. However, care should be taken to provide evidence that the source of these genes (particularly S. americanum and S. demissum) actually are non-hosts to these other Phytophthora. Currently no references or experiments are provided to support this. Similarly, S. americanum is a reported non-host of P. infestans, so the suggestion that P. infestans is adapted to S. americanum are misleading. 
+- The finding that Rpi-amr1 confers recognition of P. cactorum and P. parasitica is already published (Witek et al 2021) while the study expands upon this and demonstrates additional recognition as well as resistance conferred by Rpi-amr1, care should be taken to make this clear. 
+- The quality of writing makes it difficult to understand at times  
+
+<|ref|>text<|/ref|><|det|>[[115, 620, 215, 636]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 655, 880, 729]]<|/det|>
+26 - The term 'the corresponding Solanum NLR' is misleading, there may be multiple NLRs capable of recognising these effectors (e.g. AVR2 is recognised by the unrelated NLRs Rpi-mcq1 and R2), not all may recognise the same orthologues from other Phytophthora species - Similar in Line- 80  
+
+<|ref|>text<|/ref|><|det|>[[115, 747, 857, 784]]<|/det|>
+57- 59 - This statement directly contradicts the following paragraph and the general hypothesis of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[115, 803, 850, 838]]<|/det|>
+121 - Functionally conserved is a big assumption, conserved predicted structure does not mean that function is conserved.  
+
+<|ref|>text<|/ref|><|det|>[[115, 857, 840, 892]]<|/det|>
+163/164 - 'More natural condition' means non- transient expression? Is transgenic N. benthamiana natural?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 108, 528, 124]]<|/det|>
+190- 227 section and figure 4 - Isn't the finding obvious?  
+
+<|ref|>text<|/ref|><|det|>[[115, 127, 787, 161]]<|/det|>
+- 201/202 - "Overall, more closely related effectors were more possibly being recognized by corresponding Solanum NLRs."  
+
+<|ref|>text<|/ref|><|det|>[[115, 164, 707, 181]]<|/det|>
+- Similar effectors are more likely to be recognised than distantly related proteins  
+
+<|ref|>text<|/ref|><|det|>[[116, 200, 358, 216]]<|/det|>
+246- 251 - What does this mean?  
+
+<|ref|>text<|/ref|><|det|>[[115, 236, 662, 253]]<|/det|>
+270- 273 - This hypothesis is a repeat of the previous referenced statement  
+
+<|ref|>text<|/ref|><|det|>[[115, 274, 178, 289]]<|/det|>
+275- 294  
+
+<|ref|>text<|/ref|><|det|>[[115, 291, 872, 326]]<|/det|>
+- Do NLRs that mediate indirect recognition have lower sequence conservation than NLRs which directly recognise effectors? Are there references for this?  
+
+<|ref|>text<|/ref|><|det|>[[115, 328, 872, 363]]<|/det|>
+- 280-282 - Both convergently and divergently evolved NLRs could mediate NHR? The implication is that any NLR could mediate NHR  
+
+<|ref|>text<|/ref|><|det|>[[115, 365, 844, 400]]<|/det|>
+- 293-294 - Is there any evidence that convergently evolved NLRs are more likely to have an indirect recognition mode? Or that NLRs cannot convergently evolve direct recognition of an effector?  
+
+<|ref|>text<|/ref|><|det|>[[115, 402, 810, 437]]<|/det|>
+- How is this section relevant to the findings in the paper? Direct and indirect recognition is not distinguished elsewhere in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 475, 179, 490]]<|/det|>
+304/305  
+
+<|ref|>text<|/ref|><|det|>[[115, 493, 864, 620]]<|/det|>
+- Do you mean suppression of the NRC helper NLR? Could you correlate NRC dependency of the tested sensors with resistance observed? E.g. does Rpi-amr1-mediated recognition of P. parasitica evade suppression due to signalling through additional NRCs compared to R8 or other tested NLRs? 
+- There may be other reasons that recognition does not translate into resistance like: the sensor NLRs could also be suppressed, there are differences in the expression levels of the effectors between pathogens or due to the difference in the interaction strength between effectors from each pathogen, the effectors are overexpressed in the HR assay.  
+
+<|ref|>text<|/ref|><|det|>[[115, 657, 386, 673]]<|/det|>
+338 - N. benthamiana is not tobacco  
+
+<|ref|>text<|/ref|><|det|>[[115, 694, 393, 710]]<|/det|>
+341- 350 - Include details for P. capsici  
+
+<|ref|>text<|/ref|><|det|>[[115, 730, 393, 746]]<|/det|>
+406 - infestans rather than infestation  
+
+<|ref|>text<|/ref|><|det|>[[115, 786, 176, 801]]<|/det|>
+Figure 1  
+
+<|ref|>text<|/ref|><|det|>[[115, 803, 864, 838]]<|/det|>
+- Minor comment - It would be much easier to interpret if the species in (a) were not abbreviated, or if abbreviations were indicated in the legend  
+
+<|ref|>text<|/ref|><|det|>[[115, 840, 852, 893]]<|/det|>
+- "The presence or absence of parentheses indicates whether certain motif or domain is required or optional to be classified as each category" - It is not clear what is meant by this. Also, it appears that categories are distinguished by the presence/absence of Sig/WY, rather than the motif or domain in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 210, 105]]<|/det|>
+parentheses.  
+
+<|ref|>text<|/ref|><|det|>[[112, 125, 880, 310]]<|/det|>
+Figure 2 (a): Green is not indicated on the scale, is this a pdf formatting issue? Positions without either - /HR show where there are no orthologues identified? Was only one orthologue from each species cloned? There are 40 HR assay results shown, but 77 effectors are mentioned in the legend - in the text - "60.87% (42/69) of the tested effectors induced cell death upon co- expressed with their putative corresponding Solanum NLR". It is not clear how many effectors are represented in this figure, if multiple effectors fit into each category, it would be more informative to indicate how many induce HR with the tested NLR (b) How many lesions were measured for each set? It would be helpful to represent this in a similar way to (c). Why were some combinations not tested - e.g. Rpi- vnt1 and P. cactorum which is shown as HR in panel (a)?  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 881, 473]]<|/det|>
+Figure 3 (a) Include representative images of all lines indicated in panel (b); (b) The pathogen (P. capsici?) is not indicated in the figure or legend, the second R8 transgenic line is not resistant? Consider changing the colours or datapoint style as R8 #11 and Rpi- amr1#2 are indistinguishable. How many plants of each line were tested? (c/d/e/f) WT lesion size plots are coloured the same as R1/R8/Rpi- amr1 lines, please use a different colour to indicate the WT control. It is strange that number of inoculation sites varies between the pictures, it should be consistent between WT and transgenic lines  
+
+<|ref|>text<|/ref|><|det|>[[115, 509, 620, 526]]<|/det|>
+Figure 4: Similar to line 121, functionally conserved is an assumption.  
+
+<|ref|>text<|/ref|><|det|>[[115, 546, 863, 582]]<|/det|>
+Figure 5d: Consider rephrasing 'number of functional NLRs' to something closer to 'compatible R- gene and avr- gene pairs'. Also evolutionary distance could be changed to 'coevolving to non- adapted"  
+
+<|ref|>text<|/ref|><|det|>[[116, 638, 392, 655]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 674, 884, 857]]<|/det|>
+The manuscript by Oh et al investigates whether the conserved avirulence effectors across four Phytophthora spp. infecting different plant species were cross- recognised by immune receptors from different Solanaceous plants. They cloned 69 effectors closely related to 12 avirulence effectors into a binary vector and tested for their recognition by nine cognate immune receptors by agroinfiltration using both transient and stable transgenic Nicotiana benthamiana plants expressing these receptors. They found that some immune receptors recognised closely related effectors present in Phytophthora spp. that do not normally infect the plant species harboring those receptors. Additionally, they discovered that three of the immune receptors, tested using transgenic N. benthamiana plants, conferred broad resistance against "non- infecting" Phytophthora species. The authors concluded that these NLRs found in Solanaceous plants contribute to the non- host resistance (NHR) in this plant family.  
+
+<|ref|>text<|/ref|><|det|>[[115, 875, 856, 893]]<|/det|>
+The manuscript is mostly well written and logically laid out. Experiments were performed to excellent
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 875, 180]]<|/det|>
+standards and statistical analyses were included where applicable. The figures are of high quality (although there were some issues) with enough details for the readers to scrutinize. There are sufficient details in the materials and methods that other researchers can use to reproduce the results if need be. The authors are highly commendable for their thoroughness and robust experiments with meticulous presentation of the results.  
+
+<|ref|>text<|/ref|><|det|>[[113, 199, 881, 601]]<|/det|>
+I am convinced that the NLRs the authors have tested have the potential and capability of recognising the effectors from "non- infecting" pathogens and some do confer resistance against these pathogens when tested on a surrogate host such as N. benthamiana. However, my main concern is that the authors have not provided explicit enough answer to the question: Are these NLRs actually involved in NHR? In my opinion, the authors have not provided enough evidence to overcome the burden of proof that these NLRs are involved in NHR (e.g., Would NLR knock out in a host plant be successfully infected by a "non- infecting" pathogen in this experimental set up? Or would the "non- infecting" pathogen become infectious if the effectors that are recognised by non- host plant NLRs were knocked out? Are the NLRs and effectors expressed in the right places at the right time to the right amount? Would there be any NLRs that are involved in NHR against slightly more distant Albugo or Pythium spp?). The authors would be less controversial and more productive if they re- frame their findings along the evolutionary lines, such as evolutionary transitions, regressive evolution, diversification and differential loss of pathogenicity to host ranges etc, instead of tying them to NHR (although they can speculate about this phenomena). The fact that pathogen species as well as the host species used in this manuscript are "closely" related makes defining and resolving NHR difficult. As far as I am concerned, the infection assays conducted in N. benthamiana to show the context of NHR is confusing and not conclusive enough for authors' claims. N. benthamiana itself can be host or non- host to the pathogens they tested depending on circumstances as P. capsici and N. benthamiana might have never encountered each other in their evolutionary time until they were put together by humans. NHR could get more complicated if one considers P. capsici and P. palmivora to have wider host range than other Phytophthora species. Furthermore, the authors still cannot eliminate the possibility that other factors and genes, in addition to the NLRs, might be responsible for NHR in these interactions.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 640, 258, 655]]<|/det|>
+## Some minor issues:  
+
+<|ref|>text<|/ref|><|det|>[[115, 674, 822, 728]]<|/det|>
+- The authors would benefit from reconciling some concepts with Panstruga and Moscou (2020) https://doi.org/10.1094/MPMI-06-20-0161-CR in their introduction and discussions.- Perhaps discussions could be streamlined. It almost reads like a review article as it was written.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 767, 248, 782]]<|/det|>
+## Some corrections:  
+
+<|ref|>text<|/ref|><|det|>[[115, 802, 844, 893]]<|/det|>
+Line 32: "Solanaceae palnts" should be either "Solanaceous plants" or "Solanaceae family of plants" Line 325: "Similar with" should be "Similar to" Line 333: "more identification of NLRs" should be "identification of more NLRs" Line 338: Please delete "Tobacco" and parenthesis around N. benthamiana as it is not Tabacco plant Line 342: "rye agar plate" should be "rye sucrose agar plate"
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 339, 106]]<|/det|>
+Line 346: Please define "TDW"  
+
+<|ref|>text<|/ref|><|det|>[[115, 108, 563, 125]]<|/det|>
+Line 349: "after incubated" should be "after being incubated"  
+
+<|ref|>text<|/ref|><|det|>[[115, 127, 606, 143]]<|/det|>
+Line 415: Please state the cutoff values for bit- score and pTM score  
+
+<|ref|>text<|/ref|><|det|>[[115, 145, 882, 199]]<|/det|>
+Line 429: GV3101- it should be precisely written as GV3101 (pMP90) if this is indeed used (refer to http://www.bio.net/bionet/mm/arab- gen/2016- January/013588. html). Otherwise, GV3101 alone cannot transform plants as it is cured of Ti plasmid.  
+
+<|ref|>text<|/ref|><|det|>[[115, 200, 545, 216]]<|/det|>
+Line 432: "after adjusted" should be "after being adjusted"  
+
+<|ref|>text<|/ref|><|det|>[[115, 218, 881, 234]]<|/det|>
+Line 433: "were used for test" should be either "were used for testing" or "were used for agroinfiltration"  
+
+<|ref|>text<|/ref|><|det|>[[115, 236, 850, 252]]<|/det|>
+Line 435: "1\~4 scale indexing method" - please provide a reference, or describe the scale in details  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 875, 270]]<|/det|>
+Line 456: "For the root infection assay" - please provide the soil type used, and whether it was sterilized  
+
+<|ref|>text<|/ref|><|det|>[[115, 272, 757, 289]]<|/det|>
+Line 459: "Phenotypes were scored" - please describe specific phenotypes being scored  
+
+<|ref|>text<|/ref|><|det|>[[115, 308, 870, 343]]<|/det|>
+Figure 2a - Color scale bar next to the heat map - gradient application is wrong, e.g. 300 and 0 will have the same color according to the scale.  
+
+<|ref|>text<|/ref|><|det|>[[115, 345, 872, 380]]<|/det|>
+Figure 2b - White bar on the leaves - what does this represent? A scale bar? This should be described in the legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 382, 866, 435]]<|/det|>
+Figure 5a - does not capture the concept described in the legend very well. Perhaps it needs to be re- considered, for example, effectors and NLRs are missing. Figure 1 in their reference #1 seems to better represent the concept.  
+
+<|ref|>text<|/ref|><|det|>[[115, 437, 872, 473]]<|/det|>
+Figure 5b - what does each tick on the X- axis represent? Number of effectors? Different effector groups? This should be included in the graph. "Effectors" is not sufficient to understand the graph.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[119, 100, 442, 120]]<|/det|>
+## Reply to Reviewer's Comments  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 142, 503, 161]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 179, 879, 268]]<|/det|>
+Oh et al manuscript shows that multiple RxLR effectors are broadly conserved across Phytophthora species and most of them are recognized by corresponding Solanum NLRs. They also show that expression three of these NLRs conferred broad- spectrum resistance against multiple Phytophthora species. Based on these results, they suggest that nonhost resistance (NHR) can be mediated by NLRs.  
+
+<|ref|>text<|/ref|><|det|>[[118, 269, 879, 320]]<|/det|>
+The manuscript is well written. The topic of this manuscript would be of broad interest and the results are potentially novel. Even though the experiments are well done, the data is not presented well and that made me less convincing as described below.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 339, 278, 356]]<|/det|>
+## Major Comments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 383, 879, 467]]<|/det|>
+Q1. Line 35: Author suggests that "they have developed a homology- based approach to identify functional NLR. However, it is not clear whether this method is used for the first time by the author or the method is adopted from previously published approaches. In addition, in the abstract it will be useful to state which scientific questions were addressed and how they were addressed. This could perhaps be done in a sentence or two.  
+
+<|ref|>text<|/ref|><|det|>[[118, 467, 669, 483]]<|/det|>
+Reply: As the reviewer's concern, we revised the abstract as below:  
+
+<|ref|>text<|/ref|><|det|>[[158, 483, 879, 549]]<|/det|>
+'Moreover, considering that resistance genes against most Phytophthora species, except for P. infestans, have never been identified, a homology- based approach could provide an alternative strategy of genetic mapping for identifying functional NLRs against multiple pathogens threatening crop production.' In Line 35- 38  
+
+<|ref|>text<|/ref|><|det|>[[158, 564, 879, 662]]<|/det|>
+Additionally, we referenced several papers performed similar approaches such as (Lin et al., Mol. Plant 2022 & Witek et al., Nat. Plant 2021; Laflamme et al., Science 2020) and revised related parts (added statement about the previously reported result in Line 150- 151 / legend in Figure 2a) to tone down previous statements. We also removed 'our' words from the most part of manuscript to emphasize the approach/methods used in the study is modified/adopted from the previous methods.  
+
+<|ref|>text<|/ref|><|det|>[[157, 679, 879, 712]]<|/det|>
+We added sentence and revised abstracts to clarify our question and how it was addressed as below.  
+
+<|ref|>text<|/ref|><|det|>[[157, 712, 879, 745]]<|/det|>
+[Question in Line 21- 22] However, the evolutionary process of how plants develop receptors for recognizing wide range of non- adapted pathogens is still elusive.  
+
+<|ref|>text<|/ref|><|det|>[[158, 745, 879, 827]]<|/det|>
+[Finding and meaning in Line 32- 35] Combined results suggest that conserved effector families of Phytophthora species allow Solanaceae family of plants to recognize a wide range of pathogens via NLRs that originally reported to recognize P. infestans. Thus, NLR- mediated recognition would contribute to NHR against pathogens that possess similar repertoires of effectors.  
+
+<|ref|>text<|/ref|><|det|>[[118, 843, 879, 893]]<|/det|>
+Q2. Line 46- 59: Although author has tried to explain NHR, however, it is not clear through this paragraph how the non- host resistance works. Author can briefly explain NHR mechanism before getting into unaddressed questions in NHR research domain.  
+
+<|ref|>text<|/ref|><|det|>[[115, 893, 879, 910]]<|/det|>
+Reply: As the reviewer's concerns, we additionally describe about 'NHR' in Line 47- 50, and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[156, 100, 878, 133]]<|/det|>
+newly added reference (Panstruga and Moscou MPMI., 2020) to link the what is NHR and receptor- mediated NHR (especially NLRs) in introduction part.  
+
+<|ref|>text<|/ref|><|det|>[[118, 149, 880, 216]]<|/det|>
+Q3. Line 68- 69: This is a convoluted sentence- authors are suggested to modify the sentence to convey the message- Functional homologs of solanum NLRs recognizing effectors of Potato late blight pathogen- Phytophthora infestans were found in non- host plant chilli pepper.  
+
+<|ref|>text<|/ref|><|det|>[[118, 215, 833, 232]]<|/det|>
+Reply: As the reviewer's concerns, related part is removed from the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 880, 297]]<|/det|>
+Q4. Line 76: Acronym RxLR appears without defining it for the first time. Although, this might be an obvious terminology for a specialist in the field perhaps non- specialist reader would be left puzzled.  
+
+<|ref|>text<|/ref|><|det|>[[118, 297, 880, 329]]<|/det|>
+Reply: We added description as ' (conserved N- terminal Arg- Xaa- Leu- Arg motif)' as suggested in Line 23, and 82- 83.  
+
+<|ref|>text<|/ref|><|det|>[[118, 345, 880, 377]]<|/det|>
+Q5. Line 137: Author should clarify what is the rationale for cloning 69 out of 89 effectors chosen. Was the additional 20 effectors could not be cloned?  
+
+<|ref|>text<|/ref|><|det|>[[118, 378, 880, 443]]<|/det|>
+Reply: There were disparities between the genome information of reference strains (of Phytophthora species) and experimental (domestic/Korean) strains that we used in this study (as shown in Supplementary Table 4). We assume that several candidates were not existed (or possess SNPs in primer site) in our strain (were not amplified from PCR).  
+
+<|ref|>text<|/ref|><|det|>[[150, 459, 880, 526]]<|/det|>
+To clarify procedure and rationale, we revised method section (Line 423 - 424), and provided how we select (cut off, or detailed information about cloned effectors) in Line 415- 416, and Supplementary figure 3, 6, and supporting information about initial cloning targets (with raw data, excel file named as 'initial_sets_effector_numbering').  
+
+<|ref|>text<|/ref|><|det|>[[118, 542, 880, 673]]<|/det|>
+Q6. Line 138: Although, author mixed effectors and putative solanum NLR in a 1:1 fashion, however, simply mixing does not ensure that two Agrobacterium strains separately carrying an effector and NLR would deliver two of these molecules to a single plant cell, even though this approach is commonly used. The variation observed in cell death could be due to this issue. Perhaps this could have been avoided by cloning the effector and NLR into a single T-DNA vector for expression? I do acknowledge this increases the work significantly. Perhaps they can test few examples to see if they observe less variation in cell death.  
+
+<|ref|>text<|/ref|><|det|>[[118, 674, 880, 789]]<|/det|>
+Reply: We agree with the reviewer's suggestion and we could identify additional combination by stabilizing co- expression of effector and NLRs by cloning them into single vector. However, we are concerned about the benefits that could be gained by further elaborating screening method would not be significant because we could already obtain a plenty of candidates which exhibited consistent (at least 3 replications) and intensive cell death phenotype with commonly used 1:1 co- expression screening. We beg reviewer's generous accept about this part.  
+
+<|ref|>text<|/ref|><|det|>[[118, 805, 880, 903]]<|/det|>
+Q7. Line 167- 169: These sentences are miss leading. Author should amend the sentences to express clearly that Avr1, Avr8, or Avramr1 were transiently expressed in T0 plants expressing R1, R8, or Rpi- amr1. It is also not clear promoter and terminator used for the expression of R1, R8, or Rpi- amr1, and T- DNA backbone used for the cloning these genes. It will be useful if the author can list the sequences of the constructs used in supplementary info. In addition, it will be useful to provide representative map of construct used for the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[156, 100, 620, 116]]<|/det|>
+expression of R1, R8, or Rpi- amr1 and others in figure 3.  
+
+<|ref|>text<|/ref|><|det|>[[118, 117, 880, 166]]<|/det|>
+Reply: As the reviewer's concerns, we revised sentences in Line 171- 175. And, we added whole plasmid sequencing data (in Supplementary table 7) of vectors used in this study, and we also submitted vector map files (as a supporting information).  
+
+<|ref|>text<|/ref|><|det|>[[118, 182, 864, 198]]<|/det|>
+Q8. Line 229- 335: Discussion section could be shortened a bit and include more references.  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 880, 297]]<|/det|>
+Reply: We shortened the discussion section from \(1,219 \Rightarrow 996\) words by removing 'Convergent and divergent evolution of NLRs and NHR' section. We also revised first section 'NLRs recognizing effectors of broad- spectrum pathogens have a potential to be exploited to confer durable resistance in crops' in Line 261- 287 for the clarification, and added some references including (Schulze- Lefert et al., Trends in Plant Sci. 2011; Witek et al., Nat. Plants 2021).  
+
+<|ref|>text<|/ref|><|det|>[[118, 313, 879, 362]]<|/det|>
+Q9. Line 423: plasmid sequence of pICH31160 should be provided in the supplementary material. In addition, sequence of all 69 effectors cloned should be provided along with primers pairs used for their amplification.  
+
+<|ref|>text<|/ref|><|det|>[[118, 363, 879, 411]]<|/det|>
+Reply: As the reviewer's concern, sequences of all vectors used in this study is added in Supplementary table 7. And also, coding sequences of 69 cloned effectors are added with the primer information in Supplementary table 6.  
+
+<|ref|>text<|/ref|><|det|>[[118, 427, 879, 476]]<|/det|>
+Q10. Fig. 2a is not well explained. What does the white blank space mean? No HR? but "- " also means no HR. The scale for the heat map doesn't make sense to me. There is no dark green in the scale. I presume dark green is strong HR.  
+
+<|ref|>text<|/ref|><|det|>[[118, 477, 879, 542]]<|/det|>
+Reply: We corrected figure 2, absence of dark green part was a file converting error and we fixed it. We revised the figure legend and design of figures, described detailed information of scales, and added description about white blank spaces that mean no conserved homologs or not cloned, thus not tested.  
+
+<|ref|>text<|/ref|><|det|>[[118, 558, 879, 689]]<|/det|>
+Q11. Fig. 2b is also not well explained. There are single asterisks without connecting two box plots! What does this mean? It is statistically different than GFP control? The same applies to Supplemental figures 8, 11, and 12. There are single asterisks all over the place in some cases below the box plot! No explanation in the legend regarding what these asterisks are for! In Fig 2b, the lesion size of P. palmivora infection upon expression of R8 significantly increased instead of decreasing. The same is observed with P. capsici infection after expression of Rpi- amr1 and Rpi- blb2. The authors did not explain the reason behind this. How the R gene expression can make it more susceptible to disease?  
+
+<|ref|>text<|/ref|><|det|>[[118, 690, 879, 755]]<|/det|>
+Reply: It was not a single asterix without connecting boxes but data point which were excluded drawing boxes because they exceed threshold from average. However, we notice this type of presentation could cause misunderstanding, thus we replaced all the graph format in this manuscript (also in supplementary figures).  
+
+<|ref|>text<|/ref|><|det|>[[157, 756, 879, 903]]<|/det|>
+About the lesion size increasing: for now, we have no clear answer about this phenomenon, though it was consistently observed for several cases (Rpi- blb2 against \(P\) . capsici and R8 against \(P\) . palmivora) as reviewer's concern. At least, as shown in the Fig 2a, Rpi- blb2 and R8 cannot recognized effectors of \(P\) . capsici (Avrblb2 is not conserved) or Avr8 of \(P\) . palmivora. Thus, we could just assume both NLRs cannot function as resistance gene (because they are not activated) against each pathogen but the physiological changes occurred by over- expressing those NLRs make more infectious environment in \(N\) . benthamiana leaves to the each Phytophthora pathogen. But still, we have no clear answer about this phenomenon.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 115, 880, 247]]<|/det|>
+Q12. There are serious issues with supplemental figures 7b, 9b, and 10b. I am not sure if this is data manipulation or inadvertent error. In 7b the images of PiAvr2::R2 and PiAvramr1::Rpi-amr1 interactions are identical! The same is observed with PiAvrblb1::Rpi-blb1 and PiAvrvt1::Rpi-vnt1 interactions where the images are identical! In Fig 9b PiAvr8::R8 and PcacAvramr1::Rpi-amr1 are identical; PiAvrvt1-Rpi-vnt1 and PcacAvrvt1::Rpi-vnt1 are identical. In Fig 10b PiAvramr1::Rpi-amr1 is identical to Fig 9b. It is not clear if the experiment was repeated. I did not compare all the images across all the figures. There may be more like this.  
+
+<|ref|>text<|/ref|><|det|>[[118, 247, 880, 329]]<|/det|>
+Reply: First of all, we thanks to the reviewer for pointing out this serious flaws and give us an opportunity for revising it. We thoroughly inspect our images (even for used in other figures) and replaced duplicated pictures with proper pictures, and also provided related raw data and information when each picture was taken with the picture of research note for each corresponding date.  
+
+<|ref|>text<|/ref|><|det|>[[158, 329, 879, 411]]<|/det|>
+As shown in the submitted raw data and research scheme (supporting information file named as 'HR_test_raw_data_scheme'), although the experiments were performed properly, we regret that there was an image modification issue on editing Supplementary figures, and once again, thanks for the reviewer and editor for the concerns and devotion on revising our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 426, 880, 526]]<|/det|>
+Q13. Fig 3 is not well prepared and difficult to understand. In Fig 3 c-f is cited in the text before Fig 3a. In Fig 3c-f, I am not able to see red and green dotted lines in the graphs as mentioned in the legend! However, I do see a very light green or red shading. Maybe give different colors for R1, R8 and Rpi-amr1 since the shades given can be confusing especially between R8 and Rpi-amr1. The percentages shown in the graph is very tiny and hard to see unless zoomed. The reader will not be able to read this in a printed copy.  
+
+<|ref|>text<|/ref|><|det|>[[118, 526, 880, 576]]<|/det|>
+Reply: Figure legend are revised [dotted lines \(\Rightarrow\) shades], and all the graph in the Figure 3 are re- designed and re- colored for the clarification as the reviewer's suggestion (font size of percentage are also increased).  
+
+<|ref|>text<|/ref|><|det|>[[118, 590, 790, 607]]<|/det|>
+Q14. It is not clear what the difference between Supplemental figures 14 and 15 is.  
+
+<|ref|>text<|/ref|><|det|>[[118, 608, 879, 674]]<|/det|>
+Reply: We presented combined data on the main Figure 3, and Supplementary figure 14 and 15 were the replication of same experiment (1st and 2nd trials). To clarifying it, we described as 1st / 2nd trials in figure legends and revised design of both Supplementary Figures, and we also marked as (3rd trial) for the newly added Supplementary figure 16.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 690, 278, 705]]<|/det|>
+## Minor Comments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 722, 878, 756]]<|/det|>
+Q1. Please expand gene names when appropriate during first mention. For example, Rpi-amr3. Reply: we added 'resistance genes against P. infestans' in Line 65-66  
+
+<|ref|>text<|/ref|><|det|>[[118, 771, 822, 805]]<|/det|>
+Q2. Line 46: "now NHR" should be changed to "currently NHR" or "at present NHR". Reply: revised as 'currently NHR' in Line 45  
+
+<|ref|>text<|/ref|><|det|>[[118, 821, 616, 854]]<|/det|>
+Q3. Line 49: Need a reference. Reply: reference is added in Line 51 (Oh et al., EBC. 2022).  
+
+<|ref|>text<|/ref|><|det|>[[118, 869, 682, 903]]<|/det|>
+Q4. Line 109, Avramr3 is missing in the Supplementary Fig 3 legend. Reply: We added Avramr3 in figure legend.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 116, 500, 150]]<|/det|>
+Q5. Line 165: agrobacterium is Agrobacterium Reply: revised in the whole manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 165, 375, 199]]<|/det|>
+Q6. Line 198: Bit Score Score? Reply: revised in Line 205  
+
+<|ref|>text<|/ref|><|det|>[[118, 214, 880, 264]]<|/det|>
+Q7. Line 221: In conclusion, among our homologous effector candidates- Authors are suggested to remove "our" from the sentence. Perhaps, this can be applied through the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 264, 800, 281]]<|/det|>
+Reply: revised in the whole part of manuscript (removed 'our' word in related parts)  
+
+<|ref|>text<|/ref|><|det|>[[118, 296, 384, 312]]<|/det|>
+Q8. Line 237: needs a reference.  
+
+<|ref|>text<|/ref|><|det|>[[118, 313, 741, 330]]<|/det|>
+Reply: We added reference (Haverkort et al., Potato Res. 2016) in Line 255.  
+
+<|ref|>text<|/ref|><|det|>[[118, 345, 416, 361]]<|/det|>
+Q9. Line 241- 251: Need a reference.  
+
+<|ref|>text<|/ref|><|det|>[[118, 362, 880, 394]]<|/det|>
+Reply: Added reference in Line 266 (P. schulze- Lefert and R. panstruga et al., 2011), and revised related part.  
+
+<|ref|>text<|/ref|><|det|>[[118, 410, 880, 444]]<|/det|>
+Q10. Line 337: Plant materials and growth conditions- Author should provide details about the nutrition of the plants or if not refer to an appropriate prior study.  
+
+<|ref|>text<|/ref|><|det|>[[118, 444, 864, 461]]<|/det|>
+Reply: we added descriptions about detailed conditions for plant materials in Line 335- 338.  
+
+<|ref|>text<|/ref|><|det|>[[118, 476, 634, 510]]<|/det|>
+Q11. Line 349: "Incubated" should be changed to "incubating". Reply: revised as 'being incubated' in Line 349  
+
+<|ref|>text<|/ref|><|det|>[[118, 525, 450, 559]]<|/det|>
+Q12. Line 406, it should be P. infestans. Reply: revised as concerned, in Line 406  
+
+<|ref|>text<|/ref|><|det|>[[118, 575, 721, 608]]<|/det|>
+Q13. Line 451: Washed out thrice with what? Reply: we added detailed information about 'washing media' in Line 453.  
+
+<|ref|>text<|/ref|><|det|>[[118, 623, 761, 657]]<|/det|>
+Q14. Line 452: growth for four weeks in MS media, whether it is light or dark? Reply: under the continuous light, revised in Line 455.  
+
+<|ref|>text<|/ref|><|det|>[[118, 689, 877, 707]]<|/det|>
+Above all, we sincerely thanks to the Reviewer 1's devotion for polishing this manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 741, 504, 761]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 777, 877, 811]]<|/det|>
+"I co- reviewed this manuscript with one of the reviewers who provided the listed reports". Please recognize my contributions accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[115, 818, 840, 836]]<|/det|>
+Reply: We sincerely thanks to the Reviewer 2's contribution for revising this manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 100, 504, 120]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 135, 879, 202]]<|/det|>
+In the reviewed paper Oh et all. explored the hypothesis that NLRs conferring resistance to \(P\) infestans can also recognise effectors and thus provide resistance against other Phytophthora species. The findings in this paper are overall interesting and provide new knowledge. I have some comments for the authors to consider:  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 217, 238, 234]]<|/det|>
+## Major points:  
+
+<|ref|>text<|/ref|><|det|>[[118, 249, 880, 365]]<|/det|>
+Q1. The major hypothesis is that Rpi genes from Solanum species may confer non-host resistance to other Phytophthora species in addition to host resistance to \(P\) infestans. However, care should be taken to provide evidence that the source of these genes (particularly S. americanum and S. demissum) actually are non-hosts to these other Phytophthora species. Currently no references or experiments are provided to support this. Similarly, S. americanum is a reported non-host of \(P\) infestans, so the suggestion that \(P\) infestans is adapted to S. americanum are misleading.  
+
+<|ref|>text<|/ref|><|det|>[[118, 365, 880, 462]]<|/det|>
+Reply: We completely agree with the reviewer's concern that there is no actual evidence in this study whether the sources of the tested NLRs are nonhost or not against tested Phytophthora species (Though the pathogens used in this study could be regarded as 'normally not infecting the Solanum species harboring the tested NLRs' because it has never been reported). However, we are not able to obtain all the wild Solanum species described in this study.  
+
+<|ref|>text<|/ref|><|det|>[[151, 463, 880, 560]]<|/det|>
+Thus, we tried to tone down and reduced statements about nonhost resistance (started from the title of the manuscript, in total, the number of 'NHR' in the manuscript decreased from \(22 \Rightarrow 11\) ; and 'non- adapted' decreased from \(14 \Rightarrow 5\) ), and most of those parts remain only in discussion & very first of introduction section. As presented in the replaced title, we tried to state our result as the broad- spectrum resistance by recognizing conserved effectors but stated as these mechanisms 'would' contribute to the nonhost resistance.  
+
+<|ref|>text<|/ref|><|det|>[[151, 576, 880, 660]]<|/det|>
+We revised the related part for clarification, in Line: 277- 280 as below: Considering that S. americanum is evolutionarily more distant from potato compared to the other wild Solanum species such as S. demissum or S. bulbocastanum, we could assume that \(P\) . parasitica is relatively well adapted to potato and its closely related species but not to S. americanum.  
+
+<|ref|>text<|/ref|><|det|>[[118, 676, 880, 740]]<|/det|>
+Q2. The finding that Rpi- amr1 confers recognition of \(P\) . cactorum and \(P\) . parasitica is already published (Witek et al 2021) while the study expands upon this and demonstrates additional recognition as well as resistance conferred by Rpi- amr1, care should be taken to make this clear.  
+
+<|ref|>text<|/ref|><|det|>[[118, 741, 880, 808]]<|/det|>
+Reply: As the reviewer's concern, we added statement about the previously reported result in Line 151, 275 and referenced (Witek et al 2021; though we already cited this paper but we added statement about the result), and tried to emphasize the previous results in Figure 2a using asterix (\*) mark with Rpi- amr1/3- related results and described it in figure legends.  
+
+<|ref|>text<|/ref|><|det|>[[118, 823, 880, 889]]<|/det|>
+Q3. The quality of writing makes it difficult to understand at times Reply: We revised the whole manuscript according to the three reviewers' major/minor concerns. And corrected grammars. We sincerely thanks to the reviewers' contribution for revising the manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 100, 238, 116]]<|/det|>
+## Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[118, 131, 880, 199]]<|/det|>
+Q1. 26 - The term 'the corresponding Solanum NLR' is misleading, there may be multiple NLRs capable of recognising these effectors (e.g. AVR2 is recognised by the unrelated NLRs Rpi-mcq1 and R2), not all may recognise the same orthologues from other Phytophthora species. Similar in Line-80  
+
+<|ref|>text<|/ref|><|det|>[[118, 199, 815, 216]]<|/det|>
+Reply: we removed 'corresponding' from the Line 26 (abstract) and 86 (introduction)  
+
+<|ref|>text<|/ref|><|det|>[[118, 231, 880, 264]]<|/det|>
+Q2. 57- 59 - This statement directly contradicts the following paragraph and the general hypothesis of the paper.  
+
+<|ref|>text<|/ref|><|det|>[[118, 264, 879, 329]]<|/det|>
+Reply: Our intension was to state effectors are 'relatively variable' compared to PAMPs. Indeed, though PAMPs are conserved across the kingdom level, effectors are less broadly conserved (Phytophthora genus level in our results). We revised related part (in Line 57- 60) to clarify our intension.  
+
+<|ref|>text<|/ref|><|det|>[[118, 345, 879, 378]]<|/det|>
+Q3. 121 - Functionally conserved is a big assumption, conserved predicted structure does not mean that function is conserved.  
+
+<|ref|>text<|/ref|><|det|>[[118, 378, 879, 427]]<|/det|>
+Reply: We removed 'functionally conserved' from the related part (also in the whole manuscript), and revised as 'Combined results suggest that multiple effector families are conserved among Phytophthora species.' In Line 127- 128  
+
+<|ref|>text<|/ref|><|det|>[[118, 443, 878, 475]]<|/det|>
+Q4. 163/164 - 'More natural condition' means non- transient expression? Is transgenic N. benthamiana natural?  
+
+<|ref|>text<|/ref|><|det|>[[118, 476, 879, 510]]<|/det|>
+Reply: Revised as 'To validate Solanum NLR- mediated resistance against multiple Phytophthora species using transgenic plants' in Line 169- 170.  
+
+<|ref|>text<|/ref|><|det|>[[118, 525, 880, 574]]<|/det|>
+Q5. 190- 227 section and figure 4 - Isn't the finding obvious? 201/202 - "Overall, more closely related effectors were more possibly being recognized by corresponding Solanum NLRs." Similar effectors are more likely to be recognised than distantly related proteins  
+
+<|ref|>text<|/ref|><|det|>[[118, 575, 879, 607]]<|/det|>
+Reply: As a response to the reviewer's concern, we removed the concerned sentence and re- described the related section for the clarification (Line 198- 215 in revised manuscript).  
+
+<|ref|>text<|/ref|><|det|>[[118, 623, 421, 639]]<|/det|>
+Q6. 246- 251 - What does this mean?  
+
+<|ref|>text<|/ref|><|det|>[[118, 640, 879, 673]]<|/det|>
+Reply: We revised the related part for the clarification, in Line 261- 272, and also revised related part of Figure 5 and its legend.  
+
+<|ref|>text<|/ref|><|det|>[[118, 689, 758, 706]]<|/det|>
+Q7. 270- 273 - This hypothesis is a repeat of the previous referenced statement  
+
+<|ref|>text<|/ref|><|det|>[[118, 706, 879, 738]]<|/det|>
+Reply: We removed the repetitive statement and replace it with our own perspective about the durable resistance in Line 283- 288.  
+
+<|ref|>text<|/ref|><|det|>[[118, 754, 879, 787]]<|/det|>
+Q8. 275- 294 - Do NLRs that mediate indirect recognition have lower sequence conservation than NLRs which directly recognise effectors? Are there references for this?  
+
+<|ref|>text<|/ref|><|det|>[[118, 788, 879, 836]]<|/det|>
+Reply: As the reviewer's concerns, we realized the part named 'Convergent and divergent evolution of NLRs and NHR' was not well matched with this study. Thus, the related part is removed from the revised manuscript (in the discussion section).  
+
+<|ref|>text<|/ref|><|det|>[[118, 853, 879, 886]]<|/det|>
+Q9. 280- 282 - Both convergently and divergently evolved NLRs could mediate NHR? The implication is that any NLR could mediate NHR  
+
+<|ref|>text<|/ref|><|det|>[[118, 887, 879, 903]]<|/det|>
+Reply: As the reviewer's concerns, we realized the part named 'Convergent and divergent
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[156, 100, 878, 133]]<|/det|>
+evolution of NLRs and NHR' was not well matched with this study. Thus, the related part is removed from the revised manuscript (in the discussion section).  
+
+<|ref|>text<|/ref|><|det|>[[118, 149, 880, 214]]<|/det|>
+Q10. 293- 294 - Is there any evidence that convergently evolved NLRs are more likely to have an indirect recognition mode? Or that NLRs cannot convergently evolve direct recognition of an effector? How is this section relevant to the findings in the paper? Direct and indirect recognition is not distinguished elsewhere in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 215, 880, 264]]<|/det|>
+Reply: As the reviewer's concerns, we realized the part named 'Convergent and divergent evolution of NLRs and NHR' was not well matched with this study. Thus, the related part is removed from the revised manuscript (in the discussion section) for the uniformity.  
+
+<|ref|>text<|/ref|><|det|>[[118, 280, 880, 345]]<|/det|>
+Q11. 304/305 - Do you mean suppression of the NRC helper NLR? Could you correlate NRC dependency of the tested sensors with resistance observed? E.g. does Rpi-amr1- mediated recognition of P. parasitica evade suppression due to signaling through additional NRCs compared to R8 or other tested NLRs?  
+
+<|ref|>text<|/ref|><|det|>[[118, 346, 880, 640]]<|/det|>
+Reply: As the reviewer's comment, NRC is one of the possible candidate to be suppressed by pathogen, and similar with the reviewer's concern, we also assumed that different NRC- dependency of tested sensor NLRs could be correlated with the disparities between HR cell death against effectors and resistance against Phytophthora pathogens that we observed in Figure 2. Indeed, Rpi- amr1 (NRC2/3- dependent) cannot conferred resistance against P. capsici while R1 (NRC4- dependent) and R8 (NRC2/3/4- dependent) conferred significant resistance against P. capsici, even though all these three Solanum NLRs were able to recognize and induced cell death against effectors of P. capsici, Therefore, we hypothesized that P. capsici would suppress NRC2/3 but not NRC4 of N. benthamiana. To test this hypothesis, we used nrc4 knockout N. benthamiana. However, the result was negative (expression of R8 were still able to confer resistance against P. capsici even in nrc4 knockout N. benthamiana as shown in the submitted raw data of DLA experiment performed in '220415'using P. capsici). This result indirectly indicates that NbNRC2/3 still properly work during the P. capsici infection. As suggested by the reviewer, if we conduct similar experiments with a more variety of NRC knockout plants, we may be able to find examples that explain the correlation of NRC- dependency and the discrepancies between resistance and HR cell death phenotypes, but unfortunately, with the materials we currently have, it was not achieved.  
+
+<|ref|>text<|/ref|><|det|>[[118, 656, 880, 720]]<|/det|>
+Q12. There may be other reasons that recognition does not translate into resistance like: the sensor NLRs could also be suppressed, there are differences in the expression levels of the effectors between pathogens or due to the difference in the interaction strength between effectors from each pathogen, the effectors are overexpressed in the HR assay.  
+
+<|ref|>text<|/ref|><|det|>[[118, 721, 855, 754]]<|/det|>
+Reply: Agree, we added the several more possible reasons in the discussion part including the reviewer3's concerns in Line 300- 303.  
+
+<|ref|>text<|/ref|><|det|>[[118, 771, 464, 803]]<|/det|>
+Q13. 338 - N. benthamiana is not tobacco Reply: revised in Line 335.  
+
+<|ref|>text<|/ref|><|det|>[[118, 821, 576, 853]]<|/det|>
+Q14. 341- 350 - Include details for P. capsici Reply: added details for P. capsici in Line 343 and 346.  
+
+<|ref|>text<|/ref|><|det|>[[118, 870, 468, 902]]<|/det|>
+Q15. 406 - infestans rather than infestance Reply: revised in Line 406
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 116, 880, 149]]<|/det|>
+Q16. Figure 1 (Minor comment) - It would be much easier to interpret if the species in (a) were not abbreviated, or if abbreviations were indicated in the legend  
+
+<|ref|>text<|/ref|><|det|>[[157, 149, 880, 215]]<|/det|>
+- "The presence or absence of parentheses indicates whether certain motif or domain is required or optional to be classified as each category" - It is not clear what is meant by this. Also, it appears that categories are distinguished by the presence/absence of Sig/WY, rather than the motif or domain in parentheses.  
+
+<|ref|>text<|/ref|><|det|>[[118, 215, 880, 313]]<|/det|>
+Reply: We have revised Figure 1 based on the reviewers' suggestions. Species names were assigned in Figure 1a rather than using abbreviations. The categories in Figure 1b were classified according to the presence or absence of a signal peptide (Sig) and WY domain (WY). Additionally, we have added a note to Figure 1b indicating that detailed domain and motif information regarding homologous effectors is provided in Supplementary Table 3.  
+
+<|ref|>text<|/ref|><|det|>[[118, 330, 880, 363]]<|/det|>
+Q17. Figure 2 (a): Green is not indicated on the scale, is this a pdf formatting issue? Positions without either - /HR show where there are no orthologues identified?  
+
+<|ref|>text<|/ref|><|det|>[[118, 363, 880, 444]]<|/det|>
+Reply: The absence of green part was a formatting issue, corrected. Avr2 homologs of \(P\) capsici were missed from dataset for drawing heat-map. It's also revised (added into the heatmap). Also, we revised the figure legend and design of figures, described detailed information of scales, and added description about white blank spaces that mean no conserved homologs or not cloned, thus not tested.  
+
+<|ref|>text<|/ref|><|det|>[[118, 460, 880, 526]]<|/det|>
+Q18. Was only one orthologue from each species cloned? There are 40 HR assay results shown, but 77 effectors are mentioned in the legend - in the text - "60.87% (42/69) of the tested effectors induced cell death upon co-expressed with their putative corresponding Solanum NLR".  
+
+<|ref|>text<|/ref|><|det|>[[118, 526, 880, 559]]<|/det|>
+Reply: We revised figure and its legend to clarify how to draw this figure, and 69 was correct number, we revised it.  
+
+<|ref|>text<|/ref|><|det|>[[118, 575, 880, 624]]<|/det|>
+Q19. It is not clear how many effectors are represented in this figure, if multiple effectors fit into each category, it would be more informative to indicate how many induce HR with the tested NLR  
+
+<|ref|>text<|/ref|><|det|>[[118, 624, 880, 657]]<|/det|>
+Reply: The number of HR positive / test effectors are presented as fraction for each cases in Figure 2a as the reviewer's concern.  
+
+<|ref|>text<|/ref|><|det|>[[118, 673, 880, 721]]<|/det|>
+Q20. (b) How many lesions were measured for each set? It would be helpful to represent this in a similar way to (c). Why were some combinations not tested - e.g. Rpi-vnt1 and \(P\) . cactorum which is shown as HR in panel (a)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 722, 880, 788]]<|/det|>
+Reply: All dots presented as the reviewer's suggestion (similar with Figure 2c, for the all presented graph in main figures). However, the number of replications are different for each case and if it presented, the figures would become too complicated. Thus, detailed information (numbers) are presented in Supplementary figures 7- 12.  
+
+<|ref|>text<|/ref|><|det|>[[157, 789, 880, 853]]<|/det|>
+Not tested cases: For the cases of \(P\) . palmirova and \(P\) . cactorum, we did not test for several case (NLRs) when the corresponding effectors are not conserved in each pathogen (only R3a was tested even though \(P\) . cactorum did not possess Avr3a homologs in our criteria).  
+
+<|ref|>text<|/ref|><|det|>[[152, 854, 880, 903]]<|/det|>
+However, as the reviewer's concern, for the case of Rpi- vnt1, We have auto- activated cell death issue with p35s:Rpi- vnt1 construct. It induces moderate level of cell death on infiltrated region (weak cell death observed with naked eyes but merely detected using
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[150, 99, 879, 329]]<|/det|>
+FoBI machine – red leaves/black dead tissue images) at 2 dpi, but it kills most of infiltrated leaves at 2.5 dpi of agroinfiltration. While \(P\) . parasitica and \(P\) . capsici fully expand their lesion in \(N\) . benthamiana leaves within 2.0 days after inoculation, \(P\) . palmivora and \(P\) . cactorum take more than 3 days for the measureable lesion size. Thus, we could not test \(Rpi\) - vnt1- mediated resistance against \(P\) . cactorum in \(N\) . benthamiana (also, we could obtain other promising candidates \(R1\) , \(R8\) , and \(Rpi\) - amr1, it did not stimulate us to test \(Rpi\) - vnt1 with different/optimized experimental conditions). For the clarification, we added auto- active phenotypes of \(Rpi\) - vnt1 expressed leaves in Supplementary Figure 7a (weak cell death at 2 dpi) and 9a (severe cell death at 3 dpi), and described about these issues in legends. In addition, for the cell death test using \(Rpi\) - vnt1, we also provided the raw data (file named as 'HR_test_raw_data_scheme') with a proper control (photo taken before \(Rpi\) - vnt1 induced cell death itself, as shown in experiments performed at 230214_P. cactorum, 230309_P. parasitica, and 230406_P. sojae, respectively), and presented control cases of \(Rpi\) - vnt1 in Supplementary Figure 7a and 9a.  
+
+<|ref|>text<|/ref|><|det|>[[118, 345, 780, 363]]<|/det|>
+Q21. Figure 3 (a) Include representative images of all lines indicated in panel (b);  
+
+<|ref|>text<|/ref|><|det|>[[118, 362, 880, 429]]<|/det|>
+Reply: We could not take the pictures (as good as could be presented in the main figure) of representative images of R1 #3 and R8 #11, however we provided whole plant images in Supplementary figure 14 / 15, and newly added Supplementary figure 16, and all the pictures taken using transgenic plants (file named as 'transgenic_plant_raw_data')  
+
+<|ref|>text<|/ref|><|det|>[[118, 443, 880, 494]]<|/det|>
+Q22. Figure 3 (b) The pathogen ( \(P\) . capsici?) is not indicated in the figure or legend, the second R8 transgenic line is not resistant? Consider changing the colours or datapoint style as R8 #11 and Rpi- amr1#2 are indistinguishable. How many plants of each line were tested?  
+
+<|ref|>text<|/ref|><|det|>[[120, 494, 760, 510]]<|/det|>
+Reply: It was \(P\) . capsici, the figure and legend are revised to include \(P\) . capsici.  
+
+<|ref|>text<|/ref|><|det|>[[156, 510, 880, 673]]<|/det|>
+For the concern about the second line (#11) expressing R8, we observed characteristics of 'heterozygous line' of R8- expressing plants (especially #11) because we used T1 plants which could exhibit genetic segregation. The similar patterns were also observed in R1- expressing lines (2\~30% of plants were infected by \(P\) . capsici as wild type plants). And unfortunately, it was worse in the 1st trial (Supplementary figure 14). Thus we performed additional experiments (triplicate) as presented in Supplementary figure 15 (2nd trial) and Supplementary Figure 16 (3rd, newly performed) and observed that R8- expressing lines (#11) able to confer relatively weak resistance (compared to R1- expressing lines) against \(P\) . capsici. Indeed, this phenomenon was similarly observed in transient- expression- based experiments as shown in Figure 2c (R1 > R8, in terms of intensity of resistance).  
+
+<|ref|>text<|/ref|><|det|>[[156, 673, 877, 706]]<|/det|>
+In addition, we changed color of graph in Figure 3b, and the numbers of tested plants are presented in Figure 3a and Supplementary figure 14\~16.  
+
+<|ref|>text<|/ref|><|det|>[[118, 722, 880, 770]]<|/det|>
+Q23. (c/d/e/f) WT lesion size plots are coloured the same as R1/R8/Rpi- amr1 lines, please use a different colour to indicate the WT control. It is strange that number of inoculation sites varies between the pictures, it should be consistent between WT and transgenic lines  
+
+<|ref|>text<|/ref|><|det|>[[118, 770, 879, 820]]<|/det|>
+Reply: We changed colors, re- design, and revised legend of graph for the clear distinguish between R1 / R8 / Rpi- amr1 lines as the reviewer's concern. We also replaced images and synchronized all the numbers of inoculation sites of photos in Figure 3c\~f).  
+
+<|ref|>text<|/ref|><|det|>[[118, 836, 733, 853]]<|/det|>
+Q24. Figure 4: Similar to line 121, functionally conserved is an assumption.  
+
+<|ref|>text<|/ref|><|det|>[[118, 853, 880, 886]]<|/det|>
+Reply: We removed 'functionally conserved' as the reviewer's concern. In Line 127- 128, and also from the whole manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 98, 880, 149]]<|/det|>
+Q25. Figure 5d: Consider rephrasing 'number of functional NLRs' to something closer to 'compatible R-gene and avr-gene pairs'. Also evolutionary distance could be changed to 'coevolving to non-adapted"  
+
+<|ref|>text<|/ref|><|det|>[[118, 149, 880, 200]]<|/det|>
+Reply: Revised as reviewer's concerns. As [Number of functional NLRs => Compatible NLR/effector paris] [Evolutionary distance => Divergent time from adapted pathogen (A) of given plant (a)]  
+
+<|ref|>text<|/ref|><|det|>[[118, 230, 880, 265]]<|/det|>
+Above all, we sincerely thanks to the Reviewer 3's suggestions and devotion for revising this manuscript
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[118, 100, 504, 120]]<|/det|>
+## Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 135, 879, 315]]<|/det|>
+The manuscript by Oh et al investigates whether the conserved avirulence effectors across four Phytophthora spp. infecting different plant species were cross- recognised by immune receptors from different Solanaceous plants. They cloned 69 effectors closely related to 12 avirulence effectors into a binary vector and tested for their recognition by nine cognate immune receptors by agroinfiltration using both transient and stable transgenic Nicotiana benthamiana plants expressing these receptors. They found that some immune receptors recognised closely related effectors present in Phytophthora spp. that do not normally infect the plant species harboring those receptors. Additionally, they discovered that three of the immune receptors, tested using transgenic N. benthamiana plants, conferred broad resistance against "non- infecting" Phytophthora species. The authors concluded that these NLRs found in Solanaceous plants contribute to the non- host resistance (NHR) in this plant family.  
+
+<|ref|>text<|/ref|><|det|>[[118, 315, 879, 414]]<|/det|>
+The manuscript is mostly well written and logically laid out. Experiments were performed to excellent standards and statistical analyses were included where applicable. The figures are of high quality (although there were some issues) with enough details for the readers to scrutinize. There are sufficient details in the materials and methods that other researchers can use to reproduce the results if need be. The authors are highly commendable for their thoroughness and robust experiments with meticulous presentation of the results.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 430, 234, 446]]<|/det|>
+## Major issues:  
+
+<|ref|>text<|/ref|><|det|>[[118, 462, 880, 888]]<|/det|>
+Q1. I am convinced that the NLRs the authors have tested have the potential and capability of recognising the effectors from "non-infecting" pathogens and some do confer resistance against these pathogens when tested on a surrogate host such as N. benthamiana. However, my main concern is that the authors have not provided explicit enough answer to the question: Are these NLRs actually involved in NHR? In my opinion, the authors have not provided enough evidence to overcome the burden of proof that these NLRs are involved in NHR (e.g., Would NLR knock out in a host plant be successfully infected by a "non- infecting" pathogen in this experimental set up? Or would the "non- infecting" pathogen become infectious if the effectors that are recognised by non- host plant NLRs were knocked out? Are the NLRs and effectors expressed in the right places at the right time to the right amount? Would there be any NLRs that are involved in NHR against slightly more distant Albugo or Pythium spp?). The authors would be less controversial and more productive if they re- frame their findings along the evolutionary lines, such as evolutionary transitions, regressive evolution, diversification and differential loss of pathogenicity to host ranges etc, instead of tying them to NHR (although they can speculate about this phenomena). The fact that pathogen species as well as the host species used in this manuscript are "closely" related makes defining and resolving NHR difficult. As far as I am concerned, the infection assays conducted in N. benthamiana to show the context of NHR is confusing and not conclusive enough for authors' claims. N. benthamiana itself can be host or non- host to the pathogens they tested depending on circumstances as P. capsici and N. benthamiana might have never encountered each other in their evolutionary time until they were put together by humans. NHR could get more complicated if one considers P. capsici and P. palmivora to have wider host range than other Phytophthora species. Furthermore, the authors still cannot eliminate the possibility that other factors and genes, in addition to the NLRs, might be responsible for NHR in these interactions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 889, 877, 906]]<|/det|>
+Reply: We sincerely thanks to the reviewer's suggestion for the better presentation of our work
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[158, 99, 879, 148]]<|/det|>
+and pointing out the weak point of the manuscript. We completely agree with the reviewer's suggestion that 'to re- frame the findings along the evolutionary lines instead of tying them to NHR' would be less controversial and more productive.  
+
+<|ref|>text<|/ref|><|det|>[[158, 150, 880, 264]]<|/det|>
+Therefore, we tried to tone down and reduced statements about nonhost resistance (started from the title of the manuscript, in total, the number of 'NHR' in the manuscript decreased from \(22 \Rightarrow 11\) ; and 'non- adapted' decreased from \(14 \Rightarrow 5\) ), and those parts remain only in discussion & very first part of introduction section. As shown in the presented in the replaced title, we tried to state our result as the broad- spectrum resistance by recognizing conserved effectors of Phytophthora species but stated this kind of mechanism 'would' have potential to contribute to the nonhost resistance.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 281, 285, 296]]<|/det|>
+## Some minor issues:  
+
+<|ref|>text<|/ref|><|det|>[[118, 312, 879, 360]]<|/det|>
+Q1. The authors would benefit from reconciling some concepts with Panstruga and Moscou (2020) https://doi.org/10.1094/MPMI- 06- 20- 0161- CR in their introduction and discussions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 362, 879, 427]]<|/det|>
+Reply: We newly cited (Panstruga and Moscou et al., MPMI. 2020) and tried to adopt their concept and description into this manuscript (in Line 47- 60 of introduction part), and we also revised related part in discussion section (in Line 261- 272 of discussion) for the clarification/reframing of our previous description.  
+
+<|ref|>text<|/ref|><|det|>[[118, 444, 879, 475]]<|/det|>
+Q2. Perhaps discussions could be streamlined. It almost reads like a review article as it was written.  
+
+<|ref|>text<|/ref|><|det|>[[118, 477, 879, 525]]<|/det|>
+Reply: We shortened the discussion section from \(1,219 \Rightarrow 996\) words by removing 'Convergent and divergent evolution of NLRs and NHR' part and revised most other parts to make it streamlined and reduce description about NHR or non- adapted pathogens.  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 543, 273, 558]]<|/det|>
+## Some corrections:  
+
+<|ref|>text<|/ref|><|det|>[[118, 574, 879, 608]]<|/det|>
+Q1. Line 32: "Solanaceae palnts" should be either "Solanaceous plants" or "Solanaceae family of plants"  
+
+<|ref|>text<|/ref|><|det|>[[118, 609, 610, 625]]<|/det|>
+Reply: revised in Line 33, and also in the whole manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 641, 540, 675]]<|/det|>
+Q2. Line 325: "Similar with" should be "Similar to" Reply: revised in Line 323.  
+
+<|ref|>text<|/ref|><|det|>[[118, 690, 817, 723]]<|/det|>
+Q3. Line 333: "more identification of NLRs" should be "identification of more NLRs" Reply: revised in Line 330.  
+
+<|ref|>text<|/ref|><|det|>[[118, 740, 879, 772]]<|/det|>
+Q4. Line 338: Please delete "Tobacco" and parenthesis around \(N\) . benthamiana as it is not Tabacco plant  
+
+<|ref|>text<|/ref|><|det|>[[118, 773, 346, 788]]<|/det|>
+Reply: revised in Line 335.  
+
+<|ref|>text<|/ref|><|det|>[[118, 805, 644, 838]]<|/det|>
+Q5. Line 342: "rye agar plate" should be "rye sucrose agar plate" Reply: revised in Line 341.  
+
+<|ref|>text<|/ref|><|det|>[[118, 854, 413, 887]]<|/det|>
+Q6. Line 346: Please define "TDW" Reply: revised in Line 345.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[117, 100, 650, 134]]<|/det|>
+Q7. Line 349: "after incubated" should be "after being incubated" Reply: revised in Line 349.  
+
+<|ref|>text<|/ref|><|det|>[[117, 149, 704, 166]]<|/det|>
+Q8. Line 415: Please state the cutoff values for bit-score and pTM score  
+
+<|ref|>text<|/ref|><|det|>[[117, 166, 880, 215]]<|/det|>
+Reply: We provided cut- off in Line 415- 416 as the reviewer's concerns and we also provided initially selected effectors for cloning in submitted raw data excel file named 'initial_sets_effector_numbering'  
+
+<|ref|>text<|/ref|><|det|>[[117, 230, 880, 280]]<|/det|>
+Q9. Line 429: GV3101- it should be precisely written as GV3101 (pMP90) if this is indeed used (refer to http://www.bio.net/bionet/mm/arab- gen/2016- January/013588. html). Otherwise, GV3101 alone cannot transform plants as it is cured of Ti plasmid.  
+
+<|ref|>text<|/ref|><|det|>[[117, 280, 585, 296]]<|/det|>
+Reply: revised as commented, thanks for the notification.  
+
+<|ref|>text<|/ref|><|det|>[[117, 312, 639, 345]]<|/det|>
+Q10. Line 432: "after adjusted" should be "after being adjusted" Reply: revised in Line 433.  
+
+<|ref|>text<|/ref|><|det|>[[117, 361, 880, 394]]<|/det|>
+Q11. Line 433: "were used for test" should be either "were used for testing" or "were used for agroinfiltration"  
+
+<|ref|>text<|/ref|><|det|>[[117, 395, 345, 411]]<|/det|>
+Reply: revised in Line 435.  
+
+<|ref|>text<|/ref|><|det|>[[117, 427, 880, 460]]<|/det|>
+Q12. Line 435: "1\~4 scale indexing method" - please provide a reference, or describe the scale in details  
+
+<|ref|>text<|/ref|><|det|>[[117, 460, 880, 494]]<|/det|>
+Reply: it was our mistake, we only scored HR cell death phenotypes with positive/negative in this study, it revised as presented in Supplementary table 6, thus related part is removed.  
+
+<|ref|>text<|/ref|><|det|>[[117, 509, 880, 542]]<|/det|>
+Q13. Line 456: "For the root infection assay" - please provide the soil type used, and whether it was sterilized  
+
+<|ref|>text<|/ref|><|det|>[[117, 543, 533, 558]]<|/det|>
+Reply: details about soil is added in Line 464- 466.  
+
+<|ref|>text<|/ref|><|det|>[[117, 575, 880, 623]]<|/det|>
+Q14. Line 459: "Phenotypes were scored" - please describe specific phenotypes being scored Reply: we add descriptions about how we determined wilt/dead plants in material method in Line 462- 463.  
+
+<|ref|>text<|/ref|><|det|>[[117, 640, 880, 673]]<|/det|>
+Q15. Figure 2a - Color scale bar next to the heat map - gradient application is wrong, e.g. 300 and 0 will have the same color according to the scale.  
+
+<|ref|>text<|/ref|><|det|>[[117, 673, 876, 690]]<|/det|>
+Reply: It was an error occurred during file converting in submission procedure, we corrected.  
+
+<|ref|>text<|/ref|><|det|>[[117, 705, 880, 738]]<|/det|>
+Q16. Figure 2b - White bar on the leaves - what does this represent? A scale bar? This should be described in the legend.  
+
+<|ref|>text<|/ref|><|det|>[[117, 739, 289, 755]]<|/det|>
+Reply: It's removed.  
+
+<|ref|>text<|/ref|><|det|>[[117, 771, 880, 820]]<|/det|>
+Q17. Figure 5a - does not capture the concept described in the legend very well. Perhaps it needs to be re- considered, for example, effectors and NLRs are missing. Figure 1 in their reference #1 seems to better represent the concept.  
+
+<|ref|>text<|/ref|><|det|>[[117, 821, 880, 853]]<|/det|>
+Reply: We added NLR and effectors in the picture, and revised figure legend and related part of the manuscript in Line 261- 272.  
+
+<|ref|>text<|/ref|><|det|>[[117, 869, 880, 903]]<|/det|>
+Q18. Figure 5b - what does each tick on the X-axis represent? Number of effectors? Different effector groups? This should be included in the graph. "Effectors" is not sufficient to
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 100, 863, 135]]<|/det|>
+understand the graph.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 303, 107]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 127, 392, 144]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 163, 826, 198]]<|/det|>
+In the revised manuscript, the authors have addressed most of my concerns but needs a bit more improvement as mentioned below.  
+
+<|ref|>text<|/ref|><|det|>[[116, 218, 300, 234]]<|/det|>
+Include NHR in keywords  
+
+<|ref|>text<|/ref|><|det|>[[116, 254, 577, 271]]<|/det|>
+In Fig 2a legend, please mention blank space means not tested.  
+
+<|ref|>text<|/ref|><|det|>[[116, 290, 864, 344]]<|/det|>
+In Fig 2b, regarding the author's response to my concern about the lesion size of P. palmirova infection upon expression of R8 significantly increased instead of decreasing, the authors need to discuss this in the discussion section.  
+
+<|ref|>text<|/ref|><|det|>[[116, 364, 875, 417]]<|/det|>
+Also in fig 2b, please explain in the legend that the dots are data points. The dots look more like asterisk. It will be good to distinguish this form the asterisk. Can actual dots be used instead of asterisk? Maybe even a different colored dots.  
+
+<|ref|>text<|/ref|><|det|>[[116, 437, 830, 472]]<|/det|>
+Regarding washing media in line 453, It will be useful to clarify what is BA in liquid MS media with cefotaxime.  
+
+<|ref|>text<|/ref|><|det|>[[116, 528, 392, 545]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 564, 875, 674]]<|/det|>
+The manuscript by Oh et al is a resubmission by the authors with revisions. The authors have addressed majority of my concerns and I do not have any other issues except the title which could be re- worded so that it conveys the message more clearly, perhaps along the line of "Conserved effector families renders Phytophthora species vulnerable to recognition by Nucleotide- binding leucine- rich repeat receptors in nonhost plants" or something similar. I appreciate the authors' openness and willingness to converse with the reviewers.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 90, 366, 106]]<|/det|>
+## REPLY to REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[113, 118, 820, 135]]<|/det|>
+Above all, thanks to the anonymous reviewers for their invaluable comments on our manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[115, 146, 857, 181]]<|/det|>
+The new version of manuscript has been revised according to the reviewer's comments and replies to the reviewer's comments are followed.  
+
+<|ref|>text<|/ref|><|det|>[[116, 193, 392, 210]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 222, 827, 256]]<|/det|>
+In the revised manuscript, the authors have addressed most of my concerns but needs a bit more improvement as mentioned below.  
+
+<|ref|>text<|/ref|><|det|>[[115, 268, 325, 285]]<|/det|>
+S1: Include NHR in keywords  
+
+<|ref|>text<|/ref|><|det|>[[115, 298, 157, 313]]<|/det|>
+Done  
+
+<|ref|>text<|/ref|><|det|>[[115, 325, 603, 342]]<|/det|>
+S2: In Fig 2a legend, please mention blank space means not tested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 354, 283, 370]]<|/det|>
+Adjusted as suggested.  
+
+<|ref|>text<|/ref|><|det|>[[115, 382, 880, 435]]<|/det|>
+S3: In Fig 2b, regarding the author's response to my concern about the lesion size of P. palmiyora infection upon expression of R8 significantly increased instead of decreasing, the authors need to discuss this in the discussion section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 447, 527, 464]]<|/det|>
+Added description in Line 292- 297 of discussion session.  
+
+<|ref|>text<|/ref|><|det|>[[115, 475, 874, 529]]<|/det|>
+S4: Also in fig 2b, please explain in the legend that the dots are data points. The dots look more like asterisk. It will be good to distinguish this form the asterisk. Can actual dots be used instead of asterisk? Maybe even a different colored dot.  
+
+<|ref|>text<|/ref|><|det|>[[115, 540, 703, 558]]<|/det|>
+We increased font size of asterisk for the clear distinguishment from data points.  
+
+<|ref|>text<|/ref|><|det|>[[115, 569, 857, 604]]<|/det|>
+S5: Regarding washing media in line 453, It will be useful to clarify what is BA in liquid MS media with cefotaxime.  
+
+<|ref|>text<|/ref|><|det|>[[115, 615, 492, 632]]<|/det|>
+We added full description as (BA \(= >\) benzyladenine)  
+
+<|ref|>text<|/ref|><|det|>[[118, 644, 395, 661]]<|/det|>
+Reviewer #4 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 672, 660, 689]]<|/det|>
+The manuscript by Oh et al is a resubmission by the authors with revisions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 700, 870, 791]]<|/det|>
+The authors have addressed majority of my concerns and I do not have any other issues except the title which could be re- worded so that it conveys the message more clearly, perhaps along the line of "Conserved effector families renders Phytophthora species vulnerable to recognition by Nucleotide- binding leucine- rich repeat receptors in nonhost plants" or something similar. I appreciate the authors' openness and willingness to converse with the reviewers.  
+
+<|ref|>text<|/ref|><|det|>[[115, 802, 857, 837]]<|/det|>
+We decided below sentence as a New Title according to the reviewer's suggestion and editorial guide lines (less than 16 words)  
+
+<|ref|>text<|/ref|><|det|>[[115, 848, 872, 884]]<|/det|>
+'Conserved effector families render Phytophthora species vulnerable to recognition by NLR receptors in nonhost plants
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__43a438ad55c12c7c430994081e89ce1899b856e6563e09e4569576d8deeaaae8/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__43a438ad55c12c7c430994081e89ce1899b856e6563e09e4569576d8deeaaae8/supplementary_0_Peer Review File__43a438ad55c12c7c430994081e89ce1899b856e6563e09e4569576d8deeaaae8.mmd

@@ -0,0 +1,358 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Longitudinal functional imaging of VIP interneurons reveals sup- population specific effects of stroke that can be rescued with chemogenetic therapy  
+
+![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 --->
+
+## Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+Motaharinia et al. explored whether chemogenetic modulation of VIP+ interneurons in the somatosensory cortex of mice can be beneficial for stroke recovery. The authors provide very interesting data with longitudinal imaging of VIP+ interneurons before and several weeks after local stroke induction. The major finding is that driving VIP+ cells chemogenetically can rescue cortical excitability after stroke and fasten functional recovery. The data is of highest quality and were obtained in carefully designed experiments. I only have several minor comments/questions to clarify some findings.  
+
+1. The manuscript gives an impression that the major role of neocortical VIP+ cells is disinhibitory; however, this may not be the case. VIP+ neuronal population is highly heterogeneous, with up to \(80\%\) of VIP+ contacts formed onto GABA-negative structures (Zhou et al., 2017). This heterogeneity may result in highly variable patterns of activity observed by the authors in VIP+ cells, and needs to be carefully discussed. For example, across cortical areas, VIP+/CCK+ basket cells contact PVRs, whereas VIP+/CR+ cells contact preferentially interneurons (Guet-McCreight et al., 2020). Importantly, these two VIP+ subtypes have different intrinsic excitability, with VIP+/CR+ cells showing much higher input resistance. Accordingly, the disinhibitory VIP+/CR+ cells may represent the highly active group in this study and a preferential target for post-stroke therapy. It seems the authors are able to find VIP+ cells imaged in vivo for post-mortem IHC identification (Fig. 3B). It would be therefore possible to check whether the highly active group is expressing CR.  
+
+2. Was there any difference in evoked field response for 0.3 vs. \(0.5 \mathrm{mg / kg}\) of CNO? Why not using the same dose in all experiments? Ideally, one would use the lowest efficient dose throughout the entire study.  
+
+3. Why the therapy had started 4 days after stroke? Would a faster intervention be more beneficial? Have the authors tried different time intervals after the stroke, like 1, 2, 3 days?  
+
+4. Were there any acute affects of CNO treatment?  
+
+5. Was the beneficial effect of chemogenetic treatment persistent? Or have the authors evaluated the state of animals at a longer scale - 3-4 weeks (up to a month) after therapy cessation?  
+
+6. It is exciting to see that a single dose of CNO, with VIP activation time course of about 2 h (max up to 10 h), produces such a long-lasting (at least 23 h) effect on cortical excitability. Can you discuss a bit more the possible mechanisms of this plasticity?  
+
+7. In Discussion, the authors propose that chemogenetic manipulation can represent a relatively non-invasive way for boosting excitability. I would disagree with this point, as it requires stereotaxic surgery to deliver the viral vectors expressing the DREADD. The obvious advantage of this method is, however, its "targetability" or a possibility of focal manipulations with a local neuronal circuits in peri-infarct area.  
+
+8. Given that male and female mice were used in the study, were there any sex-specific differences in VIP+ cell activity or other reported phenomena?  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+In this manuscript, Craig Brown and colleagues have reported that chemogenetic stimulation of vasoactive intestinal peptide (VIP) interneurons enhances weakened sensory responses in the peri-infarct sensorimotor cortex of mice subjected to photothrombotic stroke and promotes the recovery of sensorimotor function. It is novel that sensory responses were disrupted by stroke
+
+<--- Page Split --->
+
+mainly within a population of highly active VIP neurons. The finding that chemogenetic stimulation accelerates the recovery of sensory responses within these highly active VIP neurons and without recruitment of neurons not previously involved in sensory responses is also intriguing. The manuscript is well written; experiments were involving a battery of imaging, electrophysiological and behavioral approaches that require significant expertise. The findings of this paper have clinical significance and support the notion that stroke recovery could be accelerated by enhancing cortical excitability through dis- inhibitory VIP interneurons. Statistical analyses are appropriate, and the level of details provided would be sufficient for other researchers to reproduce the work. The findings will be of interest to the others in the field and the broader readership. I am highly enthusiastic about this manuscript and support its publication.  
+
+Addressing several points could further improve this excellent manuscript.  
+
+1. Surprisingly, the authors did not come across dying neurons in the peri-infarct cortex during longitudinal imaging. It seems that all VIP interneurons imaged before stroke persisted for the next four weeks of 2P imaging. These findings should be discussed.  
+
+2. The peri-infarct zone in the photothrombotic stroke is relatively narrow. All 2P imaging studies were conducted 400 micrometers from the infarct border. Laser speckle images were collected. What percent of the pre-stroke blood flow was remaining in the area of 2P imaging? That could be estimated from the laser speckle images. Were these 2P images collected in the hypo-perfused cortex?  
+
+3. The infarct size is large, and spontaneous spreading depolarizations often accompany such large strokes. However, the authors did not record any large increases in the calcium signals associated with spreading depolarizations. It is not surprising since spontaneous changes in GCaMP6s fluorescence were recorded only for periods of up to 75 s after stroke, and there were no longitudinal electrophysiological recordings. It might be a missed opportunity because one possibility is that spreading depolarizations in the penumbra were the events that affected ("wreak havoc on") the predictability (fidelity) of neuronal responses to sensory stimulation (Fig. 6). Certainly, experiments addressing this possibility are beyond the experiments of this study, but such a possibility could be discussed.  
+
+1. I like the discussion that "stroke related plasticity usually involves the same circuits"(P.14L31). It might be helpful to argue that even during more severe brain injury inflicted by global ischemia, the postsynaptic dendritic membranes remain attached to axonal boutons, providing a structural basis for the recovery of the same circuits (doi.org/10.1093/cercor/bhaa134).  
+
+## Minor points:  
+
+1. What was the rationale to start CNO treatment on day four but not earlier? Also, the additional explanation on P.15 L9 "(ie. Days 4 to 6, Monday through Wednesday, before imaging on day 7)" seems unnecessary.  
+
+2. P21.L23. "... time to peak amplitude and half-width (ie. duration) of forelimb-evoked signals in the first 150ms after stimulation were measured with Clampfit 9.0 software (Molecular Devices)." Does it belong to another section, such as "Recording sensory evoked cortical field potentials"?  
+
+Sergei A. Kirov, PhD  
+
+Reviewer #3:  
+
+Remarks to the Author:  
+
+In this study by Motaharinia et al the authors report that activating VIP interneurons with Gq DREADDs after stroke to the forelimb (FL) region of somatosensory cortex can restore cortical responses (to FL stimulation) to pre- stroke levels and enhance functional recovery. This is an important study with robust findings that add significantly to our understanding of stroke recovery. It is the only paper to my knowledge that has reported on the activity of VIP neurons in the context of stroke recovery. It is also an elegant study that combines two- photon calcium imaging in vivo, mouse behavior, and DREADDs. These experiments are hard! Finally, the finding that VIP neurons could be a target for restoring circuit function and ameliorating behavioral deficits after stroke is very exciting. I was impressed by the high experimental rigor: they use CNO- only (no hM3Dq) controls and they use appropriate stats. The blood flow control to rule our DREADD effects is great too. The behavior data shows definitive results with internal replication in 2 cohorts. The
+
+<--- Page Split --->
+
+longitudinal imaging of VIP neurons over \(>4\) weeks is particularly impressive because they track the same neurons over time. I felt the paper was well written, the figures are easy to understand, and they provide appropriate references (for example of papers describing the known role of VIP cells in disinhibition of pyramidal cells). Overall, I feel the significance of this paper is high and I am enthusiastic about its publication. It is refreshing to see the use of cutting- edge tools to investigate functional circuit changes at the single cell level after stroke. Here is a list of comments/suggestions I hope the authors can address  
+
+- I guess they never show that the excitability/firing of VIP neurons in control mice is enhanced by \(\mathrm{Gq + CNO}\) . They only show indirectly with LFP that stimulation of FL elicits greater responses (from Pyr cells). I wonder if they ever saw differences with calcium imaging in VIP neurons before and 30 min after CNO.  
+
+- I was surprised that a single injection of CNO (which has such a short half-life) only 5 days a week was enough to rescue behavior? How do the authors interpret such a profound effect on the network (and behavior!)? Also, do they think the circuit is permanently restored such that, had they looked a few days after stopping CNO injections, the rescue might have persisted?  
+
+- They choose to perform photothrombotic strokes that do not completely destroy S1FL (eg Fig 1a). Presumably this is to make sure they still can find FL-responsive VIP neurons in peri-infarct cortex. The authors should make it more explicit in the text that they performed sub-total strokes, because it is likely that they would have never found any FL-responsive VIP neurons in peri-infarct cortex had they completely destroyed S1FL.  
+
+- Fig 1c shows that stroke strongly reduces FL stim evoked responses compared to sham. In Fig 1d peak amplitudes are "normalized to baseline", which means right before CNO injection. Could they show a comparison of these baseline responses (average of all the mice for stroke vs sham) to see how strongly stroke affects the FL stim evoked response (beyond the representative trace in Fig 1)  
+
+- In Fig 1d, what are the post-hoc individual p values for the effect of CNO in hM3Dq mice that did or did not receive a stroke?  
+
+- I like Suppl Fig 1a better than Fig. 1e. They might consider a swap (optional) 
+- In suppl Fig 2 please list the number of mice in each group; I don't think a t-test is appropriate here  
+
+- Fig 2d: why not show the average DF/F for all ROIs combined (1-6)? It seems like the differences between hM3Dq and control mice were not significant (otherwise all the data would be presented in the same figure panel, like a bar graph of the peak response). The text says "forepaw evoked depolarizations in peri-infarct cortex were SIGNIFICANTLY larger in amplitude ... (Fig. 2b-d)" but there are no stats provided in the text or in the figure legend. It's fine if it's only significant for some of the ROIs (e.g., 4 & 5), but this could have meaning too based on their location relative to the infarct.  
+
+- In Fig 2b-f, what is the control? Is it hM3Dq but not CNO, or is it GFP + CNO alone? Or are they combined?  
+
+- Fig 3a - is the earliest time point -6 weeks or -3 weeks? They should say in the legend what the green/blue contours of maps represent (presumably it's the 75% threshold described in page 19, line 19).  
+
+- Page 8, lines 21-24: they should probably show these data as part of suppl fig 3  
+
+- Figs 3 & 4: regarding the concern about toxicity of prolonged GCaMP expression, I agree it's reassuring that the gray traces in Fig 4e (sham) are stable. But why is the +4 wk time point missing for sham controls in Fig 4e-f?  
+
+- Fig 4b: are these example traces from a vehicle stroke mouse?  
+
+- Fig 4d: can they use different colors (or symbols) for the 3 different types of mice in these scatter plots (sham control, stroke veh and stroke Gq)?  
+
+- Fig 4e-f: It's fine to show the normalized data but can they also show the raw data for % of responsive neurons and peak amplitudes for the three groups at baseline (text has results of ANOVA but not actual data)? I ask because there does not appear to be a sustained increase in the % of VIP neurons that respond to FL stimulation after stroke, which means that no new neurons are recruited to respond to FL stim after stroke (DREADDs only maintain the original pool).  
+
+- The effects on VIP responsivity are mainly seen at 1-wk post-stroke, but behavioral effects are seen 2-7 weeks after stroke...the discussion touches on this but more could said to explain this difference.  
+
+- Fig 5b: Did the relative proportion of each of the 3 subtypes change over time (from baseline) in
+
+<--- Page Split --->
+
+sham controls and stroke animals? There should be a graph to represent that. Also, shouldn't there be a Chi- square test for all the comparisons and then follow- up 2x2 chi- sq for individual comparisons with post- hoc correction. A better description of the Chi- sq methods would be useful - Fig 6a: the issue of how predictable (or should it be 'reliable'?) neurons are seems important, but this visual representation is a bit hard to follow. Another way to show this would be to plot, for each cell, the % of stimulations it responds to at baseline and over time after stroke.
+
+<--- Page Split --->
+
+## RESPONSE TO REVIEWER COMMENTS  
+
+## Dear editors and reviewers:  
+
+We would like to thank you for the opportunity to revise our manuscript titled "Longitudinal functional imaging of VIP interneurons reveals sup- population specific effects of stroke that can be rescued with chemogenetic therapy" for publication in Nature Communications. The reviewers have provided a number of important and incisive suggestions for improving the paper. As requested by the editors and reviewers, we have conducted several new and challenging experiments to support the conclusions in the paper. These experiments include: a) electrophysiological DC recordings and in vivo GCamP6s imaging of stroke related spreading depolarizations (see Reviewer 2 comments), b) in vivo imaging of VIP sensory responses followed by post- mortem re- identification of VIP neurons that co- localize with calretinin (see Reviewer 1 comments) and c) imaging the acute effects of chemogenetic stimulation on VIP neuron responses (see Reviewer 3 comments). In addition, we have included several new data analyses at the request of the reviewers. Please note that all manuscript revisions are highlighted in red text. As a result of these recommendations and revisions, we have strengthened the evidence supporting the paper's primary findings and refined our discussion of these results.  
+
+## Reviewer #1 (Remarks to the Author):  
+
+Motaharinia et al. explored whether chemogenetic modulation of VIP+ interneurons in the somatosensory cortex of mice can be beneficial for stroke recovery. The authors provide very interesting data with longitudinal imaging of VIP+ interneurons before and several weeks after local stroke induction. The major finding is that driving VIP+ cells chemogenetically can rescue cortical excitability after stroke and fasten functional recovery. The data is of highest quality and were obtained in carefully designed experiments. I only have several minor comments/questions to clarify some findings.  
+
+RESPONSE: We thank the reviewer for their positive appraisal of our study.  
+
+1. The manuscript gives an impression that the major role of neocortical VIP+ cells is disinhibitory; however, this may not be the case. VIP+ neuronal population is highly heterogeneous, with up to \(80\%\) of VIP+ contacts formed onto GABA negative structures (Zhou et al., 2017). This heterogeneity may result in highly variable patterns of activity observed by the authors in VIP+ cells, and needs to be carefully discussed. For example, across cortical areas, VIP+/CCK+ basket cells contact PYRs, whereas VIP+/CR+ cells contact preferentially interneurons (Guet-McCreight et al., 2020). Importantly, these two VIP+ subtypes have different intrinsic excitability, with VIP+/CR+ cells showing much higher input resistance. Accordingly, the disinhibitory VIP+/CR+ cells may represent the highly active group in this study and a preferential target for post-stroke therapy. It seems the authors are able to find VIP+ cells imaged in vivo for post-mortem IHC identification (Fig. 3B). It would be therefore possible to check whether the highly active group is expressing CR.
+
+<--- Page Split --->
+
+RESPONSE: This is an interesting point, and it would certainly be fascinating to further define the neurochemical phenotype of the highly active VIP neurons. While we fully agree with the reviewer's comment about functional/neurochemical heterogeneity in the VIP population, our claim of a disinhibitory effect, at least on the network/population level, was based on our field recording data showing an increase in cortical field potential responses following DREADD based stimulation of VIP neurons. The reviewer raises the possibility that the highly active VIP neurons may express Calretinin (CR), which show much higher input resistance. Therefore to address this possibility, we installed cranial windows and longitudinally imaged 10 mice to assess VIP sensory responses, then attempted to find each individual neuron in post- mortem horizontal sections immunostained for Calretinin (CR). Since these are extremely difficult experiments with very low success rates, we could only re- locate the VIP neurons imaged in vivo in 2 mice. In these mice, we found that \(34\%\) VIP neurons expressing GCaMP6s co- localized with CR (see Supp. Fig. 6a,b). Conversely, \(31.5\%\) of CR neurons co- localized with VIP. Since there were few VIP neurons that co- localized with CR to sample from, we graphed the number of responsive trials (out of 8) as function of whether VIP neurons co- localized with CR or not (2 mice: 71 neurons were \(\mathrm{VIP + / CR - }\) and 14 neurons were \(\mathrm{VIP + / CR + }\) ). As shown in Supp. Fig. 6c, the response profile of VIP neurons that co- express calretinin versus those that do not, were not clearly different from one another. Based on these results, we conclude that the highly responsive group of VIP neurons (ie. neurons that respond to 6- 8 trials) are not exclusively the same ones that express CR. However, we agree with the reviewer that future studies probing the response profile of VIP neurons with specific neuropeptides or calcium binding proteins (CR, CCK, ChaT etc) would be informative. We have included this topic in the discussion on page 18 and presented the results of this experiment in a new Supp. Fig. 6.  
+
+2. Was there any difference in evoked field response for \(0.3\mathrm{vs.}0.5\mathrm{mg / kg}\) of CNO? Why not using the same dose in all experiments? Ideally, one would use the lowest efficient dose throughout the entire study.  
+
+RESPONSE: Ideally, yes we would have used the same dose. Since the CNO- DREADD experiments were carried out in 2 separate epochs (2013- 2015 and 2017- present), there was a mis- communication when we re- started the CNO experiments in the second epoch (trainee thought we were using \(0.5\mathrm{mg / kg}\) vs \(0.3\mathrm{mg / kg}\) ). However in both epochs, we tested and validated the CNO- DREADD effects with electrophysiology to ensure they would work as expected. We now present our analysis of peak field potential responses when mice were dosed with \(0.3\mathrm{vs.}0.5\mathrm{mg / kg}\) CNO. Importantly we show there was no significant differences between doses and now include this analysis in the results on page 5 "There was no significant difference in peak response amplitude when comparing \(0.3\mathrm{mg / kg}\) vs \(0.5\mathrm{mg / kg}\) doses of CNO (unpaired t- test, \(\mathrm{t}_{(3)} = 1.21\) , \(\mathrm{p} = 0.31\) )."  
+
+3. Why the therapy had started 4 days after stroke? Would a faster intervention be more beneficial? Have the authors tried different time intervals after the stroke, like 1, 2, 3 days?  
+
+RESPONSE: In the first 72 hours after stroke, the brain is in quite a labile and precarious state due to edema, spreading depolarizations, excitotoxicity and other potentially damaging events that can further expand the area of ischemic damage. Since our goals were to: a) focus on stroke recovery strategies rather than neuroprotection (ie. enhance function of what brain tissue remains vs. preventing ischemic cell death), and b) image the same "peri- infarct" neurons before and after stroke; we did not want to further risk aggravating ischemic cell death by chemogenetically augmenting excitability within the first 3 days after stroke. We have now included a statement in the results clarifying our rationale (page 6). Lastly, we
+
+<--- Page Split --->
+
+do agree that an interesting future study would be to test out different time points for initiating therapeutic intervention (Note: this actually part of a recent grant proposal), especially if we could extent the therapeutic window of opportunity.  
+
+4. Were there any acute affects of CNO treatment?  
+
+RESPONSE: As shown in Figure 1e and Supp. Fig. 1, the effects of CNO treatment on cortical excitability peak within the first 60-90min after injection and then, according to published literature (Alexander et al., 2009, Neuron), start to decline thereafter.  
+
+5. Was the beneficial effect of chemogenetic treatment persistent? Or have the authors evaluated the state of animals at a longer scale - 3-4 weeks (up to a month) after therapy cessation?  
+
+RESPONSE: We too were curious about this. Our behavioural data suggest that the beneficial effects did persist given that we stopped treatment at week 6, and then even one week later, the mice that received chemogenetic treatment were still significantly better than controls on the horizontal ladder test (see Fig. 2a). Furthermore, when examining forelimb evoked cortical responses with VSD imaging at 10 weeks recovery ( \(\sim 4\) weeks after treatment was stopped), the mice that received chemogenetic treatment showed significantly increased sensory-evoked cortical responses compared to controls (see Fig. 2b- e).  
+
+6. It is exciting to see that a single dose of CNO, with VIP activation time course of about \(2\mathrm{h}\) (max up to \(10\mathrm{h}\) ), produces such a long-lasting (at least \(23\mathrm{h}\) ) effect on cortical excitability. Can you discuss a bit more the possible mechanisms of this plasticity?  
+
+RESPONSE: This is a good point. The fact that the effects of chemogenetic therapy persist long after treatment has ceased suggest that up- regulating the excitability of VIP neurons has led to more permanent changes within stroke recovered circuits, presumably those downstream of VIP neurons (eg. layer 2/3 pyramidal neurons and other interneurons). Precisely how these lasting downstream changes in circuitry are accomplished remains an open question. Previous work from our lab and others has shown that promoting the restoration of cortical excitability in somatosensory cortex after stroke (via chemo- or optogenetic) is associated axonal sprouting (Wahl et al., 2017, Nature Comm) as well as the proliferation and stabilization of thalamocortical axonal boutons (Tennant et al., 2017, Nature Comm). In addition, therapies that promote the return of cortical excitability after stroke lead to changes in growth and plasticity associated gene expression (CREB, BDNF, NGF; see Cheng et al., 2014, PNAS; Caracciolo et al., 2018, Nat Comm). Although our study provides much needed insights into the effects of stroke at a cellular level, future stroke studies could dissect the contribution of other interneuron populations in stroke recovery. Furthermore, we think the stroke field could really benefit from future slice electrophysiology studies that address circuit specific changes in intrinsic excitability, spiking patterns, pre and post- synaptic forms of plasticity (LTP, LTD) using quantal analysis, paired pulse ratios, mini- analysis etc. We now include a more fulsome discussion on this topic on page 14 of the discussion.  
+
+7. In Discussion, the authors propose that chemogenetic manipulation can represent a relatively non-invasive way for boosting excitability. I would disagree with this point, as it requires stereotaxic surgery to deliver the viral vectors expressing the DREADD. The obvious advantage of this method is, however, its "targetability" or a possibility of focal manipulations with a local neuronal circuits in peri-infarct area.  
+
+RESPONSE: we agree and have removed the phrase about being "non-invasive"
+
+<--- Page Split --->
+
+8. Given that male and female mice were used in the study, were there any sex-specific differences in VIP+ cell activity or other reported phenomena?  
+
+RESPONSE: For our study, we used male VIP mice for the electrophysiology, VSD and calcium imaging experiments, as well as the first cohort of behavioural studies testing the chemogenetic therapy (see cohort 1, K.G. 2014 in Figure 2a). However as we (and science in general) have become more cognizant of considering sex in our studies, we used female VIP mice for the second cohort of behavioural studies testing the chemogenetic therapy (see Cohort 2, S.C. 2018 in Figure 2a). If we compare males and females that received the chemogenetic therapy relative to their respective control groups, the benefits of therapy for males and females (ie. comparing difference in \(\%\) correct steps between treated and controls from weeks 2- 7), were quite similar and not significantly different (2- way ANOVA; Main effect of Sex: \(\mathrm{F}_{(1,72)} = 1.31\) , \(\mathrm{p} = 0.26\) ). We have now included this analysis in the results (page 6- 7) and clarified the sex of the mice in the methods section and Figure legend 2.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+In this manuscript, Craig Brown and colleagues have reported that chemogenetic stimulation of vasoactive intestinal peptide (VIP) interneurons enhances weakened sensory responses in the perinfarct sensorimotor cortex of mice subjected to photothrombotic stroke and promotes the recovery of sensorimotor function. It is novel that sensory responses were disrupted by stroke mainly within a population of highly active VIP neurons. The finding that chemogenetic stimulation accelerates the recovery of sensory responses within these highly active VIP neurons and without recruitment of neurons not previously involved in sensory responses is also intriguing. The manuscript is well written; experiments were involving a battery of imaging, electrophysiological and behavioral approaches that require significant expertise. The findings of this paper have clinical significance and support the notion that stroke recovery could be accelerated by enhancing cortical excitability through dis- inhibitory VIP interneurons. Statistical analyses are appropriate, and the level of details provided would be sufficient for other researchers to reproduce the work. The findings will be of interest to the others in the field and the broader readership. I am highly enthusiastic about this manuscript and support its publication.  
+
+RESPONSE: We thank the reviewer for their positive comments on our study.  
+
+Addressing several points could further improve this excellent manuscript.  
+
+1. Surprisingly, the authors did not come across dying neurons in the peri-infarct cortex during longitudinal imaging. It seems that all VIP interneurons imaged before stroke persisted for the next four weeks of 2P imaging. These findings should be discussed.  
+
+RESPONSE: We think this might be an issue of semantics, but it is an important one. Since we could not control the extent of ischemic damage with micron precision, some areas we had imaged before stroke and had hoped would remain after the induction of stroke, were ultimately destroyed. These areas we considered part of the infarct core, not peri- infarct cortex. Thus, when we imaged neurons 7 days after stroke, we only focused on ones that were viable and outside of the infarct core, which we define as "peri- infarct". As mentioned below, the infarct border is quite sharp and therefore ischemic cell death was not still evolving by post- stroke day 7. In this context, it is not surprising that neurons imaged 7 days after
+
+<--- Page Split --->
+
+stroke, would persist for the remaining 4 weeks. We have clarified this point on page 25 in the methods section.  
+
+2. The peri-infarct zone in the photothrombotic stroke is relatively narrow. All 2P imaging studies were conducted 400 micrometers from the infarct border. Laser speckle images were collected. What percent of the pre-stroke blood flow was remaining in the area of 2P imaging? That could be estimated from the laser speckle images. Were these 2P images collected in the hypo-perfused cortex?  
+
+RESPONSE: In the present study, we used laser speckle imaging to define the infarct border starting 7 days after stroke, but not before stroke. Therefore, a comparison of pre vs post-stroke blood flow was not possible. However, there are several published papers including our own (Tennant et al., 2013; Sullender, et al., 2018) that show relatively normal blood flow in peri-infarct cortex (after photothrombotic stroke) by 5- 7 days post-stroke. We now include this information and cited literature on page 25 in the methods section.  
+
+3. The infarct size is large, and spontaneous spreading depolarizations often accompany such large strokes. However, the authors did not record any large increases in the calcium signals associated with spreading depolarizations. It is not surprising since spontaneous changes in GCaMP6s fluorescence were recorded only for periods of up to 75 s after stroke, and there were no longitudinal electrophysiological recordings. It might be a missed opportunity because one possibility is that spreading depolarizations in the penumbra were the events that affected ("wreak havoc on") the predictability (fidelity) of neuronal responses to sensory stimulation (Fig. 6). Certainly, experiments addressing this possibility are beyond the experiments of this study, but such a possibility could be discussed.  
+
+RESPONSE: The idea that ischemic spreading depolarization (SD) could have been occurring 7 days after stroke and thereby influence our findings is an interesting one and something we had not previously considered. Before we describe our new experiments and analyses, we should prevent any confusion by clarifying that all calcium imaging in the original submission was conducted 7 days after stroke, not in the first few minutes to hours after stroke. We apologize for any confusion and have now made this clear in the results and methods section. In order to address the possibility of SDs contaminating our calcium imaging data (collected at 7 days post-stroke), we employed two different, but complimentary approaches (n=6 mice for each experimental approach). First, we used the gold standard approach of recording direct current (DC) potentials in peri-infarct cortex immediately after stroke and 7 days recovery (n=3 mice per time point). Recording immediately after initiating photothrombosis revealed approximately 1- 2 SD's occur within the first 30 minutes after photothrombotic stroke (2 mice exhibited 2 SD each whereas 1 mouse showed just 1 SD; peak amplitude: \(12\pm 1.1\mathrm{mV}\) ; duration: \(161.8\pm 5.2\mathrm{s}\) ; Supp. Fig. 5a), which is consistent with previous studies (Risher et al., JNSCI, 2010). However, we did not observe any SDs in peri-infarct cortex 7 days after stroke (Supp Fig. 5a). To be extra careful, we also included a positive control experiment with 1mM KCl application to prove that if a SD had occurred, we would have detected it. Similarly, by imaging SD induced calcium waves in peri-infarct cortex (which are \(\sim 10\) times larger in amplitude than sensory evoked calcium transients), we typically observed an SD within the first 30min after stroke (see example shown in Supp. Fig. 5b), but never 7 days post-stroke (n=3 mice per time point). Based on these experiments and previous literature, we conclude that it is very unlikely that ongoing SD waves could have affected our calcium imaging experiments and analysis collected 7 days after stroke. We now include these results on page 10 and Supplementary Fig. 5, as well as provided methodological details on page 21 and 25 in the Methods section.
+
+<--- Page Split --->
+
+4. I like the discussion that "stroke related plasticity usually involves the same circuits" (P.14L31). It might be helpful to argue that even during more severe brain injury inflicted by global ischemia, the postsynaptic dendritic membranes remain attached to axonal boutons, providing a structural basis for the recovery of the same circuits (doi.org/10.1093/cercor/bhaa134).  
+
+RESPONSE: We thank the reviewer for this suggestion and agree that if post- synaptic structures retain their pre- synaptic partners, this could provide a structural explanation for the temporary disruption and reactivation of highly active circuits. We now include this relevant paper and explanation in the discussion section on page 16.  
+
+Minor points:  
+
+1. What was the rationale to start CNO treatment on day four but not earlier? Also, the additional explanation on P.15 L9 "(ie. Days 4 to 6, Monday through Wednesday, before imaging on day 7)" seems unnecessary.  
+
+RESPONSE: We have removed the unnecessary information. Since Reviewer 1 asked a similar question about the timing of treatment (query 3), we have copied part of our response here.  
+
+"In the first 72 hours after stroke, the brain is in quite a labile and precarious state due to edema, spreading depolarizations, excitotoxicity and other potentially damaging events that can further expand the area of ischemic damage. Since our goals were to: a) focus on stroke recovery strategies vs. neuroprotection (ie. enhance function of what brain tissue remains vs. preventing ischemic cell death), and b) image the same "peri-infarct" neurons before and after stroke; we did not want to further risk aggravating ischemic cell death by chemogenetically augmenting excitability within the first 3 days after stroke. We have now included a statement in the results clarifying our rationale (page 6)."  
+
+2. P21.L23. "... time to peak amplitude and half-width (ie. duration) of forelimb-evoked signals in the first 150ms after stimulation were measured with Clampfit 9.0 software (Molecular Devices)." Does it belong to another section, such as "Recording sensory evoked cortical field potentials"?  
+
+RESPONSE: This is correct, we analysed VSD signals in this manner.  
+
+## Reviewer #3 (Remarks to the Author):  
+
+In this study by Motaharinia et al the authors report that activating VIP interneurons with Gq DREADDs after stroke to the forelimb (FL) region of somatosensory cortex can restore cortical responses (to FL stimulation) to pre- stroke levels and enhance functional recovery. This is an important study with robust findings that add significantly to our understanding of stroke recovery. It is the only paper to my knowledge that has reported on the activity of VIP neurons in the context of stroke recovery. It is also an elegant study that combines two- photon calcium imaging in vivo, mouse behavior, and DREADDs. These experiments are hard! Finally, the finding that VIP neurons could be a target for restoring circuit function and ameliorating behavioral deficits after stroke is very exciting.  
+
+I was impressed by the high experimental rigor: they use CNO- only (no hM3Dq) controls and they use appropriate stats. The blood flow control to rule our DREADD effects is great too. The behavior data shows definitive results with internal replication in 2 cohorts. The longitudinal imaging of VIP neurons
+
+<--- Page Split --->
+
+over >4 weeks is particularly impressive because they track the same neurons over time. I felt the paper was well written, the figures are easy to understand, and they provide appropriate references (for example of papers describing the known role of VIP cells in disinhibition of pyramidal cells). Overall, I feel the significance of this paper is high and I am enthusiastic about its publication. It is refreshing to see the use of cutting- edge tools to investigate functional circuit changes at the single cell level after stroke. Here is a list of comments/suggestions I hope the authors can address  
+
+RESPONSE: We thank the reviewer for their encouraging comments on our study.  
+
+1. I guess they never show that the excitability/firing of VIP neurons in control mice is enhanced by Gq + CNO. They only show indirectly with LFP that stimulation of FL elicits greater responses (from Pyr cells). I wonder if they ever saw differences with calcium imaging in VIP neurons before and 30 min after CNO.  
+
+RESPONSE: Since there have been dozens of papers published validating the excitatory effects of hM3Dq in neurons, our primary question was whether stimulating VIP neurons would lead to a general increase in sensory evoked cortical responses. This was the first set of experiments we conducted and provided the rationale for doing the entire study. Originally in 2013- 14, the hM3Dq AAV2 was provided by Penn state core vector facility and my post- doc K.G. validated the idea that chemogenetic stimulation of VIP neurons would enhance sensory evoked responses in both sham control and stroke mice. K.G. left for maternity leave (and ultimately science in 2016), when M.M. continued the project in 2017- present. However the hM3Dq AAV was now only provided by Addgene, so we again did electrophysiology experiments to prove the hM3Dq worked as expected. We should note that there was no significant difference in peak response amplitude (following 0.3 or \(0.5\mathrm{mg / kg}\) CNO) between these two sets of validation experiments (this info is now provided on page 5 of the results). However since our sample was small for validating the effects of hM3Dq in stroke mice (see our response to query 5 below), we had 3 VIP- cre mice left in our colony that had been injected with cre- dependent hM3Dq + GCaMP6s and had recovered for \(\sim 14\) weeks from stroke in FLS1 cortex. Although imaging conditions were not ideal (this is expected many months after implantation of a cranial window, especially if a stroke is involved), we found some forelimb responsive VIP neurons and imaged their responses before and after i.p. injection of \(0.5\mathrm{mg / kg}\) CNO. These results show a significant increase in mean forelimb evoked response amplitude and # responsive trials after CNO injection and are now presented in Supp. Fig. 1f. Thus, we believe that we have validated the use of chemogenetic treatment with multiple experiments from different experimenters at different epochs in time, using different but complimentary approaches (electrophysiology, calcium/VSD imaging, behaviour).  
+
+2. I was surprised that a single injection of CNO (which has such a short half-life) only 5 days a week was enough to rescue behavior? How do the authors interpret such a profound effect on the network (and behavior!)? Also, do they think the circuit is permanently restored such that, had they looked a few days after stopping CNO injections, the rescue might have persisted?  
+
+RESPONSE: This is a good point for the discussion section. Our cortical field recording experiments indicated that the effect of CNO on network excitability lasted for at least 60 min in our VIP mice (Fig. 1 and Supp. Fig. 1), and could potentially last longer based on the work of other groups (Alexander et al., 2009, Neuron). The idea that 60 minutes of chemogenetic therapy 5 days a week can positively affect behavioural outcome is in line with our previous paper (see Tennant et al; 2017, Nature Comm) where we
+
+<--- Page Split --->
+
+provided intermittent optogenetic stimulation for 1 hour/day for 5 days a week for up to 6 weeks. Our present experiments indicate the benefits of chemogenetic therapy persist long after treatment has ceased and therefore likely involve permanent changes in circuitry. For example, the improvement in ladder walking performance is maintained at week 7 in the stimulated group even though treatment had ceased at week 6. Similarly, VSD imaging was conducted at week 10 (4 weeks after treatment had ceased) and we still see an enhancement in sensory evoked response amplitudes in the stimulated group. Given that layer 2/3 pyramidal neurons are the dominant source of signal for VSD imaging, our findings imply these layer 2/3 circuits were strengthened with chemogenetic therapy. Precisely how these lasting downstream changes in circuitry are accomplished remains an open question. Previous work from our lab and others has shown that promoting the restoration of cortical excitability in somatosensory cortex after stroke (via chemo- or opto- genetics) is associated axonal sprouting (Wahl et al., 2017, Nature Comm) as well as the proliferation and stabilization of thalamocortical axonal boutons (Tennant et al., 2017, Nature Comm). In addition, therapies that promote the return of cortical excitability after stroke lead to changes in growth or plasticity associated gene expression (CREB, BDNF, NGF; see Cheng et al., 2014, PNAS; Caracciolo et al., 2018, Nat Comm). Although our study provides much needed insights into the cellular effects of stroke, at least within VIP interneurons, future stroke studies could dissect the contribution of other interneuron populations in stroke recovery. Furthermore, we think the stroke field could really benefit from future brain slice electrophysiology studies that address circuit specific changes in intrinsic excitability, spiking patterns, pre and post- synaptic forms of plasticity (LTP, LTD) using quantal analysis, paired pulse ratios, mini- analysis etc. We now include a more fulsome discussion on this topic on page 14 of the discussion.  
+
+3. They choose to perform photothrombotic strokes that do not completely destroy S1FL (eg Fig 1a). Presumably this is to make sure they still can find FL-responsive VIP neurons in peri-infarct cortex. The authors should make it more explicit in the text that they performed sub-total strokes, because it is likely that they would have never found any FL-responsive VIP neurons in peri-infarct cortex had they completely destroyed S1FL.  
+
+RESPONSE: The reviewer is correct that we did not destroy the entire FL cortex or else there would be little or no forelimb responsive neurons to image. We have now made this clear on page 5 of the results section.  
+
+4. Fig 1c shows that stroke strongly reduces FL stem evoked responses compared to sham. In Fig 1d peak amplitudes are "normalized to baseline", which means right before CNO injection. Could they show a comparison of these baseline responses (average of all the mice for stroke vs sham) to see how strongly stroke affects the FL stem evoked response (beyond the representative trace in Fig 1)  
+
+RESPONSE: In accordance with the reviewer's suggestion, we have added the average forelimb-evoked responses at baseline and following CNO injection for both stroke and sham stroke mice. This data is now presented in Supp. Fig. 1b.  
+
+5. In Fig 1d, what are the post-hoc individual p values for the effect of CNO in hM3Dq mice that did or did not receive a stroke?  
+
+RESPONSE: As discussed in response to query 1, the excitatory effects of the hM3Dq were validated twice with electrophysiology by two different experimenters at two different points in time (K.G.
+
+<--- Page Split --->
+
+validated in 2013- 14 and M.M. validated in 2017- 18). We combined these two validation experiments together (n=7 mice) to show a very robust excitatory effect of CNO on cortical sensory responses with post- hoc p values (see Fig. 1d). We have also added a new figure showing peak forelimb evoked field amplitudes before vs after CNO injection for individual mice in each of the experimental conditions (see Supp. Fig. 1a). However, the 7d stroke data was from a small sample. Although both stroke mice expressing hM3Dq showed a very clear increase in forelimb responses 30min after CNO injection relative to before (see new Supp. Fig. 1a), concerns about sampling should be mollified by the addition of new calcium imaging data from stroke recovered mice showing an increase in response amplitude and # responsive trials following CNO treatment (see our response to query 1, data shown in Supp. Fig. 1f). Therefore, despite the small sample for that specific experiment, there are multiple converging lines of experimental evidence showing that the chemogenetic treatment increases cortical responses to sensory stimulation in sham stroke controls or stroke recovered mice.  
+
+6. I like Suppl Fig 1a better than Fig. 1e. They might consider a swap (optional)  
+
+RESPONSE: As requested by the reviewer, we have swapped these figures.  
+
+7. In suppl Fig 2 please list the number of mice in each group; I don't think a t-test is appropriate here  
+
+RESPONSE: We have now added the number of mice in each group to Supp. Fig. 2 legend  
+
+8. Fig 2d: why not show the average DF/F for all ROIs combined (1-6)? It seems like the differences between hM3Dq and control mice were not significant (otherwise all the data would be presented in the same figure panel, like a bar graph of the peak response). The text says "forepaw evoked depolarizations in peri-infarct cortex were SIGNIFICANTLY larger in amplitude ... (Fig. 2b-d)" but there are no stats provided in the text or in the figure legend. It's fine if it's only significant for some of the ROIs (e.g., 4 & 5), but this could have meaning too based on their location relative to the infarct.  
+
+RESPONSE: We agree and have now added the average forelimb evoked dF/Fo (shaded area represents SEM) for all ROIs and all mice within each group (see new traces added to Fig. 2d). The reviewer also makes a good point about the statistical description. To clarify, the omnibus statistical analysis of forelimb-evoked responses (2-way ANOVA) was conducted only on normalized data to control for considerable between-experiment variability in response amplitudes. This analysis showed a significant effect of CNO treatment which is reported in figure 2e and the accompanying legend. We have removed the word "significant" for the statement pertaining to un-normalized data and reserved it for the analysis of normalized data in Fig. 2e.  
+
+9. In Fig 2b-f, what is the control? Is it hM3Dq but not CNO, or is it \(\mathrm{GFP} + \mathrm{CNO}\) alone? Or are they combined?  
+
+RESPONSE: The stroke control group for VSD experiments was run by author K.G and consisted of mice that expressed hM3Dq, but received vehicle injections instead of CNO. We have now clarified this in the figure 2 legend.  
+
+10. Fig 3a - is the earliest time point -6 weeks or -3 weeks? They should say in the legend what the
+
+<--- Page Split --->
+
+green/blue contours of maps represent (presumably it's the \(75\%\) threshold described in page 19, line 19).  
+
+RESPONSE: - 6 weeks (before stroke) is when the AAV injection and cranial window were installed, whereas imaging began 4 weeks later (- 2 weeks before stroke). The reviewer is correct in that green/blue contours were derived form a \(75\%\) of peak amplitude threshold. We now clarify this is in the Figure 3 legend.  
+
+11. Page 8, lines 21-24: they should probably show these data as part of suppl fig 3  
+
+RESPONSE: As requested, we have inserted a new graph in Supp Fig 3a-c showing baseline \(\%\) responsive neurons, \(\%\) responsive trials and peak amplitudes for the 3 groups.  
+
+12. Figs 3 & 4: regarding the concern about toxicity of prolonged GCaMP expression, I agree it's reassuring that the gray traces in Fig 4e (sham) are stable. But why is the \(+4\) wk time point missing for sham controls in Fig 4e-f?  
+
+RESPONSE: The sham stroke control mice (grey traces) were imaged for 4 weeks (BL and 3 weeks after sham stroke). Since neural responses were quite stable over this period of time, we did not collect data for the \(+4\) week time point.  
+
+13. Fig 4b: are these example traces from a vehicle stroke mouse?  
+
+RESPONSE: The reviewer is correct as the traces are from the same cells shown in Fig. 4a. We have clarified this point in the figure legend.  
+
+14. Fig 4d: can they use different colors (or symbols) for the 3 different types of mice in these scatter plots (sham control, stroke veh and stroke Gq)?  
+
+RESPONSE: The scatterplot only shows data from stroke- affected mice that received control treatment since sham stroke or CNO treated stroke mice did not exhibit a loss of sensory responses at 1 week after stroke.  
+
+15. Fig 4e-f: It's fine to show the normalized data but can they also show the raw data for \(\%\) of responsive neurons and peak amplitudes for the three groups at baseline (text has results of ANOVA but not actual data)? I ask because there does not appear to be a sustained increase in the \(\%\) of VIP neurons that respond to FL stimulation after stroke, which means that no new neurons are recruited to respond to FL stim after stroke (DREADDs only maintain the original pool).  
+
+RESPONSE: As requested by the reviewer, we have now plotted this raw baseline data in Supp. Fig. 3a- c. The reviewer is correct in that there wasn't a significant or sustained increase in the \(\%\) responsive VIP neurons after stroke in CNO treated mice relative to pre- stroke. As shown in Fig. 4e, if this were true we would have denoted the 1 week data with a # symbol, which represents statistical comparisons of post- stroke values vs. pre- stroke. We agree with the reviewer's conclusion that new forelimb neurons are not recruited after stroke and had previously stated this in the abstract and discussion. We have now included
+
+<--- Page Split --->
+
+the following conclusion statement in the results on page 12 to make this point clear: "Further, our data argue against the idea that new forelimb responsive neurons are recruited after stroke."  
+
+16. The effects on VIP responsivity are mainly seen at 1-wk post-stroke, but behavioral effects are seen 2-7 weeks after stroke...the discussion touches on this but more could said to explain this difference.  
+
+RESPONSE: We agree that this is an interesting finding but at this point in time, we can only speculate why the behavioural effects lag by 1 week behind the VIP response differences. It could be that the post- synaptic targets of VIP neurons, or ones further downstream (eg. pyramidal neurons) are slightly slower to restore intrinsic excitability or synaptic drive. As mentioned by reviewer 1, we think future studies imaging the downstream targets of VIP neurons or whole cell recordings in brain slices in defined excitatory and inhibitory circuits would help resolve this question (see new text added to page 14 of the discussion). However, these experiments would require years of work and therefore are best reserved for future study.  
+
+17. Fig 5b: Did the relative proportion of each of the 3 subtypes change over time (from baseline) in sham controls and stroke animals? There should be a graph to represent that. Also, shouldn't there be a Chi-square test for all the comparisons and then follow-up 2x2 chi-sq for individual comparisons with post-hoc correction. A better description of the Chi-sq methods would be useful  
+
+RESPONSE: Although time dependent changes in the relative proportion of each of the 3 subtypes is depicted and analysed in Fig. 5b, we agree with the reviewer that an additional graph (perhaps simplified) showing the proportion of each subtype within each experimental group, would be informative. Therefore we have added a new graph into Figure 5 (see new Fig. 5d) that shows the \(\%\) of each subtype before stroke compared to after stroke (values for weeks 1- 3 averaged together, similar to Fig. 5b) in each of the 3 experimental groups. We now report on page 11- 12 of the results that the proportion of each subtype does not change significantly in the sham stroke controls or stroke affected mice that received chemogenetic stimulation (Note: this data further supports the idea that chemogenetic stimulation was beneficial). This conclusion was based on a chi-squared analysis where we used pre-stroke as the "expected" proportion for each neuron subtype and then compared that to the "observed" proportion averaged over 3 weeks after stroke (all \(\chi^2\) values ranged from 0.01- 0.38 and therefore none were close to significant). By contrast, stroke mice that received control stimulation exhibited significantly fewer highly active neurons ( \(\chi^2 = 4.91\) , \(*p< 0.05\) ) and more minimally active neurons after stroke ( \(\chi^2 = 13.73\) , \(**p< 0.01\) ). This finding is consistent with the idea that stroke has a dampening effect on sensory responses. Since we compared the \(\%\) cells within each experimental group (or \(\%\) neuron sub- type) and analysed the average values over 3 weeks post- stroke rather than compare each week, we believe the direct statistical comparison to Sham stroke or Pre- stroke (as shown in Fig. 5b and Fig. 5d, respectively) was justified. We have added further clarity in the statistics section of the Methods (page 27) regarding how we calculated the chi- square statistic.  
+
+18. Fig 6a: the issue of how predictable (or should it be 'reliable'?) neurons are seems important, but this visual representation is a bit hard to follow. Another way to show this would be to plot, for each cell, the \(\%\) of stimulations it responds to at baseline and over time after stroke.  
+
+RESPONSE: We agree this result was not the easiest to present or describe, but think it is quite important. We tried many different graphs/approaches and asked our colleagues what they thought was
+
+<--- Page Split --->
+
+easiest/simplest to interpret. The graph presented in Fig. 6a was the “winning” graph and therefore we wish to keep it as presented.
+
+<--- Page Split --->
+
+Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+The author have adequately addressed all critical points and clarified their observations according to the comments provided. The manuscript is largely improved.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+The authors have addressed all my previous questions. This is an excellent manuscript and I support publication of this interesting work.  
+
+Sergei A Kirov  
+
+Reviewer #3:  
+
+Remarks to the Author:  
+
+The authors have meticulously responded to all of the concerns and comments I had raised and I have no further issues. I congratulate the authors on a beautiful and compelling study.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The author have adequately addressed all critical points and clarified their observations according to the comments provided. The manuscript is largely improved.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have addressed all my previous questions. This is an excellent manuscript and I support publication of this interesting work.  
+
+Sergei A Kirov  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have meticulously responded to all of the concerns and comments I had raised and I have no further issues. I congratulate the authors on a beautiful and compelling study.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__43a438ad55c12c7c430994081e89ce1899b856e6563e09e4569576d8deeaaae8/supplementary_0_Peer Review File__43a438ad55c12c7c430994081e89ce1899b856e6563e09e4569576d8deeaaae8_det.mmd

@@ -0,0 +1,496 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 507, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[70, 110, 362, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[70, 153, 900, 241]]<|/det|>
+Longitudinal functional imaging of VIP interneurons reveals sup- population specific effects of stroke that can be rescued with chemogenetic therapy  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 780]]<|/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  
+
+<|ref|>text<|/ref|><|det|>[[57, 785, 934, 924]]<|/det|>
+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|>[[256, 84, 431, 97]]<|/det|>
+## Reviewers' Comments:  
+
+<|ref|>text<|/ref|><|det|>[[119, 113, 225, 125]]<|/det|>
+Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[119, 127, 300, 140]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[118, 140, 861, 238]]<|/det|>
+Motaharinia et al. explored whether chemogenetic modulation of VIP+ interneurons in the somatosensory cortex of mice can be beneficial for stroke recovery. The authors provide very interesting data with longitudinal imaging of VIP+ interneurons before and several weeks after local stroke induction. The major finding is that driving VIP+ cells chemogenetically can rescue cortical excitability after stroke and fasten functional recovery. The data is of highest quality and were obtained in carefully designed experiments. I only have several minor comments/questions to clarify some findings.  
+
+<|ref|>text<|/ref|><|det|>[[117, 251, 875, 420]]<|/det|>
+1. The manuscript gives an impression that the major role of neocortical VIP+ cells is disinhibitory; however, this may not be the case. VIP+ neuronal population is highly heterogeneous, with up to \(80\%\) of VIP+ contacts formed onto GABA-negative structures (Zhou et al., 2017). This heterogeneity may result in highly variable patterns of activity observed by the authors in VIP+ cells, and needs to be carefully discussed. For example, across cortical areas, VIP+/CCK+ basket cells contact PVRs, whereas VIP+/CR+ cells contact preferentially interneurons (Guet-McCreight et al., 2020). Importantly, these two VIP+ subtypes have different intrinsic excitability, with VIP+/CR+ cells showing much higher input resistance. Accordingly, the disinhibitory VIP+/CR+ cells may represent the highly active group in this study and a preferential target for post-stroke therapy. It seems the authors are able to find VIP+ cells imaged in vivo for post-mortem IHC identification (Fig. 3B). It would be therefore possible to check whether the highly active group is expressing CR.  
+
+<|ref|>text<|/ref|><|det|>[[118, 432, 871, 476]]<|/det|>
+2. Was there any difference in evoked field response for 0.3 vs. \(0.5 \mathrm{mg / kg}\) of CNO? Why not using the same dose in all experiments? Ideally, one would use the lowest efficient dose throughout the entire study.  
+
+<|ref|>text<|/ref|><|det|>[[118, 490, 829, 518]]<|/det|>
+3. Why the therapy had started 4 days after stroke? Would a faster intervention be more beneficial? Have the authors tried different time intervals after the stroke, like 1, 2, 3 days?  
+
+<|ref|>text<|/ref|><|det|>[[118, 532, 509, 546]]<|/det|>
+4. Were there any acute affects of CNO treatment?  
+
+<|ref|>text<|/ref|><|det|>[[118, 560, 868, 589]]<|/det|>
+5. Was the beneficial effect of chemogenetic treatment persistent? Or have the authors evaluated the state of animals at a longer scale - 3-4 weeks (up to a month) after therapy cessation?  
+
+<|ref|>text<|/ref|><|det|>[[118, 602, 875, 645]]<|/det|>
+6. It is exciting to see that a single dose of CNO, with VIP activation time course of about 2 h (max up to 10 h), produces such a long-lasting (at least 23 h) effect on cortical excitability. Can you discuss a bit more the possible mechanisms of this plasticity?  
+
+<|ref|>text<|/ref|><|det|>[[118, 672, 869, 743]]<|/det|>
+7. In Discussion, the authors propose that chemogenetic manipulation can represent a relatively non-invasive way for boosting excitability. I would disagree with this point, as it requires stereotaxic surgery to deliver the viral vectors expressing the DREADD. The obvious advantage of this method is, however, its "targetability" or a possibility of focal manipulations with a local neuronal circuits in peri-infarct area.  
+
+<|ref|>text<|/ref|><|det|>[[118, 757, 792, 785]]<|/det|>
+8. Given that male and female mice were used in the study, were there any sex-specific differences in VIP+ cell activity or other reported phenomena?  
+
+<|ref|>text<|/ref|><|det|>[[118, 826, 222, 839]]<|/det|>
+Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[118, 841, 298, 854]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[118, 855, 877, 911]]<|/det|>
+In this manuscript, Craig Brown and colleagues have reported that chemogenetic stimulation of vasoactive intestinal peptide (VIP) interneurons enhances weakened sensory responses in the peri-infarct sensorimotor cortex of mice subjected to photothrombotic stroke and promotes the recovery of sensorimotor function. It is novel that sensory responses were disrupted by stroke
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 874, 225]]<|/det|>
+mainly within a population of highly active VIP neurons. The finding that chemogenetic stimulation accelerates the recovery of sensory responses within these highly active VIP neurons and without recruitment of neurons not previously involved in sensory responses is also intriguing. The manuscript is well written; experiments were involving a battery of imaging, electrophysiological and behavioral approaches that require significant expertise. The findings of this paper have clinical significance and support the notion that stroke recovery could be accelerated by enhancing cortical excitability through dis- inhibitory VIP interneurons. Statistical analyses are appropriate, and the level of details provided would be sufficient for other researchers to reproduce the work. The findings will be of interest to the others in the field and the broader readership. I am highly enthusiastic about this manuscript and support its publication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 238, 686, 252]]<|/det|>
+Addressing several points could further improve this excellent manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[118, 252, 870, 293]]<|/det|>
+1. Surprisingly, the authors did not come across dying neurons in the peri-infarct cortex during longitudinal imaging. It seems that all VIP interneurons imaged before stroke persisted for the next four weeks of 2P imaging. These findings should be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 293, 874, 364]]<|/det|>
+2. The peri-infarct zone in the photothrombotic stroke is relatively narrow. All 2P imaging studies were conducted 400 micrometers from the infarct border. Laser speckle images were collected. What percent of the pre-stroke blood flow was remaining in the area of 2P imaging? That could be estimated from the laser speckle images. Were these 2P images collected in the hypo-perfused cortex?  
+
+<|ref|>text<|/ref|><|det|>[[118, 364, 876, 490]]<|/det|>
+3. The infarct size is large, and spontaneous spreading depolarizations often accompany such large strokes. However, the authors did not record any large increases in the calcium signals associated with spreading depolarizations. It is not surprising since spontaneous changes in GCaMP6s fluorescence were recorded only for periods of up to 75 s after stroke, and there were no longitudinal electrophysiological recordings. It might be a missed opportunity because one possibility is that spreading depolarizations in the penumbra were the events that affected ("wreak havoc on") the predictability (fidelity) of neuronal responses to sensory stimulation (Fig. 6). Certainly, experiments addressing this possibility are beyond the experiments of this study, but such a possibility could be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 490, 873, 546]]<|/det|>
+1. I like the discussion that "stroke related plasticity usually involves the same circuits"(P.14L31). It might be helpful to argue that even during more severe brain injury inflicted by global ischemia, the postsynaptic dendritic membranes remain attached to axonal boutons, providing a structural basis for the recovery of the same circuits (doi.org/10.1093/cercor/bhaa134).  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 560, 220, 573]]<|/det|>
+## Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[118, 574, 867, 615]]<|/det|>
+1. What was the rationale to start CNO treatment on day four but not earlier? Also, the additional explanation on P.15 L9 "(ie. Days 4 to 6, Monday through Wednesday, before imaging on day 7)" seems unnecessary.  
+
+<|ref|>text<|/ref|><|det|>[[118, 615, 867, 657]]<|/det|>
+2. P21.L23. "... time to peak amplitude and half-width (ie. duration) of forelimb-evoked signals in the first 150ms after stimulation were measured with Clampfit 9.0 software (Molecular Devices)." Does it belong to another section, such as "Recording sensory evoked cortical field potentials"?  
+
+<|ref|>text<|/ref|><|det|>[[118, 671, 275, 685]]<|/det|>
+Sergei A. Kirov, PhD  
+
+<|ref|>text<|/ref|><|det|>[[118, 728, 222, 741]]<|/det|>
+Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[118, 743, 298, 755]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[118, 756, 876, 910]]<|/det|>
+In this study by Motaharinia et al the authors report that activating VIP interneurons with Gq DREADDs after stroke to the forelimb (FL) region of somatosensory cortex can restore cortical responses (to FL stimulation) to pre- stroke levels and enhance functional recovery. This is an important study with robust findings that add significantly to our understanding of stroke recovery. It is the only paper to my knowledge that has reported on the activity of VIP neurons in the context of stroke recovery. It is also an elegant study that combines two- photon calcium imaging in vivo, mouse behavior, and DREADDs. These experiments are hard! Finally, the finding that VIP neurons could be a target for restoring circuit function and ameliorating behavioral deficits after stroke is very exciting. I was impressed by the high experimental rigor: they use CNO- only (no hM3Dq) controls and they use appropriate stats. The blood flow control to rule our DREADD effects is great too. The behavior data shows definitive results with internal replication in 2 cohorts. The
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 83, 870, 182]]<|/det|>
+longitudinal imaging of VIP neurons over \(>4\) weeks is particularly impressive because they track the same neurons over time. I felt the paper was well written, the figures are easy to understand, and they provide appropriate references (for example of papers describing the known role of VIP cells in disinhibition of pyramidal cells). Overall, I feel the significance of this paper is high and I am enthusiastic about its publication. It is refreshing to see the use of cutting- edge tools to investigate functional circuit changes at the single cell level after stroke. Here is a list of comments/suggestions I hope the authors can address  
+
+<|ref|>text<|/ref|><|det|>[[118, 195, 880, 252]]<|/det|>
+- I guess they never show that the excitability/firing of VIP neurons in control mice is enhanced by \(\mathrm{Gq + CNO}\) . They only show indirectly with LFP that stimulation of FL elicits greater responses (from Pyr cells). I wonder if they ever saw differences with calcium imaging in VIP neurons before and 30 min after CNO.  
+
+<|ref|>text<|/ref|><|det|>[[118, 252, 872, 308]]<|/det|>
+- I was surprised that a single injection of CNO (which has such a short half-life) only 5 days a week was enough to rescue behavior? How do the authors interpret such a profound effect on the network (and behavior!)? Also, do they think the circuit is permanently restored such that, had they looked a few days after stopping CNO injections, the rescue might have persisted?  
+
+<|ref|>text<|/ref|><|det|>[[118, 309, 875, 378]]<|/det|>
+- They choose to perform photothrombotic strokes that do not completely destroy S1FL (eg Fig 1a). Presumably this is to make sure they still can find FL-responsive VIP neurons in peri-infarct cortex. The authors should make it more explicit in the text that they performed sub-total strokes, because it is likely that they would have never found any FL-responsive VIP neurons in peri-infarct cortex had they completely destroyed S1FL.  
+
+<|ref|>text<|/ref|><|det|>[[118, 378, 877, 450]]<|/det|>
+- Fig 1c shows that stroke strongly reduces FL stim evoked responses compared to sham. In Fig 1d peak amplitudes are "normalized to baseline", which means right before CNO injection. Could they show a comparison of these baseline responses (average of all the mice for stroke vs sham) to see how strongly stroke affects the FL stim evoked response (beyond the representative trace in Fig 1)  
+
+<|ref|>text<|/ref|><|det|>[[118, 450, 875, 477]]<|/det|>
+- In Fig 1d, what are the post-hoc individual p values for the effect of CNO in hM3Dq mice that did or did not receive a stroke?  
+
+<|ref|>text<|/ref|><|det|>[[118, 477, 875, 520]]<|/det|>
+- I like Suppl Fig 1a better than Fig. 1e. They might consider a swap (optional) 
+- In suppl Fig 2 please list the number of mice in each group; I don't think a t-test is appropriate here  
+
+<|ref|>text<|/ref|><|det|>[[118, 520, 877, 615]]<|/det|>
+- Fig 2d: why not show the average DF/F for all ROIs combined (1-6)? It seems like the differences between hM3Dq and control mice were not significant (otherwise all the data would be presented in the same figure panel, like a bar graph of the peak response). The text says "forepaw evoked depolarizations in peri-infarct cortex were SIGNIFICANTLY larger in amplitude ... (Fig. 2b-d)" but there are no stats provided in the text or in the figure legend. It's fine if it's only significant for some of the ROIs (e.g., 4 & 5), but this could have meaning too based on their location relative to the infarct.  
+
+<|ref|>text<|/ref|><|det|>[[118, 616, 870, 644]]<|/det|>
+- In Fig 2b-f, what is the control? Is it hM3Dq but not CNO, or is it GFP + CNO alone? Or are they combined?  
+
+<|ref|>text<|/ref|><|det|>[[118, 645, 870, 686]]<|/det|>
+- Fig 3a - is the earliest time point -6 weeks or -3 weeks? They should say in the legend what the green/blue contours of maps represent (presumably it's the 75% threshold described in page 19, line 19).  
+
+<|ref|>text<|/ref|><|det|>[[118, 687, 748, 701]]<|/det|>
+- Page 8, lines 21-24: they should probably show these data as part of suppl fig 3  
+
+<|ref|>text<|/ref|><|det|>[[118, 701, 840, 742]]<|/det|>
+- Figs 3 & 4: regarding the concern about toxicity of prolonged GCaMP expression, I agree it's reassuring that the gray traces in Fig 4e (sham) are stable. But why is the +4 wk time point missing for sham controls in Fig 4e-f?  
+
+<|ref|>text<|/ref|><|det|>[[118, 743, 616, 757]]<|/det|>
+- Fig 4b: are these example traces from a vehicle stroke mouse?  
+
+<|ref|>text<|/ref|><|det|>[[118, 757, 827, 785]]<|/det|>
+- Fig 4d: can they use different colors (or symbols) for the 3 different types of mice in these scatter plots (sham control, stroke veh and stroke Gq)?  
+
+<|ref|>text<|/ref|><|det|>[[118, 785, 877, 856]]<|/det|>
+- Fig 4e-f: It's fine to show the normalized data but can they also show the raw data for % of responsive neurons and peak amplitudes for the three groups at baseline (text has results of ANOVA but not actual data)? I ask because there does not appear to be a sustained increase in the % of VIP neurons that respond to FL stimulation after stroke, which means that no new neurons are recruited to respond to FL stim after stroke (DREADDs only maintain the original pool).  
+
+<|ref|>text<|/ref|><|det|>[[118, 857, 860, 897]]<|/det|>
+- The effects on VIP responsivity are mainly seen at 1-wk post-stroke, but behavioral effects are seen 2-7 weeks after stroke...the discussion touches on this but more could said to explain this difference.  
+
+<|ref|>text<|/ref|><|det|>[[118, 898, 872, 912]]<|/det|>
+- Fig 5b: Did the relative proportion of each of the 3 subtypes change over time (from baseline) in
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 169]]<|/det|>
+sham controls and stroke animals? There should be a graph to represent that. Also, shouldn't there be a Chi- square test for all the comparisons and then follow- up 2x2 chi- sq for individual comparisons with post- hoc correction. A better description of the Chi- sq methods would be useful - Fig 6a: the issue of how predictable (or should it be 'reliable'?) neurons are seems important, but this visual representation is a bit hard to follow. Another way to show this would be to plot, for each cell, the % of stimulations it responds to at baseline and over time after stroke.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 490, 108]]<|/det|>
+## RESPONSE TO REVIEWER COMMENTS  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 137, 348, 154]]<|/det|>
+## Dear editors and reviewers:  
+
+<|ref|>text<|/ref|><|det|>[[114, 170, 872, 433]]<|/det|>
+We would like to thank you for the opportunity to revise our manuscript titled "Longitudinal functional imaging of VIP interneurons reveals sup- population specific effects of stroke that can be rescued with chemogenetic therapy" for publication in Nature Communications. The reviewers have provided a number of important and incisive suggestions for improving the paper. As requested by the editors and reviewers, we have conducted several new and challenging experiments to support the conclusions in the paper. These experiments include: a) electrophysiological DC recordings and in vivo GCamP6s imaging of stroke related spreading depolarizations (see Reviewer 2 comments), b) in vivo imaging of VIP sensory responses followed by post- mortem re- identification of VIP neurons that co- localize with calretinin (see Reviewer 1 comments) and c) imaging the acute effects of chemogenetic stimulation on VIP neuron responses (see Reviewer 3 comments). In addition, we have included several new data analyses at the request of the reviewers. Please note that all manuscript revisions are highlighted in red text. As a result of these recommendations and revisions, we have strengthened the evidence supporting the paper's primary findings and refined our discussion of these results.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 462, 425, 480]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 500, 883, 609]]<|/det|>
+Motaharinia et al. explored whether chemogenetic modulation of VIP+ interneurons in the somatosensory cortex of mice can be beneficial for stroke recovery. The authors provide very interesting data with longitudinal imaging of VIP+ interneurons before and several weeks after local stroke induction. The major finding is that driving VIP+ cells chemogenetically can rescue cortical excitability after stroke and fasten functional recovery. The data is of highest quality and were obtained in carefully designed experiments. I only have several minor comments/questions to clarify some findings.  
+
+<|ref|>text<|/ref|><|det|>[[115, 620, 679, 637]]<|/det|>
+RESPONSE: We thank the reviewer for their positive appraisal of our study.  
+
+<|ref|>text<|/ref|><|det|>[[114, 666, 883, 865]]<|/det|>
+1. The manuscript gives an impression that the major role of neocortical VIP+ cells is disinhibitory; however, this may not be the case. VIP+ neuronal population is highly heterogeneous, with up to \(80\%\) of VIP+ contacts formed onto GABA negative structures (Zhou et al., 2017). This heterogeneity may result in highly variable patterns of activity observed by the authors in VIP+ cells, and needs to be carefully discussed. For example, across cortical areas, VIP+/CCK+ basket cells contact PYRs, whereas VIP+/CR+ cells contact preferentially interneurons (Guet-McCreight et al., 2020). Importantly, these two VIP+ subtypes have different intrinsic excitability, with VIP+/CR+ cells showing much higher input resistance. Accordingly, the disinhibitory VIP+/CR+ cells may represent the highly active group in this study and a preferential target for post-stroke therapy. It seems the authors are able to find VIP+ cells imaged in vivo for post-mortem IHC identification (Fig. 3B). It would be therefore possible to check whether the highly active group is expressing CR.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 880, 452]]<|/det|>
+RESPONSE: This is an interesting point, and it would certainly be fascinating to further define the neurochemical phenotype of the highly active VIP neurons. While we fully agree with the reviewer's comment about functional/neurochemical heterogeneity in the VIP population, our claim of a disinhibitory effect, at least on the network/population level, was based on our field recording data showing an increase in cortical field potential responses following DREADD based stimulation of VIP neurons. The reviewer raises the possibility that the highly active VIP neurons may express Calretinin (CR), which show much higher input resistance. Therefore to address this possibility, we installed cranial windows and longitudinally imaged 10 mice to assess VIP sensory responses, then attempted to find each individual neuron in post- mortem horizontal sections immunostained for Calretinin (CR). Since these are extremely difficult experiments with very low success rates, we could only re- locate the VIP neurons imaged in vivo in 2 mice. In these mice, we found that \(34\%\) VIP neurons expressing GCaMP6s co- localized with CR (see Supp. Fig. 6a,b). Conversely, \(31.5\%\) of CR neurons co- localized with VIP. Since there were few VIP neurons that co- localized with CR to sample from, we graphed the number of responsive trials (out of 8) as function of whether VIP neurons co- localized with CR or not (2 mice: 71 neurons were \(\mathrm{VIP + / CR - }\) and 14 neurons were \(\mathrm{VIP + / CR + }\) ). As shown in Supp. Fig. 6c, the response profile of VIP neurons that co- express calretinin versus those that do not, were not clearly different from one another. Based on these results, we conclude that the highly responsive group of VIP neurons (ie. neurons that respond to 6- 8 trials) are not exclusively the same ones that express CR. However, we agree with the reviewer that future studies probing the response profile of VIP neurons with specific neuropeptides or calcium binding proteins (CR, CCK, ChaT etc) would be informative. We have included this topic in the discussion on page 18 and presented the results of this experiment in a new Supp. Fig. 6.  
+
+<|ref|>text<|/ref|><|det|>[[115, 479, 861, 533]]<|/det|>
+2. Was there any difference in evoked field response for \(0.3\mathrm{vs.}0.5\mathrm{mg / kg}\) of CNO? Why not using the same dose in all experiments? Ideally, one would use the lowest efficient dose throughout the entire study.  
+
+<|ref|>text<|/ref|><|det|>[[114, 543, 880, 700]]<|/det|>
+RESPONSE: Ideally, yes we would have used the same dose. Since the CNO- DREADD experiments were carried out in 2 separate epochs (2013- 2015 and 2017- present), there was a mis- communication when we re- started the CNO experiments in the second epoch (trainee thought we were using \(0.5\mathrm{mg / kg}\) vs \(0.3\mathrm{mg / kg}\) ). However in both epochs, we tested and validated the CNO- DREADD effects with electrophysiology to ensure they would work as expected. We now present our analysis of peak field potential responses when mice were dosed with \(0.3\mathrm{vs.}0.5\mathrm{mg / kg}\) CNO. Importantly we show there was no significant differences between doses and now include this analysis in the results on page 5 "There was no significant difference in peak response amplitude when comparing \(0.3\mathrm{mg / kg}\) vs \(0.5\mathrm{mg / kg}\) doses of CNO (unpaired t- test, \(\mathrm{t}_{(3)} = 1.21\) , \(\mathrm{p} = 0.31\) )."  
+
+<|ref|>text<|/ref|><|det|>[[115, 728, 850, 763]]<|/det|>
+3. Why the therapy had started 4 days after stroke? Would a faster intervention be more beneficial? Have the authors tried different time intervals after the stroke, like 1, 2, 3 days?  
+
+<|ref|>text<|/ref|><|det|>[[115, 774, 881, 895]]<|/det|>
+RESPONSE: In the first 72 hours after stroke, the brain is in quite a labile and precarious state due to edema, spreading depolarizations, excitotoxicity and other potentially damaging events that can further expand the area of ischemic damage. Since our goals were to: a) focus on stroke recovery strategies rather than neuroprotection (ie. enhance function of what brain tissue remains vs. preventing ischemic cell death), and b) image the same "peri- infarct" neurons before and after stroke; we did not want to further risk aggravating ischemic cell death by chemogenetically augmenting excitability within the first 3 days after stroke. We have now included a statement in the results clarifying our rationale (page 6). Lastly, we
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 880, 142]]<|/det|>
+do agree that an interesting future study would be to test out different time points for initiating therapeutic intervention (Note: this actually part of a recent grant proposal), especially if we could extent the therapeutic window of opportunity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 160, 490, 177]]<|/det|>
+4. Were there any acute affects of CNO treatment?  
+
+<|ref|>text<|/ref|><|det|>[[115, 187, 836, 239]]<|/det|>
+RESPONSE: As shown in Figure 1e and Supp. Fig. 1, the effects of CNO treatment on cortical excitability peak within the first 60-90min after injection and then, according to published literature (Alexander et al., 2009, Neuron), start to decline thereafter.  
+
+<|ref|>text<|/ref|><|det|>[[115, 258, 864, 293]]<|/det|>
+5. Was the beneficial effect of chemogenetic treatment persistent? Or have the authors evaluated the state of animals at a longer scale - 3-4 weeks (up to a month) after therapy cessation?  
+
+<|ref|>text<|/ref|><|det|>[[115, 303, 872, 408]]<|/det|>
+RESPONSE: We too were curious about this. Our behavioural data suggest that the beneficial effects did persist given that we stopped treatment at week 6, and then even one week later, the mice that received chemogenetic treatment were still significantly better than controls on the horizontal ladder test (see Fig. 2a). Furthermore, when examining forelimb evoked cortical responses with VSD imaging at 10 weeks recovery ( \(\sim 4\) weeks after treatment was stopped), the mice that received chemogenetic treatment showed significantly increased sensory-evoked cortical responses compared to controls (see Fig. 2b- e).  
+
+<|ref|>text<|/ref|><|det|>[[115, 425, 872, 479]]<|/det|>
+6. It is exciting to see that a single dose of CNO, with VIP activation time course of about \(2\mathrm{h}\) (max up to \(10\mathrm{h}\) ), produces such a long-lasting (at least \(23\mathrm{h}\) ) effect on cortical excitability. Can you discuss a bit more the possible mechanisms of this plasticity?  
+
+<|ref|>text<|/ref|><|det|>[[114, 489, 881, 767]]<|/det|>
+RESPONSE: This is a good point. The fact that the effects of chemogenetic therapy persist long after treatment has ceased suggest that up- regulating the excitability of VIP neurons has led to more permanent changes within stroke recovered circuits, presumably those downstream of VIP neurons (eg. layer 2/3 pyramidal neurons and other interneurons). Precisely how these lasting downstream changes in circuitry are accomplished remains an open question. Previous work from our lab and others has shown that promoting the restoration of cortical excitability in somatosensory cortex after stroke (via chemo- or optogenetic) is associated axonal sprouting (Wahl et al., 2017, Nature Comm) as well as the proliferation and stabilization of thalamocortical axonal boutons (Tennant et al., 2017, Nature Comm). In addition, therapies that promote the return of cortical excitability after stroke lead to changes in growth and plasticity associated gene expression (CREB, BDNF, NGF; see Cheng et al., 2014, PNAS; Caracciolo et al., 2018, Nat Comm). Although our study provides much needed insights into the effects of stroke at a cellular level, future stroke studies could dissect the contribution of other interneuron populations in stroke recovery. Furthermore, we think the stroke field could really benefit from future slice electrophysiology studies that address circuit specific changes in intrinsic excitability, spiking patterns, pre and post- synaptic forms of plasticity (LTP, LTD) using quantal analysis, paired pulse ratios, mini- analysis etc. We now include a more fulsome discussion on this topic on page 14 of the discussion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 785, 875, 857]]<|/det|>
+7. In Discussion, the authors propose that chemogenetic manipulation can represent a relatively non-invasive way for boosting excitability. I would disagree with this point, as it requires stereotaxic surgery to deliver the viral vectors expressing the DREADD. The obvious advantage of this method is, however, its "targetability" or a possibility of focal manipulations with a local neuronal circuits in peri-infarct area.  
+
+<|ref|>text<|/ref|><|det|>[[115, 867, 703, 885]]<|/det|>
+RESPONSE: we agree and have removed the phrase about being "non-invasive"
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 860, 125]]<|/det|>
+8. Given that male and female mice were used in the study, were there any sex-specific differences in VIP+ cell activity or other reported phenomena?  
+
+<|ref|>text<|/ref|><|det|>[[114, 135, 863, 309]]<|/det|>
+RESPONSE: For our study, we used male VIP mice for the electrophysiology, VSD and calcium imaging experiments, as well as the first cohort of behavioural studies testing the chemogenetic therapy (see cohort 1, K.G. 2014 in Figure 2a). However as we (and science in general) have become more cognizant of considering sex in our studies, we used female VIP mice for the second cohort of behavioural studies testing the chemogenetic therapy (see Cohort 2, S.C. 2018 in Figure 2a). If we compare males and females that received the chemogenetic therapy relative to their respective control groups, the benefits of therapy for males and females (ie. comparing difference in \(\%\) correct steps between treated and controls from weeks 2- 7), were quite similar and not significantly different (2- way ANOVA; Main effect of Sex: \(\mathrm{F}_{(1,72)} = 1.31\) , \(\mathrm{p} = 0.26\) ). We have now included this analysis in the results (page 6- 7) and clarified the sex of the mice in the methods section and Figure legend 2.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 346, 425, 364]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 383, 872, 638]]<|/det|>
+In this manuscript, Craig Brown and colleagues have reported that chemogenetic stimulation of vasoactive intestinal peptide (VIP) interneurons enhances weakened sensory responses in the perinfarct sensorimotor cortex of mice subjected to photothrombotic stroke and promotes the recovery of sensorimotor function. It is novel that sensory responses were disrupted by stroke mainly within a population of highly active VIP neurons. The finding that chemogenetic stimulation accelerates the recovery of sensory responses within these highly active VIP neurons and without recruitment of neurons not previously involved in sensory responses is also intriguing. The manuscript is well written; experiments were involving a battery of imaging, electrophysiological and behavioral approaches that require significant expertise. The findings of this paper have clinical significance and support the notion that stroke recovery could be accelerated by enhancing cortical excitability through dis- inhibitory VIP interneurons. Statistical analyses are appropriate, and the level of details provided would be sufficient for other researchers to reproduce the work. The findings will be of interest to the others in the field and the broader readership. I am highly enthusiastic about this manuscript and support its publication.  
+
+<|ref|>text<|/ref|><|det|>[[116, 639, 690, 656]]<|/det|>
+RESPONSE: We thank the reviewer for their positive comments on our study.  
+
+<|ref|>text<|/ref|><|det|>[[116, 686, 660, 703]]<|/det|>
+Addressing several points could further improve this excellent manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[116, 704, 875, 757]]<|/det|>
+1. Surprisingly, the authors did not come across dying neurons in the peri-infarct cortex during longitudinal imaging. It seems that all VIP interneurons imaged before stroke persisted for the next four weeks of 2P imaging. These findings should be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[114, 767, 882, 889]]<|/det|>
+RESPONSE: We think this might be an issue of semantics, but it is an important one. Since we could not control the extent of ischemic damage with micron precision, some areas we had imaged before stroke and had hoped would remain after the induction of stroke, were ultimately destroyed. These areas we considered part of the infarct core, not peri- infarct cortex. Thus, when we imaged neurons 7 days after stroke, we only focused on ones that were viable and outside of the infarct core, which we define as "peri- infarct". As mentioned below, the infarct border is quite sharp and therefore ischemic cell death was not still evolving by post- stroke day 7. In this context, it is not surprising that neurons imaged 7 days after
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 864, 123]]<|/det|>
+stroke, would persist for the remaining 4 weeks. We have clarified this point on page 25 in the methods section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 152, 880, 223]]<|/det|>
+2. The peri-infarct zone in the photothrombotic stroke is relatively narrow. All 2P imaging studies were conducted 400 micrometers from the infarct border. Laser speckle images were collected. What percent of the pre-stroke blood flow was remaining in the area of 2P imaging? That could be estimated from the laser speckle images. Were these 2P images collected in the hypo-perfused cortex?  
+
+<|ref|>text<|/ref|><|det|>[[115, 235, 878, 338]]<|/det|>
+RESPONSE: In the present study, we used laser speckle imaging to define the infarct border starting 7 days after stroke, but not before stroke. Therefore, a comparison of pre vs post-stroke blood flow was not possible. However, there are several published papers including our own (Tennant et al., 2013; Sullender, et al., 2018) that show relatively normal blood flow in peri-infarct cortex (after photothrombotic stroke) by 5- 7 days post-stroke. We now include this information and cited literature on page 25 in the methods section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 367, 880, 512]]<|/det|>
+3. The infarct size is large, and spontaneous spreading depolarizations often accompany such large strokes. However, the authors did not record any large increases in the calcium signals associated with spreading depolarizations. It is not surprising since spontaneous changes in GCaMP6s fluorescence were recorded only for periods of up to 75 s after stroke, and there were no longitudinal electrophysiological recordings. It might be a missed opportunity because one possibility is that spreading depolarizations in the penumbra were the events that affected ("wreak havoc on") the predictability (fidelity) of neuronal responses to sensory stimulation (Fig. 6). Certainly, experiments addressing this possibility are beyond the experiments of this study, but such a possibility could be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[114, 523, 880, 903]]<|/det|>
+RESPONSE: The idea that ischemic spreading depolarization (SD) could have been occurring 7 days after stroke and thereby influence our findings is an interesting one and something we had not previously considered. Before we describe our new experiments and analyses, we should prevent any confusion by clarifying that all calcium imaging in the original submission was conducted 7 days after stroke, not in the first few minutes to hours after stroke. We apologize for any confusion and have now made this clear in the results and methods section. In order to address the possibility of SDs contaminating our calcium imaging data (collected at 7 days post-stroke), we employed two different, but complimentary approaches (n=6 mice for each experimental approach). First, we used the gold standard approach of recording direct current (DC) potentials in peri-infarct cortex immediately after stroke and 7 days recovery (n=3 mice per time point). Recording immediately after initiating photothrombosis revealed approximately 1- 2 SD's occur within the first 30 minutes after photothrombotic stroke (2 mice exhibited 2 SD each whereas 1 mouse showed just 1 SD; peak amplitude: \(12\pm 1.1\mathrm{mV}\) ; duration: \(161.8\pm 5.2\mathrm{s}\) ; Supp. Fig. 5a), which is consistent with previous studies (Risher et al., JNSCI, 2010). However, we did not observe any SDs in peri-infarct cortex 7 days after stroke (Supp Fig. 5a). To be extra careful, we also included a positive control experiment with 1mM KCl application to prove that if a SD had occurred, we would have detected it. Similarly, by imaging SD induced calcium waves in peri-infarct cortex (which are \(\sim 10\) times larger in amplitude than sensory evoked calcium transients), we typically observed an SD within the first 30min after stroke (see example shown in Supp. Fig. 5b), but never 7 days post-stroke (n=3 mice per time point). Based on these experiments and previous literature, we conclude that it is very unlikely that ongoing SD waves could have affected our calcium imaging experiments and analysis collected 7 days after stroke. We now include these results on page 10 and Supplementary Fig. 5, as well as provided methodological details on page 21 and 25 in the Methods section.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 882, 161]]<|/det|>
+4. I like the discussion that "stroke related plasticity usually involves the same circuits" (P.14L31). It might be helpful to argue that even during more severe brain injury inflicted by global ischemia, the postsynaptic dendritic membranes remain attached to axonal boutons, providing a structural basis for the recovery of the same circuits (doi.org/10.1093/cercor/bhaa134).  
+
+<|ref|>text<|/ref|><|det|>[[115, 172, 880, 242]]<|/det|>
+RESPONSE: We thank the reviewer for this suggestion and agree that if post- synaptic structures retain their pre- synaptic partners, this could provide a structural explanation for the temporary disruption and reactivation of highly active circuits. We now include this relevant paper and explanation in the discussion section on page 16.  
+
+<|ref|>text<|/ref|><|det|>[[115, 253, 216, 269]]<|/det|>
+Minor points:  
+
+<|ref|>text<|/ref|><|det|>[[115, 271, 876, 324]]<|/det|>
+1. What was the rationale to start CNO treatment on day four but not earlier? Also, the additional explanation on P.15 L9 "(ie. Days 4 to 6, Monday through Wednesday, before imaging on day 7)" seems unnecessary.  
+
+<|ref|>text<|/ref|><|det|>[[115, 334, 875, 370]]<|/det|>
+RESPONSE: We have removed the unnecessary information. Since Reviewer 1 asked a similar question about the timing of treatment (query 3), we have copied part of our response here.  
+
+<|ref|>text<|/ref|><|det|>[[115, 379, 872, 500]]<|/det|>
+"In the first 72 hours after stroke, the brain is in quite a labile and precarious state due to edema, spreading depolarizations, excitotoxicity and other potentially damaging events that can further expand the area of ischemic damage. Since our goals were to: a) focus on stroke recovery strategies vs. neuroprotection (ie. enhance function of what brain tissue remains vs. preventing ischemic cell death), and b) image the same "peri-infarct" neurons before and after stroke; we did not want to further risk aggravating ischemic cell death by chemogenetically augmenting excitability within the first 3 days after stroke. We have now included a statement in the results clarifying our rationale (page 6)."  
+
+<|ref|>text<|/ref|><|det|>[[115, 528, 860, 582]]<|/det|>
+2. P21.L23. "... time to peak amplitude and half-width (ie. duration) of forelimb-evoked signals in the first 150ms after stimulation were measured with Clampfit 9.0 software (Molecular Devices)." Does it belong to another section, such as "Recording sensory evoked cortical field potentials"?  
+
+<|ref|>text<|/ref|><|det|>[[115, 592, 629, 610]]<|/det|>
+RESPONSE: This is correct, we analysed VSD signals in this manner.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 648, 425, 666]]<|/det|>
+## Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 685, 882, 831]]<|/det|>
+In this study by Motaharinia et al the authors report that activating VIP interneurons with Gq DREADDs after stroke to the forelimb (FL) region of somatosensory cortex can restore cortical responses (to FL stimulation) to pre- stroke levels and enhance functional recovery. This is an important study with robust findings that add significantly to our understanding of stroke recovery. It is the only paper to my knowledge that has reported on the activity of VIP neurons in the context of stroke recovery. It is also an elegant study that combines two- photon calcium imaging in vivo, mouse behavior, and DREADDs. These experiments are hard! Finally, the finding that VIP neurons could be a target for restoring circuit function and ameliorating behavioral deficits after stroke is very exciting.  
+
+<|ref|>text<|/ref|><|det|>[[115, 842, 860, 896]]<|/det|>
+I was impressed by the high experimental rigor: they use CNO- only (no hM3Dq) controls and they use appropriate stats. The blood flow control to rule our DREADD effects is great too. The behavior data shows definitive results with internal replication in 2 cohorts. The longitudinal imaging of VIP neurons
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 89, 876, 199]]<|/det|>
+over >4 weeks is particularly impressive because they track the same neurons over time. I felt the paper was well written, the figures are easy to understand, and they provide appropriate references (for example of papers describing the known role of VIP cells in disinhibition of pyramidal cells). Overall, I feel the significance of this paper is high and I am enthusiastic about its publication. It is refreshing to see the use of cutting- edge tools to investigate functional circuit changes at the single cell level after stroke. Here is a list of comments/suggestions I hope the authors can address  
+
+<|ref|>text<|/ref|><|det|>[[115, 225, 722, 243]]<|/det|>
+RESPONSE: We thank the reviewer for their encouraging comments on our study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 271, 880, 325]]<|/det|>
+1. I guess they never show that the excitability/firing of VIP neurons in control mice is enhanced by Gq + CNO. They only show indirectly with LFP that stimulation of FL elicits greater responses (from Pyr cells). I wonder if they ever saw differences with calcium imaging in VIP neurons before and 30 min after CNO.  
+
+<|ref|>text<|/ref|><|det|>[[113, 336, 881, 699]]<|/det|>
+RESPONSE: Since there have been dozens of papers published validating the excitatory effects of hM3Dq in neurons, our primary question was whether stimulating VIP neurons would lead to a general increase in sensory evoked cortical responses. This was the first set of experiments we conducted and provided the rationale for doing the entire study. Originally in 2013- 14, the hM3Dq AAV2 was provided by Penn state core vector facility and my post- doc K.G. validated the idea that chemogenetic stimulation of VIP neurons would enhance sensory evoked responses in both sham control and stroke mice. K.G. left for maternity leave (and ultimately science in 2016), when M.M. continued the project in 2017- present. However the hM3Dq AAV was now only provided by Addgene, so we again did electrophysiology experiments to prove the hM3Dq worked as expected. We should note that there was no significant difference in peak response amplitude (following 0.3 or \(0.5\mathrm{mg / kg}\) CNO) between these two sets of validation experiments (this info is now provided on page 5 of the results). However since our sample was small for validating the effects of hM3Dq in stroke mice (see our response to query 5 below), we had 3 VIP- cre mice left in our colony that had been injected with cre- dependent hM3Dq + GCaMP6s and had recovered for \(\sim 14\) weeks from stroke in FLS1 cortex. Although imaging conditions were not ideal (this is expected many months after implantation of a cranial window, especially if a stroke is involved), we found some forelimb responsive VIP neurons and imaged their responses before and after i.p. injection of \(0.5\mathrm{mg / kg}\) CNO. These results show a significant increase in mean forelimb evoked response amplitude and # responsive trials after CNO injection and are now presented in Supp. Fig. 1f. Thus, we believe that we have validated the use of chemogenetic treatment with multiple experiments from different experimenters at different epochs in time, using different but complimentary approaches (electrophysiology, calcium/VSD imaging, behaviour).  
+
+<|ref|>text<|/ref|><|det|>[[115, 727, 880, 799]]<|/det|>
+2. I was surprised that a single injection of CNO (which has such a short half-life) only 5 days a week was enough to rescue behavior? How do the authors interpret such a profound effect on the network (and behavior!)? Also, do they think the circuit is permanently restored such that, had they looked a few days after stopping CNO injections, the rescue might have persisted?  
+
+<|ref|>text<|/ref|><|det|>[[115, 810, 880, 896]]<|/det|>
+RESPONSE: This is a good point for the discussion section. Our cortical field recording experiments indicated that the effect of CNO on network excitability lasted for at least 60 min in our VIP mice (Fig. 1 and Supp. Fig. 1), and could potentially last longer based on the work of other groups (Alexander et al., 2009, Neuron). The idea that 60 minutes of chemogenetic therapy 5 days a week can positively affect behavioural outcome is in line with our previous paper (see Tennant et al; 2017, Nature Comm) where we
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 882, 451]]<|/det|>
+provided intermittent optogenetic stimulation for 1 hour/day for 5 days a week for up to 6 weeks. Our present experiments indicate the benefits of chemogenetic therapy persist long after treatment has ceased and therefore likely involve permanent changes in circuitry. For example, the improvement in ladder walking performance is maintained at week 7 in the stimulated group even though treatment had ceased at week 6. Similarly, VSD imaging was conducted at week 10 (4 weeks after treatment had ceased) and we still see an enhancement in sensory evoked response amplitudes in the stimulated group. Given that layer 2/3 pyramidal neurons are the dominant source of signal for VSD imaging, our findings imply these layer 2/3 circuits were strengthened with chemogenetic therapy. Precisely how these lasting downstream changes in circuitry are accomplished remains an open question. Previous work from our lab and others has shown that promoting the restoration of cortical excitability in somatosensory cortex after stroke (via chemo- or opto- genetics) is associated axonal sprouting (Wahl et al., 2017, Nature Comm) as well as the proliferation and stabilization of thalamocortical axonal boutons (Tennant et al., 2017, Nature Comm). In addition, therapies that promote the return of cortical excitability after stroke lead to changes in growth or plasticity associated gene expression (CREB, BDNF, NGF; see Cheng et al., 2014, PNAS; Caracciolo et al., 2018, Nat Comm). Although our study provides much needed insights into the cellular effects of stroke, at least within VIP interneurons, future stroke studies could dissect the contribution of other interneuron populations in stroke recovery. Furthermore, we think the stroke field could really benefit from future brain slice electrophysiology studies that address circuit specific changes in intrinsic excitability, spiking patterns, pre and post- synaptic forms of plasticity (LTP, LTD) using quantal analysis, paired pulse ratios, mini- analysis etc. We now include a more fulsome discussion on this topic on page 14 of the discussion.  
+
+<|ref|>text<|/ref|><|det|>[[115, 479, 864, 570]]<|/det|>
+3. They choose to perform photothrombotic strokes that do not completely destroy S1FL (eg Fig 1a). Presumably this is to make sure they still can find FL-responsive VIP neurons in peri-infarct cortex. The authors should make it more explicit in the text that they performed sub-total strokes, because it is likely that they would have never found any FL-responsive VIP neurons in peri-infarct cortex had they completely destroyed S1FL.  
+
+<|ref|>text<|/ref|><|det|>[[115, 580, 868, 633]]<|/det|>
+RESPONSE: The reviewer is correct that we did not destroy the entire FL cortex or else there would be little or no forelimb responsive neurons to image. We have now made this clear on page 5 of the results section.  
+
+<|ref|>text<|/ref|><|det|>[[115, 661, 882, 734]]<|/det|>
+4. Fig 1c shows that stroke strongly reduces FL stem evoked responses compared to sham. In Fig 1d peak amplitudes are "normalized to baseline", which means right before CNO injection. Could they show a comparison of these baseline responses (average of all the mice for stroke vs sham) to see how strongly stroke affects the FL stem evoked response (beyond the representative trace in Fig 1)  
+
+<|ref|>text<|/ref|><|det|>[[115, 744, 880, 797]]<|/det|>
+RESPONSE: In accordance with the reviewer's suggestion, we have added the average forelimb-evoked responses at baseline and following CNO injection for both stroke and sham stroke mice. This data is now presented in Supp. Fig. 1b.  
+
+<|ref|>text<|/ref|><|det|>[[115, 825, 867, 860]]<|/det|>
+5. In Fig 1d, what are the post-hoc individual p values for the effect of CNO in hM3Dq mice that did or did not receive a stroke?  
+
+<|ref|>text<|/ref|><|det|>[[112, 871, 857, 906]]<|/det|>
+RESPONSE: As discussed in response to query 1, the excitatory effects of the hM3Dq were validated twice with electrophysiology by two different experimenters at two different points in time (K.G.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 881, 295]]<|/det|>
+validated in 2013- 14 and M.M. validated in 2017- 18). We combined these two validation experiments together (n=7 mice) to show a very robust excitatory effect of CNO on cortical sensory responses with post- hoc p values (see Fig. 1d). We have also added a new figure showing peak forelimb evoked field amplitudes before vs after CNO injection for individual mice in each of the experimental conditions (see Supp. Fig. 1a). However, the 7d stroke data was from a small sample. Although both stroke mice expressing hM3Dq showed a very clear increase in forelimb responses 30min after CNO injection relative to before (see new Supp. Fig. 1a), concerns about sampling should be mollified by the addition of new calcium imaging data from stroke recovered mice showing an increase in response amplitude and # responsive trials following CNO treatment (see our response to query 1, data shown in Supp. Fig. 1f). Therefore, despite the small sample for that specific experiment, there are multiple converging lines of experimental evidence showing that the chemogenetic treatment increases cortical responses to sensory stimulation in sham stroke controls or stroke recovered mice.  
+
+<|ref|>text<|/ref|><|det|>[[115, 323, 700, 341]]<|/det|>
+6. I like Suppl Fig 1a better than Fig. 1e. They might consider a swap (optional)  
+
+<|ref|>text<|/ref|><|det|>[[115, 351, 667, 369]]<|/det|>
+RESPONSE: As requested by the reviewer, we have swapped these figures.  
+
+<|ref|>text<|/ref|><|det|>[[115, 397, 866, 415]]<|/det|>
+7. In suppl Fig 2 please list the number of mice in each group; I don't think a t-test is appropriate here  
+
+<|ref|>text<|/ref|><|det|>[[115, 425, 780, 444]]<|/det|>
+RESPONSE: We have now added the number of mice in each group to Supp. Fig. 2 legend  
+
+<|ref|>text<|/ref|><|det|>[[115, 471, 882, 580]]<|/det|>
+8. Fig 2d: why not show the average DF/F for all ROIs combined (1-6)? It seems like the differences between hM3Dq and control mice were not significant (otherwise all the data would be presented in the same figure panel, like a bar graph of the peak response). The text says "forepaw evoked depolarizations in peri-infarct cortex were SIGNIFICANTLY larger in amplitude ... (Fig. 2b-d)" but there are no stats provided in the text or in the figure legend. It's fine if it's only significant for some of the ROIs (e.g., 4 & 5), but this could have meaning too based on their location relative to the infarct.  
+
+<|ref|>text<|/ref|><|det|>[[115, 592, 881, 732]]<|/det|>
+RESPONSE: We agree and have now added the average forelimb evoked dF/Fo (shaded area represents SEM) for all ROIs and all mice within each group (see new traces added to Fig. 2d). The reviewer also makes a good point about the statistical description. To clarify, the omnibus statistical analysis of forelimb-evoked responses (2-way ANOVA) was conducted only on normalized data to control for considerable between-experiment variability in response amplitudes. This analysis showed a significant effect of CNO treatment which is reported in figure 2e and the accompanying legend. We have removed the word "significant" for the statement pertaining to un-normalized data and reserved it for the analysis of normalized data in Fig. 2e.  
+
+<|ref|>text<|/ref|><|det|>[[115, 761, 832, 796]]<|/det|>
+9. In Fig 2b-f, what is the control? Is it hM3Dq but not CNO, or is it \(\mathrm{GFP} + \mathrm{CNO}\) alone? Or are they combined?  
+
+<|ref|>text<|/ref|><|det|>[[115, 807, 869, 859]]<|/det|>
+RESPONSE: The stroke control group for VSD experiments was run by author K.G and consisted of mice that expressed hM3Dq, but received vehicle injections instead of CNO. We have now clarified this in the figure 2 legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 888, 841, 906]]<|/det|>
+10. Fig 3a - is the earliest time point -6 weeks or -3 weeks? They should say in the legend what the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 857, 125]]<|/det|>
+green/blue contours of maps represent (presumably it's the \(75\%\) threshold described in page 19, line 19).  
+
+<|ref|>text<|/ref|><|det|>[[115, 135, 872, 205]]<|/det|>
+RESPONSE: - 6 weeks (before stroke) is when the AAV injection and cranial window were installed, whereas imaging began 4 weeks later (- 2 weeks before stroke). The reviewer is correct in that green/blue contours were derived form a \(75\%\) of peak amplitude threshold. We now clarify this is in the Figure 3 legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 234, 730, 250]]<|/det|>
+11. Page 8, lines 21-24: they should probably show these data as part of suppl fig 3  
+
+<|ref|>text<|/ref|><|det|>[[115, 261, 812, 295]]<|/det|>
+RESPONSE: As requested, we have inserted a new graph in Supp Fig 3a-c showing baseline \(\%\) responsive neurons, \(\%\) responsive trials and peak amplitudes for the 3 groups.  
+
+<|ref|>text<|/ref|><|det|>[[115, 324, 861, 377]]<|/det|>
+12. Figs 3 & 4: regarding the concern about toxicity of prolonged GCaMP expression, I agree it's reassuring that the gray traces in Fig 4e (sham) are stable. But why is the \(+4\) wk time point missing for sham controls in Fig 4e-f?  
+
+<|ref|>text<|/ref|><|det|>[[115, 388, 880, 440]]<|/det|>
+RESPONSE: The sham stroke control mice (grey traces) were imaged for 4 weeks (BL and 3 weeks after sham stroke). Since neural responses were quite stable over this period of time, we did not collect data for the \(+4\) week time point.  
+
+<|ref|>text<|/ref|><|det|>[[115, 469, 608, 486]]<|/det|>
+13. Fig 4b: are these example traces from a vehicle stroke mouse?  
+
+<|ref|>text<|/ref|><|det|>[[115, 497, 852, 532]]<|/det|>
+RESPONSE: The reviewer is correct as the traces are from the same cells shown in Fig. 4a. We have clarified this point in the figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[115, 560, 861, 595]]<|/det|>
+14. Fig 4d: can they use different colors (or symbols) for the 3 different types of mice in these scatter plots (sham control, stroke veh and stroke Gq)?  
+
+<|ref|>text<|/ref|><|det|>[[115, 606, 867, 658]]<|/det|>
+RESPONSE: The scatterplot only shows data from stroke- affected mice that received control treatment since sham stroke or CNO treated stroke mice did not exhibit a loss of sensory responses at 1 week after stroke.  
+
+<|ref|>text<|/ref|><|det|>[[115, 688, 878, 777]]<|/det|>
+15. Fig 4e-f: It's fine to show the normalized data but can they also show the raw data for \(\%\) of responsive neurons and peak amplitudes for the three groups at baseline (text has results of ANOVA but not actual data)? I ask because there does not appear to be a sustained increase in the \(\%\) of VIP neurons that respond to FL stimulation after stroke, which means that no new neurons are recruited to respond to FL stim after stroke (DREADDs only maintain the original pool).  
+
+<|ref|>text<|/ref|><|det|>[[115, 788, 875, 893]]<|/det|>
+RESPONSE: As requested by the reviewer, we have now plotted this raw baseline data in Supp. Fig. 3a- c. The reviewer is correct in that there wasn't a significant or sustained increase in the \(\%\) responsive VIP neurons after stroke in CNO treated mice relative to pre- stroke. As shown in Fig. 4e, if this were true we would have denoted the 1 week data with a # symbol, which represents statistical comparisons of post- stroke values vs. pre- stroke. We agree with the reviewer's conclusion that new forelimb neurons are not recruited after stroke and had previously stated this in the abstract and discussion. We have now included
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 863, 123]]<|/det|>
+the following conclusion statement in the results on page 12 to make this point clear: "Further, our data argue against the idea that new forelimb responsive neurons are recruited after stroke."  
+
+<|ref|>text<|/ref|><|det|>[[115, 153, 875, 188]]<|/det|>
+16. The effects on VIP responsivity are mainly seen at 1-wk post-stroke, but behavioral effects are seen 2-7 weeks after stroke...the discussion touches on this but more could said to explain this difference.  
+
+<|ref|>text<|/ref|><|det|>[[115, 199, 872, 336]]<|/det|>
+RESPONSE: We agree that this is an interesting finding but at this point in time, we can only speculate why the behavioural effects lag by 1 week behind the VIP response differences. It could be that the post- synaptic targets of VIP neurons, or ones further downstream (eg. pyramidal neurons) are slightly slower to restore intrinsic excitability or synaptic drive. As mentioned by reviewer 1, we think future studies imaging the downstream targets of VIP neurons or whole cell recordings in brain slices in defined excitatory and inhibitory circuits would help resolve this question (see new text added to page 14 of the discussion). However, these experiments would require years of work and therefore are best reserved for future study.  
+
+<|ref|>text<|/ref|><|det|>[[115, 365, 878, 436]]<|/det|>
+17. Fig 5b: Did the relative proportion of each of the 3 subtypes change over time (from baseline) in sham controls and stroke animals? There should be a graph to represent that. Also, shouldn't there be a Chi-square test for all the comparisons and then follow-up 2x2 chi-sq for individual comparisons with post-hoc correction. A better description of the Chi-sq methods would be useful  
+
+<|ref|>text<|/ref|><|det|>[[114, 448, 881, 775]]<|/det|>
+RESPONSE: Although time dependent changes in the relative proportion of each of the 3 subtypes is depicted and analysed in Fig. 5b, we agree with the reviewer that an additional graph (perhaps simplified) showing the proportion of each subtype within each experimental group, would be informative. Therefore we have added a new graph into Figure 5 (see new Fig. 5d) that shows the \(\%\) of each subtype before stroke compared to after stroke (values for weeks 1- 3 averaged together, similar to Fig. 5b) in each of the 3 experimental groups. We now report on page 11- 12 of the results that the proportion of each subtype does not change significantly in the sham stroke controls or stroke affected mice that received chemogenetic stimulation (Note: this data further supports the idea that chemogenetic stimulation was beneficial). This conclusion was based on a chi-squared analysis where we used pre-stroke as the "expected" proportion for each neuron subtype and then compared that to the "observed" proportion averaged over 3 weeks after stroke (all \(\chi^2\) values ranged from 0.01- 0.38 and therefore none were close to significant). By contrast, stroke mice that received control stimulation exhibited significantly fewer highly active neurons ( \(\chi^2 = 4.91\) , \(*p< 0.05\) ) and more minimally active neurons after stroke ( \(\chi^2 = 13.73\) , \(**p< 0.01\) ). This finding is consistent with the idea that stroke has a dampening effect on sensory responses. Since we compared the \(\%\) cells within each experimental group (or \(\%\) neuron sub- type) and analysed the average values over 3 weeks post- stroke rather than compare each week, we believe the direct statistical comparison to Sham stroke or Pre- stroke (as shown in Fig. 5b and Fig. 5d, respectively) was justified. We have added further clarity in the statistics section of the Methods (page 27) regarding how we calculated the chi- square statistic.  
+
+<|ref|>text<|/ref|><|det|>[[115, 804, 863, 857]]<|/det|>
+18. Fig 6a: the issue of how predictable (or should it be 'reliable'?) neurons are seems important, but this visual representation is a bit hard to follow. Another way to show this would be to plot, for each cell, the \(\%\) of stimulations it responds to at baseline and over time after stroke.  
+
+<|ref|>text<|/ref|><|det|>[[115, 869, 861, 903]]<|/det|>
+RESPONSE: We agree this result was not the easiest to present or describe, but think it is quite important. We tried many different graphs/approaches and asked our colleagues what they thought was
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 852, 125]]<|/det|>
+easiest/simplest to interpret. The graph presented in Fig. 6a was the “winning” graph and therefore we wish to keep it as presented.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 293, 98]]<|/det|>
+Reviewers' Comments:  
+
+<|ref|>text<|/ref|><|det|>[[118, 113, 251, 126]]<|/det|>
+Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[118, 127, 300, 140]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[118, 140, 866, 169]]<|/det|>
+The author have adequately addressed all critical points and clarified their observations according to the comments provided. The manuscript is largely improved.  
+
+<|ref|>text<|/ref|><|det|>[[118, 210, 222, 222]]<|/det|>
+Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[118, 224, 300, 237]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[118, 238, 828, 266]]<|/det|>
+The authors have addressed all my previous questions. This is an excellent manuscript and I support publication of this interesting work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 267, 228, 280]]<|/det|>
+Sergei A Kirov  
+
+<|ref|>text<|/ref|><|det|>[[118, 322, 222, 335]]<|/det|>
+Reviewer #3:  
+
+<|ref|>text<|/ref|><|det|>[[118, 337, 300, 350]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[118, 351, 864, 379]]<|/det|>
+The authors have meticulously responded to all of the concerns and comments I had raised and I have no further issues. I congratulate the authors on a beautiful and compelling study.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 569, 304, 585]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 606, 394, 621]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 624, 880, 659]]<|/det|>
+The author have adequately addressed all critical points and clarified their observations according to the comments provided. The manuscript is largely improved.  
+
+<|ref|>text<|/ref|><|det|>[[115, 697, 394, 713]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 716, 848, 750]]<|/det|>
+The authors have addressed all my previous questions. This is an excellent manuscript and I support publication of this interesting work.  
+
+<|ref|>text<|/ref|><|det|>[[115, 753, 219, 768]]<|/det|>
+Sergei A Kirov  
+
+<|ref|>text<|/ref|><|det|>[[115, 817, 394, 833]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 863, 870]]<|/det|>
+The authors have meticulously responded to all of the concerns and comments I had raised and I have no further issues. I congratulate the authors on a beautiful and compelling study.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__43f57920d0cc1b4b5cc7738208cf1e5082ada58cf4830c9ce4fa16eabc64f49e/images_list.json

@@ -0,0 +1,317 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 Performance of the high-throughput SPH with unstained tissue from mouse brains in high-resolution mode. (a) Reconstructed phase image of a quantitative phase target. The upper inset: the corresponding one-dimensional (1D) profile of element 6 of group 6; the lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of 0.104 rads. (b) The image of a slice of unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (c)-(e) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        184,
+        85,
+        810,
+        450
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6 Reconstruction of holographic images for the unstained tissue from mouse brains with compressive sensing. The amplitude and wrapped phase images are reconstructed with different sampling ratios of \\(50\\%\\) , \\(25\\%\\) , \\(12.5\\%\\) , \\(6.25\\%\\) , and \\(3.125\\%\\) . The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        603,
+        833,
+        840
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_9.jpg",
+    "caption": "Fig. 9 Holographic performance of unstained mouse brain tissue with \\(100 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of \\(100 - \\mu \\mathrm{m}\\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        222,
+        235,
+        775,
+        504
+      ]
+    ],
+    "page_idx": 8
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_10.jpg",
+    "caption": "Fig. 10 Holographic performance of unstained mouse brain tissue with 120-μm thickness. (a) The image of a slice of 120-μm-thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is 200 μm.",
+    "footnote": [],
+    "bbox": [
+      [
+        216,
+        85,
+        772,
+        357
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_11.jpg",
+    "caption": "Fig. 11 Holographic performance of unstained mouse brain tissue with 10-μm thickness. (a) The image of a slice of 10-μm-thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is 200 μm.",
+    "footnote": [],
+    "bbox": [
+      [
+        220,
+        508,
+        775,
+        780
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_8.jpg",
+    "caption": "Fig. 8 Holographic performance of stained mouse tail tissue with \\(10 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of a \\(10 - \\mu \\mathrm{m}\\) -thick stained tissue from mouse tails, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        228,
+        179,
+        775,
+        450
+      ]
+    ],
+    "page_idx": 10
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6 Holographic results of imaging a quantitative phase resolution target under high-resolution mode. (a) The amplitude image of the quantitative phase resolution target, showing poor contrast. (b) The phase image of the quantitative phase resolution target, showing clear phase patterns. Upper inset: the corresponding one-dimensional profile of element 6 of group 6 (4.386-μm width). Lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of about 0.104 rads. The corresponding scale bar is \\(100 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        195,
+        179,
+        790,
+        377
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Fig. 7 Holographic results of imaging the phase resolution target with compressive sensing. (a)(b) The amplitude and phase images reconstructed with different sampling ratios of \\(50\\%\\) , \\(25\\%\\) , \\(12.5\\%\\) , \\(6.25\\%\\) , and \\(3.125\\%\\) . The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        188,
+        480,
+        802,
+        697
+      ]
+    ],
+    "page_idx": 15
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 Performance of the high-throughput SPH with unstained tissue from mouse brains in high-resolution mode. (a) Reconstructed phase image of a quantitative phase target. The upper inset: the corresponding one-dimensional (1D) profile of element 6 of group 6; the lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of 0.104 rads. (b) The image of a slice of unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (c)-(e) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        184,
+        87,
+        810,
+        448
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6 Reconstruction of holographic images for the unstained tissue from mouse brains with compressive sensing. The amplitude and wrapped phase images are reconstructed with different sampling ratios of \\(50\\%\\) , \\(25\\%\\) , \\(12.5\\%\\) , \\(6.25\\%\\) , and \\(3.125\\%\\) . The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        603,
+        833,
+        839
+      ]
+    ],
+    "page_idx": 17
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_9.jpg",
+    "caption": "Fig. 9 Holographic performance of unstained mouse brain tissue with \\(100 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of \\(100 - \\mu \\mathrm{m}\\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        222,
+        236,
+        773,
+        504
+      ]
+    ],
+    "page_idx": 18
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_10.jpg",
+    "caption": "Fig. 10 Holographic performance of unstained mouse brain tissue with \\(120 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of \\(120 - \\mu \\mathrm{m}\\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \\(200\\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        217,
+        85,
+        773,
+        357
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_11.jpg",
+    "caption": "Fig. 11 Holographic performance of unstained mouse brain tissue with \\(10 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of \\(10 - \\mu \\mathrm{m}\\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \\(200\\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        220,
+        508,
+        773,
+        779
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Fig. 3 Simulation results on compressive sensing under different noise levels. (a) Correlations between the reconstructed and original images as a function of different noise levels. Compressive sensing realized with different sampling ratios (SRs) were investigated, which are labeled with different colors. The yellow area represents the actual noise levels during experiments. (b) After fixing the noise level as \\(0.1\\%\\) of the measurement value, the SR-resultant effect and the noise-induced effect are represented using blue and red circles as a function of the SR. The combined effect is represented using yellow circles, exhibiting a maximum value at \\(\\mathrm{SR} = 25\\%\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        168,
+        586,
+        820,
+        765
+      ]
+    ],
+    "page_idx": 21
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6 Holographic results of imaging a quantitative phase resolution target under high-resolution mode. (a) The amplitude image of the quantitative phase resolution target, showing poor contrast. (b) The phase image of the quantitative phase resolution target, showing clear phase patterns. Upper inset: the corresponding one-dimensional profile of element 6 of group 6 (4.386- \\(\\mu \\mathrm{m}\\) width). Lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of about 0.104 rads. The corresponding scale bar is \\(100\\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        193,
+        494,
+        808,
+        692
+      ]
+    ],
+    "page_idx": 22
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_7.jpg",
+    "caption": "Fig. 7 Holographic results of imaging the phase resolution target with compressive sensing. (a)(b) The amplitude and phase images reconstructed with different sampling ratios of \\(50\\%\\) , \\(25\\%\\) , \\(12.5\\%\\) , \\(6.25\\%\\) , and \\(3.125\\%\\) . The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        189,
+        83,
+        808,
+        303
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_5.jpg",
+    "caption": "Fig. 5 Performance of the high-throughput SPH with unstained tissue from mouse brains in high-resolution mode. (a) Reconstructed phase image of a quantitative phase target. The upper inset: the corresponding one-dimensional (1D) profile of element 6 of group 6; the lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of 0.104 rads. (b) The image of a slice of unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (c)-(e) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        184,
+        85,
+        810,
+        450
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_6.jpg",
+    "caption": "Fig. 6 Reconstruction of holographic images for the unstained tissue from mouse brains with compressive sensing. The amplitude and wrapped phase images are reconstructed with different sampling ratios of \\(50\\%\\) , \\(25\\%\\) , \\(12.5\\%\\) , \\(6.25\\%\\) , and \\(3.125\\%\\) . The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        163,
+        602,
+        833,
+        838
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_9.jpg",
+    "caption": "Fig. 9 Holographic performance of unstained mouse brain tissue with \\(100 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of \\(100 - \\mu \\mathrm{m}\\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \\(200 \\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        222,
+        255,
+        774,
+        525
+      ]
+    ],
+    "page_idx": 25
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_10.jpg",
+    "caption": "Fig. 10 Holographic performance of unstained mouse brain tissue with \\(120 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of \\(120 - \\mu \\mathrm{m}\\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \\(200\\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        216,
+        85,
+        773,
+        357
+      ]
+    ],
+    "page_idx": 26
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_11.jpg",
+    "caption": "Fig. 11 Holographic performance of unstained mouse brain tissue with \\(10 - \\mu \\mathrm{m}\\) thickness. (a) The image of a slice of \\(10 - \\mu \\mathrm{m}\\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \\(200\\mu \\mathrm{m}\\) .",
+    "footnote": [],
+    "bbox": [
+      [
+        219,
+        508,
+        774,
+        780
+      ]
+    ],
+    "page_idx": 26
+  }
+]

peer_reviews/supplementary_0_Peer Review File__43f57920d0cc1b4b5cc7738208cf1e5082ada58cf4830c9ce4fa16eabc64f49e/supplementary_0_Peer Review File__43f57920d0cc1b4b5cc7738208cf1e5082ada58cf4830c9ce4fa16eabc64f49e.mmd

@@ -0,0 +1,547 @@
+
+Reviewers' Comments:  
+
+Reviewer #1: Remarks to the Author:  
+
+The authors present an optical heterodyne detection scheme for single- pixel holographic imaging. Single- pixel holography is not new (see for example references 13- 20 in the manuscript) but the role of heterodyne detection has not been studied in depth previously.  
+
+As a matter of fact, frequency beating between the two arms of the Mach- Zehnder interferometer enables phase sensing using the amplitude binary codification natural in digital micromirror devices. Neither phase- shifting (which enlarges signal acquisition), nor computer- generated hologram codification (which sacrifices spatial resolution) are compromised. Based on the above architecture, there is a claim for an improvement in the space- bandwidth- time product.  
+
+However, the above trade- off has been explored several times in the past for single- pixel imaging, both at the hardware and the software level. In Sci Rep. 8:2369 (2018) video operation at 30 frames/sec for a 128x128 image at a frame rate of 30 frames/sec is demonstrated. Extrapolation to single- pixel holography is straightforward.  
+
+Switching between large field- of- view and high- resolution mode is also claimed. However this is done at the expense of a hardware modification of the 4f optical setup which prevents real- time operation.  
+
+As the authors mention, the use of compressive sensing (the possibility of compressing during the measurement process) is an outstanding feature of single- pixel holography. The theory of compressive sensing relies on two principles: the sparsity of the signal of interest, and the incoherence between the bases employed for the measurement and the reconstruction. It should be convenient that the authors clarify if the "incoherence" condition is essential for making compressive sensing work.  
+
+Concerning, parameters in Table 1 about state- of- the art in single- pixel imaging and single- pixel holography I miss some important papers such as Opt. Express 28, 28190- 28208 (2020) and Nat. Photon. 13, 13- 20 (2019). Also, I find convenient to include some references for the single- pixel imaging () or for non- inteferometric single- pixel phase imaging (Optica 5, 164- 174 (2018)). Finally, adaptive and smart sensing should be considered for a fair comparison between the spatial and temporal trade- off in single- pixel imaging.  
+
+In summary, the approach and the theoretical description are good. However, all of the results can easily been derived from the combination of two well- established techniques: single- pixel holography and heterodyne holography. Also, the quality of experimental implementation should be improved to support the claims in the abstract concerning biological applications.  
+
+In this sense, everything is highly predictable and the paper lacks of the novelty required for publication in Nature Communications.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+In this paper, Wu et al. report single- pixel compressive holography for multi- scale imaging. Developed upon heterodyne holography and Hadamard encoding, the reported SPH system achieves a space- bandwidth- time product (SBP- T) of 41,667 pixels/s. Compared with existing prototypes, the reported system delivers superior specifications in throughput, pixel counts, and field of view (in the large FOV mode), as well as has a comparable micrometer- level spatial resolution (in the high- resolution mode). This technique is applied to image a cross- section of a mouse tail. The capability of resolving both amplitude and phase allows this technique to reveal various features on the sample.  
+
+This work can be published in Nature Communications. The main breakthrough in this work is the implementation of heterodyne detection to nullify the necessity of phase- shifting measurement.
+
+<--- Page Split --->
+
+This is a clever trick to overcome the speed limitation imposed by the DMD. The technical specifications of this system are among the best in the field. The authors have also expanded the application of this technique to bio- imaging. The technical innovations included in this work can benefit the relevant communities.  
+
+I have the following comments for the authors to address.  
+
+1. The literature review of single-pixel imaging could be strengthened. As the authors wrote in the introduction, single-pixel cameras have great performance across almost the entire spectrum range. Some previous papers for applying single-pixel imaging in infrared (e.g., Opt. Express 25, 2998-3005 (2017)), THz (e.g., Nat. Commun. 11, 2535 (2020)), and even photoacoustic imaging (e.g., Opt. Lett. 39, 430-433 (2014)) could be included. Besides, a recent trend in single-pixel cameras is 3D imaging. Representative works in this category (e.g., Nat. Commun. 7, 12010 (2016), Opt. Express 28, 29377-29389 (2020), and APL Photonics 5, 020801 (2020)) could be included.  
+
+2. In line 15-17 on Page 3: "For example, we can operate under large-FOV mode (14.9 mm × 11.1 mm) to monitor the environment [9, 10, 21] or switch to high-resolution mode (5.8 μm × 4.3 μm) to scrutinize microstructures [18, 20]." What determines the limit of the FOV and resolution in this work? Have the authors pushed these numbers to the limit already? Could the authors provide some justifications on how to get these numbers? Also, how would the spatial resolution in amplitude and phase be degraded with the compressed ratio?  
+
+3. The authors claimed that the strategy of \(768 \times 768\) pixels with \(3 \times 3\) binning from DMD was performed in holographic imaging. Why not make full use of the entire DMD face with \(768 \times 1024\) pixels and \(3 \times 4\) binning? This strategy should provide a better SNR and imaging quality.  
+
+4. In Fig. 2, lens L7 is regarded as a Fourier transformation of the input field. When loading patterns with high spatial frequency, the output field is no longer a single point but exhibits spatial distribution. Considering the finite sense area of the single-pixel detector, it is possible that not all the output energy is collected. Are there any considerations or calculations to consider this condition?  
+
+5. In the table on Page 3, the column of "refresh time" listed in this work is 48 μs, indicating that the full speed of the DMD (22 kHz) was not fully used. Why not making full use of the 22 kHz refresh rate to push the SBP-T to the limit?  
+
+6. In the provided files "data_and_supporting_files.zip" and "readme.pdf", it is mentioned that "To minimize phase drifting caused by environmental disturbance during the experiment....", could the authors provide a detailed explanation on the reasons for inserting additional Hadamard patterns and how to process these data corresponding to additional patterns?  
+
+7. In Line 17 on Page 4, Eq. (3), the first order of Hadamard-like patterns is simply a DC signal. My understanding is that if this value was not subtracted for each coefficient as the authors did, the final image would have a constant background. Nonetheless, for imaging purposes, a constant background is acceptable.  
+
+8. What's the bandwidth of the photodiode? Specifications of this component are helpful for readers to adopt this technique.  
+
+9. In Fig. 2, the DMD is not parallel to the sample plane. Will this introduce any measurement error or distortion?  
+
+10. The gold standard is missing to cross-check the presented technique. Can the authors use existing techniques (e.g., phase-contrast microscopy) to verify their results? Also, what's the measurement accuracy? This would be another important parameter that would determine the application scope of this technique.  
+
+11. In Fig. 5, the authors used red dashed circles to point out show some supplement information
+
+<--- Page Split --->
+
+in the phase image from the amplitude counterpart. This part could be elaborated on because it showcases the capability of this technique.  
+
+12. A few typos are found in the manuscript. For example, by "mussel", I believe the authors intended to say "muscle".  
+
+Reviewer #3: Remarks to the Author: Dear Authors,  
+
+In this manuscript, a novel technique for single pixel compressive holography is proposed. The proposed technique employs heterodyne interference to achieve high- throughput measurement.  
+
+The authors construct an optical system based on the proposed technique. And experimental verifications are described.  
+
+In the experimental verifications, parts of a USAF char for estimation of spatial resolution and biological tissues are reconstructed to demonstrate the usefulness of the system.  
+
+I have some questions and comments to improve the manuscript.  
+
+(1) In Fig.2 : Why the experimental set up consists of two AOMs?  
+
+I think that heterodyne interferometry can be implemented with only one AOM with easy control. Also, the authors should describe the reason why the beat frequency is set to 62,500 Hz.  
+
+(2) About Fig. 4 (a): the authors should discuss factors that the image quality at sampling ratio \(50\%\) is worse than that in more critical case (25% & 12.5%).  
+
+The discussions are important for judgement which the used algorithm for image reconstruction is Such Discussion are important in determining whether the algorithm used for image reconstruction is practical.  
+
+(3) About holographic imaging:  
+
+To show the effectiveness as holographic imaging, phase objects or three- dimensional objects should be experimentally verified.  
+
+This is because verification for phase imaging with the constructed experimental setup is not enough.  
+
+(On the other handa, That for amplitude reconstruction is quite enough to show the usefulness. )  
+
+(4) In conclusion: some performances of the proposed method are shown.  
+
+Those estimations are much important in the paper.  
+
+The authors should summarize the performance as a table like Table 1 and discuss on relations between scalability and specifications of the elemental devices in detail.  
+
+(5) Forms of reconstructed images in Figs. 3, 4, 5, and 6 : (This comment may be not so essential.)  
+
+Why area of measurement is not square like but diamond shape (rhomboid) ?
+
+<--- Page Split --->
+
+## Manuscript ID: NCOMMS-20-49700  
+
+We thank the editor for organizing the review and all three reviewers for their valuable comments. We have carefully digested the reviewers' comments and largely improved both main text and Supplementary Information in the revised submission.  
+
+The major changes to the manuscript can be summarized as the following points:  
+
+1. We improved the biological application by performing new experiments to image unstained tissue from mouse brains and supplementing corresponding holographic images in the revised submission. Slices of unstained mouse brains with various thicknesses, including \(10 \mu \mathrm{m}\) , \(80 \mu \mathrm{m}\) , \(100 \mu \mathrm{m}\) , and \(120 \mu \mathrm{m}\) , were imaged. These unstained slices of biological tissue show relatively poor contrast in amplitude but exhibit rich information with good contrast in phase. These new experimental results demonstrate the capability of the high-throughput single-pixel holography to image biological tissue that contains rich information in phase.  
+
+2. We improved the biological application by performing additional experiments to image another slice of stained tissue from mouse tails. Corresponding holographic images are provided in Supplementary Information of the revised submission.  
+
+3. We added a new experiment to quantify the performance of the holographic system in phase by imaging a quantitative phase resolution target. Using the obtained holographic images, we quantified the spatial resolution in phase, confirming its agreement with the theoretical resolution of the system. We also compared the reconstructed phase value to the actual phase value of the resolution target. The phase error was quantified to be about \(0.104\) rads \((\leq \lambda /60)\) . This new experimental result demonstrates the high-throughput single-pixel holography can quantitatively image phase objects.  
+
+4. We added a Table to summarize and compare the parameters of different operational modes, i.e., the large field-of-view (FOV) mode and the high-resolution mode, employed in this work.  
+
+5. We performed numerical simulations to study the sampling ratio of compressive sensing under the influence of measurement noises. Both numerical and experimental results indicate that the SR of \(25\%\) is likely to be the suitable choice under the current experimental condition.  
+
+6. We added detailed descriptions to explain the choice of beat frequency and the operational condition of the digital micromirror device.  
+
+7. We added a paragraph to describe the preparation procedures of these biological samples. We added Mr. Rinsen Zhang as a coauthor, who assisted in the preparation of these biological samples. Additional fundings that supported the experiments during the revision process were added in the acknowledgment.  
+
+8. Since new experimental results to image biological samples were added, we slightly modified corresponding descriptions in the abstract and discussion section. We also added the sections of Data availability, Code availability, Acknowledgments, Author contributions, Competing interests to fulfill the requirement of the journal.  
+
+A point- by- point response to the reviewers' comments is provided in the following.
+
+<--- Page Split --->
+
+## Point-by-point responses to reviewers' comments  
+
+## Reviewer #1:  
+
+The authors present an optical heterodyne detection scheme for single- pixel holographic imaging. Single- pixel holography is not new (see for example references 13- 20 in the manuscript) but the role of heterodyne detection has not been studied in depth previously.  
+
+1. As a matter of fact, frequency beating between the two arms of the Mach-Zehnder interferometer enables phase sensing using the amplitude binary codification natural in digital micromirror devices. Neither phase-shifting (which enlarges signal acquisition), nor computer-generated hologram codification (which sacrifices spatial resolution) are compromised. Based on the above architecture, there is a claim for an improvement in the space-bandwidth-time product. However, the above trade-off has been explored several times in the past for single-pixel imaging, both at the hardware and the software level. In Sci Rep. 8:2369 (2018) video operation at 30 frames/sec for a \(128 \times 128\) image at a frame rate of 30 frames/sec is demonstrated. Extrapolation to single-pixel holography is straightforward.  
+
+Response: We thank the reviewer for this comment and for suggesting this important reference. This reference reports real- time single- pixel imaging that provides intensity- based images. The claimed video frame rate (30 frames- per- second) with \(128 \times 128\) pixels is enabled by compressive sensing with a compression ratio of only \(2\%\) . A deep convolutional auto- encoder network (DCAN) was designed to optimize the sequence of the Hadamard basis and reconstruction model. Thus, the main achievement of this reference is to facilitate the imaging reconstruction process, which is completely different from the main purpose of our work (high- throughput). Moreover, for imaging acquisition, the throughput, i.e., the space- bandwidth- time product, of the system developed in this reference is solely determined by the parameters of the digital micromirror device, which is the same as those reported in previous works listed in Table 1 of the revised main text. Nonetheless, the algorithm developed in the suggested reference can be borrowed into our work to facilitate imaging reconstruction. Given enough training data with holographic images of biological samples in the future, we envision that this deep learning assisted reconstruction algorithm can effectively shorten the time cost for imaging reconstruction of our developed single- pixel holography. Thus, we added this reference and discussed this possibility of facilitating reconstruction speed using deep learning in the Discussion section of the revised main text: Due to the huge computational burden of imaging reconstruction, real- time imaging of live cells with time- varying features is still unachievable with the current holographic system. We envision that this constraint can be potentially alleviated through the assistance of deep learning and compressive sensing [62].  
+
+2. Switching between large field-of-view and high-resolution mode is also claimed. However, this is done at the expense of a hardware modification of the 4f optical setup which prevents real-time operation.  
+
+Response: We thank the reviewer for this comment and agree that the current holographic system requires modifying the 4f optical setup to switch between the large field-of-view mode and the high-resolution mode. In this work, we did not claim real-time operations that can switch between different operational modes. This kind of hardware modification is not unique to our system but is commonly seen in many microscopes as well. In addition, this hardware limitation can be conquered by using tunable lenses and an advanced electromechanical system with auto-focusing in future works. To address this issue, we added the following descriptions in the Method section of the revised main text to clarify this point: Currently, the switching between the large FOV mode and the high-resolution mode requires hardware modification of the 4f optical setup, which is inconvenient and prevents real-time operation. Nonetheless, we anticipate that a variable 4f setup that consists of tunable lenses and an advanced electromechanics system with auto-focusing can enable smooth adjustment between different operational modes in the future.  
+
+3. As the authors mention, the use of compressive sensing (the possibility of compressing during the measurement process) is an outstanding feature of single-pixel holography. The theory of compressive sensing relies on two principles: the sparsity of the signal of interest, and the incoherence between the bases employed for the measurement and the reconstruction. It should be convenient that the
+
+<--- Page Split --->
+
+authors clarify if the “incoherence” condition is essential for making compressive sensing work.  
+
+Response: We thank the reviewer for this valuable suggestion and we take the liberty to assume the incoherence is defined using formula 1.7 in the paper “Sparsity and incoherence in compressive sampling, Inverse Problems 23, 969, 2007”, written by Emmanuel Candès and Justin Romberg. Here, incoherence is the estimation for correlation between the bases for measurement (sensing modality) and reconstruction (signal model). There are two major approaches to realize compressive sensing. The first one is to employ an iterative converging algorithm with different kinds of regularization factors, such as the L1- norm and total variation (TV) factor. For this case, the employed bases for measurement and reconstruction can be either “coherence” or “incoherence”. When no prior information of the sample was provided, a random ordering was preferred, and “coherence” should generally provide good reconstruction results. Nonetheless, when an optimized ordering for a sparse signal is known prior, “incoherence” could lead to better reconstruction results with a small number of measurements. The second one is to use the same set of bases for measurement and reconstruction. An inverse transformation like inverse Hadamard transformation or inverse Fourier transformation is employed to reconstruct the image. By definition, compressive sensing that follows this rule can be categorized as “incoherence”. Our work belongs to the second approach of compressive sensing, which employs the same set of Hadamard bases for both measurement and reconstruction. A given ordering strategy of the square path was adopted to simplify the reconstruction process. Therefore, the “incoherence” condition is essential to make compressive sensing work in our case. To clarify this point, we added the following sentence in the Results section of the revised main text: “In this work, an ordering of square sampling path (detailed in the Method section) was applied for sampling, and a direct inverse fast Hadamard transformation using the same set of bases was adopted for reconstruction. Thus, incoherence is confirmed between the bases employed for the measurement and the reconstruction, which is essential to make compressive sensing work for our holographic system [61].”  
+
+4. Concerning, parameters in Table 1 about state-of-the art in single-pixel imaging and single-pixel holography I miss some important papers such as Opt. Express 28, 28190-28208 (2020) and Nat. Photon. 13, 13-20 (2019). Also, I find convenient to include some references for the single-pixel imaging () or for non-interferometric single-pixel phase imaging (Optica 5, 164-174 (2018)). Finally, adaptive and smart sensing should be considered for a fair comparison between the spatial and temporal trade-off in single-pixel imaging.  
+
+Response: We thank the reviewer for supplementing these important references. We studied the two suggested review articles on this topic (Opt. Express 28, 28190- 28208 (2020) and Nat. Photon. 13, 13- 20 (2019)). Both references were added in the revised main text as important references to prepare Tab. 1: Based on the review articles covering this topic [54, 55], Tab. 1 summarizes the performance of some representative works about SPI [8, 12- 15, 24] and SPH [16- 23, 25, 26] reported in the literature.  
+
+The non- interferometric single- pixel phase imaging (Optica 5, 164- 174 (2018)) was also added as an important reference in the Introduction section of the revised main text: Borrowing the concept of the Shack- Hartmann sensor, a non- interferometric phase image was demonstrated using a single- pixel detector [32]. However, the special requirement of using a lateral position detector sacrifices the advantages that conventional SPH holds.  
+
+As suggested by the reviewer, we also added a description to discuss adaptive and smart sensing for the spatial and temporal trade- off in single- pixel imaging in the Introduction section of the revised main text: Recently, adaptive and smart sensing with dynamic supersampling was reported to combine with compressive sensing in SPI. The enabling feature of this approach is to rapidly record fast- changing features by dynamically adapting to the evolution of the scene. Thus, it significantly shortens acquisition time without considerably sacrificing spatial information [31]. Since the performance of adaptive and smart sensing largely depend on the sparsity and types of samples, Tab. 1 only compares representative works on SPI and SPH from the perspective of the system (or hardware) rather than the algorithm. This point was clarified in the Introduction section of the revised main text: “We note here that compressive sensing, including adaptive and smart sensing, is effective in breaking the spatial and temporal trade- off for most SPI and SPH systems. However, the improvement in SBP- T is ambiguous to be quantified, especially when the target sample is not specified. Therefore, compressive sensing was not considered when estimating the parameters displayed in Tab. 1 for a fair comparison.”
+
+<--- Page Split --->
+
+5. In summary, the approach and the theoretical description are good. However, all of the results can easily been derived from the combination of two well-established techniques: single-pixel holography and heterodyne holography. Also, the quality of experimental implementation should be improved to support the claims in the abstract concerning biological applications. In this sense, everything is highly predictable and the paper lacks of the novelty required for publication in Nature Communications.  
+
+Response: We thank the reviewer for the recognition of the approach and the theoretical description of this work. We largely improved the claimed biological applications in the revised submission, by adding new experiments to imaging slices of both stained mouse tails and unstained mouse brains with various thicknesses. High-resolution holographic imaging results of these biological samples reveal rich information in both amplitude and phase, showing the prospect of our work. We believe that these newly supplement holographic results can support the claim of biological applications.  
+
+Detailed descriptions on imaging unstained biological tissue were added as a new paragraph "Holographic imaging of unstained tissue from mouse brains" in the Results section of revised main text and a section of "Additional holographic images of unstained tissue from mouse brains" in revised Supplementary Information.  
+
+In revised main text  
+
+Holographic imaging of unstained tissue from mouse brains. Having demonstrated holographic imaging of stained tissue from mouse tails, we then switched to image unstained tissue from mouse brains. Generally, images of unstained tissue exhibit low contrast in amplitude, but providing sufficient contrast through phase imaging. Before proceeding, we quantified the performance of the holographic system in phase by imaging a quantitative phase resolution target (QPT, BenchMark Tec). Figure 5(a) shows the reconstructed phase image of the resolution target, containing groups 6 and 7. To quantify the resolving power in phase, the corresponding 1D profile denoted by the black bracket within the resolution target, crossing element 6 of group 6 (4.386- μm width), is shown in the upper inset. Three verticle bars can be differentiated, indicating that the theoretical resolution under the high- resolution mode (5.80 μm × 4.31 μm) was achieved in phase. We further quantified the accuracy of the measurement in phase by estimating the phase difference between the bar enclosed by a dashed box at the bottom of the target and its adjacent background. 1D profile that crosses both the bar and the background is shown in the lower inset using blue circles, exhibiting a stepped structure. The averaged phase difference was estimated to be \(\Delta \phi_{\mathrm{exp}} \approx 1.691\) rads, yielding a phase error of only 0.104 rads ( \(\leq \lambda /60\) ) compared to the actual phase difference (1.795 rads). Detailed analysis on imaging the quantitative phase resolution target can be found in section VII of Supplementary Information. Then, we proceed to image unstained biological tissue. Figure 5(b) shows the image of an 80- μm- thick slice of unstained tissue from mouse brains, captured using a bright- field microscope. In this image, some representative types of tissue were denoted with black arrows, such as piriform cortex, perirhinal cortex, olivary pretectal nucleus, and white matter. For unstained tissue, it is usually challenging to identify different structures in amplitude, due to the insufficient contrast in transmission. Figures 5(c), (d), and (e) show a series of reconstructed holographic images for different parts of the mouse brain, indicated by three labeled diamond- shaped boxes in Fig. 5(b). Same as the one captured using the conventional microscope, these amplitude images only manifest rough outlines with low contrast. By comparison, phase images show rich details that can be hardly seen in their amplitude counterparts. For example, the piriform cortex and the perirhinal cortex are identifiable in phase images. Data used for producing Figs. 5(c)-(e) are provided in Supplementary Data 2. These results demonstrate the capability of employing SPH to image relatively transparent biological tissue.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Fig. 5 Performance of the high-throughput SPH with unstained tissue from mouse brains in high-resolution mode. (a) Reconstructed phase image of a quantitative phase target. The upper inset: the corresponding one-dimensional (1D) profile of element 6 of group 6; the lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of 0.104 rads. (b) The image of a slice of unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (c)-(e) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+For the selected area in Fig. 5(c), a series of holographic images reconstructed with compressive sensing at various SRs are illustrated in Fig. 6 as well. Similar to before, these images show the effectiveness of compressive sensing in dealing with biological tissue that contains rich information in phase. We also applied SPH to image other pieces of unstained tissues from mouse brains with different thicknesses, ranging from \(10 \mu \mathrm{m}\) to \(120 \mu \mathrm{m}\) , with details shown in section IX of Supplementary Information.  
+
+![](images/Figure_6.jpg)
+
+<center>Fig. 6 Reconstruction of holographic images for the unstained tissue from mouse brains with compressive sensing. The amplitude and wrapped phase images are reconstructed with different sampling ratios of \(50\%\) , \(25\%\) , \(12.5\%\) , \(6.25\%\) , and \(3.125\%\) . The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+In revised Supplementary Information  
+
+Additional holographic images of unstained tissue from mouse brains
+
+<--- Page Split --->
+
+To show the imaging capability for unstained tissue, we also imaged a \(100 - \mu \mathrm{m}\) - thick slice of unstained mouse brain. Figure 9(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part of the tissue contains white matter and grey matter. Since the unstained slice is relatively thick with roughly the same transmission across the imaging region, it is challenging for the conventional microscope to provide good contrast. Figures 9(b), (c), and (d) show a series of reconstructed holographic images for different parts of the mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 9(a). As expected, the features in amplitude images match well with that at corresponding areas in Fig. 9(a), which do not provide too much information with good contrast. As a comparison, phase images provide much better contrast, revealing many detailed structures that are indiscernible in their amplitude counterparts.  
+
+![](images/Figure_9.jpg)
+
+<center>Fig. 9 Holographic performance of unstained mouse brain tissue with \(100 - \mu \mathrm{m}\) thickness. (a) The image of a slice of \(100 - \mu \mathrm{m}\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+We also imaged an even thicker slice of unstained mouse brain with a thickness of \(120 \mu \mathrm{m}\) . Figure 10(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part contains the ventral part of the lateral septal nucleus, inferior colliculus, and white matter. Figures 10(b), (c), and (d) show a series of reconstructed holographic images for different parts of mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 10(a). Again, phase images provide much better contrast, compared to their amplitude counterparts.
+
+<--- Page Split --->
+![](images/Figure_10.jpg)
+
+<center>Fig. 10 Holographic performance of unstained mouse brain tissue with 120-μm thickness. (a) The image of a slice of 120-μm-thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is 200 μm. </center>  
+
+Very thin unstained tissue from mouse brains down to \(10 - \mu \mathrm{m}\) thick was also imaged using our holographic system. Figure 11(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part contains white matter and grey matter. Figures 11(b), (c), and (d) show a series of reconstructed holographic images for different parts of mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 11(a). For such a thin tissue, it is hard to see details through the amplitude images. Nonetheless, phase images still provide rich information with good contrast.  
+
+![](images/Figure_11.jpg)
+
+<center>Fig. 11 Holographic performance of unstained mouse brain tissue with 10-μm thickness. (a) The image of a slice of 10-μm-thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is 200 μm. </center>  
+
+Detailed descriptions on imaging additional slices of stained biological tissue from mouse tails were added as a new section "Additional holographic images of stained tissue from mouse tails" in revised Supplementary Information.  
+
+## Additional holographic images of stained tissue from mouse tails  
+
+To further demonstrate the imaging capability for stained biological tissue, we imaged another piece of slice from mouse tails,
+
+<--- Page Split --->
+
+which is \(10 - \mu \mathrm{m}\) thick. Figure 8 shows a bright- field image of this slice, captured using a conventional microscope. In this image, several types of tissue such as muscle, cortical bone, and cancellous bone are visually identified. Figures 8(b), (c), and (d) show a series of reconstructed holographic images for different parts of mouse tail. Again, for this stained slice, the amplitude images are in good agreement with the one shown in Fig. 8(a), manifesting great distinctions among different types of tissue. Like the one presented in Fig. 3 of main text, the reconstructed phase images are analogous to their amplitude counterparts.  
+
+![](images/Figure_8.jpg)
+
+<center>Fig. 8 Holographic performance of stained mouse tail tissue with \(10 - \mu \mathrm{m}\) thickness. (a) The image of a slice of a \(10 - \mu \mathrm{m}\) -thick stained tissue from mouse tails, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>
+
+<--- Page Split --->
+
+## Reviewer #2:  
+
+In this paper, Wu et al. report single- pixel compressive holography for multi- scale imaging. Developed upon heterodyne holography and Hadamard encoding, the reported SPH system achieves a space- bandwidth- time product (SBP- T) of 41,667 pixels/s. Compared with existing prototypes, the reported system delivers superior specifications in throughput, pixel counts, and field of view (in the large FOV mode), as well as has a comparable micrometer- level spatial resolution (in the high- resolution mode). This technique is applied to image a cross- section of a mouse tail. The capability of resolving both amplitude and phase allows this technique to reveal various features on the sample. This work can be published in Nature Communications. The main breakthrough in this work is the implementation of heterodyne detection to nullify the necessity of phase- shifting measurement. This is a clever trick to overcome the speed limitation imposed by the DMD. The technical specifications of this system are among the best in the field. The authors have also expanded the application of this technique to bio- imaging. The technical innovations included in this work can benefit the relevant communities. I have the following comments for the authors to address.  
+
+1. The literature review of single-pixel imaging could be strengthened. As the authors wrote in the introduction, single-pixel cameras have great performance across almost the entire spectrum range. Some previous papers for applying single-pixel imaging in infrared (e.g., Opt. Express 25, 2998-3005 (2017)), THz (e.g, Nat. Commun. 11, 2535 (2020)), and even photoacoustic imaging (e.g., Opt. Lett. 39, 430-433 (2014)) could be included. Besides, a recent trend in single-pixel cameras is 3D imaging. Representative works in this category (e.g., Nat. Commun. 7, 12010 (2016), Opt. Express 28, 29377-29389 (2020), and APL Photonics 5, 020801 (2020)) could be included.  
+
+Response: We thank the reviewer for providing these important references. We included them in the Introduction section of the revised main text.  
+
+a. Opt. Express 25, 2998-3005 (2017), Nat. Commun. 11, 2535 (2020), and Opt. Lett. 39, 430-433 (2014) Enabled by this property, SPI has been demonstrated with great success when operating with infrared light [1], Terahertz wave [2], and even photoacoustic signal [3].  
+
+b. Nat. Commun. 7, 12010 (2016), Opt. Express 28, 29377-29389 (2020), and APL Photonics 5, 020801 (2020) By employing various coding mechanisms including Hadamard bases [8, 11-23], Fourier bases [12, 24-26], and random patterns [27], SPI has also been extended and demonstrated with applications in full-color imaging [13], multispectral imaging [14], time-resolved imaging [15], and three-dimensional imaging [28-30].  
+
+2. In line 15-17 on Page 3: "For example, we can operate under large-FOV mode \((14.9\mathrm{mm}\times 11.1\mathrm{mm})\) to monitor the environment [9, 10, 21] or switch to high-resolution mode \((5.8\mu \mathrm{m}\times 4.3\mu \mathrm{m})\) to scrutinize microstructures [18, 20]." What determines the limit of the FOV and resolution in this work? Have the authors pushed these numbers to the limit already? Could the authors provide some justifications on how to get these numbers? Also, how would the spatial resolution in amplitude and phase be degraded with the compressed ratio?  
+
+Response: We thank the reviewer for raising this question. The field- of- view (FOV) and the resolution of the holographic system operating in different modes are determined by the parameters of the two lenses in the 4f system and the physical size of the DMD. We added a section of "Parameters in the large- FOV mode and the high- resolution mode" in revised Supplementary Information to show the detailed analysis on evaluating these numbers.  
+
+## Parameters in the large-FOV mode and the high-resolution mode  
+
+During experiments, we employed a square area with \(768 \times 768\) effective pixels of the DMD. Since the pitch of the DMD is \(19.35\mu \mathrm{m}\) (calculated above), the length of the diagonal lines along both the horizontal and vertical directions for the generated pattern are the same to be \(19.35\mu \mathrm{m} \times 768 \approx 14.9\mathrm{mm}\) .
+
+<--- Page Split --->
+
+For the large- FOV mode demonstrated in this work, the two lenses employed for the \(4f\) system are the same (AC254- 125- A, Thorlabs). Thus, the surface of the DMD is 1:1 imaged to the surface of the sample. Considering a diffraction angle \(\beta = 41.92^{\circ}\) that causes compression along the horizontal direction, the lengths of the diagonal lines along the horizontal and vertical directions are \(14.9 \mathrm{mm}\) and \(14.9 \mathrm{mm} \times \cos (41.92^{\circ}) \approx 11.1 \mathrm{mm}\) , respectively. Since a \(3 \times 3\) pixels binning was employed, the corresponding resolution along the horizontal and vertical directions are \(19.35 \mu \mathrm{m} \times 3 \approx 58.0 \mu \mathrm{m}\) and \(19.35 \mu \mathrm{m} \times \cos (41.92^{\circ}) \times 3 \approx 43.1 \mu \mathrm{m}\) , respectively. A larger FOV can be further achieved by amplifying the projected patterns using a different \(4f\) system.  
+
+For the high- resolution mode demonstrated in this work, the two lenses employed for the \(4f\) system have focal lengths of \(300 \mathrm{mm}\) (AC254- 300- A, Thorlabs) and \(30 \mathrm{mm}\) (AC254- 30- A, Thorlabs), respectively. Thus, the surface of the DMD is demagnified by a factor of 10 when imaged to the surface of the sample. In this case, both the FOV and the resolution are scaled down by a factor of 10 compared to that in the large- FOV mode. Thus, the FOV and the resolution along the horizontal and vertical directions are \(1.49 \mathrm{mm} \times 1.11 \mathrm{mm}\) and \(5.80 \mu \mathrm{m} \times 4.31 \mu \mathrm{m}\) , respectively." A finer resolution can be further achieved by using a \(4f\) system with a larger minification. However, such a \(4f\) system requires special care due to the emergence of optical aberrations.  
+
+Besides, we also added descriptions on how the resolution degrades with sampling ratios in the Results section of the revised main text: Since the square path we employed for compressive sensing follows the order from low spatial frequency to high spatial frequency, the spatial resolution is expected to degrade along with the square root of the SR.  
+
+3. The authors claimed that the strategy of \(768 \times 768\) pixels with \(3 \times 3\) binning from DMD was performed in holographic imaging. Why not making full use of entire \(768 \times 1024\) pixels with \(3 \times 4\) binning? This strategy should provide a better SNR and imaging quality.  
+
+Response: We thank the reviewer for raising this question. We confirm that making full use of the entire \(768 \times 1024\) pixels of the DMD can provide a better SNR indeed. However, such a choice will make the pattern on the surface of the DMD to be rectangular. Given a diffraction angle \(\beta = 41.92^{\circ}\) , the projected pattern on the sample turns into an asymmetric parallelogram, causing inconvenience for visualization purposes. Therefore, the effective area of the DMD is always kept as a square shape, so that the projected pattern on the surface of the sample exhibits a diamond shape. To clarify this point, we added the following sentence in the section of "Operations of the digital micromirror device (DMD)" in revised Supplementary Information: Due to this compression, a square pattern displayed by the DMD is transformed into a diamond shape. Notably, if a rectangular pattern is displayed by the DMD, this transformation turns the projected pattern into an asymmetric parallelogram shape. Thus, for visualization purposes, we restricted ourselves to use only square active areas of the DMD throughout this work.  
+
+4. In Fig. 2, lens L7 is regarded as a Fourier transformation of input field. When loading patterns with high spatial frequency, the output field is no longer a single point but exhibits spatial distribution. Considering the finite sense area of the single-pixel detector, it is possible that not all the output energy is collected. Are there any considerations or calculations to consider this condition?  
+
+Response: We thank the reviewer for raising an important question. As pointed out by the reviewer, when loading patterns with high spatial frequencies, the resulting field after passing through the collection lens is no longer a single point but exhibits spatial distribution. In this case, not all the output energy is collected if the sensing area of the single-pixel detector is not large enough. For single-pixel imaging, since the collected total energy becomes less for patterns with higher spatial frequencies, this phenomenon causes a degradation in image quality, especially for the fine details. Nonetheless, for single-pixel holography, a plane reference beam with zero spatial frequency is introduced. After passing through the collection lens, only the component with the same spatial frequency interferes between the reference beam and the signal beam. Since only the interference term is kept for image reconstruction, the single-pixel detector only needs to receive light with zero spatial frequency. As a result, the finite sensing area of the single-pixel detector does not influence the performance of single-pixel holography. This conclusion has been verified through numerical simulations in our previous studies (not shown in this work). To clarify this point, we added the following sentences in the Discussion section of the revised main text: In practice,
+
+<--- Page Split --->
+
+when loading patterns with high spatial frequencies, not all light can be collected by the photodetector if the sensing area is not large enough. This phenomenon can degrade the quality of the reconstructed images in SPI, especially for fine details. Nonetheless, it is surprising to find that SPH is immune to this effect. This observation is because only the interference term with zero spatial frequency contributes to the reconstruction process of SPH.  
+
+5. In the table on Page 3, the column of "refresh time" listed in this work is \(48 \mu \mathrm{s}\) , indicating that the full speed of the DMD (22k) was not fully used. Why not making full use of the 22k refresh rate to push the SBP-T to the limit?  
+
+Response: We thank the reviewer for this careful catch. The reason we chose a refresh time of \(48 \mu \mathrm{s}\) , i.e., a refresh rate of \(20.8 \mathrm{kHz}\) , is for the convenience of batch processing during imaging reconstruction. Given a sampling rate of \(1.25 \mathrm{Ms / s}\) of the data acquisition card and a beat frequency at \(62,500 \mathrm{Hz}\) , we can get 60 data points for the duration of one Hadamard pattern. Under this condition, 3 periods of heterodyne signals were measured and processed. In contrast, if we used the full speed of the DMD, we would get on average \(56.8\) data points for the duration of one Hadamard pattern. This choice only increases the data rate by \(5\%\) but causes considerable difficulties in batch processing data. To clarify this point, we added the following sentences in the Method section of the revised main text: "We note here that this DMD can support a shorter refresh time down to \(45 \mu \mathrm{s}\) ( \(22 \mathrm{kHz}\) refresh rate), corresponding to on average \(56.8\) data points for the duration of one Hadamard pattern. Although such a choice can gain an increase in data rate by roughly \(5\%\) , it causes difficulties in batch processing to reconstruct images."  
+
+6. In the provided files "data_and_supporting_files.zip" and "readme.pdf", it is mentioned that "To minimize phase drifting caused by environmental disturbance during the experiment....", could the authors provide a detailed explanation on the reasons of inserting additional Hadamard patterns and how to process these data corresponding to additional patterns?  
+
+Response: We thank the reviewer for catching this detail and we would like to clarify this point here. From the theoretical perspective, there is no need to insert such patterns. In practice, systems employing heterodyne holography are very sensitive to physical jitters and ambient noises, causing phase drifting during the data acquisition process. Such an effect can significantly deteriorate the quality of reconstructed holographic images. Moreover, based on Eq. (3) in revised main text that the determination of the complex coefficient for the \(n\) - th order Hadamard basis requires subtracting the measurement from the \(1^{\mathrm{st}}\) order Hadamard pattern. Thus, we consider using the \(1^{\mathrm{st}}\) order of Hadamard-like pattern for calibration naturally. Specifically, the \(1^{\mathrm{st}}\) order of Hadamard-like pattern was inserted as a tracking base after displaying 16 different orders of Hadamard-like patterns. In this way, the drifted phase can be corrected over time by reference to the measurement of the \(1^{\mathrm{st}}\) order of Hadamard-like pattern. We provided a detailed description to clarify this point in the section of "Heterodyne holography" in revised Supplementary Information: We note that the second term in Eq. (S6) is simply a constant value. Keeping this term during imaging reconstruction only imposes a constant background to the reconstructed images, which is acceptable for visualization. In practice, this term can be used to correct phase drifting that occurred during the data acquisition process. Specifically, in this work, the \(1^{\mathrm{st}}\) order of Hadamard-like pattern was inserted as the tracking base after displaying every 16 orders of Hadamard-like patterns. The measurement of this recurring pattern was used to update the second term in Eq. (S6) over time.  
+
+7. In line 17 on Page 4, Eq(3), the first order of Hadamard-like patterns is simply a DC signal. My understanding is that if this value was not subtracted for each coefficient as the authors did, the final image would have a constant background. Nonetheless, for imaging purposes, a constant background is acceptable.  
+
+Response: We thank the reviewer for pointing this out. From the theoretical perspective, this term can be kept as the reviewer suggested. In practice, as shown in the response to Comment 6, this term was utilized as a tracking base to correct for the phase drifting that occurred during the data acquisition process. Detailed explanations are provided in revised Supplementary Information, which can be found in
+
+<--- Page Split --->
+
+the response to Comment 6.  
+
+8. What's the bandwidth of the photodiode? Specifications of this component are helpful for readers to adopt this technique.  
+
+Response: We thank the reviewer for this kind suggestion. We added the specification of the photodiode when describing the experimental setup in the Method section of the revised main text: ... by a photodiode (DET10A2, Thorlabs) with a bandwidth of 350 MHz, which was then ...  
+
+9. In Fig. 2, the DMD is not parallel to the sample plane. Will this introduce any measurement error or distortion?  
+
+Response: We thank the reviewer for asking this question. To make the illumination beam and the diffracted beam lie in the same horizontal plane, the rotational axis is perpendicular to the plane of the optical table. In this condition, the surface of the DMD is not orthogonal to neither the illuminating beam nor the diffracted beam to maximize the diffraction efficiency in this optical setup. The analysis of this choice has been analyzed and provided with details in the section of "Operations of the digital micromirror device (DMD)" in revised Supplementary Information. As a result, such an oblique configuration may induce a phase ramp across the projected image. Nonetheless, the calibration process we described in the section of "The procedure of removing phase contaminations induced from the optical system" can conveniently remove this affection. To clarify this issue, we added the following sentences in the section of "Operations of the digital micromirror device (DMD)" in revised Supplementary Information: "This diffracted angle \(\beta\) will cause the displayed pattern not perpendicular to the propagation direction, inducing a phase ramp across the projected pattern. Nonetheless, the calibration process to correct phase contamination we described above can conveniently ease this affection. Moreover, this configuration also causes compression along the horizontal direction."  
+
+10. The gold standard is missing to cross-check the presented technique. Can the authors use existing techniques (e.g., phase-contrast microscopy) to verify their results? Also, what's the measurement accuracy? This would be another important parameter that would determine the application scope of this technique.  
+
+Response: We thank the reviewer for making this valuable suggestion. Unfortunately, we do not have a phase-contrast microscope in the lab. Meanwhile, the phase-contrast microscope cannot provide quantitative results to determine the accuracy of the measured phase values either. To overcome this weakness in the original submission, we performed new experiments by imaging a quantitative phase resolution target (QPT, BenchMark Tec) as a benchmark. The measurement accuracy in phase was quantified. We added the section of "Holographic results of a quantitative phase resolution target" in revised Supplementary Information, including both detailed descriptions and figures, to quantify the measurement accuracy of the holographic system in terms of phase.  
+
+## Holographic results of a quantitative phase resolution target  
+
+To quantify the imaging capability in phase, a quantitative phase resolution target (QPT, BenchMark Tec) was imaged. This type of phase target is fabricated by coating transparent materials on a piece of glass. Specifically, phase patterns with the same size as groups 6- 7 of the USAF standard resolution target are provided, allowing us to gauge the resolution in phase. While operating under the high- resolution mode, Fig. 6 shows the reconstructed amplitude and phase images of this phase target. Due to the low contrast in transmission, the amplitude image shown in Fig. 6(a) is vague. Nevertheless, the phase image shown in Fig. 6(b) exhibits great performance, manifesting delicate phase patterns that are almost blind in the amplitude counterpart. Specifically, we also quantified that the smallest structure that can be distinguished in phase is element 6 of group 6 (4.386- \(\mu \mathrm{m}\) width), with a corresponding 1D profile displayed in the upper inset. This result agrees with the theoretical resolution under the high- resolution mode, which is \(5.80 \mu \mathrm{m} \times 4.31 \mu \mathrm{m}\) .  
+
+Next, appraising whether the reconstructed phase value is quantitatively correct is another important issue. Given a coating thickness of \(250 \mathrm{nm}\) and a refractive index of 1.52, the phase difference between the phase pattern and the background is estimated to be \(\Delta \phi \approx 1.795\) rads for the green light. Here, we examined the bar located at the bottom of the phase target, enclosed in a dashed
+
+<--- Page Split --->
+
+rectangular. To minimize statistical errors, the same area was imaged four times. 1D profile that crosses both the bar and the background is depicted in the lower inset using blue circles, exhibiting a stepped structure. The errorbars represent the standard deviation of four independent measurements, resulting in an averaged fluctuation of about 0.137 rads. The averaged phase difference was estimated to be \(\Delta \phi_{\mathrm{exp}} \approx 1.691\) rads, giving a phase error of only 0.104 rads \((\leq \lambda /60)\) . These results confirm that the developed SPH is quantitatively accurate to retrieve phase, showing prospects in biophotonics.  
+
+![](images/Figure_6.jpg)
+
+<center>Fig. 6 Holographic results of imaging a quantitative phase resolution target under high-resolution mode. (a) The amplitude image of the quantitative phase resolution target, showing poor contrast. (b) The phase image of the quantitative phase resolution target, showing clear phase patterns. Upper inset: the corresponding one-dimensional profile of element 6 of group 6 (4.386-μm width). Lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of about 0.104 rads. The corresponding scale bar is \(100 \mu \mathrm{m}\) . </center>  
+
+Moreover, Fig. 7 shows the holographic results with compressive sensing. Similarly, when the SR decreases, the detailed structure gradually becomes obscure. Nonetheless, the holographic result reconstructed when \(\mathrm{SR} = 12.5\%\) is still acceptable.  
+
+![](images/Figure_7.jpg)
+
+<center>Fig. 7 Holographic results of imaging the phase resolution target with compressive sensing. (a)(b) The amplitude and phase images reconstructed with different sampling ratios of \(50\%\) , \(25\%\) , \(12.5\%\) , \(6.25\%\) , and \(3.125\%\) . The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+Besides, we also make Fig. 6(b) as part of the newly added Fig. 5 in the Results section of the revised main text. Detailed descriptions on imaging a quantitative phase resolution target in the revised main text can be found in the Response to Comment 11.  
+
+11. In Fig. 5, the authors used red dashed circles to point out show some supplement information in the phase image from the amplitude counterpart. This part could be elaborated on because it showcases the capability of this technique.  
+
+Response: We thank the reviewer for this valuable suggestion. To showcase case the capability of this holographic technique that phase images can become a good supplement to the amplitude counterpart, we performed many new experiments on phase objects and unstained biological tissue. These holographic images confirmed that in many conditions, phase images can provide rich information with good contrast that is hardly seen in their amplitude counterparts. Detailed descriptions on imaging a quantitative phase resolution
+
+<--- Page Split --->
+
+target was added as a new section "Holographic results of a quantitative phase resolution target" in revised Supplementary Information, which can be found in the Response to Comment 10. Detailed descriptions on imaging unstained biological tissue were added as a new paragraph "Holographic imaging of unstained tissue from mouse brains" in the Results section of revised main text and a new section of "Additional holographic images of unstained tissue from mouse brains" in revised Supplementary Information.  
+
+In revised main text  
+
+Holographic imaging of unstained tissue from mouse brains. Having demonstrated holographic imaging of stained tissue from mouse tails, we then switched to image unstained tissue from mouse brains. Generally, images of unstained tissue exhibit low contrast in amplitude, but providing sufficient contrast through phase imaging. Before proceeding, we quantified the performance of the holographic system in phase by imaging a quantitative phase resolution target (QPT, BenchMark Tec). Figure 5(a) shows the reconstructed phase image of the resolution target, containing groups 6 and 7. To quantify the resolving power in phase, the corresponding 1D profile denoted by the black bracket within the resolution target, crossing element 6 of group 6 (4.386- μm width), is shown in the upper inset. Three verticle bars can be differentiated, indicating that the theoretical resolution under the high- resolution mode (5.80 μm × 4.31 μm) was achieved in phase. We further quantified the accuracy of the measurement in phase by estimating the phase difference between the bar enclosed by a dashed box at the bottom of the target and its adjacent background. 1D profile that crosses both the bar and the background is shown in the lower inset using blue circles, exhibiting a stepped structure. The averaged phase difference was estimated to be \(\Delta \phi_{\mathrm{exp}} \approx 1.691\) rads, yielding a phase error of only 0.104 rads (≤ \(\lambda /60\) ) compared to the actual phase difference (1.795 rads). Detailed analysis on imaging the quantitative phase resolution target can be found in section VII of Supplementary Information. Then, we proceed to image unstained biological tissue. Figure 5(b) shows the image of an 80- μm- thick slice of unstained tissue from mouse brains, captured using a bright- field microscope. In this image, some representative types of tissue were denoted with black arrows, such as piriform cortex, perirhinal cortex, olivary pretectal nucleus, and white matter. For unstained tissue, it is usually challenging to identify different structures in amplitude, due to the insufficient contrast in transmission. Figures 5(c), (d), and (e) show a series of reconstructed holographic images for different parts of the mouse brain, indicated by three labeled diamond- shaped boxes in Fig. 5(b). Same as the one captured using the conventional microscope, these amplitude images only manifest rough outlines with low contrast. By comparison, phase images show rich details that can be hardly seen in their amplitude counterparts. For example, the piriform cortex and the perirhinal cortex are identifiable in phase images. Data used for producing Figs. 5(c)-(e) are provided in Supplementary Data 2. These results demonstrate the capability of employing SPH to image relatively transparent biological tissue.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Fig. 5 Performance of the high-throughput SPH with unstained tissue from mouse brains in high-resolution mode. (a) Reconstructed phase image of a quantitative phase target. The upper inset: the corresponding one-dimensional (1D) profile of element 6 of group 6; the lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of 0.104 rads. (b) The image of a slice of unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (c)-(e) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+For the selected area in Fig. 5(c), a series of holographic images reconstructed with compressive sensing at various SRs are illustrated in Fig. 6 as well. Similar to before, these images show the effectiveness of compressive sensing in dealing with biological tissue that contains rich information in phase. We also applied SPH to image other pieces of unstained tissues from mouse brains with different thicknesses, ranging from \(10 \mu \mathrm{m}\) to \(120 \mu \mathrm{m}\) , with details shown in section IX of Supplementary Information.  
+
+![](images/Figure_6.jpg)
+
+<center>Fig. 6 Reconstruction of holographic images for the unstained tissue from mouse brains with compressive sensing. The amplitude and wrapped phase images are reconstructed with different sampling ratios of \(50\%\) , \(25\%\) , \(12.5\%\) , \(6.25\%\) , and \(3.125\%\) . The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+In revised Supplementary Information  
+
+Additional holographic images of unstained tissue from mouse brains
+
+<--- Page Split --->
+
+To show the imaging capability for unstained tissue, we also imaged a \(100 - \mu \mathrm{m}\) - thick slice of unstained mouse brain. Figure 9(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part of the tissue contains white matter and grey matter. Since the unstained slice is relatively thick with roughly the same transmission across the imaging region, it is challenging for the conventional microscope to provide good contrast. Figures 9(b), (c), and (d) show a series of reconstructed holographic images for different parts of the mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 9(a). As expected, the features in amplitude images match well with that at corresponding areas in Fig. 9(a), which do not provide too much information with good contrast. As a comparison, phase images provide much better contrast, revealing many detailed structures that are indiscernible in their amplitude counterparts.  
+
+![](images/Figure_9.jpg)
+
+<center>Fig. 9 Holographic performance of unstained mouse brain tissue with \(100 - \mu \mathrm{m}\) thickness. (a) The image of a slice of \(100 - \mu \mathrm{m}\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+We also imaged an even thicker slice of unstained mouse brain with a thickness of \(120 \mu \mathrm{m}\) . Figure 10(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part contains the ventral part of the lateral septal nucleus, inferior colliculus, and white matter. Figures 10(b), (c), and (d) show a series of reconstructed holographic images for different parts of mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 10(a). Again, phase images provide much better contrast, compared to their amplitude counterparts.
+
+<--- Page Split --->
+![](images/Figure_10.jpg)
+
+<center>Fig. 10 Holographic performance of unstained mouse brain tissue with \(120 - \mu \mathrm{m}\) thickness. (a) The image of a slice of \(120 - \mu \mathrm{m}\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \(200\mu \mathrm{m}\) . </center>  
+
+Very thin unstained tissue from mouse brains down to \(10 - \mu \mathrm{m}\) thick was also imaged using our holographic system. Figure 11(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part contains white matter and grey matter. Figures 11(b), (c), and (d) show a series of reconstructed holographic images for different parts of mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 11(a). For such a thin tissue, it is hard to see details through the amplitude images. Nonetheless, phase images still provide rich information with good contrast.  
+
+![](images/Figure_11.jpg)
+
+<center>Fig. 11 Holographic performance of unstained mouse brain tissue with \(10 - \mu \mathrm{m}\) thickness. (a) The image of a slice of \(10 - \mu \mathrm{m}\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \(200\mu \mathrm{m}\) . </center>  
+
+12. A few typos are found in the manuscript. For example, by “mussel”, I believe the authors intended to say “muscle”.  
+
+Response: We thank the reviewer for catching it out. We carefully went through the revised submission and corrected such kinds of typos as many as we could.
+
+<--- Page Split --->
+
+## Reviewer #3:  
+
+In this manuscript, a novel technique for single pixel compressive holography is proposed. The proposed technique employs heterodyne interference to achieve high- throughput measurement. The authors construct an optical system based on the proposed technique. And experimental verifications are described. In the experimental verifications, parts of a USAF chart for estimation of spatial resolution and biological tissues are reconstructed to demonstrate the usefulness of the system.  
+
+I have some questions and comments to improve the manuscript.  
+
+1. In Fig.2: Why the experimental set up consists of two AOMs? I think that heterodyne interferometry can be implemented with only one AOM with easy control. Also, the authors should describe the reason why the beat frequency is set to 62,500 Hz.  
+
+Response: We thank the reviewer for asking this important question. The AOM we used in the system is purchased from Intraaction company with a model number AOM- 505AF1, which supports modulation frequencies in the range of 40 to 60 MHz. In this condition, if only one AOM was used, the beat frequency had to be set within the range of 40 to 60 MHz, which is far beyond the bandwidth of the data acquisition card we had (1.25Ms/s). Therefore, we chose to use the two AOMs to lower down the beat frequency to the range that the data acquisition card can handle. We clarified this point by adding the following information in the Method section of the revised main text: ...(AOM- 505AF1, Intraaction, Optical frequency shift range: \(\pm 40 \sim 60 \mathrm{MHz}\) ). The reason why the beat frequency was set to 62,500 Hz is the compromise of the sampling rate of the data acquisition card (1.25Ms/s) and the refresh time of the DMD (48 \(\mu \mathrm{s}\) ). The number of beating cycles is desired to be an integer for each displayed pattern. In this condition, a total number of 60 data points were measured for one displayed Hadamard pattern. For a beating frequency of 62,500 Hz, these 60 data points can be evenly distributed into 3 beating cycles. Indeed, there are many other choices for beating frequencies such as 125,000 Hz and 250,000 Hz, distributing these 60 data points into 6 and 12 beating cycles, respectively. Notably, for the same number of data points for one displayed Hadamard pattern, the quality of the reconstructed signal should not be sensitive on the choice of the beating frequency, provided the Nyquist sampling criterion was followed. Thus, we added the following sentence in the Method section of the revised main text: What's more, the choice of the beating frequency is also not unique. In practice, there are many choices for beating frequencies such as 125,000 Hz and 250,000 Hz, distributing these 60 data points into 6 and 12 beating cycles, respectively. Notably, for the same number of data points for one displayed Hadamard pattern, the quality of the reconstructed signal should not be sensitive on the choice of the beating frequency, provided the Nyquist sampling criterion was followed. An integer number of beating cycles for each displayed pattern is also desired for computational convenience.  
+
+2. About Fig. 4 (a): the authors should discuss factors that the image quality at sampling ratio \(50\%\) is worse than that in more critical case (25% & 12.5%). The discussions are important for judgement which the used algorithm for image reconstruction is. Such Discussion are important in determining whether the algorithm used for image reconstruction is practical.  
+
+Response: We thank the reviewer for this valuable suggestion. In general, Hadamard- like patterns with higher orders contain higher spatial frequency, so that information carried by these orders is more sensitive to measurement noises. Thus, although compressive sensing with a smaller sampling ratio may lose fine details in the reconstructed image, it is more robust to measurement noises. This is the reason why the image quality at a sampling ratio of \(50\%\) is worse than that at sampling ratios of \(25\%\) and \(12.5\%\) . As suggested by the reviewer, we added a few sentences in the Results section of the revised main text to explain this observation and added the section of "Investigations on compressive sensing under different noise levels" in revised Supplementary Information to quantitatively study this effect.  
+
+In revised main text:  
+
+We note that this observation is because information carried by Hadamard- like patterns with higher orders is more sensitive to measurement noises. A quantitative study on how measurement noises affect the image quality of the reconstructed image under different choices of SRs can be found in section V of Supplementary Information. Thus, the SR of \(25\%\) might be a suitable choice for the noise
+
+<--- Page Split --->
+
+level in current experimental conditions (the standard deviation of the measurement noise is around \(0.1\%\) of the measured value).  
+
+## In revised Supplementary Information:  
+
+## "Investigations on compressive sensing under different noise levels  
+
+In practice, measurement noises contributed from shot noises of light and electronic noises of equipment, causing degradation in image quality. Using numerical tools, here, we quantify how this effect impacts the performance of compressive sensing with different sampling ratios (SRs).  
+
+The original holographic image \(S\) was set with a resolution of \(256 \times 256\) . We adopted a direct inverse basis transformation with a square path for compressive sensing, as described in main text. For simplicity, measurement noises were simulated as white Gaussian noises with various standard deviations [1]. In particular, noise levels ranging from 0 to \(1\%\) of the averaged measurement values were considered. For compressive sensing, various SRs at \(3.125\%\) , \(6.25\%\) , \(12.5\%\) , \(25\%\) , \(50\%\) , and \(100\%\) were employed to reconstruct images \(O_{CS}\) under different noise levels. A first- order correlation function between the reconstructed and original images, defined as \(|\langle O_{CS}, S \rangle | / (|\langle O_{CS}, O_{CS} \rangle | |\langle S, S \rangle |)^{1 / 2}\) , was used to quantify the similarities between these two holographic images.  
+
+Figure 3(a) plots correlations as a function of noise levels when holographic images were reconstructed at different SRs. For all cases, the correlations decrease as the noises become large. However, correlations correspond to different SRs decay at different speeds. In general, the larger the SR is, the faster the correlation decays. Thus, although the reconstruction process with a small SR performs worse for small measurement noises, it turns out to be better when measurement noises become large. This observation indicates that compressive sensing with a smaller SR, which concentrates more on the lower spatial frequency, is more robust to measurement noises. In our experiment, the noise level is around \(0.1\%\) of the measurement, denoted as a yellow area in the figure. To identify which SR is the most suitable one under the current experimental condition, we then fixed the noise level of \(0.1\%\) in the following simulation. Under such conditions, blue circles in Fig. 3(b) describe the decreased correlation solely induced by the noise, while red circles describe the evolution of the correlation purely resulted from the changing of SRs in compressive sensing. Notably, the trends of these two effects indicate that there exists a suitable SR under the current noise level. By multiplying these two effects, yellow circles represent the combined effect. One could see that the maximum correlation is achieved at \(\mathrm{SR} = 25\%\) . This result explains why the reconstructed images with \(\mathrm{SR} = 100\%\) and \(50\%\) look worse than those with \(\mathrm{SR} = 25\%\) and \(12.5\%\) in main text. Nonetheless, we emphasize here that although a large SR is susceptible to measurement noises, it still provides a high- resolution image, as the decreased correlation is mainly contributed from the noisy background but not from the fine feature.  
+
+![](images/Figure_3.jpg)
+
+<center>Fig. 3 Simulation results on compressive sensing under different noise levels. (a) Correlations between the reconstructed and original images as a function of different noise levels. Compressive sensing realized with different sampling ratios (SRs) were investigated, which are labeled with different colors. The yellow area represents the actual noise levels during experiments. (b) After fixing the noise level as \(0.1\%\) of the measurement value, the SR-resultant effect and the noise-induced effect are represented using blue and red circles as a function of the SR. The combined effect is represented using yellow circles, exhibiting a maximum value at \(\mathrm{SR} = 25\%\) . </center>  
+
+3. About holographic imaging: To show the effectiveness as holographic imaging, phase objects or three-dimensional objects should be experimentally verified. This is because verification for phase imaging with the constructed experimental setup is not enough. (On the other hands, that for amplitude reconstruction is quite enough to show the usefulness.)
+
+<--- Page Split --->
+
+Response: We thank the reviewer for raising this valuable question. To show the effectiveness of holographic imaging, we supplemented a series of new experiments to image a quantitative phase resolution target (Benchmark) and several slices of unstained biological tissue from mouse brains. These holographic images show rich information in phase that is hardly discernable in the amplitude counterpart.  
+
+We added the section of "Holographic results of a quantitative phase resolution target" in revised Supplementary Information, including both detailed descriptions and figures, to quantify the measurement accuracy of the holographic system in phase values.  
+
+## Holographic results of a quantitative phase resolution target  
+
+To quantify the imaging capability in phase, a quantitative phase resolution target (QPT, BenchMark Tec) was imaged. This type of phase target is fabricated by coating transparent materials on a piece of glass. Specifically, phase patterns with the same size as groups 6- 7 of the USAF standard resolution target are provided, allowing us to gauge the resolution in phase. While operating under the high- resolution mode, Fig. 6 shows the reconstructed amplitude and phase images of this phase target. Due to the low contrast in transmission, the amplitude image shown in Fig. 6(a) is vague. Nevertheless, the phase image shown in Fig. 6(b) exhibits great performance, manifesting delicate phase patterns that are almost blind in the amplitude counterpart. Specifically, we also quantified that the smallest structure that can be distinguished in phase is element 6 of group 6 (4.386- \(\mu \mathrm{m}\) width), with a corresponding 1D profile displayed in the upper inset. This result agrees with the theoretical resolution under the high- resolution mode, which is \(5.80\mu \mathrm{m}\times 4.31\mu \mathrm{m}\)  
+
+Next, appraising whether the reconstructed phase value is quantitatively correct is another important issue. Given a coating thickness of \(250\mathrm{nm}\) and a refractive index of 1.52, the phase difference between the phase pattern and the background is estimated to be \(\Delta \phi \approx 1.795\) rads for the green light. Here, we examined the bar located at the bottom of the phase target, enclosed in a dashed rectangular. To minimize statistical errors, the same area was imaged four times. 1D profile that crosses both the bar and the background is depicted in the lower inset using blue circles, exhibiting a stepped structure. The errorbars represent the standard deviation of four independent measurements, resulting in an averaged fluctuation of about 0.137 rads. The averaged phase difference was estimated to be \(\Delta \phi_{\mathrm{exp}}\approx 1.691\) rads, giving a phase error of only \(0.104\mathrm{rads}(\leq \lambda /60)\) . These results confirm that the developed SPH is quantitatively accurate to retrieve phase, showing prospects in biophotonics.  
+
+![](images/Figure_6.jpg)
+
+<center>Fig. 6 Holographic results of imaging a quantitative phase resolution target under high-resolution mode. (a) The amplitude image of the quantitative phase resolution target, showing poor contrast. (b) The phase image of the quantitative phase resolution target, showing clear phase patterns. Upper inset: the corresponding one-dimensional profile of element 6 of group 6 (4.386- \(\mu \mathrm{m}\) width). Lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of about 0.104 rads. The corresponding scale bar is \(100\mu \mathrm{m}\) . </center>  
+
+Moreover, Fig. 7 shows the holographic results with compressive sensing. Similarly, when the SR decreases, the detailed structure gradually becomes obscure. Nonetheless, the holographic result reconstructed when \(\mathrm{SR} = 12.5\%\) is still acceptable.
+
+<--- Page Split --->
+![](images/Figure_7.jpg)
+
+<center>Fig. 7 Holographic results of imaging the phase resolution target with compressive sensing. (a)(b) The amplitude and phase images reconstructed with different sampling ratios of \(50\%\) , \(25\%\) , \(12.5\%\) , \(6.25\%\) , and \(3.125\%\) . The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+Detailed descriptions on imaging unstained biological tissue were added as a new paragraph "Holographic imaging of unstained tissue from mouse brains" in the Results section of revised main text and a section of "Additional holographic images of unstained tissue from mouse brains" in revised Supplementary Information.  
+
+In revised main text  
+
+Holographic imaging of unstained tissue from mouse brains. Having demonstrated holographic imaging of stained tissue from mouse tails, we then switched to image unstained tissue from mouse brains. Generally, images of unstained tissue exhibit low contrast in amplitude, but providing sufficient contrast through phase imaging. Before proceeding, we quantified the performance of the holographic system in phase by imaging a quantitative phase resolution target (QPT, BenchMark Tec). Figure 5(a) shows the reconstructed phase image of the resolution target, containing groups 6 and 7. To quantify the resolving power in phase, the corresponding 1D profile denoted by the black bracket within the resolution target, crossing element 6 of group 6 (4.386- \(\mu \mathrm{m}\) width), is shown in the upper inset. Three verticle bars can be differentiated, indicating that the theoretical resolution under the high- resolution mode (5.80 \(\mu \mathrm{m} \times 4.31 \mu \mathrm{m}\) ) was achieved in phase. We further quantified the accuracy of the measurement in phase by estimating the phase difference between the bar enclosed by a dashed box at the bottom of the target and its adjacent background. 1D profile that crosses both the bar and the background is shown in the lower inset using blue circles, exhibiting a stepped structure. The averaged phase difference was estimated to be \(\Delta \phi_{\mathrm{exp}} \approx 1.691\) rads, yielding a phase error of only 0.104 rads \((\leq \lambda /60)\) compared to the actual phase difference (1.795 rads). Detailed analysis on imaging the quantitative phase resolution target can be found in section VII of Supplementary Information. Then, we proceed to image unstained biological tissue. Figure 5(b) shows the image of an 80- \(\mu \mathrm{m}\) - thick slice of unstained tissue from mouse brains, captured using a bright- field microscope. In this image, some representative types of tissue were denoted with black arrows, such as piriform cortex, perirhinal cortex, olivary pretectal nucleus, and white matter. For unstained tissue, it is usually challenging to identify different structures in amplitude, due to the insufficient contrast in transmission. Figures 5(c), (d), and (e) show a series of reconstructed holographic images for different parts of the mouse brain, indicated by three labeled diamond- shaped boxes in Fig. 5(b). Same as the one captured using the conventional microscope, these amplitude images only manifest rough outlines with low contrast. By comparison, phase images show rich details that can be hardly seen in their amplitude counterparts. For example, the piriform cortex and the perirhinal cortex are identifiable in phase images. Data used for producing Figs. 5(c)- (e) are provided in Supplementary Data 2. These results demonstrate the capability of employing SPH to image relatively transparent biological tissue.
+
+<--- Page Split --->
+![](images/Figure_5.jpg)
+
+<center>Fig. 5 Performance of the high-throughput SPH with unstained tissue from mouse brains in high-resolution mode. (a) Reconstructed phase image of a quantitative phase target. The upper inset: the corresponding one-dimensional (1D) profile of element 6 of group 6; the lower inset: quantitative analysis on the accuracy of the reconstructed phase values, resulting in a phase error of 0.104 rads. (b) The image of a slice of unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (c)-(e) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+For the selected area in Fig. 5(c), a series of holographic images reconstructed with compressive sensing at various SRs are illustrated in Fig. 6 as well. Similar to before, these images show the effectiveness of compressive sensing in dealing with biological tissue that contains rich information in phase. We also applied SPH to image other pieces of unstained tissues from mouse brains with different thicknesses, ranging from \(10 \mu \mathrm{m}\) to \(120 \mu \mathrm{m}\) , with details shown in section IX of Supplementary Information.  
+
+![](images/Figure_6.jpg)
+
+<center>Fig. 6 Reconstruction of holographic images for the unstained tissue from mouse brains with compressive sensing. The amplitude and wrapped phase images are reconstructed with different sampling ratios of \(50\%\) , \(25\%\) , \(12.5\%\) , \(6.25\%\) , and \(3.125\%\) . The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+In revised Supplementary Information
+
+<--- Page Split --->
+
+## Additional holographic images of unstained tissue from mouse brains  
+
+To show the imaging capability for unstained tissue, we also imaged a \(100 - \mu \mathrm{m}\) - thick slice of unstained mouse brain. Figure 9(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part of the tissue contains white matter and grey matter. Since the unstained slice is relatively thick with roughly the same transmission across the imaging region, it is challenging for the conventional microscope to provide good contrast. Figures 9(b), (c), and (d) show a series of reconstructed holographic images for different parts of the mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 9(a). As expected, the features in amplitude images match well with that at corresponding areas in Fig. 9(a), which do not provide too much information with good contrast. As a comparison, phase images provide much better contrast, revealing many detailed structures that are indiscernible in their amplitude counterparts.  
+
+![](images/Figure_9.jpg)
+
+<center>Fig. 9 Holographic performance of unstained mouse brain tissue with \(100 - \mu \mathrm{m}\) thickness. (a) The image of a slice of \(100 - \mu \mathrm{m}\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the unstained tissue. The corresponding scale bar is \(200 \mu \mathrm{m}\) . </center>  
+
+We also imaged an even thicker slice of unstained mouse brain with a thickness of \(120 \mu \mathrm{m}\) . Figure 10(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part contains the ventral part of the lateral septal nucleus, inferior colliculus, and white matter. Figures 10(b), (c), and (d) show a series of reconstructed holographic images for different parts of mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 10(a). Again, phase images provide much better contrast, compared to their amplitude counterparts.
+
+<--- Page Split --->
+![](images/Figure_10.jpg)
+
+<center>Fig. 10 Holographic performance of unstained mouse brain tissue with \(120 - \mu \mathrm{m}\) thickness. (a) The image of a slice of \(120 - \mu \mathrm{m}\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \(200\mu \mathrm{m}\) . </center>  
+
+Very thin unstained tissue from mouse brains down to \(10 - \mu \mathrm{m}\) thick was also imaged using our holographic system. Figure 11(a) shows a bright- field image of this slice, captured using a conventional microscope. The imaged part contains white matter and grey matter. Figures 11(b), (c), and (d) show a series of reconstructed holographic images for different parts of mouse brain, denoted by three labeled diamond- shaped boxes in Fig. 11(a). For such a thin tissue, it is hard to see details through the amplitude images. Nonetheless, phase images still provide rich information with good contrast.  
+
+![](images/Figure_11.jpg)
+
+<center>Fig. 11 Holographic performance of unstained mouse brain tissue with \(10 - \mu \mathrm{m}\) thickness. (a) The image of a slice of \(10 - \mu \mathrm{m}\) -thick unstained tissue from mouse brains, captured using a conventional microscope. Three diamond-shaped boxes represent the area being imaged by the holographic system. (b)-(d) The reconstructed amplitude and phase images for different parts of the stained tissue. The corresponding scale bar is \(200\mu \mathrm{m}\) . </center>  
+
+4. In conclusion: some performances of the proposed method are shown. Those estimations are much important in the paper. The authors should summarize the performance as a table like Table 1 and discuss on relations between scalability and specifications of the elemental devices in detail.
+
+<--- Page Split --->
+
+**Response:** We thank the reviewer for this valuable suggestion. In the Method section of the revised main text, we summarized the performance of three operational modes in Tab. 2 and discussed the relations between scalability and specifications in detail: As a final remark, we briefly compare the performance of several operational modes developed in this work, which are summarized in Tab. 2. Two large-FOV modes and one high-resolution mode were demonstrated with the same illumination device, beat frequency, and the sampling rate of DAC. Thus, the same SBP- \(T\) was achieved for all three operational modes, calculated by multiplying the reciprocal of refresh time of the illumination device by 2. Moreover, the two large-FOV modes share the same \(4f\) system, while the only difference is the binning strategy. Due to this reason, the lateral resolution and the FOV of these two modes are simply scaled by a factor of 1.5. As for the high-resolution mode, a different \(4f\) system was used. Therefore, the lateral resolution and the FOV of this mode are scaled to be ten times smaller than that in the first large-FOV mode. Detailed evaluations on these parameters can be found in section X of Supplementary Information. Currently, the switching between the large FOV mode and the high-resolution mode requires hardware modification of the \(4f\) optical setup, which is inconvenient and prevents real-time operation. Nonetheless, we anticipate that a variable \(4f\) setup that consists of tunable lenses and an advanced electromechanics system with auto-focusing can enable smooth adjustment between different operational modes in the future.
+
+Table. 2 List of the parameters in both the large-FOV mode and the high-resolution mode.
+
+<table><tr><td colspan="2" rowspan="2"></td><td rowspan="2">Large-FOV mode 1<br>(Main Text)</td><td rowspan="2">Large-FOV mode 2<br>(Supplementary Information)</td><td colspan="2" rowspan="2">High-resolution mode 1<br>(Main Text & Supplementary Information)</td></tr><tr></tr><tr><td colspan="2">Illumination</td><td>Pixel size (μm)</td><td colspan="3">13.68</td></tr><tr><td colspan="2">device</td><td>Refresh time (ms)</td><td colspan="3">0.048</td></tr><tr><td colspan="2">Beating frequency (Hz)</td><td colspan="4">62,500</td></tr><tr><td colspan="2">Sampling rate of DAC (Ms/s)</td><td colspan="4">1.25</td></tr><tr><td colspan="2">Strategy of binning pixels</td><td>768x768 pixels with \(3x3\)binning</td><td>512x512 pixels with \(2x2\)binning</td><td colspan="2">768x768 pixels with \(3x3\)binning</td></tr><tr><td rowspan="2">Lens in \(4f\) system</td><td>front model</td><td>AC254-125-A,Thorlabs(125mm)</td><td>AC254-125-A,Thorlabs(125mm)</td><td colspan="2">AC254-300-A,Thorlabs(300mm)</td></tr><tr><td>rear model</td><td>AC254-125-A,Thorlabs(125mm)</td><td>AC254-125-A,Thorlabs(125mm)</td><td colspan="2">AC254-030-A,Thorlabs(30mm)</td></tr><tr><td colspan="2">Lateral resolution (μm)</td><td>\(58.0x43.1\)</td><td>\(38.7x28.8\)</td><td colspan="2">\(5.80x4.31\)</td></tr><tr><td colspan="2">FOV (mm)</td><td>\(14.9x11.1\)</td><td>\(9.91x7.37\)</td><td colspan="2">\(1.49x1.11\)</td></tr><tr><td colspan="2">SBP-\(T\)(pixels/s)</td><td colspan="4">41,666.6</td></tr></table>
+
+5. Forms of reconstructed images in Figs. 3, 4, 5, and 6: (This comment may be not so essential.) Why area of measurement is not square like but diamond shape (rhomboid)?
+
+**Response:** We thank the reviewer for raising a good question. To make the illumination beam and the diffracted beam lie in the same horizontal plane, the rotational axis is perpendicular to the plane of the optical table. Thus, the effective area of the DMD, i.e., a square area, should be rotated in plane by \(45^{\circ }\) . Moreover, to maximize the diffraction efficiency of the two-dimensional blazed grating, a diffraction angle \(\beta =41.92^{\circ }\) was chosen. This diffraction compresses the dimension along the horizontal direction, transforming the projected pattern from a square shape to a diamond shape. Detailed analysis can be found in the section of **“Operations of the digital micromirror device (DMD)”** in revised Supplementary Information. To clarify this issue and avoid confusion, we added the following sentences at the end of the section: Due to this compression, a square pattern displayed by the DMD is transformed into a diamond shape.Notably, if a rectangular pattern is displayed by the DMD, this transformation turns the projected pattern into an asymmetric parallelogram shape. Thus, for visualization purposes, we restricted ourselves to use only square active areas of the DMD throughout this work.
+
+<--- Page Split --->
+
+Reviewers' Comments:  
+
+Reviewer #1: Remarks to the Author:  
+
+The authors present a revised version of the optical heterodyne detection scheme for single- pixel holographic imaging along the lines raised by the different reviewers.  
+
+The main scientific message of the manuscript has now been emphasized and clarity of the manuscript has improved so that is easy to follow. The study extends current knowledge about single- pixel holography.  
+
+Technical data have been improved and new experiments have been performed. Laboratory experiments are now at the heart of the manuscript and have been rationally designed to support the claims made in the paper. Also, previously published material have been cited in accordance with reviewers' suggestions.  
+
+Thus, I recommend acceptance of the manuscript in Nature Communications.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+In this revised manuscript, the authors have addressed all my comments. It is always good to see the authors spent time directly answering the comments with new data. I have to admit that I am impressed by the author's diligence in performing many additional experiments, which have considerably improved the quality of an already excellent article. In these new results, the imaging of unstained mouse brain tissue makes the technique clearly stand out from existing methods (e.g., HE histology). In this regard, the proposed method, thanks to its scalable FOV and relatively high speed, has the promising potential of becoming a generic tool for many biomedical applications. Thus, I recommend publish it as is.  
+
+Reviewer #3: Remarks to the Author: Dear authors,  
+
+In this revised version of the manuscript and supplementary informations, the authors are considered to respond to my questions and comments clearly. The revised set of the manuscript is much improved in comparison with the original one.
+
+<--- Page Split --->
+
+## Point-by-point responses to reviewers' comments  
+
+## Reviewer #1:  
+
+The authors present a revised version of the optical heterodyne detection scheme for single- pixel holographic imaging along the lines raised by the different reviewers.  
+
+The main scientific message of the manuscript has now been emphasized and clarity of the manuscript has improved so that is easy to follow. The study extends current knowledge about single- pixel holography.  
+
+Technical data have been improved and new experiments have been performed. Laboratory experiments are now at the heart of the manuscript and have been rationally designed to support the claims made in the paper. Also, previously published material have been cited in accordance with reviewers' suggestions.  
+
+Thus, I recommend acceptance of the manuscript in Nature Communications.  
+
+Response: We are glad that our reply addresses the reviewer's question and also thank the reviewer for recognizing our work. The reviewer's unique insight on the reconstructive algorithm about single- pixel imaging enables a deeper comprehension in the realm of compressive sensing and adaptive smart sensing. Some representative references the reviewer listed also broaden our horizon about real- time demonstrations in single- pixel imaging through deep learning.
+
+<--- Page Split --->
+
+## Reviewer #2:  
+
+In this revised manuscript, the authors have addressed all my comments. It is always good to see the authors spent time directly answering the comments with new data. I have to admit that I am impressed by the author's diligence in performing many additional experiments, which have considerably improved the quality of an already excellent article. In these new results, the imaging of unstained mouse brain tissue makes the technique clearly stand out from existing methods (e.g., HE histology). In this regard, the proposed method, thanks to its scalable FOV and relatively high speed, has the promising potential of becoming a generic tool for many biomedical applications. Thus, I recommend publish it as is.  
+
+Response: We are happy that the revised submission addresses all the reviewer's comments and also thank the reviewer for recognizing our work. The reviewer provides lots of detailed comments, such as verification of the correctness of phase imaging and the clarification of certain choices of parameters during experiments. These comments are critical and help us improve the demonstration of the imaging capability of our holographic system.
+
+<--- Page Split --->
+
+## Reviewer #3:  
+
+Dear authors,  
+
+In this revised version of the manuscript and supplementary information, the authors are considered to respond to my questions and comments clearly.  
+
+The revised set of the manuscript is much improved in comparison with the original one.  
+
+Regards,
+
+<--- Page Split --->
+
+Response: We are grateful that the revised submission satisfactorily responds to the questions and comments of the reviewer and thank the reviewer for recognizing our work. The reviewer's suggestion is very helpful, enabling us to improve the capability of the holographic system by adding experiments on phase resolution targets and relatively transparent samples. The reviewer's comments also urge us to consider compressive sensing under different noise levels.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__441781baa1b2dedeba52491b135464cc88bacc38a173fe0d5b9f563ad3051d30/images_list.json

@@ -0,0 +1,63 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "1 (Fig. 2 in Grillitsch et al, 2011-PMID: 21820081), note the presence of TAG 54:6, 54:5 in yeast cells, implying that one of the constituent fatty acids must contain more than one C=C bonds in their structures.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Revision Fig. 1",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 39
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Revision Fig. 2",
+    "footnote": [],
+    "bbox": [
+      [
+        470,
+        120,
+        720,
+        370
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_3.jpg",
+    "caption": "Revision Fig. 3",
+    "footnote": [],
+    "bbox": [
+      [
+        188,
+        409,
+        432,
+        600
+      ]
+    ],
+    "page_idx": 40
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_4.jpg",
+    "caption": "Revision Fig. 4",
+    "footnote": [],
+    "bbox": [
+      [
+        480,
+        409,
+        722,
+        465
+      ]
+    ],
+    "page_idx": 51
+  }
+]

peer_reviews/supplementary_0_Peer Review File__441f9407302393914095b7edc3ad47804b8a172b40d660d04f980b77da7e41d6/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__441f9407302393914095b7edc3ad47804b8a172b40d660d04f980b77da7e41d6/supplementary_0_Peer Review File__441f9407302393914095b7edc3ad47804b8a172b40d660d04f980b77da7e41d6.mmd

@@ -0,0 +1,101 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Symmetries in quantum networks lead to no- go theorems for entanglement distribution and to verification techniques  
+
+![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 --->
+
+Reviewers' Comments:  
+
+Reviewer #1:  
+
+Remarks to the Author:  
+
+The submitted manuscript deals with a very timely topic, the characterization of quantum correlations that can be observed in networks. Namely, the network geometry severely restricts types of quantum correlations. The authors focus on density matrices, i.e. quantum states, and try to answer how we can determine what kind of quantum states can be generated with a given network. Up to now the main method for approaching this problem was using inflation techniques. However, the technique is applicable only for networks involving a rather small number of nodes and edges. In this manuscript the authors suggest using symmetry to make the use of inflation more economical. They show that the method works very well for stabilizer states and those with bosonic or fermionic symmetry.  
+
+I like the problem the authors consider. The method might be quite useful in further research, and I think that this paper should be highlighted by publication in Nature Communications. It considers a fairly unexplored problem and provides a concrete step forward. In addition, the paper is well written and easy to read and understand even by non- specialists.  
+
+Reviewer #2:  
+
+Remarks to the Author:  
+
+This manuscript is concerned with the quantum networks scenario which consists of many spatially separated observers that are linked by different sources of quantum correlations and can perform local operations on their shares of distributed quantum states. Moreover, the implementation of these local channels is coordinated by classical correlations distributed to the observers by a single source. On the other hand, the authors impose one constraints on the considered scenario that no classical communication between parties is allowed. This assumption is well motivated in the manuscript and at the same time makes the considered scenario highly nontrivial.  
+
+Quantum networks constitute a paradigmatic framework in quantum information as it models for instance quantum communication protocols or quantum internet. For this reason characterization of quantum correlations that can be prepared within quantum networks under various assumptions is gaining a lot of attention recently. At the same time this problem is extremely difficult to study both from analytical and numerical perspectives and therefore most of the results on distribution of entangled states concern networks of the simplest topologies such the triangle networks.  
+
+In the present manuscript, by employing certain symmetries that play a prominent role in quantum information or physics in general, which are the permutational symmetry and invariance under the action of certain stabilizer groups, the authors are able to go significantly beyond what has been known so far. Precisely, they provide two no- go theorems for distribution of entanglement in quantum networks that can be summarized as follows: (i) a certain large class of N- particle stabilizer states cannot be prepared in quantum networks with bipartite sources, (ii) no symmetric entangled mixed states can be prepared in networks with (N- 1)- partite sources. These statements are then used to design methods of certifying that a given link of the network functions properly.  
+
+I find this paper very interesting and important. By nicely combining certain symmetries with the inflation method it significantly moves our understanding of quantum networks forward. Consequently, I am happy to recommend it for publication in Nature Communications. There are a few suggestions and corrections that can be taken into account while preparing the final version:  
+
+1. It is not entirely clear, at least to me, whether Observation 2 concerns the symmetric states of arbitrary local dimension, or similarly to Observation 1 only the qubit ones. I think it would be useful
+
+<--- Page Split --->
+
+to reveal that information in the main text.  
+
+2. I don't quite understand why the authors cited [26] in a footnote. If Ref. [26] contains similar statements to those made in Observation 1 (as the authors state in the footnote), it would better to mention that explicitly in the text. In fact, the authors could add a sentence or two with a short comparison between the present results and those of [26].  
+
+3. I think it would be useful to add a few sentences commenting on how the results on the stabilizer states would change if the authors considered sources that link more parties than two.  
+
+4. In Supplementary Note there are some problems with references to Figures and Equations [between Eqs. (3) and (4) and Eqs. (40) and (41)].  
+
+5. There is redundant word 'Interestingly' in the third sentence of the section 'Certifying network links'.  
+
+6. The last sentence of the work: 'result open' \(\rightarrow\) 'result opens'.
+
+<--- Page Split --->
+
+## REPLY TO REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The submitted manuscript deals with a very timely topic, the characterization of quantum correlations that can be observed in networks. Namely, the network geometry severely restricts types of quantum correlations. The authors focus on density matrices, i.e. quantum states, and try to answer how we can determine what kind of quantum states can be generated with a given network. Up to now the main method for approaching this problem was using inflation techniques. However, the technique is applicable only for networks involving a rather small number of nodes and edges. In this manuscript the authors suggest using symmetry to make the use of inflation more economical. They show that the method works very well for stabilizer states and those with bosonic or fermionic symmetry.  
+
+I like the problem the authors consider. The method might be quite useful in further research, and I think that this paper should be highlighted by publication in Nature Communications. It considers a fairly unexplored problem and provides a concrete step forward. In addition, the paper is well written and easy to read and understand even by non- specialists.  
+
+Our reply: We would like to thank the reviewer for the time they took to read and comment on our manuscript. We are glad the reviewer liked our manuscript and recommended it for publication.  
+
+Reviewer #2 (Remarks to the Author):  
+
+This manuscript is concerned with the quantum networks scenario which consists of many spatially separated observers that are linked by different sources of quantum correlations and can perform local operations on their shares of distributed quantum states. Moreover, the implementation of these local channels is coordinated by classical correlations distributed to the observers by a single source. On the other hand, the authors impose one constraints on the considered scenario that no classical communication between parties is allowed. This assumption is well motivated in the manuscript and at the same time makes the considered scenario highly nontrivial.  
+
+Quantum networks constitute a paradigmatic framework in quantum information as it models for instance quantum communication protocols or quantum Internet. For this reason characterization of quantum correlations that can be prepared within quantum networks under various assumptions is gaining a lot of attention recently. At the same time this problem is extremely difficult to study both from analytical and numerical perspectives and therefore most of the results on distribution of entangled states concern networks of the simplest topologies such the triangle networks.  
+
+In the present manuscript, by employing certain symmetries that play a prominent role in quantum information or physics in general, which are the permutational symmetry and invariance under the action of certain stabilizer groups, the authors are able to go significantly beyond what has been known so far. Precisely, they provide two no- go theorems for distribution of entanglement in quantum networks that can be summarized as follows: (i) a certain large class of N- particle stabilizer states cannot be prepared in quantum networks with bipartite sources, (ii) no symmetric entangled mixed states can be prepared in networks with (N- 1)- partite sources. These statements are then used to design methods of certifying that a given link of the network functions properly.  
+
+I find this paper very interesting and important. By nicely combining certain symmetries with the
+
+<--- Page Split --->
+
+inflation method it significantly moves our understanding of quantum networks forward. Consequently, I am happy to recommend it for publication in Nature Communications. There are a few suggestions and corrections that can be taken into account while preparing the final version:  
+
+Our reply: We would like to thank the reviewer for the time they took to read and evaluate our paper. We are glad the reviewer found our manuscript interesting and recommended it for publication.  
+
+1. It is not entirely clear, at least to me, whether Observation 2 concerns the symmetric states of arbitrary local dimension, or similarly to Observation 1 only the qubit ones. I think it would be useful to reveal that information in the main text.  
+
+Our reply: We thank the reviewer for pointing this out, it indeed holds for arbitrary local dimensions. A sentence has been added above Observation 2 to clarify this point.  
+
+2. I don't quite understand why the authors cited [26] in a footnote. If Ref. [26] contains similar statements to those made in Observation 1 (as the authors state in the footnote), it would better to mention that explicitly in the text. In fact, the authors could add a sentence or two with a short comparison between the present results and those of [26].  
+
+Our reply: We thank the reviewer for addressing this. The footnote has been removed and a paragraph has been added to compare our results to the ones of Ref. [26].  
+
+3. I think it would be useful to add a few sentences commenting on how the results on the stabilizer states would change if the authors considered sources that link more parties than two.  
+
+Our reply: We thank the reviewers for this comment. We added a paragraph (page 4, above Permutational symmetry) commenting on this. It turns out that our method based on an anticommuting relation might be generalized to more than bipartite sources. We present in the supplemental material two different examples of the method applied to networks with tripartite sources, which are generalizations of the GHZ- method for the triangle and the \(C_4\) - method for the square network.  
+
+4. In Supplementary Note there are some problems with references to Figures and Equations [between Eqs. (3) and (4) and Eqs. (40) and (41)].  
+
+5. There is redundant word 'Interestingly' in the third sentence of the section 'Certifying network links'.  
+
+6. The last sentence of the work: 'result open' \(\rightarrow\) 'result opens'.  
+
+Our reply: We thank the reviewer for those last three comments, they have been corrected.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__441f9407302393914095b7edc3ad47804b8a172b40d660d04f980b77da7e41d6/supplementary_0_Peer Review File__441f9407302393914095b7edc3ad47804b8a172b40d660d04f980b77da7e41d6_det.mmd

@@ -0,0 +1,139 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 507, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[106, 155, 880, 211]]<|/det|>
+Symmetries in quantum networks lead to no- go theorems for entanglement distribution and to verification techniques  
+
+<|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|>text<|/ref|><|det|>[[116, 90, 286, 103]]<|/det|>
+Reviewers' Comments:  
+
+<|ref|>text<|/ref|><|det|>[[116, 120, 217, 133]]<|/det|>
+Reviewer #1:  
+
+<|ref|>text<|/ref|><|det|>[[116, 135, 291, 148]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 150, 875, 268]]<|/det|>
+The submitted manuscript deals with a very timely topic, the characterization of quantum correlations that can be observed in networks. Namely, the network geometry severely restricts types of quantum correlations. The authors focus on density matrices, i.e. quantum states, and try to answer how we can determine what kind of quantum states can be generated with a given network. Up to now the main method for approaching this problem was using inflation techniques. However, the technique is applicable only for networks involving a rather small number of nodes and edges. In this manuscript the authors suggest using symmetry to make the use of inflation more economical. They show that the method works very well for stabilizer states and those with bosonic or fermionic symmetry.  
+
+<|ref|>text<|/ref|><|det|>[[115, 283, 871, 343]]<|/det|>
+I like the problem the authors consider. The method might be quite useful in further research, and I think that this paper should be highlighted by publication in Nature Communications. It considers a fairly unexplored problem and provides a concrete step forward. In addition, the paper is well written and easy to read and understand even by non- specialists.  
+
+<|ref|>text<|/ref|><|det|>[[115, 387, 216, 400]]<|/det|>
+Reviewer #2:  
+
+<|ref|>text<|/ref|><|det|>[[116, 402, 291, 415]]<|/det|>
+Remarks to the Author:  
+
+<|ref|>text<|/ref|><|det|>[[115, 417, 881, 521]]<|/det|>
+This manuscript is concerned with the quantum networks scenario which consists of many spatially separated observers that are linked by different sources of quantum correlations and can perform local operations on their shares of distributed quantum states. Moreover, the implementation of these local channels is coordinated by classical correlations distributed to the observers by a single source. On the other hand, the authors impose one constraints on the considered scenario that no classical communication between parties is allowed. This assumption is well motivated in the manuscript and at the same time makes the considered scenario highly nontrivial.  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 863, 625]]<|/det|>
+Quantum networks constitute a paradigmatic framework in quantum information as it models for instance quantum communication protocols or quantum internet. For this reason characterization of quantum correlations that can be prepared within quantum networks under various assumptions is gaining a lot of attention recently. At the same time this problem is extremely difficult to study both from analytical and numerical perspectives and therefore most of the results on distribution of entangled states concern networks of the simplest topologies such the triangle networks.  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 878, 760]]<|/det|>
+In the present manuscript, by employing certain symmetries that play a prominent role in quantum information or physics in general, which are the permutational symmetry and invariance under the action of certain stabilizer groups, the authors are able to go significantly beyond what has been known so far. Precisely, they provide two no- go theorems for distribution of entanglement in quantum networks that can be summarized as follows: (i) a certain large class of N- particle stabilizer states cannot be prepared in quantum networks with bipartite sources, (ii) no symmetric entangled mixed states can be prepared in networks with (N- 1)- partite sources. These statements are then used to design methods of certifying that a given link of the network functions properly.  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 880, 834]]<|/det|>
+I find this paper very interesting and important. By nicely combining certain symmetries with the inflation method it significantly moves our understanding of quantum networks forward. Consequently, I am happy to recommend it for publication in Nature Communications. There are a few suggestions and corrections that can be taken into account while preparing the final version:  
+
+<|ref|>text<|/ref|><|det|>[[115, 864, 866, 894]]<|/det|>
+1. It is not entirely clear, at least to me, whether Observation 2 concerns the symmetric states of arbitrary local dimension, or similarly to Observation 1 only the qubit ones. I think it would be useful
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 435, 104]]<|/det|>
+to reveal that information in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[115, 119, 861, 180]]<|/det|>
+2. I don't quite understand why the authors cited [26] in a footnote. If Ref. [26] contains similar statements to those made in Observation 1 (as the authors state in the footnote), it would better to mention that explicitly in the text. In fact, the authors could add a sentence or two with a short comparison between the present results and those of [26].  
+
+<|ref|>text<|/ref|><|det|>[[115, 194, 861, 225]]<|/det|>
+3. I think it would be useful to add a few sentences commenting on how the results on the stabilizer states would change if the authors considered sources that link more parties than two.  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 810, 270]]<|/det|>
+4. In Supplementary Note there are some problems with references to Figures and Equations [between Eqs. (3) and (4) and Eqs. (40) and (41)].  
+
+<|ref|>text<|/ref|><|det|>[[115, 283, 839, 314]]<|/det|>
+5. There is redundant word 'Interestingly' in the third sentence of the section 'Certifying network links'.  
+
+<|ref|>text<|/ref|><|det|>[[115, 329, 587, 344]]<|/det|>
+6. The last sentence of the work: 'result open' \(\rightarrow\) 'result opens'.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[93, 66, 399, 82]]<|/det|>
+## REPLY TO REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[93, 98, 408, 115]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[92, 130, 904, 280]]<|/det|>
+The submitted manuscript deals with a very timely topic, the characterization of quantum correlations that can be observed in networks. Namely, the network geometry severely restricts types of quantum correlations. The authors focus on density matrices, i.e. quantum states, and try to answer how we can determine what kind of quantum states can be generated with a given network. Up to now the main method for approaching this problem was using inflation techniques. However, the technique is applicable only for networks involving a rather small number of nodes and edges. In this manuscript the authors suggest using symmetry to make the use of inflation more economical. They show that the method works very well for stabilizer states and those with bosonic or fermionic symmetry.  
+
+<|ref|>text<|/ref|><|det|>[[92, 294, 900, 362]]<|/det|>
+I like the problem the authors consider. The method might be quite useful in further research, and I think that this paper should be highlighted by publication in Nature Communications. It considers a fairly unexplored problem and provides a concrete step forward. In addition, the paper is well written and easy to read and understand even by non- specialists.  
+
+<|ref|>text<|/ref|><|det|>[[93, 376, 880, 427]]<|/det|>
+Our reply: We would like to thank the reviewer for the time they took to read and comment on our manuscript. We are glad the reviewer liked our manuscript and recommended it for publication.  
+
+<|ref|>text<|/ref|><|det|>[[93, 458, 410, 475]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[92, 491, 902, 608]]<|/det|>
+This manuscript is concerned with the quantum networks scenario which consists of many spatially separated observers that are linked by different sources of quantum correlations and can perform local operations on their shares of distributed quantum states. Moreover, the implementation of these local channels is coordinated by classical correlations distributed to the observers by a single source. On the other hand, the authors impose one constraints on the considered scenario that no classical communication between parties is allowed. This assumption is well motivated in the manuscript and at the same time makes the considered scenario highly nontrivial.  
+
+<|ref|>text<|/ref|><|det|>[[92, 622, 900, 739]]<|/det|>
+Quantum networks constitute a paradigmatic framework in quantum information as it models for instance quantum communication protocols or quantum Internet. For this reason characterization of quantum correlations that can be prepared within quantum networks under various assumptions is gaining a lot of attention recently. At the same time this problem is extremely difficult to study both from analytical and numerical perspectives and therefore most of the results on distribution of entangled states concern networks of the simplest topologies such the triangle networks.  
+
+<|ref|>text<|/ref|><|det|>[[92, 754, 902, 888]]<|/det|>
+In the present manuscript, by employing certain symmetries that play a prominent role in quantum information or physics in general, which are the permutational symmetry and invariance under the action of certain stabilizer groups, the authors are able to go significantly beyond what has been known so far. Precisely, they provide two no- go theorems for distribution of entanglement in quantum networks that can be summarized as follows: (i) a certain large class of N- particle stabilizer states cannot be prepared in quantum networks with bipartite sources, (ii) no symmetric entangled mixed states can be prepared in networks with (N- 1)- partite sources. These statements are then used to design methods of certifying that a given link of the network functions properly.  
+
+<|ref|>text<|/ref|><|det|>[[92, 902, 881, 920]]<|/det|>
+I find this paper very interesting and important. By nicely combining certain symmetries with the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[92, 66, 891, 115]]<|/det|>
+inflation method it significantly moves our understanding of quantum networks forward. Consequently, I am happy to recommend it for publication in Nature Communications. There are a few suggestions and corrections that can be taken into account while preparing the final version:  
+
+<|ref|>text<|/ref|><|det|>[[92, 131, 880, 181]]<|/det|>
+Our reply: We would like to thank the reviewer for the time they took to read and evaluate our paper. We are glad the reviewer found our manuscript interesting and recommended it for publication.  
+
+<|ref|>text<|/ref|><|det|>[[92, 197, 870, 246]]<|/det|>
+1. It is not entirely clear, at least to me, whether Observation 2 concerns the symmetric states of arbitrary local dimension, or similarly to Observation 1 only the qubit ones. I think it would be useful to reveal that information in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[92, 262, 833, 296]]<|/det|>
+Our reply: We thank the reviewer for pointing this out, it indeed holds for arbitrary local dimensions. A sentence has been added above Observation 2 to clarify this point.  
+
+<|ref|>text<|/ref|><|det|>[[92, 311, 894, 378]]<|/det|>
+2. I don't quite understand why the authors cited [26] in a footnote. If Ref. [26] contains similar statements to those made in Observation 1 (as the authors state in the footnote), it would better to mention that explicitly in the text. In fact, the authors could add a sentence or two with a short comparison between the present results and those of [26].  
+
+<|ref|>text<|/ref|><|det|>[[92, 393, 857, 427]]<|/det|>
+Our reply: We thank the reviewer for addressing this. The footnote has been removed and a paragraph has been added to compare our results to the ones of Ref. [26].  
+
+<|ref|>text<|/ref|><|det|>[[92, 442, 896, 475]]<|/det|>
+3. I think it would be useful to add a few sentences commenting on how the results on the stabilizer states would change if the authors considered sources that link more parties than two.  
+
+<|ref|>text<|/ref|><|det|>[[92, 491, 886, 590]]<|/det|>
+Our reply: We thank the reviewers for this comment. We added a paragraph (page 4, above Permutational symmetry) commenting on this. It turns out that our method based on an anticommuting relation might be generalized to more than bipartite sources. We present in the supplemental material two different examples of the method applied to networks with tripartite sources, which are generalizations of the GHZ- method for the triangle and the \(C_4\) - method for the square network.  
+
+<|ref|>text<|/ref|><|det|>[[92, 606, 840, 639]]<|/det|>
+4. In Supplementary Note there are some problems with references to Figures and Equations [between Eqs. (3) and (4) and Eqs. (40) and (41)].  
+
+<|ref|>text<|/ref|><|det|>[[92, 655, 872, 688]]<|/det|>
+5. There is redundant word 'Interestingly' in the third sentence of the section 'Certifying network links'.  
+
+<|ref|>text<|/ref|><|det|>[[92, 705, 587, 722]]<|/det|>
+6. The last sentence of the work: 'result open' \(\rightarrow\) 'result opens'.  
+
+<|ref|>text<|/ref|><|det|>[[90, 737, 845, 755]]<|/det|>
+Our reply: We thank the reviewer for those last three comments, they have been corrected.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__4434c62cfd8163014f10888e9efa5b5818b5458a31008d13e8e813f04a96a9ab/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__4434c62cfd8163014f10888e9efa5b5818b5458a31008d13e8e813f04a96a9ab/supplementary_0_Peer Review File__4434c62cfd8163014f10888e9efa5b5818b5458a31008d13e8e813f04a96a9ab.mmd

@@ -0,0 +1,623 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Internal states as a source of subject- dependent movement variability are represented by large- scale brain networks  
+
+![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 authors sought to link latent factors ("internal states") to motor performance via neural activity. 10 participants with stereo electroencephalographic (SEEG) recordings performed a reaching task using a robotic manipulandum. The speed (fast, slow) with which participants were to complete their reach varied across trials. Additionally, on \(20\%\) of trials a perturbation was applied either toward or away from the target. Through state- space representation modeling, the authors identified two key internal states: the error state, which tracks past errors and the perturbed state, which tracks past perturbations, to predict variability in both reaction times and speed errors - - how much the actual speed on a given trial deviated from the target speed. The authors claim that spectral signals in the dorsal attention network (DAN) track the error state whereas the default mode network (DN) tracks the perturbation state and conclude that these networks regulate motor strategy through internal states.  
+
+Although the question of how past experience influences motor learning would be of interest to the field, the interpretations made in this manuscript are not supported by the analyses and results. The limited number of participants - understandable due to the population - severely limits conclusions drawn regarding across subject variability. A number of methodological details are absent or challenging to understand. Below I outline my concerns.  
+
+1. Absence of direct statistical tests.  
+
+The authors make claims regarding the neural substrate of error and perturbation states without the appropriate statistical tests. "Remarkably, we then found that these internal states were linked to encoding in large- scale brain networks, DAN and DN, respectively." (Page 22) and "For the first time, our results implicate DAN as a network as encoding tracking history" (Page 23). However, to make such claims, a direct statistical test must be performed on DAN vs. at least one other region, and the other region cannot have been derived from visual inspection of a figure. The presence of an effect in a particular region does not indicate that the effect is specific or distinctive to that region. The same is true for frequency specificity; the authors highlight both high gamma (100- 200 Hz) and low frequency (<15Hz) effects without proper statistical comparisons across frequencies.  
+
+2. An overwhelming lack of statistical reporting and claims that appear to be made on the basis of visual inspection rather than appropriate statistical testing.
+
+<--- Page Split --->
+
+Every statistical test reported in the manuscript needs to include report of a test statistic, degrees of freedom, p- value, and an effect size. These can be reported in a table but it is insufficient to report p- values alone. Additionally, post- hoc tests must be reported for any follow up analyses. Correction for multiple comparison must also be performed and reported. Many of the reported results cannot be evaluated due to insufficient statistical information. For example, the authors report, "Indeed, subjects reacted more quickly for fast trials than slow trials \((p = 0.014,\) ANOVA). They also reacted more slowly for trials with an upward motion to the target \((p = 0.013,\) ANOVA)." (Page 8). F- statistics, degrees of freedom, and effect sizes must be reported. What type of ANOVA was used and what were the factors? Furthermore, it seems that these results should have been found via a post- hoc t- test following a multiway ANOVA. Means and standard deviation of the individual conditions (e.g. RT for upward trials) should also be reported. There is an over reliance on figures to support claims rather than statistical tests (e.g. "Visually, the estimate follows key features, including sudden jumps between trials and gradual changes such as between trials 100 and 125. For example, the estimate on trial 109 (black triangle) matches what was observed, which was that subject 6 reacted faster than average."). Multiple figures also specifically highlight Subject 6 and it is unclear how/why this participant was selected.  
+
+3. The connectivity analysis is confusing and challenging to interpret.  
+
+The authors describe the connectivity analysis as  
+
+"To calculate connectivity, we began by averaged the neural activity of each group of clusters using its time- frequency window across the epochs it spans for each channel, trial, and subject. Then, we found the magnitude of the Pearson correlation value between each pair the averaged neural activity of group of clusters and channels for each subject across trials. Pairs of channels in the same group of clusters were then averaged within each subject. These values represent the connectivity strength of the pair of regions for each subject."  
+
+It's not clear what is being correlated exactly. The authors use the term "pair of regions," but it is unclear what that means or how regions are defined. How many pairs of regions were identified for each participant and how many correlations were performed? I am concerned about the number of statistical tests that were performed without correction for multiple comparisons. Furthermore, the connectivity analysis was only performed on top performers. This choice was insufficiently justified (Page 19, lines 6- 8), limits the ability to generalize the current findings beyond this group of participants, and renders claims such as "...exhibits stronger functional connectivity for top performers" (Page 3, lines 2- 3) inaccurate since there are no direct comparisons.  
+
+4. The confluence of multiple factors makes any results difficult to generalize.
+
+<--- Page Split --->
+
+Although intracranial EEG can present a unique opportunity and researchers leveraging such data are limited by clinical necessities, the choice to investigate large scale networks seems suboptimal in a small population with (necessarily) variable recording locations. The number of subjects with a recording in any given region is small (the maximum appears to be 6). The number of perturbation trials is low, and appears to be imbalanced when separately considering toward vs. away, making fitting of the perturbation state more challenging. Dividing the participants into top and bottom performers further reduces the sample size and is hard to justify when the original N is already so low. The authors state that "Top performers benefited the most from adding internal states to RT" but fail to mention that RT is much better modeled for top performers ( \(\sim 20 - 60\%\) coefficient of determination) than for bottom performers ( \(\sim 3 - 5\%\) coefficient of determination) in both models, including the model without states added. In general it seems that more participants would be needed for many of the claims that are made.  
+
+5. Inappropriate frequency band selection.  
+
+The authors state, "The frequency band was identified by matching the frequency bins to frequency bands commonly defined in literature" (Page 46). It's unclear what this means, exactly, but it is not appropriate to define bands by visual inspection of time- frequency spectrograms. Bands should be defined a priori without respect to the current data.  
+
+Reviewer #2 (Remarks to the Author):  
+
+Thank you for the opportunity to review "Internal states as a source of subject- dependent movement variability and their representation by large- scale networks" by Breault and colleagues. The manuscript describes an SEEG study looking at how movement variability is encoded in the brain. The authors first fit an internal state model to participant behavioural data in a reaching task and find evidence of two states, related to errors and perturbations. They then localize these states to the well- studied dorsal attention and default mode networks.  
+
+The study addresses an important question that is likely to be of wide interest to the field. The authors carefully triangulate towards a coherent explanation of how internal states govern movement variability, shedding light on the neural mechanisms underlying speed- accuracy trade- offs. However, as I outline below, there are several conceptual and methodological concerns that should be addressed before recommending publication.
+
+<--- Page Split --->
+
+1. "Such reflections form latent factors called internal states that induce variability of movement and behavior to improve performance." This is minor, but there is a general tendency in the narrative, particularly in the Abstract and Introduction, to talk about internal states as a real biological entity. I think this is misleading, because all work on states has the starting assumption that ongoing behaviour/neural activity can be partitioned into states. I don't think the narrative would suffer if the authors discussed internal states more as a theoretical and methodological construct, rather than as a real biological phenomenon.  
+
+2. There is a considerable literature on the role of signal variability in noninvasive human imaging that could potentially be relevant, e.g.  
+
+Garrett, D. D., Samanez-Larkin, G. R., MacDonald, S. W., Lindenberger, U., McIntosh, A. R., & Grady, C. L. (2013). Moment-to-moment brain signal variability: a next frontier in human brain mapping?. Neuroscience & Biobehavioral Reviews, 37(4), 610-624.  
+
+McIntosh, A. R., Kovacevic, N., & Itier, R. J. (2008). Increased brain signal variability accompanies lower behavioral variability in development. PLoS computational biology, 4(7), e1000106.  
+
+3. There is also a large literature on dynamic states in human imaging, e.g.  
+
+Lurie, D. J., Kessler, D., Bassett, D. S., Betzel, R. F., Breakspear, M., Kheilholz, S., ... & Calhoun, V. D. (2020). Questions and controversies in the study of time-varying functional connectivity in resting fMRI. Network neuroscience, 4(1), 30-69.  
+
+4. Why standard deviation as opposed to the coefficient of variation, which takes mean scaling differences into account?  
+
+5. My main methodological concern is the arbitrary (mean) splitting of participants into top and bottom performers. This choice is unnecessary and potentially suboptimal as it transforms an inherently continuous variable into a categorical one. All the analyses could then be performed using correlations (with bootstrapping) rather than ANOVAs.  
+
+6. Connectivity estimated as correlated power. Why not use an established method that focuses on coherence/synchrony or on correlating amplitude envelopes?
+
+<--- Page Split --->
+
+7. Is there any cross-validation for the fitted internal states model? The authors compare their model with a pure linear model and show that the former is better, but is there any way to demonstrate that the fitted model helps to predict future values of RT?  
+
+8. "First, a Notch filter was applied using notches located at the fundamental frequency of 60 Hz with bandwidth at the \(-1\) dB point set to \(3 \text{Hz}\) ". The word "notch" does not need capitalizing. Also, did the authors include filters at subsequent harmonics of \(60 \text{Hz}\) ?  
+
+9. The authors assign regions to frequency bands, but does this method account for the fact that the power spectrum has a \(1 / f\) shape and low frequencies will naturally tend to have greater power than higher frequencies?  
+
+10. The other major methodological concern is how network specificity is established for the reported effects (e.g. localizing states to DAN or DMN), Namely, the comparison was clearly made visually/qualitatively, but the different networks all have different size, spatial coverage, etc. This means that some networks are more likely to get "hits" than others. The authors should confirm that each network is enriched for a particular state by computing the mean in the network and then comparing this to a null distribution where network labels are permuted.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The main topic addressed by this paper is to examine how the brain represents internal states during motor processing, and to reveal the neural substrate of these patterns. Specifically, this paper utilizes state- space models to link internal states (past errors and perturbations) with motor performance and neural activity, and to identify variations in performance across subjects and trials. While performing a goal- directed center- out reaching task, ten human subjects' electrophysiological data and behavioral data were collected. The paper concluded that large- scale brain networks in the dorsal attention and default networks reflect the neural basis of strategy to regulate movement variability through internal states (past error and perturbation states, respectively) to improve motor performance.  
+
+Overall, the idea of examining the neural basis of internal states in regulating an individual's motor performance is interesting. The paper is well written and has many innovative ideas. However, my enthusiasm for the paper is lowered the impact of the neural data analyses is rather limited. The neural power analyses, which found signals at particular frequencies that correlated with aspects of behavior, were not described in a way that lets the reader make specific inferences about the functional contributions of those regions to behavior. The synchrony analyses were improved in this way, but still,
+
+<--- Page Split --->
+
+there was not enough detail to make a compelling case about the functional or mechanistic roles of particular regions or networks to support motor processing. Further, I have significant issues with the methods and data analysis that the authors presented.  
+
+Specific critiques:  
+
+The behavioral data were not presented with enough detail for me to understand the differences across good versus bad subjects, as well as to explain differences in the subjects that the model fit well versus poorly. I would suggest expanding figure 1 and the related sections of the text to separately analyze and interpret the behavioral characteristics of the group data for the subjects with different levels of absolute performance and for those who the model did or did not fit well.  
+
+The neural data analyses for figures 4, 5, and 6 are not illustrated with adequate statistical justification. I could not tell which neural signals correlated with behavior in a statistically robust fashion, and there was a lack of adjustment for multiple comparisons. Overall this section of the paper was largely qualitative, which is inadequate. This section of the paper should rigorously show which neural signals, across frequencies, correlated with behavior, and include both within- and across subject comparisons.  
+
+The analysis described in Figure 2 was hard to understand because it was focused so much on just subject 6.  
+
+The behavioral data in Figure 2 are hard to interpret because panels B- D are visually very dense. Panel E seems to indicate that the model fit quality is bimodal, with a cluster of four subjects who have four model fits and others who show rather good model fits. Does this undermine the results?  
+
+Figure 4C is hard to interpret because the statistic is calculated based on the correlation but the colors reflect a different value, mean power. It would be much easier to interpret this plot if the same value was plotted as indicated in the statistic.  
+
+Figure 6 is hard to interpret, is there a more intuitive or data- driven way to order the labels around the plots?  
+
+The analysis on Page 20, Line 30 and Fig 6 could benefit from a more quantitative comparison between network connectivity during "error state" and "perturbed state" rather than qualitative comparisons. There are various quantitative measures that the authors might use for this network comparison
+
+<--- Page Split --->
+
+Perhaps it is a minor point, but I found the conclusion of the paper confusing. How are the results relevant for the type of behavioral data that could be collected from smartphones? There might be an interesting idea here but it needs more length to spell out.  
+
+Please explain in the text that discuss in the text why was it was valid for the subjects to be divided based on a median split of RT? Some "top performers" may be doing relatively poorly in the task.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors sought to link latent factors ("internal states") to motor performance via neural activity. 10 participants with stereo electroencephalographic (SEEG) recordings performed a reaching task using a robotic manipulandum. The speed (fast, slow) with which participants were to complete their reach varied across trials. Additionally, on \(20\%\) of trials a perturbation was applied either toward or away from the target. Through state- space representation modeling, the authors identified two key internal states: the error state, which tracks past errors and the perturbed state, which tracks past perturbations, to predict variability in both reaction times and speed errors - - how much the actual speed on a given trial deviated from the target speed. The authors claim that spectral signals in the dorsal attention network (DAN) track the error state whereas the default mode network (DN) tracks the perturbation state and conclude that these networks regulate motor strategy through internal states.  
+
+Although the question of how past experience influences motor learning would be of interest to the field, the interpretations made in this manuscript are not supported by the analyses and results. The limited number of participants - understandable due to the population - severely limits conclusions drawn regarding across subject variability. A number of methodological details are absent or challenging to understand. Below I outline my concerns.  
+
+Authors: We thank the reviewer for their helpful comments and positive feedback. As detailed below we have revised the manuscript to clarify the rigorous statistical methods that were applied in our analyses and made changes in the main document as well as supplemental material. In addition, we provide a point- by- point response to each of the reviewer's specific comments and indicate where we revised the manuscript.  
+
+## 1. Absence of direct statistical tests.  
+
+The authors make claims regarding the neural substrate of error and perturbation states without the appropriate statistical tests. "Remarkably, we then found that these internal states were linked to encoding in large- scale brain networks, DAN and DN, respectively." (Page 22) and "For the first time, our results implicate DAN as a network as encoding tracking history" (Page 23). However, to make such claims, a direct statistical test must be performed on DAN vs. at least one other region, and the other region cannot have been derived from visual inspection of a figure. The presence of an effect in a particular region does not indicate that the effect is specific or distinctive to that region. The same is true for frequency specificity; the authors highlight both high gamma (100- 200 Hz) and low frequency (<15Hz) effects without proper statistical comparisons across frequencies.
+
+<--- Page Split --->
+
+Authors: Thank you for pointing out that we were not clear about the rigor of our neural analysis. To clarify, all of our neural results were derived using an unsupervised and data- driven approach, which were supported by statistical tests at all steps. First, the raw SEEG data was preprocessed into spectral data of normalized power in the time and frequency domains. The spectral data were then sliced into trials, which were then sliced into epochs. Next, we used a non- parametric cluster statistical test (Maris & Oostenveld, 2007) to find windows of time and frequency in which the spectral power correlated with each internal state across trials for each region across the population. This method is agnostic to time and frequency and avoids the burden of needing to correct for multiple comparisons at this level. These windows are called clusters, and each has an associated cluster statistic (Figure 4c). Finally, we corrected for multiple comparisons (one comparison for each brain region for each epoch for each internal state) using a false- discovery rate (FDR) of \(q = 0.015\) . We hope this summary helps clarify our procedure and have revised the manuscript as detailed below.  
+
+Specifically, we now clarify that the results that we report in our paper are only those that pass the FDR above (see Tables 2 and 3 for statistics from nonparametric cluster analysis). As a result, this approach narrowed down our scope from 40 possible regions to just 12 for the error state and 14 for the perturbation state. Based on this data- driven approach, we noticed that the error state was populated by regions in DAN while the perturbation state was populated by regions in DN (Figure 5a- b).  
+
+Thus, our approach was entirely reliant on our statistical approach, however, we appreciate that this was not explained. To be more clear, we have revised our manuscript as follows:  
+
+## 1. Page 16-17:  
+
+- "To identify the neural correlates amongst these regions with an unsupervised and data-driven method, we used a non-parametric cluster statistic [36] (see Methods) between the spectral data of each region and each internal state across the population (Fig. 4b). This method finds windows of time (relative to epoch onset) and frequency (between 1 and 200 Hz) in which the spectral power of a region significantly correlates with the internal state across trials as demonstrated in Fig. 4c. The statistic provided us with two sets of neural correlates, one for each internal state."  
+
+- "Our modeling results showed that subjects weighed internal states differently, primarily based on their session performance. We suspected that this would be reflected in the brain for each subject by how well these regions encoded the states through the strength of their neural correlates. The degree to which a subject encodes an internal state in a region, which we called the encoding
+
+<--- Page Split --->
+
+strength, was quantified by correlating the average power within the time- frequency window (from the population statistic) to the state on a trial- by- trial basis (see Methods). Fig. 4d shows an example of how the encoding strength is obtained for a channel in subject 6 using the neural correlate from Fig. 4c.  
+
+2. Figure 4, panel b,c,d  
+
+2. An overwhelming lack of statistical reporting and claims that appear to be made on the basis of visual inspection rather than appropriate statistical testing.  
+
+Every statistical test reported in the manuscript needs to include a report of a test statistic, degrees of freedom, p-value, and an effect size. These can be reported in a table but it is insufficient to report p-values alone. Additionally, post-hoc tests must be reported for any follow up analyses. Correction for multiple comparisons must also be performed and reported. Many of the reported results cannot be evaluated due to insufficient statistical information. For example, the authors report, "Indeed, subjects reacted more quickly for fast trials than slow trials ( \(p = 0.014\) , ANOVA). They also reacted more slowly for trials with an upward motion to the target ( \(p = 0.013\) , ANOVA)." (Page 8). F-statistics, degrees of freedom, and effect sizes must be reported. What type of ANOVA was used and what were the factors? Furthermore, it seems that these results should have been found via a post-hoc t-test following a multi-way ANOVA. Means and standard deviation of the individual conditions (e.g. RT for upward trials) should also be reported. There is an over reliance on figures to support claims rather than statistical tests (e.g. "Visually, the estimate follows key features, including sudden jumps between trials and gradual changes such as between trials 100 and 125. For example, the estimate on trial 109 (black triangle) matches what was observed, which was that subject 6 reacted faster than average."). Multiple figures also specifically highlight Subject 6 and it is unclear how/why this participant was selected.  
+
+Authors: Thank you for your suggestions. We agree that our manuscript required more detailed reporting of statistics. We have directly addressed these concerns in our revised manuscript. Specifically, we have added more detailed statistics in our text to include the statistical test, test statistic, degrees of freedom, p- value and effect size when appropriate (pages 7, 8, 9, 11, 12, 14, 18, 20) as well as added tables of statistics (ANOVA, post- hoc test, etc.) for behavior across conditions (Supplementary Tables 3, 4, 5, 6, 7). For example, the claim on page 8 stated as "they reacted more slowly for trials when the target was up compared to when the target was down (post- hoc Tukey's: \(p = 0.0037\) ) or right (post- hoc Tukey's: \(p = 0.016\) )" are now supported by post- hoc test reported in the text but detailed in Supplementary Tables 4 and 5.  
+
+The claim about sudden and gradual changes characterizing the conditions and states respectively can be captured using their averaged absolute second derivative (which we've added as Supplementary Figure 7a), in which conditions
+
+<--- Page Split --->
+
+had significantly larger values than states for both the RT (paired- sample t- test: t(9)=4.96, p=0.0008) and SE (paired- sample t- test: t(9)=6.91, p=7.02e- 05). We added this result as well (page 11, "Conditions accounted for the trial- by- trial changes and states accounted for the gradual changes across all subjects (Supplementary Figure 7a).")  
+
+Regarding subject 6, we now note that we focused on this subject to exemplify key points where we wanted the reader to gain intuition through visuals, which are backed up by the statistics as well. We added the following line to frame the reader for this thinking: pages 10—11 "Though we used subject 6 to demonstrate the intuition behind our model, our general observations are applicable to the population, who are shown in Supplementary Fig. 1—Supplementary Fig. 5." We have also incorporated other subjects into the interpretation of our model results in this section as well:  
+
+- Page 11, "With the exception of RT for subject 1, we consistently found a significant relationship between the estimated and observed RT (Supplementary Fig. 1) and SE (Supplementary Fig. 2) for all subjects."- Page 12, "In RT, we also observed states that monotonically increased over trials for subjects 4, 5, 7, and 10 (Supplementary Fig. 3). This state reproduced subjects progressively reacting slower due to fatigue (Supplementary Fig. 1)"- Page 12, "Based on the structure, we can also see what influenced the RT states to be monotonic for subjects 4, 5, 7, and 10. Supplementary Figure 5 shows that the primary contributor was the perturbed state."  
+
+3. The connectivity analysis is confusing and challenging to interpret.  
+
+The authors describe the connectivity analysis as  
+
+"To calculate connectivity, we began by averaged the neural activity of each group of clusters using its time- frequency window across the epochs it spans for each channel, trial, and subject. Then, we found the magnitude of the Pearson correlation value between each pair the averaged neural activity of group of clusters and channels for each subject across trials. Pairs of channels in the same group of clusters were then averaged within each subject. These values represent the connectivity strength of the pair of regions for each subject."  
+
+It's not clear what is being correlated exactly. The authors use the term "pair of regions," but it is unclear what that means or how regions are defined. How many pairs of regions were identified for each participant and how many correlations were performed? I am concerned about the number of statistical tests that were performed without correction for multiple comparisons. Furthermore, the connectivity analysis was only performed on top performers. This choice was insufficiently justified (Page 19, lines 6- 8), limits the ability to generalize the current findings beyond this group of participants, and renders
+
+<--- Page Split --->
+
+claims such as "...exhibits stronger functional connectivity for top performers" (Page 3, lines 2- 3) inaccurate since there are no direct comparisons.  
+
+Authors: We appreciate the reviewer's comment and agree that the original presentation of connectivity was unclear. We have since added panels b,c, and d to Figure 6 to help explain the method we used to arrive at the connectivity results.  
+
+Continuing from our response to concern 1, once the clusters (windows of time and frequency) were statistically identified, we calculated the average power in each cluster for each trial (visualized in Figure 4b, new Figure 6a). This gave us a signal of how the power in the cluster varied over trials for each channel in a region. This naturally led us to run a connectivity analysis between these signals, via Pearson's correlation, for each subject (new Figure 6b). After obtaining correlations between each pair of channels, the correlations for common pairs were averaged together to get subject connectivity (i.e., average all the correlations in Figure 6b). "Pairs" refers to the fact that connectivity measures the interaction of two channels. For example Figure 6a- c shows how we arrived at subject connectivity for the pair IPS R and MTG R for subject 6. We have revised the connectivity methods to make this clearer (page 51- 53).  
+
+The number of correlations for each subject varied based on the number of regions that were identified using the non- parametric cluster statistic. But, to minimize the number of statistical tests, we only ran correlations within each internal state result. That is, we ran connectivity analysis for regions related to the error state and perturbed state separately. The following table summarizes the total number of correlations ran for each subject and internal state:  
+
+<table><tr><td>Subject</td><td>Error state</td><td>Perturb state</td></tr><tr><td>1</td><td>820</td><td>1830</td></tr><tr><td>2</td><td>45</td><td>528</td></tr><tr><td>3</td><td>15</td><td>91</td></tr><tr><td>4</td><td>595</td><td>630</td></tr><tr><td>5</td><td>0</td><td>0</td></tr><tr><td>6</td><td>153</td><td>325</td></tr><tr><td>7</td><td>66</td><td>153</td></tr><tr><td>8</td><td>3</td><td>28</td></tr><tr><td>9</td><td>91</td><td>300</td></tr></table>
+
+<--- Page Split --->
+
+
+<table><tr><td>10</td><td>703</td><td>2775</td></tr></table>
+
+which then averaged together to form unique pairs of regions for each subject into:
+
+<table><tr><td>Subject</td><td>Error state</td><td>Perturb state</td></tr><tr><td>1</td><td>52</td><td>132</td></tr><tr><td>2</td><td>3</td><td>48</td></tr><tr><td>3</td><td>4</td><td>20</td></tr><tr><td>4</td><td>43</td><td>61</td></tr><tr><td>5</td><td>0</td><td>0</td></tr><tr><td>6</td><td>14</td><td>43</td></tr><tr><td>7</td><td>5</td><td>10</td></tr><tr><td>8</td><td>0</td><td>1</td></tr><tr><td>9</td><td>26</td><td>73</td></tr><tr><td>10</td><td>41</td><td>116</td></tr></table>
+
+Though the connectivity in Figure 6d-e appears to only show top performers, the connectivity analysis on performance was actually performed across all subjects.Originally, we found that top performers had high connectivity between the pairs of regions shown in Figure 6d-e whereas the bottom performers had high connectivity between the pairs of regions in Supplementary Figure 9. Since then,we have removed the binary groups of subjects based on session performance and updated Supplementary Figure 9 to show all of the performance connectivity results.
+
+# 4. The confluence of multiple factors makes any results difficult to generalize.
+
+Although intracranial EEG can present a unique opportunity and researchers leveraging such data are limited by clinical necessities, the choice to investigate large scale networks seems suboptimal in a small population with (necessarily) variable recording locations. The number of subjects with a recording in any given region is small (the maximum appears to be 6). The number of perturbation trials is low, and appears to be imbalanced when separately considering toward vs. away, making fitting of the perturbation state more challenging. Dividing the participants into top and bottom
+
+<--- Page Split --->
+
+performers further reduces the sample size and is hard to justify when the original N is already so low. The authors state that "Top performers benefited the most from adding internal states to RT" but fail to mention that RT is much better modeled for top performers ( \(\sim 20 - 60\%\) coefficient of determination) than for bottom performers ( \(\sim 3 - 5\%\) coefficient of determination) in both models, including the model without states added. In general it seems that more participants would be needed for many of the claims that are made.  
+
+Authors: Thank you for your comment. Indeed, we are limited in the number of subjects, trials, regions, etc. Although the complete coverage cannot be obtained with any method of invasive monitoring, the spatial and temporal resolution with recordings from intracortical electrodes are optimal and highly superior to fMRI studies These challenges are part of the limitations but also correspond to the main advantages of the method. Similar data has been highly accepted and published in literature [Sacre et al. 2019, PNAS]. We also addressed these concerns in the Discussion, under Study limitations:  
+
+- Page 31: "At this time, the only ethical method to record from the brain necessary for our study using SEEG depth electrodes in humans is while they are implanted for clinical purposes."  
+
+As for perturbations, you are correct that the number of perturbation trials is low compared to unperturbed trials. If we were trying to predict whether a trial was going to be perturbed or not, then we would indeed be handicapped by the limited number of trials. However, the perturbed state is the accumulation of history fit using all trials. Therefore, the validity of the perturbed state does not depend on the number of perturbed trials. For example, if a patient never experienced a perturbation trial, then their perturbed state would be 0 for all trials. We also added the following to the discussion:  
+
+- Page 31 "Second, our behavioral data was limited by the design of the motor task and trial conditions, including two speeds, four directions, and few perturbation trials. Since internal states rely on the accumulation of history across trials, the validity of a state such as the perturbed state does not depend on the number of perturbed trials but rather the overall number of trials. For example, if no trials were perturbed, then the perturbed state would remain 0."  
+
+Finally, we agree that dividing subjects based on performance was rather limiting and we no longer do that. As such we have redone Figures 1d, 2e, 3, 5c, 5e.  
+
+5. Inappropriate frequency band selection.  
+
+The authors state, "The frequency band was identified by matching the frequency bins to frequency bands commonly defined in literature" (Page 46). It's unclear what this means, exactly, but it is not appropriate to define bands by visual inspection of
+
+<--- Page Split --->
+
+time- frequency spectrograms. Bands should be defined a priori without respect to the current data.  
+
+Authors: We did not want to a priori any frequency bands because there is variability between subjects and their frequency bands [Haegens et al. 2014, Neurolmage]. This is why we chose a data- driven approach using the non- parametric cluster statistic, which was agnostic to frequency bands as it found windows of spectral power containing bins of time and frequency. Frequency bins were later mapped to frequency bands for interpretation only so that we could compare our results to existing literature that uses this language. A similar method of mapping frequency bins to bands was used for other work in our lab that used the nonparametric cluster statistic [Sacre et al. 2019, PNAS].
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+Thank you for the opportunity to review "Internal states as a source of subject- dependent movement variability and their representation by large- scale networks" by Breault and colleagues. The manuscript describes an SEEG study looking at how movement variability is encoded in the brain. The authors first fit an internal state model to participant behavioural data in a reaching task and find evidence of two states, related to errors and perturbations. They then localize these states to the well- studied dorsal attention and default mode networks.  
+
+The study addresses an important question that is likely to be of wide interest to the field. The authors carefully triangulate towards a coherent explanation of how internal states govern movement variability, shedding light on the neural mechanisms underlying speed- accuracy trade- offs. However, as I outline below, there are several conceptual and methodological concerns that should be addressed before recommending publication.  
+
+Authors: We thank the reviewer for their positive feedback and helpful comments. In this revision, we provide a point- by- point response to each of the reviewer's specific comments and indicate where we revised the manuscript.  
+
+1. "Such reflections form latent factors called internal states that induce variability of movement and behavior to improve performance." This is minor, but there is a general tendency in the narrative, particularly in the Abstract and Introduction, to talk about internal states as a real biological entity. I think this is misleading, because all work on states has the starting assumption that ongoing behaviour/neural activity can be partitioned into states. I don't think the narrative would suffer if the authors discussed internal states more as a theoretical and methodological construct, rather than as a real biological phenomenon.  
+
+Authors: Thank you for this comment. By internal states, we mean latent variables (hidden and/or unmeasurable) that influence behavior. Internal states vary over time, on a trial-by-trial basis, and are not discrete. Based on the reviewer's comment, we now appreciate that the term "states" is also used to describe discrete events (e.g. planning, moving) to partition behavior. This is not what we mean when we use the term "states". In systems theory, states are variables (often latent) that carry "memory" of a system that influences behavior when provided an input stimulus. That is, the behavior output is a function of both the stimulus and state. Since we construct state-space models, we used states in the context of systems theory. Indeed our two states represent the accumulation of past events (e.g. past perturbations and past errors), i.e., memory, which we show improved prediction of behavior (e.g. RT). To address the reviewer's concern, we now use the suggested language that internal states are our methodological constructs of "memory" that we believe influence behavior (page 3 "These factors are commonly represented as a
+
+<--- Page Split --->
+
+methodological construct of memory called internal states."). We thank the reviewer for this constructive critique.  
+
+2. There is a considerable literature on the role of signal variability in noninvasive human imaging that could potentially be relevant, e.g.  
+
+Garrett, D. D., Samanez-Larkin, G. R., MacDonald, S. W., Lindenberger, U., McIntosh, A. R., & Grady, C. L. (2013). Moment-to-moment brain signal variability: a next frontier in human brain mapping?. Neuroscience & Biobehavioral Reviews, 37(4), 610-624.  
+
+McIntosh, A. R., Kovacevic, N., & Itier, R. J. (2008). Increased brain signal variability accompanies lower behavioral variability in development. PLoS computational biology, 4(7), e1000106.  
+
+Authors: We agree that including the observations of signal variability from noninvasive human imaging studies would be valuable. We have added these points to the introduction:  
+
+- page 3 "However, there is emerging evidence that this variability is purposefully orchestrated by the brain to facilitate learning and adaptation [McIntosh 2008, Garrett 2013]."  
+
+3. There is also a large literature on dynamic states in human imaging, e.g.  
+
+Lurie, D. J., Kessler, D., Bassett, D. S., Betzel, R. F., Breakspear, M., Kheilholz, S., ... & Calhoun, V. D. (2020). Questions and controversies in the study of time-varying functional connectivity in resting fMRI. Network neuroscience, 4(1), 30-69.  
+
+Authors: We agree and have added relevant studies that use noninvasive human imaging to our introduction:  
+
+- page 5, "These studies report the occurrence of co-varying behavioral and neurological variability [McIntosh 2008, Garrett 2013, Lurie 2020]"  
+
+4. Why standard deviation as opposed to the coefficient of variation, which takes mean scaling differences into account?  
+
+Authors: We chose to quantify variability using the standard deviation instead of the coefficient of variation for Figure 1d because it is a statistic of variability. We acknowledge that keeping the standard deviation as it incorporates the mean, but the mean is also an important differentiator of variability across subjects.  
+
+5. My main methodological concern is the arbitrary (mean) splitting of participants into top and bottom performers. This choice is unnecessary and potentially suboptimal as it transforms an inherently continuous variable into a categorical one. All the analyses could then be performed using correlations (with bootstrapping) rather than ANOVAs.
+
+<--- Page Split --->
+
+Authors: We thank the reviewer for this comment. We agree and have redone our analyses using correlations instead of splitting subjects into groups. This has changed the following results:  
+
+- Figure 1d. We found that STD of SE significantly correlated with session performance.- Figure 3 with accompanying results on pages 13-15. We found the same general relationships we previously discussed in that the weight of the error state significantly correlated with session performance. We also found a significant correlation between the weight of speed in RT and session performance (Supplementary Figure 6c).- Figure 5 c, e with accompanying results on pages 18-20. We found a significant correlation between the encoding strength and session performance for both the error state and perturbed state. This suggests that higher performance may be led by neural power that modulates with the internal states.- Connectivity results between pages 20-23. We found regions whose encoding strength modulates with internal states are also functionally connected. But the strength by which a pair of regions covaries is also dependent on session performance (Supplementary Figure 9). For example, connections that correspond to higher performance are shown in Figure 6d-e.  
+
+6. Connectivity estimated as correlated power. Why not use an established method that focuses on coherence/synchrony or on correlating amplitude envelopes?  
+
+Authors: There are several methods for computing functional connectivity. We chose to start with the simplest and most widely used metric for connectivity: correlation. Other methods such as coherence/synchrony or amplitude envelopes would not work here because clusters varied in size (time and frequency).  
+
+7. Is there any cross-validation for the fitted internal states model? The authors compare their model with a pure linear model and show that the former is better, but is there any way to demonstrate that the fitted model helps to predict future values of RT?  
+
+Authors: Cross- validating our internal state models proved difficult, as the internal states are history- dependent. However, since the focus of our paper was on how internal states influence behavior as opposed to trying to predict behavior, our final models were fitted using all data. To address your point, we implemented a 10- fold cross- validation using the internal states previously fitted and found that adding internal states significantly improved the model's ability to predict RTs (paired- sample t- test: \(\mathrm{t}(9) = 2.45\) , \(p = 0.037\) ) but not SE (Supplementary Figure 7c). Adding internal states did improve the model's ability to predict RT and SE for all but two subjects.  
+
+8. "First, a Notch filter was applied using notches located at the fundamental frequency
+
+<--- Page Split --->
+
+of 60 Hz with bandwidth at the \(- 1\) dB point set to \(3\mathrm{Hz}\) . The word "notch" does not need capitalizing. Also, did the authors include filters at subsequent harmonics of 60Hz?  
+
+Authors: Thank you for this comment. The filters did include higher harmonics of 60 Hz. The methods have been updated to make this clear (page 47, "First, a notch filter was applied to the raw voltage data using notches located at the fundamental frequency of 60 Hz and its higher harmonics with the bandwidth at the \(- 1\) dB point set to \(3\mathrm{Hz}\) .").  
+
+9. The authors assign regions to frequency bands, but does this method account for the fact that the power spectrum has a 1/f shape and low frequencies will naturally tend to have greater power than higher frequencies?  
+
+Authors: Again, thank you for this comment. Indeed, this method does account for the reviewer's concern since each frequency bin was independently normalized using the z-score across the entire session. Regardless, the non-parametric cluster statistic is invariant to these problems.  
+
+10. The other major methodological concern is how network specificity is established for the reported effects (e.g. localizing states to DAN or DMN), Namely, the comparison was clearly made visually/qualitatively, but the different networks all have different size, spatial coverage, etc. This means that some networks are more likely to get "hits" than others. The authors should confirm that each network is enriched for a particular state by computing the mean in the network and then comparing this to a null distribution where network labels are permuted.  
+
+Authors: Our analysis was constrained by the coverage of the implanted depth electrodes across our 10 subjects (seen in Fig 4a). Electrodes were not assigned to networks visually but rather using a rigorous method based on an electrode localization algorithm. The coordinate of each electrode relative to their MRI using an electrode localization method established by Stolk et al. 2018, by fusing the subject's MRI (pre- processed using Freesurfer) with their CT (which contacts the coordinates of the electrodes). From there, each electrode was separately mapped to an atlas label (such as intraparietal sulcus L using Destrieux atlas [Destrieux et al. 2010]) and a network label (such as DAN using Yeo atlas [Yeo et al. 2011]) using established the cortical atlases mentioned. This process is described in detail on page 48 of our manuscript. Across all subjects, we had 512 contacts in 40 unique region labels that mapped to 7 networks viable for analysis. These 512 electrodes are mapped to the networks as seen in panel a (below). We acknowledge that our results could be skewed by the disproportionate coverage of electrodes across the main large- scale brain networks. Despite this, we still found a large proportion of electrodes in the underrepresented network of DAN to be encoding the error state seen in panel b (below).
+
+<--- Page Split --->
+![PLACEHOLDER_20_0]
+
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+The main topic addressed by this paper is to examine how the brain represents internal states during motor processing, and to reveal the neural substrate of these patterns. Specifically, this paper utilizes state- space models to link internal states (past errors and perturbations) with motor performance and neural activity, and to identify variations in performance across subjects and trials. While performing a goal- directed center- out reaching task, ten human subjects' electrophysiological data and behavioral data were collected. The paper concluded that large- scale brain networks in the dorsal attention and default networks reflect the neural basis of strategy to regulate movement variability through internal states (past error and perturbation states, respectively) to improve motor performance.  
+
+Overall, the idea of examining the neural basis of internal states in regulating an individual's motor performance is interesting. The paper is well written and has many innovative ideas. However, my enthusiasm for the paper is lowered the impact of the neural data analyses is rather limited. The neural power analyses, which found signals at particular frequencies that correlated with aspects of behavior, were not described in a way that lets the reader make specific inferences about the functional contributions of those regions to behavior. The synchrony analyses were improved in this way, but still, there was not enough detail to make a compelling case about the functional or mechanistic roles of particular regions or networks to support motor processing. Further, I have significant issues with the methods and data analysis that the authors presented.  
+
+Authors: We thank the reviewer for their positive feedback and helpful comments. In this revision, we provide a point- by- point response to each of the reviewer's specific comments and indicate where we revised the manuscript.  
+
+## Specific critiques:  
+
+1. The behavioral data were not presented with enough detail for me to understand the differences across good versus bad subjects, as well as to explain differences in the subjects that the model fit well versus poorly. I would suggest expanding figure 1 and the related sections of the text to separately analyze and interpret the behavioral characteristics of the group data for the subjects with different levels of absolute performance and for those who the model did or did not fit well.  
+
+Authors: We thank the reviewer for their concern. We have since removed grouping subjects by performance and replaced it with the spectrum of subject performance. We believe Figure 1d shows that good performers have less variable RT and SE compared to bad performers, the latter of which is statistically supported. In our revised manuscript, we have also added detailed statistics that confirmed subject differences in RT and SE:  
+
+- Page 9, "Figure 1d shows how subject's variability of RT (top) and SE (bottom) is related to their performance. Specifically, we found that
+
+<--- Page Split --->
+
+## subjects with higher session performance had less variable SE (Pearson's correlation: \(r = -0.88\) , \(p = 7.42 \times 10^{-4}\) )."  
+
+We used our model to examine the relationship between session performance and conditions/states, via how subjects weighted model variables. For example, we go into detail about the relationship between the internal states and session performance in Figure 3. Supplementary Figure 6 also shows the comparison between trial conditions and session performance. The only other significant relationship was between the weight of speed, where higher performance correlated with larger weights on speed (Pearson's correlation: \(r = 0.91\) , \(p = 2.36e - 04\) ), which is brought up in the Results:  
+
+- Page 15, "Another key strategy was speed, where larger magnitudes of weights on speed correlated with high session performance (Supplementary Figure 6c)."  
+
+2. The neural data analyses for figures 4, 5, and 6 are not illustrated with adequate statistical justification. I could not tell which neural signals correlated with behavior in a statistically robust fashion, and there was a lack of adjustment for multiple comparisons. Overall this section of the paper was largely qualitative, which is inadequate. This section of the paper should rigorously show which neural signals, across frequencies, correlated with behavior, and include both within- and across subject comparisons.  
+
+Authors: Thank you for your comment. The other reviewers raised similar concerns so we have revised the manuscript as detailed below to more clearly explain our statistical analyses. All of our neural results were derived using an unsupervised and data- driven approach, which were supported by statistical tests at all steps. First, the raw SEEG data was preprocessed into spectral data of normalized power in the time and frequency domains. The spectral data were then sliced into trials, which were then sliced into epochs. Next, we used a non- parametric cluster statistical test [Maris & Oostenveld, 2007] to find windows of time and frequency in which the spectral power correlated with each internal state across trials for each region across the population. This method is agnostic to time and frequency. These windows are called clusters, and each has an associated cluster statistic (Figure 4c). Finally, we corrected for multiple comparisons (one comparison for each brain region for each epoch for each internal state) using a false- discovery rate. What we report in our paper are only the positive results (see Tables 2 and 3 for statistics from nonparametric cluster analysis). As a result, this approach narrowed down our scope from 40 possible regions to just 12 for the error state and 14 for the perturbation state.  
+
+Based on this data- driven approach, we noticed that the error state was populated by regions in DAN while the perturbation state was populated by regions in DN (Figure 5a- b). If anything, our approach was extreme in that we used all trials, all regions, all frequencies, etc. We reinforced these results to also show that the strength of which subjects encoded the internal states (via the average power in a cluster) depended on their session performance (Figure 5c- f).
+
+<--- Page Split --->
+
+Once the clusters (windows of time and frequency) were statistically identified, we calculated the average power in each cluster for each trial (visualized in Figure 4b). This gave us a signal of how the power in the cluster varied over trials for each region. This naturally led us to run a connectivity analysis between these signals across regions, via Pearson's correlation, for each subject. The number of correlations for each subject varied based on the number of regions that were identified using the non- parametric cluster statistic. But, to minimize the number of statistical tests, we only ran correlations within each internal state result. That is, we ran connectivity analysis for regions related to the error state and perturbed state separately.  
+
+Accordingly, based on the reviewer's feedback, we have updated figures and revised the text to make it more apparent that our results were data- driven and supported by statistics.  
+
+- Page 16-17:  
+
+- "To identify the neural correlates amongst these regions with an unsupervised and data-driven method, we used a non-parametric cluster statistic [36] (see Methods) between the spectral data of each region and each internal state across the population (Fig. 4b). This method finds windows of time (relative to epoch onset) and frequency (between 1 and 200 Hz) in which the spectral power of a region significantly correlates with the internal state across trials as demonstrated in Fig. 4c. The statistic provided us with two sets of neural correlates, one for each internal state."  
+
+- "Our modeling results showed that subjects weighed internal states differently, primarily based on their session performance. We suspected that this would be reflected in the brain for each subject by how well these regions encoded the states through the strength of their neural correlates. The degree to which a subject encodes an internal state in a region, which we called the encoding strength, was quantified by correlating the average power within the time-frequency window (from the population statistic) to the state on a trial-by-trial basis (see Methods). Fig. 4d shows an example of how the encoding strength is obtained for a channel in subject 6 using the neural correlate from Fig. 4c."  
+
+3. The analysis described in Figure 2 was hard to understand because it was focused so much on just subject 6.  
+
+Authors: We chose to focus the majority of Figure 2 on subject 6 to help the reader gain intuition about the data and have revised the text to make this point clear. The qualities we described for subject 6 were found across all subjects, whose equivalent figures are given in Supplementary Figures 1–5. We have
+
+<--- Page Split --->
+
+subsequently revised the results related to Figure 2 to make their implications more generally applicable beyond subject 6.  
+
+- Pages 10-11 "Though we used subject 6 to demonstrate the intuition behind our model, our general observations are applicable to the population, who are shown in Supplementary Fig. 1 - Supplementary Fig. 5."- Page 11, "Conditions accounted for the trial-by-trial changes and states accounted for the gradual changes across all subjects (Supplementary Figure 7)."- Page 11, "With the exception of RT for subject 1, we consistently found a significant relationship between the estimated and observed RT (Supplementary Fig. 1) and SE (Supplementary Fig. 2) for all subjects."- Page 12, "In RT, we also observed states that monotonically increased over trials for subjects 4, 5, 7, and 10 (Supplementary Fig. 3). This state reproduced subjects progressively reacting slower due to fatigue (Supplementary Fig. 1)"- Page 12, "Based on the structure, we can also see what influenced the RT states to be monotonic for subjects 4, 5, 7, and 10. Supplementary Figure 5 shows that the primary contributor was the perturbed state."  
+
+4. The behavioral data in Figure 2 are hard to interpret because panels B-D are visually very dense. Panel E seems to indicate that the model fit quality is bimodal, with a cluster of four subjects who have four model fits and others who show rather good model fits. Does this undermine the results?  
+
+Authors: The bimodality of Figure 2e does not undermine the results. This stems from the fact that models were fit for subjects which led to varying optimal model performance. We discuss the implications of these observations to our results:  
+
+- Page 13, "For example, the goodness-of-fit for subject 6 (black) improved by 16%. However, the model structure fits some subjects better than others. Specifically, subjects 1-4 had outlying model performance for RT (Figure2e(left)) compared to other subjects. This is notable because these subjects also had the lowest session performance. Because their model performance was low prior to adding internal states, this indicates that their RT varied by factors other than speed and direction which could also explain their low session performance. Despite this, adding internal states also improved the deviance (Supplementary Figure 7b) and 10-fold cross-validation using the fitted internal states for RT (Supplementary Figure 7c). This, in addition to the fact that adding internal states improved their model fit and the inlying goodness-of-fit of their SE model still merits the validity of interpreting their RT model"
+
+<--- Page Split --->
+
+5. Figure 4C is hard to interpret because the statistic is calculated based on the correlation but the colors reflect a different value, mean power. It would be much easier to interpret this plot if the same value was plotted as indicated in the statistic.  
+
+Authors: We agree that the original version of Figure 4 was difficult to interpret. We now use Figure 4 to help the readers understand our data- driven approach using non- parametric cluster statistics to identify windows of time and frequency in the spectrum that correlated with internal states. Figure 4 now includes spectrums (Fig. 4b) as a qualitative example as well as the t- statistic (Fig. 4c) as a quantitative example.  
+
+6. Figure 6 is hard to interpret, is there a more intuitive or data-driven way to order the labels around the plots?  
+
+Authors: We feel that Figure 6 is similar to how connectivity is displayed in other literature. However, we added panels to Figure 6 (a- c) to help explain the method we used to arrive at the connectivity results. The ordering of the labels is data driven in that the regions are ordered by when their cluster appears during the trial (which follows Tables 2 and 3).  
+
+- Page 21, "We quantified functional connectivity by correlating the average power within the time-frequency window, used for encoding strength, on a trial-by-trial basis for each channel per subject (Supplementary Figures 6a and 6b)."  
+- Page 21, "For example, subjects with high performance will exhibit higher subject connectivity strength between key regions (Figures 6c)."  
+
+7. The analysis on Page 20, Line 30 and Fig 6 could benefit from a more quantitative comparison between network connectivity during "error state" and "perturbed state" rather than qualitative comparisons. There are various quantitative measures that the authors might use for this network comparison  
+
+Authors: The network connectivity results were derived from quantitative measurements. We've elaborated in Figure 6 to show how we quantitatively arrived at our connectivity result. We also added population connectivity in Supplementary Figure 8, though between subject variability was the focus of our paper and not population. We also added Supplementary Figure 9 to showcase all of the performance-modulated networks and not just the ones that lead to higher performance (seen in Figure 6d- e).  
+
+8. Perhaps it is a minor point, but I found the conclusion of the paper confusing. How are the results relevant for the type of behavioral data that could be collected from smartphones? There might be an interesting idea here but it needs more length to spell out.
+
+<--- Page Split --->
+
+Authors: We agree that our conclusion could be made more straightforward. We have since edited to make our point clearer.  
+
+- Page 32, "Our results raise the possibility that the underlying history of measured behaviors could be used to make inferences about a person's brain state without needing to collect electrophysiological data, saving time and money in the health field for personalized medicine [81] or business ventures such as sports [82]"  
+
+9. Please explain in the text that discuss in the text why was it was valid for the subjects to be divided based on a median split of RT? Some "top performers" may be doing relatively poorly in the task.  
+
+Authors: Subjects were initially divided into "top" and "bottom" performers based on the average session performance of the population, which was \(51\%\) . After much consideration, we've decided to broaden our analysis by analyzing subjects as a spectrum based on their session performance instead of binary groups. This has changed the following results:  
+
+- Figure 1d. We found that STD of SE significantly correlated with session performance.- Figure 3 with accompanying results on pages 13-15. We found the same general relationships we previously discussed in that the weight of the error state significantly correlated with session performance. We also found a significant correlation between the weight of speed in RT and session performance (Supplementary Figure 6c).- Figure 5 c, e with accompanying results on pages 18-20. We found a significant correlation between the encoding strength and session performance for both the error state and the perturbed state. This suggests that higher performance may be led by neural power that modulates with the internal states.- Connectivity results between pages 20-23. We found regions whose encoding strength modulates with internal states are also functionally connected. But the strength by which a pair of regions covariates is also dependent on session performance (Supplementary Figure 9). For example, connections that correspond to higher performance are shown in Figure 6d-e.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+In this resubmission, the authors attempt to show that internal states account for behavior in a motor task. Unfortunately, they failed to address my previous concerns. The persistent statistical issues throughout the manuscript prevent interpretation of the findings.  
+
+The neural analysis methods utilized - - including non- parametric cluster analysis - - are clear, however, they still do not support the claims made by the authors. The authors state  
+
+"Specifically, the Dorsal Attention Network (DAN) and Default Network (DN) were linked to the error and perturbed state, respectively" (page 6)  
+
+"Remarkably, we then found that these internal states were linked to encoding in large- scale brain networks, DAN and DN, respectively" (page 25)  
+
+"For the first time, our results implicate DAN as a network as encoding tracking history" (page 26- 27)  
+
+"We also found that the DAN encoded error history throughout the trial using persistent activity in frequencies above \(100 \text{Hz}\) " (page 27)  
+
+To make these claims, the authors must perform direct tests between DAN and another network(s), and across different frequency bands. If one network shows a significant effect and another network does not show a significant effect (as determined by the cluster analysis), this does not mean that there is necessarily a difference between the two networks. DAN could show an effect at \(p = 0.04\) and DMN could show an effect at \(p = 0.10\) , but the two networks need not differ from each other. This inference problem is explained in Makin & Orban de Xivry (2019) "Ten common statistical mistakes to watch out for when writing or reviewing a manuscript." Elife ("Interpreting comparisons between two effects without directly comparing them").  
+
+The authors make multiple references to figures to support their claims. Figures are for visualization, but statistics are required to make any/all claims. These are absent from the manuscript.  
+
+The authors added degrees of freedom (DFs) and test statistics to the behavioral results, but the DFs do not appear to be correct. The authors report ANOVA with DFs of (1,1296), (3,1296) and (5,1296). Based on the levels and factors, the DFs should be (1,9) for main effect of speed; (3, 27) for main effect of
+
+<--- Page Split --->
+
+direction and (3,27) for the interaction between speed and direction (there is also a lack of standard terminology, e.g. "main effect" and "interaction," to communicate the specific tests being performed).  
+
+As defined in the manuscript, connectivity is performed by correlating all pairs of electrodes, yet the authors make reference to different regions/networks, which presumably do not include all electrodes. The number of statistical tests performed (based on the response letter) is quite large and it is unclear whether any corrections were applied.  
+
+The authors continue to focus on subject 6 with little justification and refer the reader to supplemental figures for the full set of data, which again must be supplemented by statistics. The connectivity correlation shown in Figure 6 is clearly driven by the fact that only three subjects are included and one subject is an outlier in comparison to the other two. In general the correlation results (the majority of the manuscript) seem likely to be driven by outliers given the small sample size.  
+
+Frequency band selection is still not defined. The authors state "as commonly defined in literature," however, one or more citations must be provided, as different definitions exist in the literature.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have comprehensively addressed my concerns and I recommend publication.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have done a very good job responding to my comments and made a number of improvements to the manuscript. I think the paper is significantly improved and suitable for publication.  
+
+My only remaining comment is that I think the authors should clarify the text and caption related to Figure 6C. It was a bit hard to understand how this figure related to the rest of the analyses and how this figure was generated. I eventually figured out how they generated this figure, but they should clarify the sample size that contributed to this analysis and how this sample was generated. This analyses around Figure 6 are mentioned in the abstract, so I think it is especially important for the relevant analyses to be explained clearly.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+In this resubmission, the authors attempt to show that internal states account for behavior in a motor task. Unfortunately, they failed to address my previous concerns. The persistent statistical issues throughout the manuscript prevent interpretation of the findings.  
+
+Authors: Thank you for your help and patience as we continue to correct our manuscript to properly adhere to the rigorous statistical standards expected for any scientific work. Based on your comments, we decided to consult a statistician about our analysis methods. We believe the quality of our findings has greatly improved as a result and assure the review that we have now properly addressed any remaining concerns that you may have about the statistical validity of the manuscript. The manuscript is marked based on items that were removed (red text with streakthrough) or added (blue text).  
+
+1. The neural analysis methods utilized -- including non-parametric cluster analysis -- are clear, however, they still do not support the claims made by the authors. The authors state  
+
+"Specifically, the Dorsal Attention Network (DAN) and Default Network (DN) were linked to the error and perturbed state, respectively" (page 6) "Remarkably, we then found that these internal states were linked to encoding in large-scale brain networks, DAN and DN, respectively" (page 25) "For the first time, our results implicate DAN as a network as encoding tracking history" (page 26- 27) "We also found that the DAN encoded error history throughout the trial using persistent activity in frequencies above 100 Hz" (page 27)  
+
+To make these claims, the authors must perform direct tests between DAN and another network(s), and across different frequency bands. If one network shows a significant effect and another network does not show a significant effect (as determined by the cluster analysis), this does not mean that there is necessarily a difference between the two networks. DAN could show an effect at \(p = 0.04\) and DMN could show an effect at \(p = 0.10\) , but the two networks need not differ from each other. This inference problem is explained in Makin & Orban de Xivry (2019) "Ten common statistical mistakes to watch out for when writing or reviewing a manuscript." ELife ("Interpreting comparisons between two effects without directly comparing them").  
+
+Authors: We thank the reviewer for pointing this mistake out! The reference offers the following solutions:  
+
+"Researchers should compare groups directly when they want to contrast them (and reviewers should point authors to Nieuwenhuis et al., 2011 for a clear explanation of the problem and its impact). The correlations in the two groups can be compared with Monte Carlo simulations (Wilcox and Tian, 2008). For
+
+<--- Page Split --->
+
+group comparisons, ANOVA might be suitable. Although non- parametric statistics offers some tools (e.g., Leys and Schumann, 2010), these require more thought and customisation."  
+
+To determine whether there is a statistically significant relationship (p- value \(< 0.5\) ) between the specific large- scale network in a given frequency band and the internal states per trial, we correlated (Spearman's correlation, consistent with the type of correlation used for the non- parametric cluster statistic) the average power of the cluster for each trial aggregated over all subjects from each region in either DAN or DN that we found to encode each internal state, which we also visualized through scatter plots:  
+
+![PLACEHOLDER_30_0]
+  
+
+The top row represents the error state and the bottom row represents the perturbed state, where j represents the number of regions plotted. The left plot displays DAN and DN aggregated, the middle plot displays only the regions in DAN and the right plot displays only the regions in DN. We found all the correlations to be significant (p- value \(< 0.5\) ), i.e., DAN and DN encode both internal states. What we can see is that, of the total regions in DAN+DN, the majority of regions encoding the error state were in DAN (70%) and the majority of regions encoding the perturbed state were in DN (~77%). This supports the claim in our manuscript in which the error state is primarily encoded by DAN and the perturbed state is encoded primarily by DN.  
+
+Additionally, we were also interested in the relationship between the specific regions in each network and the internal state. Thus, we also examined the correlation (Spearman's correlation) between internal states and the regions within each network aggregating subjects:
+
+<--- Page Split --->
+![PLACEHOLDER_31_0]
+  
+
+The figure above represents the error state, where each plot represents the average power of a cluster for each trial across all subjects from the region in the title (n represents the number of subjects aggregated in the scatter plot). The top row represents the regions from DAN that encoded the error state (j refers to the specific number of regions) whereas the bottom row represents the regions from DN. We found each region to be significantly correlated (p- value \(< 0.5\) ) with the internal state. Using the correlation value as a measurement for how well a region encoded the internal state, we found that the regions in DN (bottom row, median of 0.16) encoded the error state as well as the regions in DAN (top row, median of 0.19) supported through a statistical test (Mann- Whitney \(U = 39\) , \(p = 1\) ).  
+
+![PLACEHOLDER_31_1]
+  
+
+All regions encoding the perturbed state were significantly correlated (p- value \(< 0.5\) ) as well (using Spearman's correlation). Similarly, for the perturbed state, the regions in DAN (top row, median \(= 0.17\) ) that also encoded it did so as well as the regions in DN (bottom row, median \(= 0.12\) ) (Mann- Whitney \(U = 26\) , \(p = 0.47\) ).  
+
+Again, what we can draw from both of these figures is that more regions in DAN encoded the error state than DN (7 regions in DAN vs. 3 regions in DN) and more regions in DN encoded the perturbed state than DAN (3 regions in DAN vs. 10 regions in DN). Therefore, the ultimate conclusion we can make from our results is not that DAN exclusively encodes the error state but rather that the
+
+<--- Page Split --->
+
+majority of regions that encoded the error state were in DAN and the majority of regions that encoded the perturbed state were in DN.  
+
+What we cannot draw from these relationships is the varying degree to which subjects actually encode the internal states, i.e. the basis for our claim that encoding scales by performance. Below, we found the correlation between encoding regions with internal states (and their p- value) varied based on subject performance, most easily summarized by aggregating all the regions subjects had recordings from for either DAN or DN, visualized in a scatter plot:
+
+<--- Page Split --->
+![PLACEHOLDER_33_0]
+  
+
+For the error state (left) and perturbed state (right), we found that higher- performing subjects had higher correlation values for both the regions in DAN and DN (where j represents the number of regions aggregated per subject). The correlations also revealed that the higher- performing subjects were more likely to have a significant correlation (p- value \(< 0.5\) ) than the lower- performing subjects.
+
+<--- Page Split --->
+
+To check whether these results were not simply skewed by possible disproportionate coverage of the electrodes to any particular networks, we compared the actual coverage of the networks from population to the coverage found to encode either internal state:  
+
+![PLACEHOLDER_34_0]
+  
+
+As you can see, the networks from our results do not match the proportions of networks from the population.  
+
+With that said, we modified the language in our manuscript to better reflect the more rigorous analyses above regarding results:  
+
+- Page 1 (Abstract): "The dorsal attention network primarily encoded past errors in frequencies above 100 Hz, suggesting a role in modulating attention based on tracking recent performance in working memory. The default network primarily encoded past perturbations in frequencies below 15 Hz suggesting a role in achieving robust performance in an uncertain environment."- Page 6 (Introduction): "Specifically, the Dorsal Attention Network (DAN) and Default Network (DN) were linked to the error and perturbed state; respectively."- Page 18 (Results): "Some regions from other networks also appeared, such as the angular gyrus right (AG R) and posterior-dorsal cingulate gyrus right (dPCC R) in DN, the supramarginal gyrus right (SMG R) from the Ventral Attention Network (VAN), as well as cuneus right (Cu R) in the Visual Network (Visual). 'However, the majority of regions that encoded the error state were located in DAN.'- Page 20 (Results): "Overall, the majority of regions we found to encode the perturbed state were in DN."- Page 20 (Results): "This figure also shows regions from other networks, which include the DAN in dark blue, VAN in light blue, Somatomotor in green, and Visual in yellow."- Page 25 (Discussion): "Remarkably, we then found that these internal states were linked to encoding in large-scale brain networks, DAN and DN, respectively."
+
+<--- Page Split --->
+
+- Page 26 (Discussion): "We found that our subjects learned the task by monitoring their history of errors across trials to decide where to allocate attention primarily through the DAN."- Page 28 (Discussion): "Finally, we found that the encoding strength of regions in the DAN and functional connectivity between these regions, along with the visual network, scaled based on the subject's overall performance."- Page 29 (Discussion): "We found that regions whose activity correlated with the perturbed state were primarily in the DN."- Page 30 (Discussion): "Taken together, these findings suggest that high performers react to uncertainty by heightening vigilance through activation and connectivity in DN in conjunction with the VAN and the visual network."  
+
+2. The authors make multiple references to figures to support their claims. Figures are for visualization, but statistics are required to make any/all claims. These are absent from the manuscript.  
+
+Authors: After carefully reviewing our revised manuscript, we are confident that it contains the appropriate statistics for any/all claims that were made. In particular, in this version of our manuscript, we have added the p-values and sample size to the performance connectivity analysis in Supplementary Figure 8 which was summarized in Figure 6.  
+
+3. The authors added degrees of freedom (DFs) and test statistics to the behavioral results, but the DFs do not appear to be correct. The authors report ANOVAs with DFs of (1,1296), (3,1296) and (5,1296). Based on the levels and factors, the DFs should be (1,9) for main effect of speed; (3, 27) for main effect of direction and (3,27) for the interaction between speed and direction (there is also a lack of standard terminology, e.g. "main effect" and "interaction," to communicate the specific tests being performed).  
+
+Authors: Thank you for your suggestions. After consulting with Emery Brown, a statistician, we have corrected the reporting of the statistics used on our behavioral analysis to use the correct degrees of freedom:  
+
+- Page 8 (Results): "That is, there was a main effect of subject on RT (three-way ANOVA: \(\mathbf{F}(9,27) = 7.24\) , \(\mathbf{p} = 2.85 \times 10^{-3}\) , partial eta^2 = 0.87) and SE (three-way ANOVA: \(\mathbf{F}(9,45) = 13.17\) , \(\mathbf{p} = 4.09 \times 10^{-12}\) , partial eta^2 = 0.62)."  
+- Page 8 (Results): "Indeed, there was a main effect of speed on RT (three-way ANOVA: \(\mathbf{F}(1,9) = 9.74\) , \(\mathbf{p} = 0.012\) , partial eta^2 = 0.52), meaning subjects reacted more quickly for fast trials than slow trials. We also found a main effect between the location of the target and how quickly subjects
+
+<--- Page Split --->
+
+reacted (three- way ANOVA: \(\mathbf{F}(3,27) = 4.59\) , \(p = 0.0012\) , partial eta^2=0.32)."  
+
+- Page 9 (Results): "We found a main effect between the type of perturbation and SE (three-way ANOVA: \(\mathbf{F}(5,45) = 23.23\) , \(p = 9.56 \times 10^{-12}\) , partial eta^2=0.71), with the exception of unperturbed compared to towards trials regardless of speed (Supplementary Table 10)."  
+
+We also modified our results and methods to better communicate the statistical tests being used:  
+
+- Page 8 (Results): "That is, there was a main effect of subject on RT (three-way ANOVA: \(\mathbf{F}(9,27) = 7.24\) , \(p = 2.85 \times 10^{-3}\) , partial eta^2=0.87) and SE (three-way ANOVA: \(\mathbf{F}(9,45) = 13.17\) , \(p = 4.09 \times 10^{-12}\) , partial eta^2=0.62)."  
+- Page 8 (Results): "Indeed, there was a main effect of speed on RT (three-way ANOVA: \(\mathbf{F}(1,9) = 9.74\) , \(p = 0.012\) , partial eta^2=0.52), meaning subjects reacted more quickly for fast trials than slow trials. We also found a main effect between the location of the target and how quickly subjects reacted (three-way ANOVA: \(\mathbf{F}(3,27) = 4.59\) , \(p = 0.0012\) , partial eta^2=0.32)."  
+- Page 9 (Results): "However, there was not a significant interaction between speed and direction."  
+- Page 9 (Results): "We found a main effect between type of perturbation and SE (three-way ANOVA: \(\mathbf{F}(5,45) = 23.23\) , \(p = 9.56 \times 10^{-12}\) , partial eta^2=0.71), with the exception of unperturbed compared to towards trials regardless of speed (Supplementary Table 10)."  
+- Page 40 (Methods): "To test the main effects and interactions that subjects and trial conditions had on RT and SE, we constructed a multi-way ANOVA for each."  
+- Page 11 (Supplementary): "Supplementary Table 4 | Three-way ANOVA of main effects and interactions that subject, speed, and direction have on reaction time. (df = degrees of freedom, \(\mathbf{SS} = \mathbf{Sum}\) of Squares, \(\mathbf{MS} = \mathbf{Mean}\) Squares, \(\mathbf{EMS} = \mathbf{Expected}\) Mean Squares, \(\mathbf{F} = \mathbf{F}\) -statistic)"  
+- Page 12 (Supplementary): "Supplementary Table 5 | Post-hoc Tukey's comparison for the main effect of subject on reaction time."  
+- Page 13 (Supplementary): "Supplementary Table 6 | Post-hoc Tukey's comparison for the main effect of direction on reaction time."  
+- Page 14 (Supplementary): "Supplementary Table 7 | Post-hoc Tukey's comparison for the interaction between speed and direction on reaction time."  
+- Page 15 (Supplementary): "Supplementary Table 8 | Three-way ANOVA of main effects and interactions that subject, type of perturbation (speed, pert.), and reaction time (RT) have on speed error. (df = degrees of freedom, \(\mathbf{SS} = \mathbf{Sum}\) of Squares, \(\mathbf{MS} = \mathbf{Mean}\) Squares, \(\mathbf{EMS} = \mathbf{Expected}\) Mean Squares, \(\mathbf{F} = \mathbf{F}\) -statistic)"  
+- Page 16 (Supplementary): "Supplementary Table 9 | Post-hoc Tukey's comparison for the main effect of subject on speed error."
+
+<--- Page Split --->
+
+- Page 17 (Supplementary): "Supplementary Table 10 | Post-hoc Tukey's comparison for the main effect of type of perturbation on speed error."  
+
+4. As defined in the manuscript, connectivity is performed by correlating all pairs of electrodes, yet the authors make reference to different regions/networks, which presumably do not include all electrodes. The number of statistical tests performed (based on the response letter) is quite large and it is unclear whether any corrections were applied.  
+
+Authors: Prior to performing any additional analysis, we corrected the nonparametric cluster statistic results for multiple comparisons (one comparison for each brain region for each epoch for each internal state) using a false- discovery rate (FDR) of \(q = 0.015\) . We only reported and used the clusters that passed the FDR. For connectivity, to limit the number of correlations performed, the normalized power across trials was averaged across electrodes in the same region for each subject. Hence, the sample size used for correlations within each subject (i.e., between regions for connectivity) was limited to the number of trials they performed.  
+
+5. The authors continue to focus on subject 6 with little justification and refer the reader to supplemental figures for the full set of data, which again must be supplemented by statistics.  
+
+Authors: We again note that we chose subject 6 as a representative example of the entire dataset:  
+
+1. Their performance is around average, 
+2. The dynamics of their internal states provide key points that aid in interpreting the impact internal states have on behavior for the population, 
+3. Their neural coverage provided a basis for transitioning from regions to large-scale brain networks using the error state and DAN as an example.  
+
+Using the same subject throughout the paper and figures provided consistency for narrative purposes and is justified by statistics stated in the manuscript. We have added the following statements to the manuscript to convey the justification of using subject 6 while conserving the narrative:  
+
+- Page 11 (Results): "To illustrate this, consider as an example subject 6, whose session performance was around average across all subjects. Figure 2b shows the RT across all trials for this subject, with their observed behavior in gray and their estimated behavior in black."- Page 11 (Results): "Continuing with subject 6 as an example, Figure 2c shows their estimated RT separated into the trial conditions (orange) and internal states (purple)."
+
+<--- Page Split --->
+
+- Page 11 (Results): "Generally, conditions accounted for the trial-by-trial changes and states accounted for the gradual changes across all subjects (Supplementary Fig. 7a)."- Page 12 (Results): "Following subject 6, Figure 2d shows their error state (blue) and perturbed state (red) across all trials."  
+
+6. The connectivity correlation shown in Figure 6 is clearly driven by the fact that only three subjects are included and one subject is an outlier in comparison to the other two. In general the correlation results (the majority of the manuscript) seem likely to be driven by outliers given the small sample size.  
+
+Authors: First, we want to emphasize that the connectivity results are preliminary results, due to the limited sample size as you pointed out. The main findings of our manuscript are (i) explaining variability with states and (ii) identifying regions that belong to large-scale networks are encoding these states.  
+
+We also felt that it was important to be as transparent as possible about the statistics we used to get to our results. Therefore, we now provide the p-values and sample sizes for the performance connectivity analysis (from Figure 6) in Supplementary Figure 8. We also edited our manuscript to emphasize the preliminary aspect of the performance- related correlation results:  
+
+- Page 1 (Abstract): "Moreover, these networks were preliminarily found to more strongly encoded internal states and be more functionally connected in higher performing subjects, whose learning strategy was to respond by countering with behavior that opposed accumulating error."- Page 6 (Introduction): "We also have preliminary evidence linking the encoding strength and functional connectivity of these networks back to subject performance and strategy."- Page 30 (Discussion): "Our preliminary findings suggest that high performers effectively implement their new semantic knowledge about the environment to explore different approaches to prepare for the possibility of future perturbations."- Page 28 (Discussion): "Finally, we found that the encoding strength of regions in the DAN and functional connectivity between these regions, along with the visual network, scaled based on the subject's overall performance. Though these results are preliminary, we linked this with observations that subjects have different strategies."- Page 32 (Discussion): "Finally, we want to emphasize that the performance-related results should be taken as preliminary as the sample sizes for these statistics were limited to the number of subjects implanted in the same regions. Hence, studies with larger sample sizes are needed to make any conclusive statements."- Page 54 (Methods): "To identify preliminarily pairs of regions whose connectivity encodes the internal states and performance, we related the
+
+<--- Page Split --->
+
+subject connectivity strength to session performance for each pair across subjects the Pearson's correlation value (Fig. 6c)."  
+
+- Page 54 (Methods): "Supplementary Figure 8 b,e and Supplementary Figure 8 c,f show the p-values and sample sizes used when calculating the performance connectivity strength."  
+
+7. Frequency band selection is still not defined. The authors state "as commonly defined in literature," however, one or more citations must be provided, as different definitions exist in the literature.  
+
+Authors: Thank you for bringing this to our attention. The new manuscript now contains references for each frequency band:  
+
+- Page 50 (Methods): "delta (1-4 Hz) [98], theta (4-8 Hz) [98], alpha (8-15 Hz) [98], beta (15-30 Hz) [98], low gamma (30-60 Hz) [99], high gamma (60-100 Hz) [99], and hyper gamma (100-200 Hz) [100]"  
+
+98. Prerau, M. J., Brown, R. E., Bianchi, M. T., Ellenbogen, J. M. & Purdon, P. L. Sleep neurophysiological dynamics through the lens of multitaper spectral analysis. Physiology 32, 60-92 (2017).  
+99. Sani, O. G. et al. Mood variations decoded from multi-site intracranial human brain activity. Nature Biotechnology 36, 954-961 (2018).  
+100. Crone, N. E., Sinai, A. & Korzeniewska, A. High-frequency gamma oscillations and human brain mapping with electrocorticography. Progress in Brain Research 159, 275-295 (2006).
+
+<--- Page Split --->
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have done a very good job responding to my comments and made a number of improvements to the manuscript. I think the paper is significantly improved and suitable for publication.  
+
+Authors: We thank the reviewer for their positive reception to our revisions. We addressed their final comment in this latest revision and have indicated where we revised the manuscript. The manuscript is marked based on items that were removed (red text with streakthrough) or added (blue text).  
+
+1. My only remaining comment is that I think the authors should clarify the text and caption related to Figure 6C. It was a bit hard to understand how this figure related to the rest of the analyses and how this figure was generated. I eventually figured out how they generated this figure, but they should clarify the sample size that contributed to this analysis and how this sample was generated. This analyses around Figure 6 are mentioned in the abstract, so I think it is especially important for the relevant analyses to be explained clearly.  
+
+Authors: Thank you for reviewing our manuscript again. We agree that the sample size is an important point for our preliminary connectivity results. As such, we provided more description about the sample size used for the performance connectivity analysis in the methods:  
+
+- Page 54 (Methods): "Any pair that did not include both a subject above and below average session performance was excluded. We also excluded pairs that were not represented by at least three subjects."- Page 54 (Methods): "Supplementary Figure 8 b,e and Supplementary Figure 8 c,f show the p-values and sample sizes used when calculating the performance connectivity strength."  
+
+We also explicitly provide the sample size used to get Figure 6 in Supplementary Figure 8 c,f on page 6 of Supplementary.
+
+<--- Page Split --->
+
+Reviewers' comments:  
+
+Reviewer #1 (Remarks to the Author):  
+
+Although the authors have revised their manuscript, my concerns regarding statistics and analysis approaches remain. The majority of reported findings are not supported by direct tests or group statistics, instead the authors present qualitative comparisons and/or rely on individual subject data, severely limiting the interpretability and generalizability of the findings.
+
+<--- Page Split --->
+
+Review 1 Comment: Although the authors have revised their manuscript, my concerns regarding statistics and analysis approaches remain. The majority of reported findings are not supported by direct tests or group statistics, instead the authors present qualitative comparisons and/or rely on individual subject data, severely limiting the interpretability and generalizability of the findings."  
+
+Authors: In our responses over three revisions, we consulted an expert statistician (PhD in Statistics who also conducts studies in neuroscience), and carefully revised our statistics. We adapted our figures and ran multiple analyses to address reviewers' concerns over the year (including direct tests and group statistics). We also clarified that none of our results were based on qualitative comparisons. We had originally highlighted one subject throughout in figures to show raw data and to help better understand what our models capture; and made this clear that it was just an example and that rigorous quantitative comparisons were performed throughout the study and reported. The interpretability and generalizability of our findings has been clarified in the final revision.  
+
+After the second cycle, reviewers 2 and 3 accepted our manuscript for publication, while reviewer 1 still had two concerns. The first concern was about the validity of our claims. We ran the additional analysis as requested and have added this to our revised manuscript along with an additional figure to support our claims. The second concern was about clarifying the use of an example to illustrate general observations, which we have subsequently addressed.  
+
+We believe the quality of our findings has greatly improved as a result and assure the reviewer that we have now properly addressed any remaining concerns.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__44ac714467bf52e85c6116012157571d06f97bede43922ad63033d05e6843059/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__44ac714467bf52e85c6116012157571d06f97bede43922ad63033d05e6843059/supplementary_0_Peer Review File__44ac714467bf52e85c6116012157571d06f97bede43922ad63033d05e6843059.mmd

@@ -0,0 +1,47 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Ataluren Binds to Multiple Protein Synthesis Apparatus Sites and Competitively Inhibits Release Factor Dependent Termination  
+
+![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 --->
+
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors did an excellent job replying to the comments from the first review. However, there are still questions about some of the interpretation of results.  
+
+## Comments:  
+
+1) The authors did a nice job explaining how the settled in on the ataluren binding sites for study to 3 sites. However, it cannot be disregarded that ataluren was shown to bind to 22 sites. Indeed, these results suggests that while ataluren is shown here to inhibit translate termination in this reductionist assay, it does so in a rather non-specific manor, as only \(13.6\%\) of the binding sites contribute to translation termination activity.  
+
+2) The authors do a nice job of highlighting studies where ataluren has somewhat succeeded in promoting PTC readthrough. However, ataluren is far from a success in the clinic and there are several studies that show it has no significant effect in vivo. The authors should do the literature and clinical trials justice in ataluren's failures as well. Further, the manuscript would benefit from a section that describes how individual PTCs (position, sequence context, etc) could be more or less difficult to readthrough by TRIDS. This discussion is common in most PTC suppression studies. Do the same rules apply to TRIDS as they do for aminoglycosides for instance?  
+
+Reviewer #2 (Remarks to the Author):  
+
+It appears that many errors were caught during this peer review process. The authors have now addressed at least to some degree of satisfaction all of the comments and suggestions of both Referees of this work in their revised manuscript
+
+<--- Page Split --->
+
+Response to referee #1:  
+
+Please address the remaining concerns of referee #1:  
+
+1. The authors did a nice job explaining how the settled in on the ataluren binding sites for study to 3 sites. However, it cannot be disregarded that ataluren was shown to bind to 22 sites. Indeed, these results suggests that while ataluren is shown here to inhibit translate termination in this reductionist assay, it does so in a rather non-specific manor, as only \(13.6\%\) of the binding sites contribute to translation termination activity.  
+
+Non- site- specific labeling, including from the reactive photoproduct formed from the photoaffinity label in solution, is always present in PAL experiments. This is especially true in the present case because the relatively low affinity binding of AzAt to its target sites necessitated using a high concentration of AzAt in the experiments presented in Figs. 2d and 2e. This is why the results in Tables 1 and S1 demonstrating the site- specificity of the labeling of RNAse H Fragments I- III and the lack of specificity in the labeling of RNAse H fragments IV and V are so important. We do not consider the relatively low \(13.6\%\) labeling of the site- specifically labeled sites as problematic, because the large majority of the remaining \(86.4\%\) labeling, spread over the entire ribosomal RNA, is likely to come either from very weakly binding sites, none of which are anywhere close to eRF1, or from solution.  
+
+2) The authors do a nice job of highlighting studies where ataluren has somewhat succeeded in promoting PTC readthrough. However, ataluren is far from a success in the clinic and there are several studies that show it has no significant effect in vivo. The authors should do the literature and clinical trials justice in ataluren's failures as well. Further, the manuscript would benefit from a section that describes how individual PTCs (position, sequence context, etc) could be more or less difficult to readthrough by TRIDS. This discussion is common in most PTC suppression studies. Do the same rules apply to TRIDS as they do for aminoglycosides for instance?  
+
+In response to the comment regarding literature and clinical trials, we have added significant text to paragraph 1 of the Introduction, (lines 35- 37, 41- 46) which also includes seven new literature references (#s 4- 6, 9, 16- 18). We do not think it necessary to add a section on how position and sequence context of a nonsense codon affect readthrough, as these variables were not investigated in this manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__44ac714467bf52e85c6116012157571d06f97bede43922ad63033d05e6843059/supplementary_0_Peer Review File__44ac714467bf52e85c6116012157571d06f97bede43922ad63033d05e6843059_det.mmd

@@ -0,0 +1,62 @@
+<|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, 866, 250]]<|/det|>
+Ataluren Binds to Multiple Protein Synthesis Apparatus Sites and Competitively Inhibits Release Factor Dependent Termination  
+
+<|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|>text<|/ref|><|det|>[[115, 90, 844, 141]]<|/det|>
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 168, 300, 185]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[115, 225, 393, 241]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 281, 878, 317]]<|/det|>
+The authors did an excellent job replying to the comments from the first review. However, there are still questions about some of the interpretation of results.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 358, 201, 373]]<|/det|>
+## Comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 412, 857, 504]]<|/det|>
+1) The authors did a nice job explaining how the settled in on the ataluren binding sites for study to 3 sites. However, it cannot be disregarded that ataluren was shown to bind to 22 sites. Indeed, these results suggests that while ataluren is shown here to inhibit translate termination in this reductionist assay, it does so in a rather non-specific manor, as only \(13.6\%\) of the binding sites contribute to translation termination activity.  
+
+<|ref|>text<|/ref|><|det|>[[114, 543, 865, 670]]<|/det|>
+2) The authors do a nice job of highlighting studies where ataluren has somewhat succeeded in promoting PTC readthrough. However, ataluren is far from a success in the clinic and there are several studies that show it has no significant effect in vivo. The authors should do the literature and clinical trials justice in ataluren's failures as well. Further, the manuscript would benefit from a section that describes how individual PTCs (position, sequence context, etc) could be more or less difficult to readthrough by TRIDS. This discussion is common in most PTC suppression studies. Do the same rules apply to TRIDS as they do for aminoglycosides for instance?  
+
+<|ref|>text<|/ref|><|det|>[[115, 740, 393, 755]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 795, 873, 849]]<|/det|>
+It appears that many errors were caught during this peer review process. The authors have now addressed at least to some degree of satisfaction all of the comments and suggestions of both Referees of this work in their revised manuscript
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[116, 92, 307, 108]]<|/det|>
+Response to referee #1:  
+
+<|ref|>text<|/ref|><|det|>[[116, 126, 584, 145]]<|/det|>
+Please address the remaining concerns of referee #1:  
+
+<|ref|>text<|/ref|><|det|>[[115, 160, 879, 250]]<|/det|>
+1. The authors did a nice job explaining how the settled in on the ataluren binding sites for study to 3 sites. However, it cannot be disregarded that ataluren was shown to bind to 22 sites. Indeed, these results suggests that while ataluren is shown here to inhibit translate termination in this reductionist assay, it does so in a rather non-specific manor, as only \(13.6\%\) of the binding sites contribute to translation termination activity.  
+
+<|ref|>text<|/ref|><|det|>[[114, 265, 861, 457]]<|/det|>
+Non- site- specific labeling, including from the reactive photoproduct formed from the photoaffinity label in solution, is always present in PAL experiments. This is especially true in the present case because the relatively low affinity binding of AzAt to its target sites necessitated using a high concentration of AzAt in the experiments presented in Figs. 2d and 2e. This is why the results in Tables 1 and S1 demonstrating the site- specificity of the labeling of RNAse H Fragments I- III and the lack of specificity in the labeling of RNAse H fragments IV and V are so important. We do not consider the relatively low \(13.6\%\) labeling of the site- specifically labeled sites as problematic, because the large majority of the remaining \(86.4\%\) labeling, spread over the entire ribosomal RNA, is likely to come either from very weakly binding sites, none of which are anywhere close to eRF1, or from solution.  
+
+<|ref|>text<|/ref|><|det|>[[114, 490, 881, 633]]<|/det|>
+2) The authors do a nice job of highlighting studies where ataluren has somewhat succeeded in promoting PTC readthrough. However, ataluren is far from a success in the clinic and there are several studies that show it has no significant effect in vivo. The authors should do the literature and clinical trials justice in ataluren's failures as well. Further, the manuscript would benefit from a section that describes how individual PTCs (position, sequence context, etc) could be more or less difficult to readthrough by TRIDS. This discussion is common in most PTC suppression studies. Do the same rules apply to TRIDS as they do for aminoglycosides for instance?  
+
+<|ref|>text<|/ref|><|det|>[[115, 648, 883, 736]]<|/det|>
+In response to the comment regarding literature and clinical trials, we have added significant text to paragraph 1 of the Introduction, (lines 35- 37, 41- 46) which also includes seven new literature references (#s 4- 6, 9, 16- 18). We do not think it necessary to add a section on how position and sequence context of a nonsense codon affect readthrough, as these variables were not investigated in this manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__44bd8119853e2fcc7fb044a9f959e97bc3b0440d36ba5953911b7e794c249100/images_list.json

@@ -0,0 +1,145 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_0.jpg",
+    "caption": "Figure R1. a) Schematic of circuit for the modified tanh activation function using a \\(n\\) -type MoS2 memtransistor (T1) and a \\(V\\) -doped \\(p\\) -type WSe2 memtransistor (T2), where the input voltage ( \\(V_{S}\\) ) is applied to the gate terminal of T1 and T2. b) The transfer characteristics of the circuit (solid line) i.e., output voltage ( \\(V_{O}\\) ) versus \\(V_{S}\\) , closely models the tanh activation function (dotted line). c) Schematic of circuit for the modified sigmoid activation function and its d) transfer characteristics.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_1.jpg",
+    "caption": "Figure R2. a) Schematic of the GRNG-based synapse (left). The input to the synapse, \\(V_{\\mathrm{in}}\\) is applied as \\(+V_{\\mathrm{in}}\\) and \\(-V_{\\mathrm{in}}\\) to the memtransistors, \\(T_{+}\\) and \\(T_{-}\\) with conductance \\(G_{+}\\) and \\(G_{- }\\) (modulated using \\(V_{G_{+}}\\) and \\(V_{G_{- }}\\) ), respectively. The effective conductance of this synapse is given by \\(G_{eff} = G_{+} - G_{- }\\) , allowing positive and negative conductance. On the right side is its implementation in LTSpice, where both memtransistors are implemented using resistors. b) Device-to-device variation across 40 \\(\\mathrm{MoS_2}\\) memtransistors for different program and erase voltages. Relationship between the change in the drain current \\((\\Delta I_{\\mathrm{DS}})\\) as a result of c) erase and d) program operation and the starting current \\((I_{\\mathrm{DS,start}})\\) . e) Histogram of \\(\\Delta I_{\\mathrm{DS}} / I_{\\mathrm{DS,start}}\\) for the erase operation, following a Gaussian distribution.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        128,
+        872,
+        480
+      ]
+    ],
+    "page_idx": 7
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_2.jpg",
+    "caption": "Figure R3. a) Training and testing curves for 300 epochs of the BNN constructed to classify the PIMA Indians diabetes dataset. b) Accuracy and predicative accuracy as function of model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. c) Total entropy, aleatoric entropy and epistemic entropy as a function of model variation. d) Accuracy and predictive accuracy as a function of input variation. e) Total entropy, aleatoric entropy and epistemic entropy as a function of input variation.",
+    "footnote": [],
+    "bbox": [
+      [
+        125,
+        93,
+        870,
+        389
+      ]
+    ],
+    "page_idx": 9
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_3.jpg",
+    "caption": "Figure R4. a) Schematic of the GRNG-based synapse (left). The input to the synapse, \\(V_{\\mathrm{in}}\\) is applied as \\(+V_{\\mathrm{in}}\\) and \\(-V_{\\mathrm{in}}\\) to the memtransistors, \\(T_{+}\\) and \\(T_{- }\\) with conductance \\(G_{+}\\) and \\(G_{- }\\) (modulated using \\(V_{G_{+}}\\) and \\(V_{G_{- }}\\) ), respectively. The effective conductance of this synapse is given by \\(G_{eff} = G_{+} - G_{- }\\) , allowing positive and negative conductance. On the right side is its implementation in LTSpice, where both memtransistors are implemented using resistors. b) Device-to-device variation across 40 MoS2 memtransistors for different program and erase voltages. Relationship between the change in the drain current \\((\\Delta I_{\\mathrm{DS}})\\) as a result of c) erase and d) program operation and the starting current \\((I_{\\mathrm{DS,start}})\\) . e) Histogram of \\(\\Delta I_{\\mathrm{DS}} / I_{\\mathrm{DS,start}}\\) for the erase operation, following a Gaussian distribution.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        92,
+        876,
+        416
+      ]
+    ],
+    "page_idx": 12
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_4.jpg",
+    "caption": "Figure R5. Endurance characteristics of an \\(\\mathrm{MoS_2}\\) memtransistor for 2000 cycles.",
+    "footnote": [],
+    "bbox": [
+      [
+        352,
+        474,
+        636,
+        646
+      ]
+    ],
+    "page_idx": 16
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_5.jpg",
+    "caption": "Figure R6. a) Training and testing curves for 300 epochs of the BNN constructed to classify the PIMA Indians diabetes dataset. b) Accuracy and predicative accuracy as function of model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. c) Total entropy, aleatoric entropy and epistemic entropy as a function of model variation. d) Accuracy and predictive accuracy as a function of input variation. e) Total entropy, aleatoric entropy and epistemic entropy as a function of input variation.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        300,
+        870,
+        599
+      ]
+    ],
+    "page_idx": 19
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_6.jpg",
+    "caption": "Figure R7. a) Schematic of circuit for the modified tanh activation function using a \\(n\\) -type \\(\\mathrm{MoS}_2\\) memtransistor (T1) and a \\(V\\) -doped \\(p\\) -type WSe2 memtransistor (T2), where the input voltage \\((V_{S})\\) is applied to the gate terminal of T1 and T2. b) The transfer characteristics of the circuit (solid line) i.e., output voltage \\((V_{O})\\) versus \\(V_{S}\\) , closely models the tanh activation function (dotted line). a) Schematic of circuit for the modified sigmoid activation function and its b) transfer characteristics.",
+    "footnote": [],
+    "bbox": [
+      [
+        122,
+        195,
+        874,
+        344
+      ]
+    ],
+    "page_idx": 20
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_7.jpg",
+    "caption": "Figure R8. a) Accuracy and predicative accuracy as function of synaptic variation. Here, the effect of variation in synaptic devices and sense resistors is demonstrated. b) Accuracy and predicative accuracy as function of total model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. As evident, the inclusion of activation function does not significantly impact the accuracy.",
+    "footnote": [],
+    "bbox": [
+      [
+        223,
+        685,
+        787,
+        840
+      ]
+    ],
+    "page_idx": 23
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_8.jpg",
+    "caption": "Figure R1. a) Accuracy and predicative accuracy as function of variation in synapse. b) Accuracy and predicative accuracy as function of variation in neuron. The variation in neuron includes variation in the circuit for activation function and the sense resistor.",
+    "footnote": [],
+    "bbox": [
+      [
+        400,
+        570,
+        880,
+        707
+      ]
+    ],
+    "page_idx": 24
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_unknown_9.jpg",
+    "caption": "Figure R2. a) Endurance characteristics of an \\(\\mathrm{MoS_2}\\) memtransistor demonstrating various conductance states for a total of 20,000 cycles. b) Moving mean and c) moving standard deviation for these different states.",
+    "footnote": [],
+    "bbox": [
+      [
+        123,
+        699,
+        870,
+        848
+      ]
+    ],
+    "page_idx": 32
+  }
+]

peer_reviews/supplementary_0_Peer Review File__44bd8119853e2fcc7fb044a9f959e97bc3b0440d36ba5953911b7e794c249100/supplementary_0_Peer Review File__44bd8119853e2fcc7fb044a9f959e97bc3b0440d36ba5953911b7e794c249100.mmd

@@ -0,0 +1,524 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Two- dimensional Materials- based Probabilistic Synapses and Reconfigurable Neurons for Measuring Inference Uncertainty Using Bayesian Neural Networks  
+
+![](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):  
+
+This paper describes the use of a previously demonstrated MoS2 nonvolatile transistor for use in neural network processing circuits. The paper implements a Bayesian Neural Net using MoS2 neurons and synapses. The BNN is used to explore uncertainty quantification. The accuracy is assessed using LTSpice on the IRIS dataset and with certain configurations is found to be reasonably high.  
+
+The topic of novel devices to implement BNNs is of interest to the community, as is the quantification of uncertainty in neural inference. This paper is very well written and well organized. However, before considering for publication in Nature Communications, there are some key issues which should be considered to clarify the novelty and significance of the results.  
+
+1. The authors spend a significant amount of time comparing the devices to two terminal memristors, and listing the advantages in that respect. However, the MoS2 transistor as described is a three terminal charge trapping memory, and hence not particularly different from a standard three terminal charge trapping memory. While it is impressive that the authors have developed a CTM devices that incorporates MoS2, it appears the circuits would all be possible to implement in a standard, commercial CTM memory, such as TANOS, SONOS, etc. The random distributions also seem likely an effect of charge trapping, which would be possible in a standard CTM. Please elaborate on whether this is the case, or there is something significantly different about the physics of the MoS2 that cannot be replicated in a standard CTM memory.  
+
+2. The "tanh circuit" created by using T1 as a resistive load is a common drain amplifier, which is a standard transistor amplifier circuit with a well-known transfer function. I would recommend acknowledging this because it seems a bit like this is being presented as a novel circuit (which would seem odd to the microelectronic circuits community). I have also seen similar common drain circuit implemented with floating gate devices, giving similar VTC behavior, such as in Fig 5 of the referenced paper below.  
+
+3. For the LTSpice simulations, can you elaborate on the tanh function implementation; is this just modeled with a single resistor and NMOs? If the VTC looks reasonable (i.e. Fig 4h) then why is this a major source of inaccuracy? If this circuit is so inaccurate, why would it be used rather than the referenced previously used NMOs/PMOS version in ref 39?
+
+<--- Page Split --->
+
+4. A minor, optional note on the amplifier circuits, the more common convention is to have the high voltage on the top of the circuit such that current flows down. (It is opposite of this in Fig 4(g) and Supplemental Fig 7(a).  
+
+5. How was device to device variation modeled in LTSpice? Is there an experimental link between physical nonidealities and the modeling factors should be presented? Also, do the authors know the programming error of the devices, i.e. the delta between the expected and actual conductance after programming?  
+
+6. What is the gate voltage (or gate voltage range) of the synapses during read? What operating region is the transistor in at this gate voltage? It sounds like these are modulated, but in LTSpice, if you are treating the two device pair as a resistor, I am assuming the gate to source and drain to source voltages are such that both devices are operating in the linear part of the triode region, such that a resistor is a suitable element to represent the pair. However this is unclear from the description.  
+
+7. The authors should model something larger than the IRIS dataset, which is quite far from a dataset and network that would be used in a real world application. It has been observed that have seen that excellent accuracy on small datasets does not translate to high accuracy on larger datasets. There is little if any practical application for using a custom, efficient accelerator to process a dataset as small as IRIS. I would understand the very small dataset if this was an experimental demonstration, but given that it is a simulation only, the authors should include something larger.  
+
+8. Although the title conveys an interesting prospect, it does not seem well connected to the content and I would suggest modifying it. The work aims to quantify uncertainty, but does not "avoid inference inaccuracy".  
+
+9. Some of the figure captions are far too small to read. The labels on Fig 5d are equivalent to about a 4pt font.  
+
+## Reference:  
+
+S. Shah, H. Toreyin, J. Hasler, and A. Natarajan, "Models and Techniques for Temperature Robust Systems on a Reconfigurable Platform," Journal of Low Power Electronics and Applications, vol. 7, no. 3, p. 21, Aug. 2017, doi: 10.3390/jlpea7030021.
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+In this work, Sebastian and Das introduce a MoS2- based device that can provide synaptic and neuronal functions to implement hardware Bayesian neural networks. These uses are illustrated in a simulation of a crossbar circuit. The device in itself is similar to previous works by the same group, but it is used in a new manner. The idea of exploiting stochastic effect inherent to device physics to implement probabilistic AI, as is done here, is a very strong concept, which has started to develop in recent literature. The device introduced by the authors has some nice features. However, I have concerns about the manuscript and some questions about the meaning of the work.  
+
+On a general note, I actually do not understand the benefits of using 2- D materials in this project. The authors explain: "The choice of MoS2 as the element of memtransistor is motivated by recent demonstrations highlighting the technological viability of 2D materials [27- 29] and their wide scale adoption in brain- inspired computing [30- 34]."  
+
+This is too vague, in my opinion. In fact, the adoption of 2- D materials in brain- inspired computing is not that wide- scale: Refs 30- 34 all have S. Das as senior author, if I am not mistaken.  
+
+The title of the article, "A Neural Network Accelerator to Avoid Inference Inaccuracy", does not seem related to the paper results. The authors do not present a neural network accelerator, and the paper never really talks about avoiding inference inaccuracy. This was a little surprising.  
+
+The authors should position this new work more clearly with regards to their paper "Gaussian synapses for probabilistic neural networks" (Ref 30), which has some similar keywords and ideas (but is different).  
+
+The programming voltages are very high, similar to FLASH memory, and much higher than memristors. The authors should benchmark their device with alternative approaches (FLASH, non- 2D memristors, phase change memory...).  
+
+I have concerns about the proposed crossbar architecture. I understand that for each presented input, the crossbar devices need to be reprogrammed multiple times to provide a distribution at the
+
+<--- Page Split --->
+
+output (once per sample). Having to reprogram the crossbar numerous times to perform a single inference seems an enormous energy cost. Due to this concern, the paper should include an energy analysis with some benchmarks, in my opinion.  
+
+These multiple programming operations also raise the issue of device endurance, which should be discussed.  
+
+Also, if the synapses need to be reprogrammed each sample, we need additional memory arrays storing the mean value and standard deviations of each weight. This is an important cost, and this would also limit the energy efficiency of the authors' approach, as important data movement will be involved.  
+
+A significant concern is that Fig 5 shows no Bayesian result: we only see the final accuracy, but no distributions.  
+
+Also, why are Bayesian neural networks useful for this example? I think that conventional neural networks get excellent accuracy on this task.  
+
+The degradation in accuracy between software and LTSpice simulation is quite severe, even for a very simple task, due to the neuron behavior. This is an important limitation. Can this problem be fixed?  
+
+Fig 5g with \(0\%\) variation shows a test accuracy that is practically \(100\%\) . However, in the text, the accuracy is said to be \(93.78\%\) . This is a major concern.  
+
+Does Fig 5g include the effects of variability of the neuron devices? I understand that it does not, and that would be a problem, as this variability might have worse effects than the one of synapses.  
+
+The caption of Figure 5 lacks details.  
+
+The methods section should include the methods associated with Figure 5.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+This paper describes the use of a previously demonstrated MoS2 nonvolatile transistor for use in neural network processing circuits. The paper implements a Bayesian Neural Net using MoS2 neurons and synapses. The BNN is used to explore uncertainty quantification. The accuracy is assessed using LTSpice on the IRIS dataset and with certain configurations is found to be reasonably high.  
+
+The topic of novel devices to implement BNNs is of interest to the community, as is the quantification of uncertainty in neural inference. This paper is very well written and well organized. However, before considering for publication in Nature Communications, there are some key issues which should be considered to clarify the novelty and significance of the results.  
+
+We are happy to know that the reviewer finds our manuscript is well written and is of interest to the community. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+1. The authors spend a significant amount of time comparing the devices to two terminal memristors, and listing the advantages in that respect. However, the MoS2 transistor as described is a three terminal charge trapping memory, and hence not particularly different from a standard three terminal charge trapping memory. While it is impressive that the authors have developed a CTM devices that incorporates MoS2, it appears the circuits would all be possible to implement in a standard, commercial CTM memory, such as TANOS, SONOS, etc. The random distributions also seem likely an effect of charge trapping, which would be possible in a standard CTM. Please elaborate on whether this is the case, or there is something significantly different about the physics of the MoS2 that cannot be replicated in a standard CTM memory.  
+
+We would like to thank the reviewer for raising this point. We agree that our Bayesian neural network (BNN) architecture can be implemented with standard three- terminal charge- trap flash memories such as TaN- Al2O3- Si3N4- SiO2- Si (TANOS), Si- SiO2- Si3N4- SiO2- Si (SONOS) [1- 3]. The random distributions are observed as an effect of random nature of charge trapping which is
+
+<--- Page Split --->
+
+typically observed in charge- trap memory (CTM) devices [4, 5]. Hence, our implementation can be easily translated to other CTM devices. Our \(\mathrm{MoS_2}\) memtransistors offer an alternative to other CTM, such as TANOS and SONOS. The atomically thin two- dimensional (2D) semiconductors allow geometric miniaturization of field- effect transistors (FETs) without any loss of electrostatic integrity. The scalability of FETs is captured through the screening length \((\lambda_{\mathrm{SC}})\) , shown in Eq. R1, which represents the competition between gate and drain potential for control of the channel charge.  
+
+\[\lambda_{\mathrm{SC}} = \sqrt{\frac{\epsilon_{\mathrm{s}}}{\epsilon_{\mathrm{ox}}}} t_{\mathrm{s}}t_{\mathrm{ox}} \quad (R1)\]  
+
+Here, \(t_{\mathrm{s}}\) and \(t_{\mathrm{ox}}\) are the thicknesses and \(\epsilon_{\mathrm{s}}\) and \(\epsilon_{\mathrm{ox}}\) are the dielectric constants of the semiconducting channel and the insulating oxide, respectively. To avoid short channel effects the channel length of an FET \((L_{\mathrm{CH}})\) has to be at least three times higher than the screening length, i.e. \(\mathrm{L_{CH} > 3\lambda_{SC}}\) [6]. The atomically thin semiconducting monolayers allows extreme scalability, as they offer \(t_{\mathrm{s}}\) lower than \(1\mathrm{nm}\) . Among various semiconducting 2D materials, \(\mathrm{MoS_2}\) has gained the most attention owing to its dominant n- type transport, stability, and ease of high- quality growth [7- 10]. Hence, we have utilized \(\mathrm{MoS_2}\) memtransistors for the implementation of BNN.  
+
+We have clarified that the circuits can also be implemented with standard flash memories in the main manuscript.  
+
+2. The "tanh circuit" created by using T1 as a resistive load is a common drain amplifier, which is a standard transistor amplifier circuit with a well-known transfer function. I would recommend acknowledging this because it seems a bit like this is being presented as a novel circuit (which would seem odd to the microelectronic circuits community). I have also seen similar common drain circuit implemented with floating gate devices, giving similar VTC behavior, such as in Fig 5 of the referenced paper below.  
+
+We agree with the reviewer that this is a common- drain amplifier circuit. While the output characteristics of the common- drain amplifier circuit is well- known [11], to the best of our knowledge, it has not been used in the context of neuromorphic computing in order to implement the tanh/sigmoid activation function. As demonstrated in the manuscript, this simple and energy
+
+<--- Page Split --->
+
+efficient circuit can be used to implement the tanh activation function which is otherwise implemented using look- up- tables (LUT) consisting of numerous transistors [3].  
+
+We have clarified that the output characteristics is well- known and we have added the reference suggested by the reviewer in the main manuscript.  
+
+3. For the LTSpice simulations, can you elaborate on the tanh function implementation; is this just modeled with a single resistor and NMOS? If the VTC looks reasonable (i.e. Fig 4h) then why is this a major source of inaccuracy? If this circuit is so inaccurate, why would it be used rather than the referenced previously used NMOS/PMOS version in ref 39?  
+
+![](images/Figure_unknown_1.jpg)
+
+<center>Figure R1. a) Schematic of circuit for the modified tanh activation function using a \(n\) -type MoS2 memtransistor (T1) and a \(V\) -doped \(p\) -type WSe2 memtransistor (T2), where the input voltage ( \(V_{S}\) ) is applied to the gate terminal of T1 and T2. b) The transfer characteristics of the circuit (solid line) i.e., output voltage ( \(V_{O}\) ) versus \(V_{S}\) , closely models the tanh activation function (dotted line). c) Schematic of circuit for the modified sigmoid activation function and its d) transfer characteristics. </center>  
+
+We thank the reviewer for raising this question. As the reviewer rightly points out, the tanh activation function is modelled using a resistor and an \(n\) - type FET, which results in an asymmetry in the transfer function. We agree that an implementation using the integration of an \(n\) - type FET and a \(p\) - type FET would result in better accuracy. Hence, we demonstrate a circuit for a modified tanh (m- tanh) activation function using a \(n\) - type MoS2 memtransistor (T1) and a V- doped \(p\) - type WSe2 memtransistor (T2) as shown in Fig. R1a. The transfer function of the circuit i.e., output voltage ( \(V_{O}\) ) versus input voltage ( \(V_{S}\) ) closely follows the tanh activation function as shown in Fig. R1b. The maximum of the m- tanh activation function is determined by the drain voltage ( \(V_{\mathrm{DD}}\) ). Here, \(V_{S}\) is applied to the gate of T1 and T2. T1 and T2 are programmed to ensure that the m- tanh function passes through the origin. Note that when \(V_{S} = - 2 \mathrm{~V}\) , T1 operates in the off- state and T2
+
+<--- Page Split --->
+
+operates in the on- state, resulting in \(V_{\mathrm{O}} = - V_{\mathrm{DD}}\) , whereas for \(V_{\mathrm{S}} = 2 \mathrm{V}\) , T1 operates in the on- state and becomes more conductive than T2, which results in \(V_{\mathrm{O}} = V_{\mathrm{DD}}\) . While this output characteristics is well- known [3, 12], to the best of our knowledge it has not been used to implement activation functions, as the activation functions are typically implemented using look- up- tables [3]. Additionally, modified sigmoid activation function can be realized by applying 0 V to the drain terminal of T1, as shown in Fig. R1c and Fig. R1d.  
+
+We have revised the implementation of the activation function and added this discussion in the main manuscript.  
+
+4. A minor, optional note on the amplifier circuits, the more common convention is to have the high voltage on the top of the circuit such that current flows down. (It is opposite of this in Fig 4(g) and Supplemental Fig 7(a).  
+
+We thank the reviewer for this suggestion. We have changed the figures to follow this common convention.  
+
+5. How was device to device variation modeled in LTSpice? Is there an experimental link between physical nonidealities and the modeling factors should be presented? Also, do the authors know the programming error of the devices, i.e. the delta between the expected and actual conductance after programming?  
+
+We are happy to provide further clarification. The synapses consisting of 2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) are modelled using resistors, as shown in Fig. R2a. Here, \(\mathrm{T_{+}}\) is modelled as a resistor which draws from Gaussian conductance distribution given by \(G_{+} \sim N(\mu_{G_{+}}, \sigma_{G_{+}})\) and \(\mathrm{T_{- }}\) is modelled as a resistor set to a constant conductance of \(G_{- }\) . This results in the synapse having an effective conductance distribution given by \(G_{eff} \sim N(\mu_{G_{eff}}, \sigma_{G_{eff}})\) , where, \(\mu_{G_{eff}} = \mu_{G_{+}} - G_{- }\) and \(\sigma_{G_{eff}} = \sigma_{G_{+}}\) , enabling independent control of the mean and standard deviation of \(G_{eff}\) .
+
+<--- Page Split --->
+![](images/Figure_unknown_2.jpg)
+
+<center>Figure R2. a) Schematic of the GRNG-based synapse (left). The input to the synapse, \(V_{\mathrm{in}}\) is applied as \(+V_{\mathrm{in}}\) and \(-V_{\mathrm{in}}\) to the memtransistors, \(T_{+}\) and \(T_{-}\) with conductance \(G_{+}\) and \(G_{- }\) (modulated using \(V_{G_{+}}\) and \(V_{G_{- }}\) ), respectively. The effective conductance of this synapse is given by \(G_{eff} = G_{+} - G_{- }\) , allowing positive and negative conductance. On the right side is its implementation in LTSpice, where both memtransistors are implemented using resistors. b) Device-to-device variation across 40 \(\mathrm{MoS_2}\) memtransistors for different program and erase voltages. Relationship between the change in the drain current \((\Delta I_{\mathrm{DS}})\) as a result of c) erase and d) program operation and the starting current \((I_{\mathrm{DS,start}})\) . e) Histogram of \(\Delta I_{\mathrm{DS}} / I_{\mathrm{DS,start}}\) for the erase operation, following a Gaussian distribution. </center>  
+
+While the cycle- to- cycle variation in programming is useful for the generation of Gaussian random numbers, the device- to- device variation is undesirable. Fig. R2b demonstrates the device- to- device variation across 40 \(\mathrm{MoS_2}\) memtransistors while performing program/erase operation. Here, the progressively higher programming voltage pulses \((V_{\mathrm{P}})\) followed by progressively higher erase voltage pulses \((V_{\mathrm{E}})\) are applied to the \(\mathrm{MoS_2}\) memtransistors and the drain- to- source current \((I_{\mathrm{DS}})\) is measured at a read voltage \((V_{\mathrm{R}})\) of 0 V. This is repeated for 2 cycles, to demonstrate cycle- to- cycle variation. While the cycle- to- cycle variation is beneficial for a BNN, device- to- device variation is detrimental to its operation. Fig. R2c and Fig. R2d demonstrates the relationship between the change in current \((\Delta I_{\mathrm{DS}})\) as a result of erase or program operation and the starting current \((I_{\mathrm{DS,start}})\) before each erase and program operation, respectively. They follow a power law relationship given by Eq. R2.
+
+<--- Page Split --->
+
+\[\Delta I_{\mathrm{DS}} = p * I_{\mathrm{DS,start}}^{q}\]  
+
+Here, \(p\) is the scaling factor and \(q\) is the exponential factor. If we account for this dependence, the device- to- device variation can be reduced. \(\Delta I_{\mathrm{DS}} / I_{\mathrm{DS,start}}^{q}\) follows a Gaussian distribution, as shown representatively in Fig. R2e for the erase operation.  
+
+In order to evaluate the effect of device- to- device variation, we implement up to \(10\%\) variation in the parameters \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) . In line with the above discussion, they are drawn from Gaussian distributions where the mean is given by expected \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) and standard deviation of up to \(10\%\) is considered. In the revised manuscript, we have also modelled the effect of device- to- device variation in the neurons. For this, we assume up to \(10\%\) variation in the threshold voltage for the \(n\) - type FET \((V_{\mathrm{TH},n})\) and the \(p\) - type FET \((V_{\mathrm{TH},p})\) . We have also implemented up to \(10\%\) variation in resistance of the sense resistor.  
+
+We have included this discussion in the supplementary information and main manuscript.  
+
+6. What is the gate voltage (or gate voltage range) of the synapses during read? What operating region is the transistor in at this gate voltage? It sounds like these are modulated, but in LTSpice, if you are treating the two-device pair as a resistor, I am assuming the gate to source and drain to source voltages are such that both devices are operating in the linear part of the triode region, such that a resistor is a suitable element to represent the pair. However, this is unclear from the description.  
+
+We are happy to provide further clarification. A gate- to- source voltage \((V_{\mathrm{GS}})\) or read voltage \((V_{\mathrm{R}})\) of \(0 \mathrm{~V}\) is used during the read step. \(V_{\mathrm{R}}\) of \(0\) is used in order to avoid read- disturb caused by a positive or negative \(V_{\mathrm{GS}}\) . As the reviewer rightly pointed out, both the \(V_{\mathrm{GS}}\) and drain- to- source voltage \((V_{\mathrm{DS}})\) are used such that the transistors are operating in the linear part of the triode region. Hence, they are modelled using resistors.  
+
+We have added this clarification in the main manuscript.
+
+<--- Page Split --->
+
+7. The authors should model something larger than the IRIS dataset, which is quite far from a dataset and network that would be used in a real-world application. It has been observed that have seen that excellent accuracy on small datasets does not translate to high accuracy on larger datasets. There is little if any practical application for using a custom, efficient accelerator to process a dataset as small as IRIS. I would understand the very small dataset if this was an experimental demonstration, but given that it is a simulation only, the authors should include something larger.  
+
+We appreciate the suggestion by the reviewer. We have now implemented a BNN classifier to classify the PIMA Indian diabetes dataset. This dataset consists of nine parameters such as number of pregnancies, glucose levels, insulin levels, body mass index, age, etc. To classify this dataset, we use a fully connected \(8 \times 10 \times 2\) BNN i.e., it has an input layer with 8 neurons, one hidden layer with 10 neurons, and an output layer with 2 neurons. The dataset with 767 instances is divided into 720 for training and 47 for testing. Bayes by Backprop algorithm, with a Gaussian prior is used to train the synaptic weight distributions [13, 14]. The BNN is trained off- chip for 300 epochs as shown in Fig. R3a, to obtain train accuracy of \(75.41\%\) and test accuracy of \(80.85\%\) . Similar accuracy numbers have been reported in prior works [15- 17].  
+
+Following Eq. R3, the output of the BNN accelerator is sampled \(Z = 100\) times to obtain a predictive distribution, and its mean is used to make the classification.  
+
+\[p(y^{*}|x^{*},D)\approx \frac{1}{Z}\sum_{z = 1}^{Z}p(y^{*}|x^{*},W^{z}) \quad (R3)\]  
+
+Here, \(x^{*}\) and \(y^{*}\) are the test input and output, respectively, D is the training data, and \(W^{z}\) represents the \(z^{\mathrm{th}}\) Monte Carlo weight sample. The softmax of predictive distribution can be used to calculate the uncertainty in classification or entropy given by Eq. R4. [18, 19].  
+
+\[H(p(\hat{y}^{*}|x^{*},D)) = -\sum_{j = 1}^{J}p_{j}(\hat{y}^{*}|x^{*},D)*\log \left(p_{j}(\hat{y}^{*}|x^{*},D)\right) \quad (R4)\]  
+
+Here, \(\hat{y}^{*}\) is the softmax output and \(J\) is the number of output classes. The entropy can be decomposed into epistemic entropy i.e., uncertainty in model and aleatoric entropy i.e., uncertainty in data as shown in Eq. R5.  
+
+\[H(p(\hat{y}^{*}|x^{*},D)) = \Pi \big(p(\hat{y}^{*}|x^{*},D)\big) + E_{W\sim q(W;\theta)}[H(p(\hat{y}^{*}|x^{*},W))] \quad (R5)\]
+
+<--- Page Split --->
+![](images/Figure_unknown_3.jpg)
+
+<center>Figure R3. a) Training and testing curves for 300 epochs of the BNN constructed to classify the PIMA Indians diabetes dataset. b) Accuracy and predicative accuracy as function of model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. c) Total entropy, aleatoric entropy and epistemic entropy as a function of model variation. d) Accuracy and predictive accuracy as a function of input variation. e) Total entropy, aleatoric entropy and epistemic entropy as a function of input variation. </center>  
+
+Here, on the right- hand side, the first term represents epistemic entropy and the second term represents aleatoric entropy. Aleatoric entropy is the average entropy for fixed weights and hence the uncertainty arises from the data. Epistemic entropy can be obtained by subtracting aleatoric entropy from total entropy, following Eq. R5.  
+
+The BNN accelerator to classify PIMA Indian diabetes dataset is evaluated using LTSpice simulations. The synapses consisting of 2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) are modelled using resistors. Here, \(\mathrm{T_{+}}\) is modelled as a resistor which draws from Gaussian conductance distribution given by \(G_{+}\sim N(\mu_{G_{+}},\sigma_{G_{+}})\) and \(\mathrm{T_{- }}\) is modelled as a resistor set to a constant conductance of \(G_{- }\) . The sense transistor is modelled using a resistor. The tanh activation function is implemented using the combination of an \(n\) - type FET and a \(p\) - type FET. Using the BNN accelerator, we are able replicate the test accuracy of \(80.85\%\) . In order to evaluate the effect of device- to- device variation in memtransistors, we implement up to \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the tanh function: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) - type FET and the \(p\) - type FET, and the conductance of the sense resistor \(G_{\mathrm{S}}\) . These parameters are drawn from Gaussian distributions where the mean is given by
+
+<--- Page Split --->
+
+expected parameter value and standard deviation of up to \(10\%\) is considered. Fig. R3b shows the effect of device- to- device (model) variation on the test accuracy and predictive accuracy. The predictive accuracy demonstrates how well the predictions of the expected correct classes are made. Here, the BNN is simulated and averaged over 5 runs. While, we observe a decrease in the test accuracy, it is not seen to significantly impact the operation of the BNN and an accuracy of \(\approx 60\%\) is maintained for \(10\%\) variation. In a BNN, we can use entropy estimation and entropy decomposition to quantify uncertainty and to find its source, respectively. Fig. R3c shows the total entropy, aleatoric entropy, and epistemic entropy as a function of model variation, calculated using Eq. R4 and Eq. R5. Here, we would expect the aleatoric uncertainty to remain unchanged and total entropy to increase. However, their extraction is impacted by the increased model variation. Nevertheless, as expected the epistemic entropy increases as the model variation increases. As shown in Fig. R3d and Fig. R3e, an increase in the input variation results in the degradation of accuracy, along with an increase in the total and aleatoric entropy, while a constant epistemic entropy is maintained, as expected. Hence, in addition to the estimation of total entropy using a BNN, with uncertainty decomposition various sources of entropy can be identified.  
+
+We have changed the classification dataset and added this discussion in the main manuscript.  
+
+8. Although the title conveys an interesting prospect, it does not seem well connected to the content and I would suggest modifying it. The work aims to quantify uncertainty, but does not "avoid inference inaccuracy".  
+
+In line with the reviewer's suggestion, we have changed the title to "A Bayesian Neural Network to Quantify Inference Inaccuracy".  
+
+9. Some of the figure captions are far too small to read. The labels on Fig 5d are equivalent to about a 4pt font.  
+
+We have ensured that the figure labels are larger and easier to read.
+
+<--- Page Split --->
+
+Reviewer #2 (Remarks to the Author):  
+
+In this work, Sebastian and Das introduce a MoS2- based device that can provide synaptic and neuronal functions to implement hardware Bayesian neural networks. These uses are illustrated in a simulation of a crossbar circuit. The device in itself is similar to previous works by the same group, but it is used in a new manner. The idea of exploiting stochastic effect inherent to device physics to implement probabilistic AI, as is done here, is a very strong concept, which has started to develop in recent literature. The device introduced by the authors has some nice features. However, I have concerns about the manuscript and some questions about the meaning of the work.  
+
+We are happy to know that the reviewer appreciates the idea of exploiting stochasticity in devices to implement probabilistic computing demonstrated in the paper. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+1. On a general note, I actually do not understand the benefits of using 2-D materials in this project. The authors explain: "The choice of MoS2 as the element of memtransistor is motivated by recent demonstrations highlighting the technological viability of 2D materials [27-29] and their wide scale adoption in brain-inspired computing [30-34]." This is too vague, in my opinion. In fact, the adoption of 2-D materials in brain-inspired computing is not that wide-scale: Refs 30-34 all have S. Das as senior author, if I am not mistaken.  
+
+We are happy to add further references supporting our statement. Various memristors, transistors and other devices based on 2D materials have been explored for neuromorphic computing applications [20- 22]. 2D materials have also been explored for optoelectronic synapses enabled by their optically active monolayers [23, 24]. Also, note that the atomically thin 2D semiconductors allow geometric miniaturization of FETs without any loss of electrostatic integrity. The scalability of FETs is captured through \(\lambda_{\mathrm{SC}}\) , shown in Eq. R6, which represents the competition between gate and drain potential for control of the channel charge.  
+
+\[\lambda_{\mathrm{SC}} = \sqrt{\frac{\epsilon_{\mathrm{s}}}{\epsilon_{\mathrm{ox}}}} t_{\mathrm{s}}t_{\mathrm{ox}} \quad (R6)\]
+
+<--- Page Split --->
+
+Here, \(t_{\mathrm{s}}\) and \(t_{\mathrm{ox}}\) are the thicknesses and \(\epsilon_{\mathrm{s}}\) and \(\epsilon_{\mathrm{ox}}\) are the dielectric constants of the semiconducting channel and the insulating oxide, respectively. To avoid short channel effects, \(L_{\mathrm{CH}}\) has to be at least three times higher than the screening length, i.e. \(\mathrm{L}_{\mathrm{CH}} > 3\lambda_{\mathrm{SC}}\) [6]. The atomically thin semiconducting monolayers allows extreme scalability, as they offer \(t_{\mathrm{s}}\) lower than \(1 \mathrm{nm}\) . Among various semiconducting 2D materials, \(\mathrm{MoS}_2\) has gained the most attention owing to its dominant n- type transport, stability, and ease of high- quality growth [7- 10]. Hence, we have utilized \(\mathrm{MoS}_2\) memtransistors for the implementation of BNN.  
+
+We have added more references to support our statement in the main manuscript.  
+
+2. The title of the article, "A Neural Network Accelerator to Avoid Inference Inaccuracy", does not seem related to the paper results. The authors do not present a neural network accelerator, and the paper never really talks about avoiding inference inaccuracy. This was a little surprising.  
+
+In line with the reviewer's suggestion, we have changed the title to "A Bayesian Neural Network to Quantify Inference Inaccuracy".  
+
+3. The authors should position this new work more clearly with regards to their paper "Gaussian synapses for probabilistic neural networks" (Ref 30), which has some similar keywords and ideas (but is different).  
+
+A major difference between our implementation of a probabilistic neural network (PNN) [25] and BNN is the nature of the Gaussian synapse. To implement the PNN, we design a Gaussian synapse can mimics the Gaussian function. Fig. R4a shows the schematic of the two transistor Gaussian synapse based on the integration of \(n\) - type \(\mathrm{MoS}_2\) and \(p\) - type BP back- gated FETs. Figure R4b also shows the equivalent circuit diagram for the Gaussian synapse, which simply consists of two variable resistors in series. The two variable resistors, i.e., \(R_{\mathrm{MoS}_2}\) and \(R_{\mathrm{BP}}\) correspond to the \(\mathrm{MoS}_2\) and BP FETs. Figure R4c shows the experimentally measured transfer characteristics i.e., \(I_{\mathrm{DS}}\) versus \(V_{\mathrm{GS}}\) of the Gaussian synapse for different \(V_{\mathrm{DS}}\) . Here, the total current flowing through the series combination of \(\mathrm{MoS}_2\) and BP FETs is measured. Clearly, the transfer characteristics
+
+<--- Page Split --->
+![](images/Figure_unknown_4.jpg)
+
+<center>Figure R4. a) Schematic of the GRNG-based synapse (left). The input to the synapse, \(V_{\mathrm{in}}\) is applied as \(+V_{\mathrm{in}}\) and \(-V_{\mathrm{in}}\) to the memtransistors, \(T_{+}\) and \(T_{- }\) with conductance \(G_{+}\) and \(G_{- }\) (modulated using \(V_{G_{+}}\) and \(V_{G_{- }}\) ), respectively. The effective conductance of this synapse is given by \(G_{eff} = G_{+} - G_{- }\) , allowing positive and negative conductance. On the right side is its implementation in LTSpice, where both memtransistors are implemented using resistors. b) Device-to-device variation across 40 MoS2 memtransistors for different program and erase voltages. Relationship between the change in the drain current \((\Delta I_{\mathrm{DS}})\) as a result of c) erase and d) program operation and the starting current \((I_{\mathrm{DS,start}})\) . e) Histogram of \(\Delta I_{\mathrm{DS}} / I_{\mathrm{DS,start}}\) for the erase operation, following a Gaussian distribution. </center>  
+
+resemble a Gaussian function which can be modeled. The emergence of Gaussian transfer characteristics can be explained using the experimentally measured transfer characteristics of its constituents, i.e., the MoS2 FET and the BP FET, as shown in Figure R4d and Figure R4d, respectively. MoS2 FETs exhibit unipolar \(n\) - type characteristics, whereas, BP FETs are predominantly p- type with large work function contact metals such as Ni [7, 26- 28].  
+
+For the implementation of the BNN, we design synapses that can draw random numbers from a Gaussian distribution. Fig. R4e shows the design of our GRNG- based synapse with independent control over its mean \((\mu)\) and standard deviation \((\sigma)\) , using two MoS2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) . Here, the input to the synapse, \(V_{\mathrm{in}}\) is applied as \(+V_{\mathrm{in}}\) and \(- V_{\mathrm{in}}\) to \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) , respectively, as shown in Fig. R4e. The current at the output node, \(I_{\mathrm{out}}\) is then given by sum of currents through \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) i.e., \(I_{\mathrm{T_{+}}}\) and \(I_{\mathrm{T_{- }}}\) , according to the KCL. To control \(\mu_{G_{\mathrm{eff}}}\) and \(\sigma_{G_{\mathrm{eff}}}\) , \(\mathrm{T_{+}}\) is subjected to successive erase- program- read pulse cycles, while \(\mathrm{T_{- }}\) is programmed to a given state and subsequently only read, using the waveforms shown in Fig. R4f. This results in \(G_{+}\) being drawn from a Gaussian distribution, with \(\mu_{G_{+}} = 5 \mathrm{nS}\) and \(\sigma_{G_{+}} = 0.49 \mathrm{nS}\) i.e., \(G_{+} \sim N\) and \(G_{- }\) having a constant value of \(\approx\)
+
+<--- Page Split --->
+
+8.89 nS, as shown in Fig. R4f. \(G_{\mathrm{eff}}\) is expected to be drawn from a distribution with \(\sigma_{G_{\mathrm{eff}}} = \sigma_{G_{+}}\) and \(\mu_{G_{\mathrm{eff}}} = \mu_{G_{+}}\cdot G_{-}\) . This is confirmed by our measurements as shown in Fig. R4g, \(G_{\mathrm{eff}}\sim N(- 3.9,0.49)\) nS. As evident, while we have some similar keywords in both works, even the nature of synapse is very different between our PNN and BNN work. Additionally, PNNs and BNNs are both different kind of networks based on different algorithms for training such as expectation maximization and backpropagation, respectively with different network structure. In terms of the structure and operation, BNNs are closer to deep artificial neural networks, with the addition of synapses being represented by probability distributions.  
+
+4. The programming voltages are very high, similar to FLASH memory, and much higher than memristors. The authors should benchmark their device with alternative approaches (FLASH, non-2D memristors, phase change memory...).  
+
+We are happy to benchmark our devices with other alternative approaches. Table R1 shows the comparison in energy consumption between different memory technologies such as memristor, phase- change memory (PCM), NAND and our 2D memtransistors [29]. While the program/erase voltages and times for the transistor memory technologies are high, the energy consumption is similar as the programming current is much lower for the transistor technologies ( \(\approx 10^{- 11}\) A) compared to memristors and PCM. Also note that, while our demonstration uses 2D memtransistors, it can also be implemented using commercial NAND technologies.  
+
+<table><tr><td colspan="5">Table R1. Memory Technologies Energy Benchmark</td></tr><tr><td></td><td>Memristor</td><td>PCM</td><td>NAND</td><td>2D Memtransistor</td></tr><tr><td>Cell Elements</td><td>1T1R</td><td>1T1R</td><td>1T</td><td>1T</td></tr><tr><td>Read Time (ns)</td><td>&amp;lt;50</td><td>&amp;lt;60</td><td>&amp;lt;50</td><td>&amp;lt;50</td></tr><tr><td>Read Voltage (V)</td><td>&amp;lt;3</td><td>3</td><td>2</td><td>0</td></tr><tr><td>Program/Erase Time (ns)</td><td>&amp;lt;250</td><td>60</td><td>10⁶</td><td>10⁵</td></tr><tr><td>Program/Erase Voltage (V)</td><td>&amp;lt;3</td><td>3</td><td>15</td><td>13</td></tr><tr><td>Program/Erase Energy (fJ)</td><td>&amp;lt;50</td><td>6 × 10³</td><td>10</td><td>10</td></tr></table>
+
+<--- Page Split --->
+
+We have included this table in the supplementary information.  
+
+5. I have concerns about the proposed crossbar architecture. I understand that for each presented input, the crossbar devices need to be reprogrammed multiple times to provide a distribution at the output (once per sample). Having to reprogram the crossbar numerous times to perform a single inference seems an enormous energy cost. Due to this concern, the paper should include an energy analysis with some benchmarks, in my opinion.  
+
+We thank the reviewer for their recommendation. We have now included an energy analysis. The total energy consumption of the BNN accelerator is given by Eq. R7.  
+
+\[\begin{array}{l}{{E_{\mathrm{Total}}=E_{\mathrm{syn}}+E_{\mathrm{sense}}+E_{\mathrm{tanh}}+E_{\mathrm{P}}+E_{\mathrm{E}}}}\\ {{=\sum_{N}\bigl[\bigl(I_{\mathrm{T}_{+}}+I_{\mathrm{T}_{-}}\bigr)V_{\mathrm{in}}t\bigr]_{\mathrm{syn}}+\bigl[I_{\mathrm{S}}V_{\mathrm{S}}t\bigr]_{\mathrm{sense}}+\bigl[(I_{\mathrm{T}_{1}}+I_{\mathrm{T}_{2}})V_{\mathrm{DD}}t\bigr]_{\mathrm{tanh}}+E_{\mathrm{P}}+E_{\mathrm{E}}}}\\ {{E_{\mathrm{P}}+E_{\mathrm{E}}=\sum_{N-100}t(I_{\mathrm{P}}V_{\mathrm{P}}+I_{\mathrm{E}}V_{\mathrm{E}})+\sum_{100}\bigl[Q t(I_{\mathrm{P}}V_{\mathrm{P}}+I_{\mathrm{E}}V_{\mathrm{E}}) \bigr]_{\mathrm{P,E}}}}\end{array} \quad (R7)\]  
+
+Here, \(E_{\mathrm{Total}}\) represents the total energy consumption per test sample. \(E_{\mathrm{syn}}\) , \(E_{\mathrm{sense}}\) , and \(E_{\mathrm{tanh}}\) represents the total energy consumption by the synapses, sense resistors, and the tanh circuit, respectively, during the inference step. \(I_{\mathrm{S}}\) represents the current through the sense resistor and \(N\) is the total number of devices. \(E_{\mathrm{P}}\) and \(E_{\mathrm{E}}\) represents the energy consumption for programming and erasing operation on these components. \(I_{\mathrm{P}}\) and \(I_{\mathrm{E}}\) respectively, are the gate currents associated with programming and erase operations. Here, only half of the synaptic devices (100 devices) needs to be programmed and erased for \(Z = 100\) Monte Carlo samples. The rest of the synaptic devices, sense transistors, and tanh circuit components are programmed and erased once and subsequently only used for inference. For \(E_{\mathrm{P}}\) and \(E_{\mathrm{E}}\) evaluations, maximum \(I_{\mathrm{P}}\) , \(V_{\mathrm{P}}\) , \(I_{\mathrm{E}}\) , \(V_{\mathrm{E}}\) values expected are used. Using Eq. R7, \(E_{\mathrm{syn}}\) of \(18 \mathrm{nJ}\) , \(E_{\mathrm{sense}}\) of \(0.025 \mathrm{nJ}\) , and \(E_{\mathrm{tanh}}\) of \(0.012 \mathrm{nJ}\) are obtained from LTSpice simulations and \(E_{\mathrm{P}} + E_{\mathrm{E}}\) of \(0.34 \mathrm{nJ}\) is estimated for a total of \(Z = 100\) Monte Carlo samples, resulting in an \(E_{\mathrm{Total}}\) of \(18.37 \mathrm{nJ}\) . Note that while the program/erase voltages are high the energy consumption associated with them is low as the gate currents are on the order of pA and since the transistors are biased with a drain voltage of \(0 \mathrm{V}\) during programming/erasing, there is no drain current.
+
+<--- Page Split --->
+
+We have included the energy analysis main manuscript.  
+
+## 6. These multiple programming operations also raise the issue of device endurance, which should be discussed.  
+
+The reviewer raises an important point. For the use of \(\mathrm{MoS_2}\) memtransistors for BNN, it's important to evaluate its endurance. Fig. R5 shows the endurance characteristics of a \(\mathrm{MoS_2}\) memtransistor for a total of 2000 cycles. Here, for each cycle the device is switched between a high current state and a low current state by using a program voltage \((V_{\mathrm{P}})\) of - 11 V and erase voltage \((V_{\mathrm{E}})\) of 11 V. The current values are read for \(V_{\mathrm{R}}\) of 0 V. Minimal degradation is observed after 2000 cycles and the distinction between the high and low states are maintained.  
+
+We have included the endurance characteristics in the main manuscript.  
+
+![](images/Figure_unknown_5.jpg)
+
+<center>Figure R5. Endurance characteristics of an \(\mathrm{MoS_2}\) memtransistor for 2000 cycles. </center>  
+
+7. Also, if the synapses need to be reprogrammed each sample, we need additional memory arrays storing the mean value and standard deviations of each weight. This is an important cost, and this would also limit the energy efficiency of the authors' approach, as important data movement will be involved.  
+
+We are happy to provide further clarification. We propose to use a scheme where all devices are erased with a high \(V_{\mathrm{E}}\) of 13 V. Following that, each device must be programmed with the separate
+
+<--- Page Split --->
+
+\(V_{\mathrm{P}}\) values to set them to their expected states. Hence, we agree that we will need to store the \(V_{\mathrm{P}}\) values to obtain the required mean and standard deviation.  
+
+We have included this clarification in the main manuscript.  
+
+8. A significant concern is that Fig 5 shows no Bayesian result: we only see the final accuracy, but no distributions.  
+
+![](images/Figure_unknown_6.jpg)
+
+<center>Figure R6. a) Training and testing curves for 300 epochs of the BNN constructed to classify the PIMA Indians diabetes dataset. b) Accuracy and predicative accuracy as function of model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. c) Total entropy, aleatoric entropy and epistemic entropy as a function of model variation. d) Accuracy and predictive accuracy as a function of input variation. e) Total entropy, aleatoric entropy and epistemic entropy as a function of input variation. </center>  
+
+We are happy to include more Bayesian results. We have now implemented a BNN classifier to classify the PIMA Indian diabetes dataset. This dataset consists of nine parameters such as number of pregnancies, glucose levels, insulin levels, body mass index, and age. To classify this dataset, we use a fully connected \(8 \times 10 \times 2\) BNN i.e., it has an input layer with 8 neurons, one hidden layer with 10 neurons, and an output layer with 2 neurons. The dataset with 767 instances is divided into 720 for training and 47 for testing. Bayes by Backprop algorithm, with a Gaussian prior is used to train the synaptic weight distributions [13, 14]. The BNN is trained off- chip for 300 epochs as
+
+<--- Page Split --->
+
+shown in Fig. R6a, to obtain train accuracy of \(75.41\%\) and test accuracy of \(80.85\%\) . Similar accuracy numbers have been reported in prior works [15- 17].  
+
+Following Eq. R8, the output of the BNN accelerator is sampled \(Z = 100\) times to obtain a predictive distribution, and its mean is used to make the classification.  
+
+\[p(y^{*}|x^{*},D)\approx \frac{1}{Z}\sum_{z = 1}^{Z}p(y^{*}|x^{*},W^{t}) \quad [R8]\]  
+
+Here, \(x^{*}\) and \(y^{*}\) are the test input and output, respectively, D is the training data, and \(W^{z}\) represents the \(z^{\mathrm{th}}\) Monte Carlo weight sample. The softmax of predictive distribution can be used to calculate the uncertainty in classification or entropy given by Eq. R9. [18, 19].  
+
+\[H(p(\hat{y}^{*}|x^{*},D)) = -\sum_{j = 1}^{J}p_{j}(\hat{y}^{*}|x^{*},D)*\log \left(p_{j}(\hat{y}^{*}|x^{*},D)\right) \quad [R9]\]  
+
+Here, \(\hat{y}^{*}\) is the softmax output and \(J\) is the number of output classes. The entropy can be decomposed into epistemic entropy i.e., uncertainty in model and aleatoric entropy i.e., uncertainty in data as shown in Eq. R10.  
+
+\[H(p(\hat{y}^{*}|x^{*},D)) = \Pi \big(p(\hat{y}^{*}|x^{*},D)\big) + E_{W\sim q(W;\theta)}[H(p(\hat{y}^{*}|x^{*},W))] \quad [R10]\]  
+
+Here, on the right- hand side, the first term represents epistemic entropy and the second term represents aleatoric entropy. Aleatoric entropy is the average entropy for fixed weights and hence the uncertainty arises from the data. Epistemic entropy can be obtained by subtracting aleatoric entropy from total entropy, following Eq. R10.  
+
+The BNN accelerator to classify PIMA Indian diabetes dataset is evaluated using LTSpice simulations. The synapses consisting of 2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) are modelled using resistors. Here, \(\mathrm{T_{+}}\) is modelled as a resistor which draws from Gaussian conductance distribution given by \(G_{+}\sim N(\mu_{G_{+}},\sigma_{G_{+}})\) and \(\mathrm{T_{- }}\) is modelled as a resistor set to a constant conductance of \(G_{- }\) . The sense transistor is modelled using a sense resistor. The neuron is implemented using the combination of an \(n\) - type FET and a \(p\) - type FET. Using the BNN accelerator, we are able replicate the test accuracy of \(80.85\%\) . In order to evaluate the effect of device- to- device variation in memtransistors, we implement up to \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the neuron: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) - type FET and the \(p\) - type FET, and the conductance of the sense resistor \(G_{S}\) . These parameters are drawn from Gaussian distributions where the mean is given by expected parameter value and
+
+<--- Page Split --->
+
+standard deviation of up to \(10\%\) is considered. Fig. R6b shows the effect of device- to- device (model) variation on the test accuracy and predictive accuracy. The predictive accuracy demonstrates how well the predictions of the expected correct classes are made. Here, the BNN is simulated and averaged over 5 runs. While, we observe a decrease in the test accuracy, it is not seen to significantly impact the operation of the BNN and an accuracy of \(\approx 60\%\) is maintained for \(10\%\) variation. In a BNN, we can use entropy estimation and entropy decomposition to quantify uncertainty and to find its source, respectively. Fig. R6c shows the total entropy, aleatoric entropy, and epistemic entropy as a function of model variation, calculated using Eq. R4 and Eq. R5. Here, we would expect the aleatoric uncertainty to remain unchanged and total entropy to increase. However, their extraction is impacted by the increased model variation. Nevertheless, as expected the epistemic entropy increases as the model variation increases. As shown in Fig. R6d and Fig. R6e, an increase in the input variation results in the degradation of accuracy, along with an increase in the total and aleatoric entropy, while a constant epistemic entropy is maintained, as expected. Hence, in addition to the estimation of total entropy using a BNN, with uncertainty decomposition various sources of entropy/uncertainty can be identified.  
+
+We have added this discussion in the main manuscript.  
+
+## 9. Also, why are Bayesian neural networks useful for this example? I think that conventional neural networks get excellent accuracy on this task.  
+
+We thank the reviewer for their suggestion. As discussed in detail for the previous question, we have now implemented BNN classification on the PIMA Indian diabetes dataset, where we have a test accuracy of \(80.85\%\) . This dataset has typically demonstrated similar accuracy numbers as evident from prior works [15- 17], and hence serves better to demonstrate uncertainty estimation.  
+
+We have now implemented BNN to classify the PIMA Indian dataset and revised the main manuscript.
+
+<--- Page Split --->
+
+10. The degradation in accuracy between software and LTSpice simulation is quite severe, even for a very simple task, due to the neuron behavior. This is an important limitation. Can this problem be fixed?  
+
+![](images/Figure_unknown_7.jpg)
+
+<center>Figure R7. a) Schematic of circuit for the modified tanh activation function using a \(n\) -type \(\mathrm{MoS}_2\) memtransistor (T1) and a \(V\) -doped \(p\) -type WSe2 memtransistor (T2), where the input voltage \((V_{S})\) is applied to the gate terminal of T1 and T2. b) The transfer characteristics of the circuit (solid line) i.e., output voltage \((V_{O})\) versus \(V_{S}\) , closely models the tanh activation function (dotted line). a) Schematic of circuit for the modified sigmoid activation function and its b) transfer characteristics. </center>  
+
+We thank the reviewer for raising this question. As the reviewer rightly points out, when the tanh activation function is modelled using a resistor and an \(n\) - type FET, there is an asymmetric transfer function, which leads to a degradation in the accuracy. To avoid this, we have now implemented the tanh activation function using an \(n\) - type FET and a \(p\) - type FET, which enables us to replicate the test accuracy of \(80.85\%\) in LTSpice simulations. For the tanh activation function, we demonstrate a circuit for a modified tanh (m- tanh) activation function using a \(n\) - type \(\mathrm{MoS}_2\) memtransistors (T1) and a V- doped \(p\) - type WSe2 memtransistor (T2) as shown in Fig. R7a. The transfer function of the circuit i.e., output voltage \((V_{O})\) versus input voltage \((V_{S})\) closely follows the tanh activation function as shown in Fig. R7b. The maximum of the m- tanh activation function is determined by the drain voltage \((V_{\mathrm{DD}})\) . Here, \(V_{S}\) is applied to the gate of T1 and T2. T1 and T2 are programmed to ensure that the m- tanh function passes through the origin. Note that when \(V_{S} = - 2 \mathrm{V}\) , T1 operates in the off- state and T2 operates in the on- state, resulting in \(V_{O} = - V_{\mathrm{DD}}\) , whereas for \(V_{S} = 2 \mathrm{V}\) , T1 operates in the on- state and becomes more conductive than T2, which results in \(V_{O} = V_{\mathrm{DD}}\) . While this output characteristics is well- known [3, 12], to the best of our knowledge it has not been used to implement activation functions, as the activation functions are typically implemented using look- up- tables [3]. Additionally, modified sigmoid activation function can be realized by applying \(0 \mathrm{V}\) to the drain terminal of T1, as shown in Fig. R7c and Fig. R7d.
+
+<--- Page Split --->
+
+We have revised the implementation of the activation function and added this discussion in the main manuscript.  
+
+11. Fig 5g with \(0\%\) variation shows a test accuracy that is practically \(100\%\) . However, in the text, the accuracy is said to be \(93.78\%\) . This is a major concern.  
+
+We are happy to provide further clarification. Here, the degradation of accuracy was a result of the asymmetry in the tanh activation function implemented using and \(n\) - type FET and a resistor. By implementing a neuron with a symmetric tanh activation function, using the integration of an \(n\) - type FET and a \(p\) - type FET enabled us to replicate the test accuracy of \(80.85\%\) in LTSpice simulations.  
+
+12. Does Fig 5g include the effects of variability of the neuron devices? I understand that it does not, and that would be a problem, as this variability might have worse effects than the one of synapses.  
+
+The reviewer raises an excellent point. Hence, we have now included the effect of neuron variation as well. In order to evaluate the effect of device- to- device variation in memtransistors, we implement up to \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the neuron: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) - type FET and the \(p\) - type FET, and the conductance of the sense resistor \(G_{S}\) . These parameters are drawn from Gaussian distributions where the mean is given by expected parameter value and  
+
+![](images/Figure_unknown_8.jpg)
+
+<center>Figure R8. a) Accuracy and predicative accuracy as function of synaptic variation. Here, the effect of variation in synaptic devices and sense resistors is demonstrated. b) Accuracy and predicative accuracy as function of total model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. As evident, the inclusion of activation function does not significantly impact the accuracy. </center>
+
+<--- Page Split --->
+
+standard deviation of up to \(10\%\) is considered. To evaluate the added impact of neuron variation, Fig. R8a shows the effect of device- to- device variation when only the weights and sense resistors are considered and Fig. R8b shows the effect of device- to- device variation when weights, sense resistors and neurons are considered. Here, to obtain each data point, the BNN is simulated and averaged over 5 runs. We can clearly see that the variation in neurons does not significantly impact the accuracy.  
+
+In error analysis, we have now included the effect neuron variation as well.  
+
+13. The caption of Figure 5 lacks details. The methods section should include the methods associated with Figure 5.  
+
+We have now included an extensive discussion in the Methods section to provide further details on Fig. 5.
+
+<--- Page Split --->
+
+## References
+
+<--- Page Split --->
+
+[12] A. Sebastian, S. Das, and S. Das, "An Annealing Accelerator for Ising Spin Systems Based on In- Memory Complementary 2D FETs," \*Adv Mater\*, vol. 34, p. e2107076, Jan 2022. [13] J. Antoran, "Bayesian- Neural- Networks," 2019. [14] C. Blundell, J. Cornebise, K. Kavukcuoglu, and D. Wierstra, "Weight Uncertainty in Neural Networks," \*ArXiv\*, vol. abs/1505.05424, 2015. [15] Q. Zou, K. Qu, Y. Luo, D. Yin, Y. Ju, and H. Tang, "Predicting Diabetes Mellitus With Machine Learning Techniques," \*Front Genet\*, vol. 9, p. 515, 2018. [16] R. Vaishali, R. Sasikala, S. Ramasubbareddy, S. Remya, and S. Nalluri, "Genetic algorithm based feature selection and MOE Fuzzy classification algorithm on Pima Indians Diabetes dataset," pp. 1- 5, 2017. [17] R. Z. Islamic, "Diagnosis of Diabetes in Female Population of Pima Indian Heritage with Ensemble of BP Neural Network and SVM," 2012. [18] Y. Kwon, J.- H. Won, B. J. Kim, and M. C. Paik, "Uncertainty quantification using Bayesian neural networks in classification: Application to biomedical image segmentation," \*Computational Statistics & Data Analysis\*, vol. 142, p. 106816, 2020. [19] A. Kendall and Y. Gal, "What uncertainties do we need in Bayesian deep learning for computer vision?," presented at the Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, California, USA, 2017. [20] T. J. Ko, H. Li, S. A. Mofid, C. Yoo, E. Okogbue, S. S. Han, et al., "Two- Dimensional Near- Atom- Thickness Materials for Emerging Neuromorphic Devices and Applications," \*iScience\*, vol. 23, p. 101676, Nov 20 2020. [21] K. C. Kwon, J. H. Baek, K. Hong, S. Y. Kim, and H. W. Jang, "Memristive Devices Based on Two- Dimensional Transition Metal Chalcogenides for Neuromorphic Computing," \*Nanomicro Lett\*, vol. 14, p. 58, Feb 5 2022. [22] J. Bian, Z. Cao, and P. Zhou, "Neuromorphic computing: Devices, hardware, and system application facilitated by two- dimensional materials," \*Applied Physics Reviews\*, vol. 8, p. 041313, 2021. [23] L. Mennel, J. Symonowicz, S. Wachter, D. K. Polyushkin, A. J. Molina- Mendoza, and T. Mueller, "Ultrafast machine vision with 2D material neural network image sensors," \*Nature\*, vol. 579, pp. 62- 66, Mar 2020.
+
+<--- Page Split --->
+
+[24] P. Wu, T. He, H. Zhu, Y. Wang, Q. Li, Z. Wang, et al., "Next- generation machine vision systems incorporating two- dimensional materials: Progress and perspectives," InfoMat, vol. 4, 2021. [25] A. Sebastian, A. Pannone, S. Subbulakshmi Radhakrishnan, and S. Das, "Gaussian synapses for probabilistic neural networks," Nat Commun, vol. 10, p. 4199, Sep 13 2019. [26] S. Das, M. Demarteau, and A. Roelofs, "Ambipolar phosphorene field effect transistor," ACS nano, vol. 8, pp. 11730- 11738, 2014. [27] S. Das, J. A. Robinson, M. Dubey, H. Terrones, and M. Terrones, "Beyond Graphene: Progress in Novel Two- Dimensional Materials and van der Waals Solids," Annual Review of Materials Research, Vol 45, vol. 45, pp. 1- 27, 2015. [28] D. S. Schulman, A. J. Arnold, and S. Das, "Contact engineering for 2D materials and devices," Chem Soc Rev, Mar 2 2018. [29] O. Kavehei, "Memristive devices and circuits for computing, memory, and neuromorphic applications," 2012.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+I have reviewed the response and new manuscript, including the new tanh circuit. This version has addressed many of my previous concerns in their detailed response. However, I still would consider the following two issues with the manuscript, in determining if it is impactful enough to publish in Nature Communications.  
+
+The most significant issue with the paper is that there does not seem to be an advantage to using a MoS2 charge trapping device for this work. The same body of work could have been demonstrated using measured data from a SONOS, TANOS, or other modern, commercial device (perhaps a floating gate cell) – almost certainly with more consistent and stable electrical behavior due to maturity (including consistent random noise). I agree with the standard scaling argument for MoS2 transistors (as logic devices) that the authors presented in their response, but this scaling argument is not really relevant to how the device is being used in the paper. Hence, this MoS2 is not needed to achieve the functionality demonstrated in the paper. My general impression is that the paper is implying that this novel (but previously demonstrated) MoS2 device is enabling this new functionality – the paper must be clear that a new device is - not- required to achieve the presented functionality. Hence, it also follows that the novelty of the paper must be judged on the new circuit and application concepts, and whether these results constitute publication in Nature Comm.  
+
+In general, I did find the circuit and relevant mathematical analysis presented in Fig 5 and the relevant equations compelling and well thought out, albeit an unusual circuit. I would comment that one significant challenge of this configuration will be that it is very dependent on the programming accuracy of the resistance of the three devices at the end/bottom of each column, and programming error of these "neurons" will lead to significantly greater accuracy degradation than the same error would cause in the synapses. Put slightly differently, any noise or variation in those column end devices will essentially have equal footing to the sum of all the noise from synapses in the column.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have very significantly improved the quality of the manuscript. I still have a few comments.
+
+<--- Page Split --->
+
+- The endurance of 2000 is in fact low, as each inference will consume many of these cycles. Also, the endurance experiment only validates that the ON and OFF states can be distinguished, not the fact the probabilistic synapse behavior is retained. This experiment is important, I think. The endurance issue should also be clearly highlighted as one of the major limitations of the approach.  
+
+- The title of the paper is still not adapted. All Bayesian neural networks quantify accuracy, and this particular aspect is not very developed in the manuscript. A more relevant title could be something like "Two-dimensional materials-based synapses for Bayesian neural network".  
+
+- The comparison with Ref 30 that the authors gave me in their answer should be included in the manuscript.
+
+<--- Page Split --->
+
+## Reviewer #1 (Remarks to the Author):  
+
+I have reviewed the response and new manuscript, including the new tanh circuit. This version has addressed many of my previous concerns in their detailed response. However, I still would consider the following two issues with the manuscript, in determining if it is impactful enough to publish in Nature Communications.  
+
+We are happy to know that we have addressed your previous concerns. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+1. The most significant issue with the paper is that there does not seem to be an advantage to using a MoS2 charge trapping device for this work. The same body of work could have been demonstrated using measured data from a SONOS, TANOS, or other modern, commercial device (perhaps a floating gate cell) – almost certainly with more consistent and stable electrical behavior due to maturity (including consistent random noise). I agree with the standard scaling argument for MoS2 transistors (as logic devices) that the authors presented in their response, but this scaling argument is not really relevant to how the device is being used in the paper. Hence, this MoS2 is not needed to achieve the functionality demonstrated in the paper. My general impression is that the paper is implying that this novel (but previously demonstrated) MoS2 device is enabling this new functionality – the paper must be clear that a new device is -not- required to achieve the presented functionality. Hence, it also follows that the novelty of the paper must be judged on the new circuit and application concepts, and whether these results constitute publication in Nature Comm.  
+
+We are happy to provide further clarification. We included the following statements in the main manuscript to clarify that MoS2 memristors can be replaced with devices with charge- trap memory. "Here, the random distributions are observed as an effect of random nature of charge trapping which is typically observed in charge- trap memory devices [1, 2]. Hence, these memtransistors can also be replaced with standard three- terminal charge- trap flash memories such as TaN- Al2O3- Si3N4- SiO2- Si (TANOS) and Si- SiO2- Si3N4- SiO2- Si (SONOS) [3- 5]." Note that MoS2 is being used here as it has emerged as a potential alternative to Si in recent years. Hence, MoS2 and other 2D materials have been used as channel material for devices with floating- gate to demonstrate
+
+<--- Page Split --->
+
+non- volatile memory [6- 9]. Hence, we have utilized MoS₂ memtransistors for the implementation of BNN.  
+
+2. In general, I did find the circuit and relevant mathematical analysis presented in Fig 5 and the relevant equations compelling and well thought out, albeit an unusual circuit. I would comment that one significant challenge of this configuration will be that it is very dependent on the programming accuracy of the resistance of the three devices at the end/bottom of each column, and programming error of these "neurons" will lead to significantly greater accuracy degradation than the same error would cause in the synapses. Put slightly differently, any noise or variation in those column end devices will essentially have equal footing to the sum of all the noise from synapses in the column.  
+
+We are happy to know that the reviewer finds the circuit and mathematical analysis compelling and well thought out. We agree that evaluating the programming error introduced by the neuron must be analyzed. In order to evaluate the effect of device- to- device variation in memtransistors, we implement \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the neuron: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) -  
+
+drawn from Gaussian distributions where the mean is given by expected parameter value and standard deviation of \(10\%\) is considered. We separately evaluate the effect of variation in synapses and neurons on the classification  
+
+![](images/Figure_unknown_9.jpg)
+
+<center>Figure R1. a) Accuracy and predicative accuracy as function of variation in synapse. b) Accuracy and predicative accuracy as function of variation in neuron. The variation in neuron includes variation in the circuit for activation function and the sense resistor. </center>  
+
+accuracy. As evident from Fig. R1, variation in synapse is much more detrimental to the performance of the BNN compared to the variation in neuron.  
+
+In the revised manuscript, we have clarified that variation in synapses is more detrimental to the operation of the BNN.
+
+<--- Page Split --->
+
+## Reviewer #2 (Remarks to the Author):  
+
+The authors have very significantly improved the quality of the manuscript. I still have a few comments.  
+
+We are happy to know that the reviewer finds that the manuscript has improved. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+1. The endurance of 2000 is in fact low, as each inference will consume many of these cycles. Also, the endurance experiment only validates that the ON and OFF states can be distinguished, not the fact the probabilistic synapse behavior is retained. This experiment is important, I think. The endurance issue should also be clearly highlighted as one of the major limitations of the approach.  
+
+The reviewer raises an important point regarding low endurance and retention of synaptic behavior. For the use of \(\mathrm{MoS_2}\) memtransistors as a synapse for a BNN, it's important to ensure that the random number generation is maintained across multiple cycles. Hence using \(\mathrm{MoS_2}\) memtransistors, we demonstrate the endurance characteristics/random number generation for 20,000 cycles in Fig. R2a. The gate is subjected to successive erase- program- read pulses, and hence each cycle shown in Fig. R2a includes the effect of a program and erase operation. Here, multiple current levels consisting of high and low current states (achieved with programming voltages, \(V_{\mathrm{P}}\) of - 8.5 V and - 8 V, respectively) for different drain- to- source voltages ( \(V_{\mathrm{DS}}\) ) are demonstrated. Additionally, to demonstrate the retention of the random number generation, the moving mean \((\mu)\) and moving standard deviation \((\sigma)\) for these states are demonstrated in Fig. R2b  
+
+![PLACEHOLDER_33_0]
+
+<center>Figure R2. a) Endurance characteristics of an \(\mathrm{MoS_2}\) memtransistor demonstrating various conductance states for a total of 20,000 cycles. b) Moving mean and c) moving standard deviation for these different states. </center>
+
+<--- Page Split --->
+
+and Fig. R2c. The moving \(\mu\) and \(\sigma\) are obtained across 100 samples at a time. Clearly the random number generation is stable, and the expected synaptic behaviors of increasing \(\mu\) and \(\sigma\) as a function of \(V_{\mathrm{DS}}\) and \(|V_{\mathrm{P}}|\) is retained after numerous measurements. However, we agree with the reviewer that the endurance must be improved for \(\mathrm{MoS}_2\) memtransistors and we have clarified this in the main text.  
+
+We have included this discussion in the main manuscript and supplementary information.  
+
+2. The title of the paper is still not adapted. All Bayesian neural networks quantify accuracy, and this particular aspect is not very developed in the manuscript. A more relevant title could be something like "Two-dimensional materials-based synapses for Bayesian neural network".  
+
+We thank the reviewer for their suggestion. We have now changed the title to "Two- dimensional materials- based synapses for Bayesian neural network".  
+
+3. The comparison with Ref 30 that the authors gave me in their answer should be included in the manuscript.  
+
+We have included this discussion in the manuscript.
+
+<--- Page Split --->
+
+## References  
+
+[1] L. Danial, V. Gupta, E. Pikhay, Y. Roizin, and S. Kvatinsky, "Modeling a Floating- Gate Memristive Device for Computer Aided Design of Neuromorphic Computing," pp. 472- 477, 2020. [2] C. Monzio Compagnoni, R. Gusmeroli, A. S. Spinelli, and A. Visconti, "Analytical Model for the Electron- Injection Statistics During Programming of Nanoscale and Flash Memories," IEEE Transactions on Electron Devices, vol. 55, pp. 3192- 3199, 2008. [3] J. K. Han, J. Oh, G. J. Yun, D. Yoo, M. S. Kim, J. M. Yu, et al., "Cointegration of single- transistor neurons and synapses by nanoscale CMOS fabrication for highly scalable neuromorphic hardware," Sci Adv, vol. 7, Aug 2021. [4] S. Hwang, J. Yu, G. H. Lee, M. S. Song, J. Chang, K. K. Min, et al., "Capacitor- Based Synaptic Devices for Hardware Spiking Neural Networks," IEEE Electron Device Letters, vol. 43, pp. 549- 552, 2022. [5] T. P. Xiao, C. H. Bennett, B. Feinberg, S. Agarwal, and M. J. Marinella, "Analog architectures for neural network acceleration based on non- volatile memory," Applied Physics Reviews, vol. 7, p. 031301, 2020. [6] D. Li, M. Chen, Z. Sun, P. Yu, Z. Liu, P. M. Ajayan, et al., "Two- dimensional non- volatile programmable p- n junctions," Nat Nanotechnol, vol. 12, pp. 901- 906, Sep 2017. [7] E. Zhang, W. Wang, C. Zhang, Y. Jin, G. Zhu, Q. Sun, et al., "Tunable charge- trap memory based on few- layer MoS2," ACS Nano, vol. 9, pp. 612- 9, Jan 27 2015. [8] C. Liu, X. Yan, J. Wang, S. Ding, P. Zhou, and D. W. Zhang, "Eliminating Overerase Behavior by Designing Energy Band in High- Speed Charge- Trap Memory Based on WSe2," Small, vol. 13, May 2017. [9] Q. Feng, F. Yan, W. Luo, and K. Wang, "Charge trap memory based on few- layer black phosphorus," Nanoscale, vol. 8, pp. 2686- 92, Feb 7 2016.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+I appreciate the authors work modeling the neural net error as a function of variation in the neuron and synapses and found that an important addition to the manuscript. However, in the model of neuron variability vs accuracy, I have found an inconsistency that I've been unable to reconcile. The finding of Fig R1b shows the sense conductance Gs (which should be specific to each column, as Gsj) can vary by \(1000\%\) with a negligible impact on accuracy - which is a very surprising finding and merits a detailed explanation. Mathematically, for this to be true, the Gs term in the denominator of the first summation term in eqn 4a must be negligible compared to the other term (sum) in the denominator. However, this contradicts the original purpose of Gsj correction - which was to correct the error associated with each column, stated as "Gsj(k) is used to make sure that each column of the crossbar array has the same \(\gamma (k)\) ". If Gsj, which represents the inaccuracy of each column, is negligible, then why is this compensation scheme even needed? The proposal of the compensation term implies that Gs is large enough to significantly affect the denominator of 4a. But the plot in R1b implies that Gs has a negligible affect on the accuracy. Given the strong effect of Vthn and Vthp on the output, a similar argument could be made for the very small Vt variation which was varied in fig R1b. Perhaps more details on how specifically the \(10\%\) variation was modeled would clarify this result. In any case, I strongly recommend checking and clarifying this before the article is published - specifically assert whether Gs term has a non- negligible effect on the accuracy and is needed for column compensation, or if Gs is negligible and does not need to be adjusted between columns.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have addressed my last comments well.
+
+<--- Page Split --->
+
+## COMMENTS TO AUTHOR:  
+
+Reviewer #1 (Remarks to the Author):  
+
+I appreciate the authors work modeling the neural net error as a function of variation in the neuron and synapses and found that an important addition to the manuscript. However, in the model of neuron variability vs accuracy, I have found an inconsistency that I've been unable to reconcile. The finding of Fig R1b shows the sense conductance Gs (which should be specific to each column, as Gsj) can vary by \(1000\%\) with a negligible impact on accuracy - which is a very surprising finding and merits a detailed explanation. Mathematically, for this to be true, the Gs term in the denominator of the first summation term in eqn 4a must be negligible compared to the other term (sum) in the denominator. However, this contradicts the original purpose of Gsj correction - which was to correct the error associated with each column, stated as "Gsj(k) is used to make sure that each column of the crossbar array has the same \(\gamma (k)\) ." If Gsj, which represents the inaccuracy of each column, is negligible, then why is this compensation scheme even needed? The proposal of the compensation term implies that Gs is large enough to significantly affect the denominator of 4a. But the plot in R1b implies that Gs has a negligible affect on the accuracy. Given the strong effect of Vthn and Vthp on the output, a similar argument could be made for the very small Vt variation which was varied in fig R1b. Perhaps more details on how specifically the \(10\%\) variation was modeled would clarify this result. In any case, I strongly recommend checking and clarifying this before the article is published - specifically assert whether Gs term has a non- negligible effect on the accuracy and is needed for column compensation, or if Gs is negligible and does not need to be adjusted between columns.  
+
+We are happy to know that the reviewer appreciates our response. We are happy to provide further clarification on the relevance of the sense transistor. Note that the primary purpose of the sense transistor is to perform a current- to- voltage conversion. Here, the current through a column
+
+<--- Page Split --->
+
+(corresponding to the dot- product of input and weights) is converted to a proportional voltage given by Eq. R1.  
+
+\[V_{\mathrm{S}j}^{(k)} = \frac{\sum_{i = 1}^{M}V_{i}^{(k)}G_{ij}^{(k)}}{G_{\mathrm{S}j}^{(k)} + \sum_{i = 1}^{M}G_{ij}^{(k)}} \quad (R1)\]  
+
+Here, \(V_{i}^{(k)}\) is the input voltage, \(G_{ij}^{(k)}\) is the conductance of synapse, and \(G_{\mathrm{S}j}^{(k)}\) is the conductance of the sense transistor for the \(i^{\mathrm{th}}\) row, \(j^{\mathrm{th}}\) column and \(k^{\mathrm{th}}\) layer. The current- to- voltage conversion allows us to seamlessly integrate the circuit for m- tanh activation function into the crossbar array, as the input to the m- tanh activation function is a voltage. The secondary purpose of the sense transistor is to ensure that the nonideality in Eq. R1 i.e., the denominator term is consistent between different columns. Note, that for our implementation, the difference in \(\sum_{i = 1}^{M}G_{ij}^{(k)}\) between different columns is not significant and hence the \(G_{\mathrm{S}j}^{(k)}\) correction in our case is small in magnitude. Additionally, note that each synapse is represented by two devices instead one and hence their combined conductance leads to a much higher magnitude of \(\sum_{i = 1}^{M}G_{ij}^{(k)}\) compared to \(G_{\mathrm{S}j}^{(k)}\) . In a different network implementation, where \(\sum_{i = 1}^{M}G_{ij}^{(k)}\) is significantly different between columns, the impact error in \(G_{\mathrm{S}j}^{(k)}\) on the classification accuracy will be higher.  
+
+We believe that the impact of variation the tanh activation (implemented through variation in \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) ) is low due to the binary nature of the classification problem that we have chosen. For a multiclass classification problem, the impact of the variation in activation function for is expected to be larger. In fact, in the initial version of the manuscript, we inspected a multiclass problem, where the impact of variation in the activation function was seen to be higher. To evaluate the effect of model variation, the parameters are drawn from Gaussian distributions where the mean is given by expected parameter value and standard deviation of up to \(10\%\) is considered.
+
+<--- Page Split --->
+
+This is achieved using gaussian random number generation in Python, with the expected parameter value and percentage of standard deviation. We have verified our implementation to ensure that the variation in the parameters is modelled correctly.
+
+<--- Page Split --->
+
+Reviewer #1 (Remarks to the Author):  
+
+I thank the authors for clarifying the previous response. The authors made a clarification that seems to suggest a change in the manuscript text, but I did not see a change, and hence the statement in current version of the manuscript is not completely correct. I would recommend fixing this before publication.  
+
+Specifically, the authors have written in the rebuttal: "We believe that the impact of variation the tanh activation (implemented through variation in VVTH, nn and VVTH, pp) is low due to the binary nature of the classification problem that we have chosen. For a multiclass classification problem, the impact of the variation in activation function for is expected to be larger."  
+
+This would render the statement in the manuscript incorrect: "In model variation, it was determined that the variation is synapses is more determinantal to the performance of BNN compared to variations in the neuron." From the rebuttal statement, the authors appear to believe this is not always true, but depends on the nature of the classification problem. I would recommend the authors correct this before publication.
+
+<--- Page Split --->
+
+## COMMENTS TO AUTHOR:  
+
+Reviewer #1 (Remarks to the Author):  
+
+I thank the authors for clarifying the previous response. The authors made a clarification that seems to suggest a change in the manuscript text, but I did not see a change, and hence the statement in the current version of the manuscript is not completely correct. I would recommend fixing this before publication.  
+
+Specifically, the authors have written in the rebuttal: "We believe that the impact of variation the tanh activation (implemented through variation in VTH, nn and VTH, pp) is low due to the binary nature of the classification problem that we have chosen. For a multiclass classification problem, the impact of the variation in activation function for is expected to be larger."  
+
+This would render the statement in the manuscript incorrect: "In model variation, it was determined that the variation is synapses is more determinantal to the performance of BNN compared to variations in the neuron." From the rebuttal statement, the authors appear to believe this is not always true, but depends on the nature of the classification problem. I would recommend the authors correct this before publication.  
+
+We thank the reviewer for this suggestion. We have made this clarification in the main manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__44bd8119853e2fcc7fb044a9f959e97bc3b0440d36ba5953911b7e794c249100/supplementary_0_Peer Review File__44bd8119853e2fcc7fb044a9f959e97bc3b0440d36ba5953911b7e794c249100_det.mmd

@@ -0,0 +1,685 @@
+<|ref|>title<|/ref|><|det|>[[61, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[68, 110, 362, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[88, 155, 890, 295]]<|/det|>
+Two- dimensional Materials- based Probabilistic Synapses and Reconfigurable Neurons for Measuring Inference Uncertainty Using Bayesian Neural Networks  
+
+<|ref|>image<|/ref|><|det|>[[57, 732, 240, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[250, 732, 911, 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  
+
+<|ref|>text<|/ref|><|det|>[[56, 784, 936, 924]]<|/det|>
+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|>[[118, 144, 404, 159]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 204, 870, 295]]<|/det|>
+This paper describes the use of a previously demonstrated MoS2 nonvolatile transistor for use in neural network processing circuits. The paper implements a Bayesian Neural Net using MoS2 neurons and synapses. The BNN is used to explore uncertainty quantification. The accuracy is assessed using LTSpice on the IRIS dataset and with certain configurations is found to be reasonably high.  
+
+<|ref|>text<|/ref|><|det|>[[118, 339, 866, 409]]<|/det|>
+The topic of novel devices to implement BNNs is of interest to the community, as is the quantification of uncertainty in neural inference. This paper is very well written and well organized. However, before considering for publication in Nature Communications, there are some key issues which should be considered to clarify the novelty and significance of the results.  
+
+<|ref|>text<|/ref|><|det|>[[117, 453, 880, 617]]<|/det|>
+1. The authors spend a significant amount of time comparing the devices to two terminal memristors, and listing the advantages in that respect. However, the MoS2 transistor as described is a three terminal charge trapping memory, and hence not particularly different from a standard three terminal charge trapping memory. While it is impressive that the authors have developed a CTM devices that incorporates MoS2, it appears the circuits would all be possible to implement in a standard, commercial CTM memory, such as TANOS, SONOS, etc. The random distributions also seem likely an effect of charge trapping, which would be possible in a standard CTM. Please elaborate on whether this is the case, or there is something significantly different about the physics of the MoS2 that cannot be replicated in a standard CTM memory.  
+
+<|ref|>text<|/ref|><|det|>[[117, 660, 875, 770]]<|/det|>
+2. The "tanh circuit" created by using T1 as a resistive load is a common drain amplifier, which is a standard transistor amplifier circuit with a well-known transfer function. I would recommend acknowledging this because it seems a bit like this is being presented as a novel circuit (which would seem odd to the microelectronic circuits community). I have also seen similar common drain circuit implemented with floating gate devices, giving similar VTC behavior, such as in Fig 5 of the referenced paper below.  
+
+<|ref|>text<|/ref|><|det|>[[118, 813, 872, 884]]<|/det|>
+3. For the LTSpice simulations, can you elaborate on the tanh function implementation; is this just modeled with a single resistor and NMOs? If the VTC looks reasonable (i.e. Fig 4h) then why is this a major source of inaccuracy? If this circuit is so inaccurate, why would it be used rather than the referenced previously used NMOs/PMOS version in ref 39?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 114, 872, 168]]<|/det|>
+4. A minor, optional note on the amplifier circuits, the more common convention is to have the high voltage on the top of the circuit such that current flows down. (It is opposite of this in Fig 4(g) and Supplemental Fig 7(a).  
+
+<|ref|>text<|/ref|><|det|>[[118, 210, 875, 283]]<|/det|>
+5. How was device to device variation modeled in LTSpice? Is there an experimental link between physical nonidealities and the modeling factors should be presented? Also, do the authors know the programming error of the devices, i.e. the delta between the expected and actual conductance after programming?  
+
+<|ref|>text<|/ref|><|det|>[[118, 326, 870, 435]]<|/det|>
+6. What is the gate voltage (or gate voltage range) of the synapses during read? What operating region is the transistor in at this gate voltage? It sounds like these are modulated, but in LTSpice, if you are treating the two device pair as a resistor, I am assuming the gate to source and drain to source voltages are such that both devices are operating in the linear part of the triode region, such that a resistor is a suitable element to represent the pair. However this is unclear from the description.  
+
+<|ref|>text<|/ref|><|det|>[[118, 478, 877, 587]]<|/det|>
+7. The authors should model something larger than the IRIS dataset, which is quite far from a dataset and network that would be used in a real world application. It has been observed that have seen that excellent accuracy on small datasets does not translate to high accuracy on larger datasets. There is little if any practical application for using a custom, efficient accelerator to process a dataset as small as IRIS. I would understand the very small dataset if this was an experimental demonstration, but given that it is a simulation only, the authors should include something larger.  
+
+<|ref|>text<|/ref|><|det|>[[118, 630, 878, 684]]<|/det|>
+8. Although the title conveys an interesting prospect, it does not seem well connected to the content and I would suggest modifying it. The work aims to quantify uncertainty, but does not "avoid inference inaccuracy".  
+
+<|ref|>text<|/ref|><|det|>[[118, 728, 870, 762]]<|/det|>
+9. Some of the figure captions are far too small to read. The labels on Fig 5d are equivalent to about a 4pt font.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 807, 202, 822]]<|/det|>
+## Reference:  
+
+<|ref|>text<|/ref|><|det|>[[118, 837, 854, 890]]<|/det|>
+S. Shah, H. Toreyin, J. Hasler, and A. Natarajan, "Models and Techniques for Temperature Robust Systems on a Reconfigurable Platform," Journal of Low Power Electronics and Applications, vol. 7, no. 3, p. 21, Aug. 2017, doi: 10.3390/jlpea7030021.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 175, 405, 190]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[117, 233, 877, 362]]<|/det|>
+In this work, Sebastian and Das introduce a MoS2- based device that can provide synaptic and neuronal functions to implement hardware Bayesian neural networks. These uses are illustrated in a simulation of a crossbar circuit. The device in itself is similar to previous works by the same group, but it is used in a new manner. The idea of exploiting stochastic effect inherent to device physics to implement probabilistic AI, as is done here, is a very strong concept, which has started to develop in recent literature. The device introduced by the authors has some nice features. However, I have concerns about the manuscript and some questions about the meaning of the work.  
+
+<|ref|>text<|/ref|><|det|>[[118, 404, 861, 477]]<|/det|>
+On a general note, I actually do not understand the benefits of using 2- D materials in this project. The authors explain: "The choice of MoS2 as the element of memtransistor is motivated by recent demonstrations highlighting the technological viability of 2D materials [27- 29] and their wide scale adoption in brain- inspired computing [30- 34]."  
+
+<|ref|>text<|/ref|><|det|>[[118, 490, 872, 525]]<|/det|>
+This is too vague, in my opinion. In fact, the adoption of 2- D materials in brain- inspired computing is not that wide- scale: Refs 30- 34 all have S. Das as senior author, if I am not mistaken.  
+
+<|ref|>text<|/ref|><|det|>[[118, 568, 878, 622]]<|/det|>
+The title of the article, "A Neural Network Accelerator to Avoid Inference Inaccuracy", does not seem related to the paper results. The authors do not present a neural network accelerator, and the paper never really talks about avoiding inference inaccuracy. This was a little surprising.  
+
+<|ref|>text<|/ref|><|det|>[[118, 665, 880, 719]]<|/det|>
+The authors should position this new work more clearly with regards to their paper "Gaussian synapses for probabilistic neural networks" (Ref 30), which has some similar keywords and ideas (but is different).  
+
+<|ref|>text<|/ref|><|det|>[[118, 762, 878, 816]]<|/det|>
+The programming voltages are very high, similar to FLASH memory, and much higher than memristors. The authors should benchmark their device with alternative approaches (FLASH, non- 2D memristors, phase change memory...).  
+
+<|ref|>text<|/ref|><|det|>[[118, 859, 875, 894]]<|/det|>
+I have concerns about the proposed crossbar architecture. I understand that for each presented input, the crossbar devices need to be reprogrammed multiple times to provide a distribution at the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 868, 138]]<|/det|>
+output (once per sample). Having to reprogram the crossbar numerous times to perform a single inference seems an enormous energy cost. Due to this concern, the paper should include an energy analysis with some benchmarks, in my opinion.  
+
+<|ref|>text<|/ref|><|det|>[[118, 181, 860, 216]]<|/det|>
+These multiple programming operations also raise the issue of device endurance, which should be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[118, 260, 873, 332]]<|/det|>
+Also, if the synapses need to be reprogrammed each sample, we need additional memory arrays storing the mean value and standard deviations of each weight. This is an important cost, and this would also limit the energy efficiency of the authors' approach, as important data movement will be involved.  
+
+<|ref|>text<|/ref|><|det|>[[118, 375, 852, 410]]<|/det|>
+A significant concern is that Fig 5 shows no Bayesian result: we only see the final accuracy, but no distributions.  
+
+<|ref|>text<|/ref|><|det|>[[118, 423, 847, 459]]<|/det|>
+Also, why are Bayesian neural networks useful for this example? I think that conventional neural networks get excellent accuracy on this task.  
+
+<|ref|>text<|/ref|><|det|>[[118, 502, 852, 555]]<|/det|>
+The degradation in accuracy between software and LTSpice simulation is quite severe, even for a very simple task, due to the neuron behavior. This is an important limitation. Can this problem be fixed?  
+
+<|ref|>text<|/ref|><|det|>[[118, 599, 842, 634]]<|/det|>
+Fig 5g with \(0\%\) variation shows a test accuracy that is practically \(100\%\) . However, in the text, the accuracy is said to be \(93.78\%\) . This is a major concern.  
+
+<|ref|>text<|/ref|><|det|>[[118, 677, 880, 713]]<|/det|>
+Does Fig 5g include the effects of variability of the neuron devices? I understand that it does not, and that would be a problem, as this variability might have worse effects than the one of synapses.  
+
+<|ref|>text<|/ref|><|det|>[[120, 757, 391, 773]]<|/det|>
+The caption of Figure 5 lacks details.  
+
+<|ref|>text<|/ref|><|det|>[[118, 817, 680, 834]]<|/det|>
+The methods section should include the methods associated with Figure 5.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 439, 108]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 115, 884, 240]]<|/det|>
+This paper describes the use of a previously demonstrated MoS2 nonvolatile transistor for use in neural network processing circuits. The paper implements a Bayesian Neural Net using MoS2 neurons and synapses. The BNN is used to explore uncertainty quantification. The accuracy is assessed using LTSpice on the IRIS dataset and with certain configurations is found to be reasonably high.  
+
+<|ref|>text<|/ref|><|det|>[[113, 245, 884, 370]]<|/det|>
+The topic of novel devices to implement BNNs is of interest to the community, as is the quantification of uncertainty in neural inference. This paper is very well written and well organized. However, before considering for publication in Nature Communications, there are some key issues which should be considered to clarify the novelty and significance of the results.  
+
+<|ref|>text<|/ref|><|det|>[[114, 401, 884, 474]]<|/det|>
+We are happy to know that the reviewer finds our manuscript is well written and is of interest to the community. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+<|ref|>text<|/ref|><|det|>[[113, 505, 884, 762]]<|/det|>
+1. The authors spend a significant amount of time comparing the devices to two terminal memristors, and listing the advantages in that respect. However, the MoS2 transistor as described is a three terminal charge trapping memory, and hence not particularly different from a standard three terminal charge trapping memory. While it is impressive that the authors have developed a CTM devices that incorporates MoS2, it appears the circuits would all be possible to implement in a standard, commercial CTM memory, such as TANOS, SONOS, etc. The random distributions also seem likely an effect of charge trapping, which would be possible in a standard CTM. Please elaborate on whether this is the case, or there is something significantly different about the physics of the MoS2 that cannot be replicated in a standard CTM memory.  
+
+<|ref|>text<|/ref|><|det|>[[114, 793, 884, 893]]<|/det|>
+We would like to thank the reviewer for raising this point. We agree that our Bayesian neural network (BNN) architecture can be implemented with standard three- terminal charge- trap flash memories such as TaN- Al2O3- Si3N4- SiO2- Si (TANOS), Si- SiO2- Si3N4- SiO2- Si (SONOS) [1- 3]. The random distributions are observed as an effect of random nature of charge trapping which is
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 87, 884, 265]]<|/det|>
+typically observed in charge- trap memory (CTM) devices [4, 5]. Hence, our implementation can be easily translated to other CTM devices. Our \(\mathrm{MoS_2}\) memtransistors offer an alternative to other CTM, such as TANOS and SONOS. The atomically thin two- dimensional (2D) semiconductors allow geometric miniaturization of field- effect transistors (FETs) without any loss of electrostatic integrity. The scalability of FETs is captured through the screening length \((\lambda_{\mathrm{SC}})\) , shown in Eq. R1, which represents the competition between gate and drain potential for control of the channel charge.  
+
+<|ref|>equation<|/ref|><|det|>[[113, 272, 877, 325]]<|/det|>
+\[\lambda_{\mathrm{SC}} = \sqrt{\frac{\epsilon_{\mathrm{s}}}{\epsilon_{\mathrm{ox}}}} t_{\mathrm{s}}t_{\mathrm{ox}} \quad (R1)\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 334, 884, 514]]<|/det|>
+Here, \(t_{\mathrm{s}}\) and \(t_{\mathrm{ox}}\) are the thicknesses and \(\epsilon_{\mathrm{s}}\) and \(\epsilon_{\mathrm{ox}}\) are the dielectric constants of the semiconducting channel and the insulating oxide, respectively. To avoid short channel effects the channel length of an FET \((L_{\mathrm{CH}})\) has to be at least three times higher than the screening length, i.e. \(\mathrm{L_{CH} > 3\lambda_{SC}}\) [6]. The atomically thin semiconducting monolayers allows extreme scalability, as they offer \(t_{\mathrm{s}}\) lower than \(1\mathrm{nm}\) . Among various semiconducting 2D materials, \(\mathrm{MoS_2}\) has gained the most attention owing to its dominant n- type transport, stability, and ease of high- quality growth [7- 10]. Hence, we have utilized \(\mathrm{MoS_2}\) memtransistors for the implementation of BNN.  
+
+<|ref|>text<|/ref|><|det|>[[114, 544, 883, 590]]<|/det|>
+We have clarified that the circuits can also be implemented with standard flash memories in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 621, 884, 775]]<|/det|>
+2. The "tanh circuit" created by using T1 as a resistive load is a common drain amplifier, which is a standard transistor amplifier circuit with a well-known transfer function. I would recommend acknowledging this because it seems a bit like this is being presented as a novel circuit (which would seem odd to the microelectronic circuits community). I have also seen similar common drain circuit implemented with floating gate devices, giving similar VTC behavior, such as in Fig 5 of the referenced paper below.  
+
+<|ref|>text<|/ref|><|det|>[[114, 806, 884, 905]]<|/det|>
+We agree with the reviewer that this is a common- drain amplifier circuit. While the output characteristics of the common- drain amplifier circuit is well- known [11], to the best of our knowledge, it has not been used in the context of neuromorphic computing in order to implement the tanh/sigmoid activation function. As demonstrated in the manuscript, this simple and energy
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 135]]<|/det|>
+efficient circuit can be used to implement the tanh activation function which is otherwise implemented using look- up- tables (LUT) consisting of numerous transistors [3].  
+
+<|ref|>text<|/ref|><|det|>[[114, 166, 883, 213]]<|/det|>
+We have clarified that the output characteristics is well- known and we have added the reference suggested by the reviewer in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 245, 884, 344]]<|/det|>
+3. For the LTSpice simulations, can you elaborate on the tanh function implementation; is this just modeled with a single resistor and NMOS? If the VTC looks reasonable (i.e. Fig 4h) then why is this a major source of inaccuracy? If this circuit is so inaccurate, why would it be used rather than the referenced previously used NMOS/PMOS version in ref 39?  
+
+<|ref|>image<|/ref|><|det|>[[123, 376, 875, 528]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 540, 877, 595]]<|/det|>
+<center>Figure R1. a) Schematic of circuit for the modified tanh activation function using a \(n\) -type MoS2 memtransistor (T1) and a \(V\) -doped \(p\) -type WSe2 memtransistor (T2), where the input voltage ( \(V_{S}\) ) is applied to the gate terminal of T1 and T2. b) The transfer characteristics of the circuit (solid line) i.e., output voltage ( \(V_{O}\) ) versus \(V_{S}\) , closely models the tanh activation function (dotted line). c) Schematic of circuit for the modified sigmoid activation function and its d) transfer characteristics. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 627, 884, 888]]<|/det|>
+We thank the reviewer for raising this question. As the reviewer rightly points out, the tanh activation function is modelled using a resistor and an \(n\) - type FET, which results in an asymmetry in the transfer function. We agree that an implementation using the integration of an \(n\) - type FET and a \(p\) - type FET would result in better accuracy. Hence, we demonstrate a circuit for a modified tanh (m- tanh) activation function using a \(n\) - type MoS2 memtransistor (T1) and a V- doped \(p\) - type WSe2 memtransistor (T2) as shown in Fig. R1a. The transfer function of the circuit i.e., output voltage ( \(V_{O}\) ) versus input voltage ( \(V_{S}\) ) closely follows the tanh activation function as shown in Fig. R1b. The maximum of the m- tanh activation function is determined by the drain voltage ( \(V_{\mathrm{DD}}\) ). Here, \(V_{S}\) is applied to the gate of T1 and T2. T1 and T2 are programmed to ensure that the m- tanh function passes through the origin. Note that when \(V_{S} = - 2 \mathrm{~V}\) , T1 operates in the off- state and T2
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 242]]<|/det|>
+operates in the on- state, resulting in \(V_{\mathrm{O}} = - V_{\mathrm{DD}}\) , whereas for \(V_{\mathrm{S}} = 2 \mathrm{V}\) , T1 operates in the on- state and becomes more conductive than T2, which results in \(V_{\mathrm{O}} = V_{\mathrm{DD}}\) . While this output characteristics is well- known [3, 12], to the best of our knowledge it has not been used to implement activation functions, as the activation functions are typically implemented using look- up- tables [3]. Additionally, modified sigmoid activation function can be realized by applying 0 V to the drain terminal of T1, as shown in Fig. R1c and Fig. R1d.  
+
+<|ref|>text<|/ref|><|det|>[[114, 273, 883, 319]]<|/det|>
+We have revised the implementation of the activation function and added this discussion in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[114, 351, 884, 425]]<|/det|>
+4. A minor, optional note on the amplifier circuits, the more common convention is to have the high voltage on the top of the circuit such that current flows down. (It is opposite of this in Fig 4(g) and Supplemental Fig 7(a).  
+
+<|ref|>text<|/ref|><|det|>[[114, 455, 883, 502]]<|/det|>
+We thank the reviewer for this suggestion. We have changed the figures to follow this common convention.  
+
+<|ref|>text<|/ref|><|det|>[[113, 533, 884, 634]]<|/det|>
+5. How was device to device variation modeled in LTSpice? Is there an experimental link between physical nonidealities and the modeling factors should be presented? Also, do the authors know the programming error of the devices, i.e. the delta between the expected and actual conductance after programming?  
+
+<|ref|>text<|/ref|><|det|>[[113, 664, 884, 826]]<|/det|>
+We are happy to provide further clarification. The synapses consisting of 2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) are modelled using resistors, as shown in Fig. R2a. Here, \(\mathrm{T_{+}}\) is modelled as a resistor which draws from Gaussian conductance distribution given by \(G_{+} \sim N(\mu_{G_{+}}, \sigma_{G_{+}})\) and \(\mathrm{T_{- }}\) is modelled as a resistor set to a constant conductance of \(G_{- }\) . This results in the synapse having an effective conductance distribution given by \(G_{eff} \sim N(\mu_{G_{eff}}, \sigma_{G_{eff}})\) , where, \(\mu_{G_{eff}} = \mu_{G_{+}} - G_{- }\) and \(\sigma_{G_{eff}} = \sigma_{G_{+}}\) , enabling independent control of the mean and standard deviation of \(G_{eff}\) .
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[123, 128, 872, 480]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 486, 876, 585]]<|/det|>
+<center>Figure R2. a) Schematic of the GRNG-based synapse (left). The input to the synapse, \(V_{\mathrm{in}}\) is applied as \(+V_{\mathrm{in}}\) and \(-V_{\mathrm{in}}\) to the memtransistors, \(T_{+}\) and \(T_{-}\) with conductance \(G_{+}\) and \(G_{- }\) (modulated using \(V_{G_{+}}\) and \(V_{G_{- }}\) ), respectively. The effective conductance of this synapse is given by \(G_{eff} = G_{+} - G_{- }\) , allowing positive and negative conductance. On the right side is its implementation in LTSpice, where both memtransistors are implemented using resistors. b) Device-to-device variation across 40 \(\mathrm{MoS_2}\) memtransistors for different program and erase voltages. Relationship between the change in the drain current \((\Delta I_{\mathrm{DS}})\) as a result of c) erase and d) program operation and the starting current \((I_{\mathrm{DS,start}})\) . e) Histogram of \(\Delta I_{\mathrm{DS}} / I_{\mathrm{DS,start}}\) for the erase operation, following a Gaussian distribution. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 601, 884, 888]]<|/det|>
+While the cycle- to- cycle variation in programming is useful for the generation of Gaussian random numbers, the device- to- device variation is undesirable. Fig. R2b demonstrates the device- to- device variation across 40 \(\mathrm{MoS_2}\) memtransistors while performing program/erase operation. Here, the progressively higher programming voltage pulses \((V_{\mathrm{P}})\) followed by progressively higher erase voltage pulses \((V_{\mathrm{E}})\) are applied to the \(\mathrm{MoS_2}\) memtransistors and the drain- to- source current \((I_{\mathrm{DS}})\) is measured at a read voltage \((V_{\mathrm{R}})\) of 0 V. This is repeated for 2 cycles, to demonstrate cycle- to- cycle variation. While the cycle- to- cycle variation is beneficial for a BNN, device- to- device variation is detrimental to its operation. Fig. R2c and Fig. R2d demonstrates the relationship between the change in current \((\Delta I_{\mathrm{DS}})\) as a result of erase or program operation and the starting current \((I_{\mathrm{DS,start}})\) before each erase and program operation, respectively. They follow a power law relationship given by Eq. R2.
+
+<--- Page Split --->
+<|ref|>equation<|/ref|><|det|>[[114, 88, 281, 111]]<|/det|>
+\[\Delta I_{\mathrm{DS}} = p * I_{\mathrm{DS,start}}^{q}\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 116, 884, 192]]<|/det|>
+Here, \(p\) is the scaling factor and \(q\) is the exponential factor. If we account for this dependence, the device- to- device variation can be reduced. \(\Delta I_{\mathrm{DS}} / I_{\mathrm{DS,start}}^{q}\) follows a Gaussian distribution, as shown representatively in Fig. R2e for the erase operation.  
+
+<|ref|>text<|/ref|><|det|>[[113, 220, 885, 409]]<|/det|>
+In order to evaluate the effect of device- to- device variation, we implement up to \(10\%\) variation in the parameters \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) . In line with the above discussion, they are drawn from Gaussian distributions where the mean is given by expected \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) and standard deviation of up to \(10\%\) is considered. In the revised manuscript, we have also modelled the effect of device- to- device variation in the neurons. For this, we assume up to \(10\%\) variation in the threshold voltage for the \(n\) - type FET \((V_{\mathrm{TH},n})\) and the \(p\) - type FET \((V_{\mathrm{TH},p})\) . We have also implemented up to \(10\%\) variation in resistance of the sense resistor.  
+
+<|ref|>text<|/ref|><|det|>[[113, 440, 819, 461]]<|/det|>
+We have included this discussion in the supplementary information and main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 492, 885, 645]]<|/det|>
+6. What is the gate voltage (or gate voltage range) of the synapses during read? What operating region is the transistor in at this gate voltage? It sounds like these are modulated, but in LTSpice, if you are treating the two-device pair as a resistor, I am assuming the gate to source and drain to source voltages are such that both devices are operating in the linear part of the triode region, such that a resistor is a suitable element to represent the pair. However, this is unclear from the description.  
+
+<|ref|>text<|/ref|><|det|>[[113, 675, 885, 804]]<|/det|>
+We are happy to provide further clarification. A gate- to- source voltage \((V_{\mathrm{GS}})\) or read voltage \((V_{\mathrm{R}})\) of \(0 \mathrm{~V}\) is used during the read step. \(V_{\mathrm{R}}\) of \(0\) is used in order to avoid read- disturb caused by a positive or negative \(V_{\mathrm{GS}}\) . As the reviewer rightly pointed out, both the \(V_{\mathrm{GS}}\) and drain- to- source voltage \((V_{\mathrm{DS}})\) are used such that the transistors are operating in the linear part of the triode region. Hence, they are modelled using resistors.  
+
+<|ref|>text<|/ref|><|det|>[[114, 835, 560, 855]]<|/det|>
+We have added this clarification in the main manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 266]]<|/det|>
+7. The authors should model something larger than the IRIS dataset, which is quite far from a dataset and network that would be used in a real-world application. It has been observed that have seen that excellent accuracy on small datasets does not translate to high accuracy on larger datasets. There is little if any practical application for using a custom, efficient accelerator to process a dataset as small as IRIS. I would understand the very small dataset if this was an experimental demonstration, but given that it is a simulation only, the authors should include something larger.  
+
+<|ref|>text<|/ref|><|det|>[[113, 298, 885, 528]]<|/det|>
+We appreciate the suggestion by the reviewer. We have now implemented a BNN classifier to classify the PIMA Indian diabetes dataset. This dataset consists of nine parameters such as number of pregnancies, glucose levels, insulin levels, body mass index, age, etc. To classify this dataset, we use a fully connected \(8 \times 10 \times 2\) BNN i.e., it has an input layer with 8 neurons, one hidden layer with 10 neurons, and an output layer with 2 neurons. The dataset with 767 instances is divided into 720 for training and 47 for testing. Bayes by Backprop algorithm, with a Gaussian prior is used to train the synaptic weight distributions [13, 14]. The BNN is trained off- chip for 300 epochs as shown in Fig. R3a, to obtain train accuracy of \(75.41\%\) and test accuracy of \(80.85\%\) . Similar accuracy numbers have been reported in prior works [15- 17].  
+
+<|ref|>text<|/ref|><|det|>[[114, 533, 883, 579]]<|/det|>
+Following Eq. R3, the output of the BNN accelerator is sampled \(Z = 100\) times to obtain a predictive distribution, and its mean is used to make the classification.  
+
+<|ref|>equation<|/ref|><|det|>[[114, 584, 880, 640]]<|/det|>
+\[p(y^{*}|x^{*},D)\approx \frac{1}{Z}\sum_{z = 1}^{Z}p(y^{*}|x^{*},W^{z}) \quad (R3)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 646, 883, 720]]<|/det|>
+Here, \(x^{*}\) and \(y^{*}\) are the test input and output, respectively, D is the training data, and \(W^{z}\) represents the \(z^{\mathrm{th}}\) Monte Carlo weight sample. The softmax of predictive distribution can be used to calculate the uncertainty in classification or entropy given by Eq. R4. [18, 19].  
+
+<|ref|>equation<|/ref|><|det|>[[119, 722, 875, 784]]<|/det|>
+\[H(p(\hat{y}^{*}|x^{*},D)) = -\sum_{j = 1}^{J}p_{j}(\hat{y}^{*}|x^{*},D)*\log \left(p_{j}(\hat{y}^{*}|x^{*},D)\right) \quad (R4)\]  
+
+<|ref|>text<|/ref|><|det|>[[114, 789, 883, 860]]<|/det|>
+Here, \(\hat{y}^{*}\) is the softmax output and \(J\) is the number of output classes. The entropy can be decomposed into epistemic entropy i.e., uncertainty in model and aleatoric entropy i.e., uncertainty in data as shown in Eq. R5.  
+
+<|ref|>equation<|/ref|><|det|>[[115, 866, 877, 890]]<|/det|>
+\[H(p(\hat{y}^{*}|x^{*},D)) = \Pi \big(p(\hat{y}^{*}|x^{*},D)\big) + E_{W\sim q(W;\theta)}[H(p(\hat{y}^{*}|x^{*},W))] \quad (R5)\]
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[125, 93, 870, 389]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[126, 396, 878, 465]]<|/det|>
+<center>Figure R3. a) Training and testing curves for 300 epochs of the BNN constructed to classify the PIMA Indians diabetes dataset. b) Accuracy and predicative accuracy as function of model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. c) Total entropy, aleatoric entropy and epistemic entropy as a function of model variation. d) Accuracy and predictive accuracy as a function of input variation. e) Total entropy, aleatoric entropy and epistemic entropy as a function of input variation. </center>  
+
+<|ref|>text<|/ref|><|det|>[[114, 489, 883, 588]]<|/det|>
+Here, on the right- hand side, the first term represents epistemic entropy and the second term represents aleatoric entropy. Aleatoric entropy is the average entropy for fixed weights and hence the uncertainty arises from the data. Epistemic entropy can be obtained by subtracting aleatoric entropy from total entropy, following Eq. R5.  
+
+<|ref|>text<|/ref|><|det|>[[113, 593, 884, 889]]<|/det|>
+The BNN accelerator to classify PIMA Indian diabetes dataset is evaluated using LTSpice simulations. The synapses consisting of 2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) are modelled using resistors. Here, \(\mathrm{T_{+}}\) is modelled as a resistor which draws from Gaussian conductance distribution given by \(G_{+}\sim N(\mu_{G_{+}},\sigma_{G_{+}})\) and \(\mathrm{T_{- }}\) is modelled as a resistor set to a constant conductance of \(G_{- }\) . The sense transistor is modelled using a resistor. The tanh activation function is implemented using the combination of an \(n\) - type FET and a \(p\) - type FET. Using the BNN accelerator, we are able replicate the test accuracy of \(80.85\%\) . In order to evaluate the effect of device- to- device variation in memtransistors, we implement up to \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the tanh function: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) - type FET and the \(p\) - type FET, and the conductance of the sense resistor \(G_{\mathrm{S}}\) . These parameters are drawn from Gaussian distributions where the mean is given by
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 886, 476]]<|/det|>
+expected parameter value and standard deviation of up to \(10\%\) is considered. Fig. R3b shows the effect of device- to- device (model) variation on the test accuracy and predictive accuracy. The predictive accuracy demonstrates how well the predictions of the expected correct classes are made. Here, the BNN is simulated and averaged over 5 runs. While, we observe a decrease in the test accuracy, it is not seen to significantly impact the operation of the BNN and an accuracy of \(\approx 60\%\) is maintained for \(10\%\) variation. In a BNN, we can use entropy estimation and entropy decomposition to quantify uncertainty and to find its source, respectively. Fig. R3c shows the total entropy, aleatoric entropy, and epistemic entropy as a function of model variation, calculated using Eq. R4 and Eq. R5. Here, we would expect the aleatoric uncertainty to remain unchanged and total entropy to increase. However, their extraction is impacted by the increased model variation. Nevertheless, as expected the epistemic entropy increases as the model variation increases. As shown in Fig. R3d and Fig. R3e, an increase in the input variation results in the degradation of accuracy, along with an increase in the total and aleatoric entropy, while a constant epistemic entropy is maintained, as expected. Hence, in addition to the estimation of total entropy using a BNN, with uncertainty decomposition various sources of entropy can be identified.  
+
+<|ref|>text<|/ref|><|det|>[[115, 507, 848, 527]]<|/det|>
+We have changed the classification dataset and added this discussion in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[114, 559, 884, 631]]<|/det|>
+8. Although the title conveys an interesting prospect, it does not seem well connected to the content and I would suggest modifying it. The work aims to quantify uncertainty, but does not "avoid inference inaccuracy".  
+
+<|ref|>text<|/ref|><|det|>[[114, 663, 883, 709]]<|/det|>
+In line with the reviewer's suggestion, we have changed the title to "A Bayesian Neural Network to Quantify Inference Inaccuracy".  
+
+<|ref|>text<|/ref|><|det|>[[114, 741, 883, 787]]<|/det|>
+9. Some of the figure captions are far too small to read. The labels on Fig 5d are equivalent to about a 4pt font.  
+
+<|ref|>text<|/ref|><|det|>[[114, 821, 644, 840]]<|/det|>
+We have ensured that the figure labels are larger and easier to read.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 439, 108]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 114, 884, 319]]<|/det|>
+In this work, Sebastian and Das introduce a MoS2- based device that can provide synaptic and neuronal functions to implement hardware Bayesian neural networks. These uses are illustrated in a simulation of a crossbar circuit. The device in itself is similar to previous works by the same group, but it is used in a new manner. The idea of exploiting stochastic effect inherent to device physics to implement probabilistic AI, as is done here, is a very strong concept, which has started to develop in recent literature. The device introduced by the authors has some nice features. However, I have concerns about the manuscript and some questions about the meaning of the work.  
+
+<|ref|>text<|/ref|><|det|>[[114, 350, 883, 422]]<|/det|>
+We are happy to know that the reviewer appreciates the idea of exploiting stochasticity in devices to implement probabilistic computing demonstrated in the paper. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+<|ref|>text<|/ref|><|det|>[[113, 454, 884, 606]]<|/det|>
+1. On a general note, I actually do not understand the benefits of using 2-D materials in this project. The authors explain: "The choice of MoS2 as the element of memtransistor is motivated by recent demonstrations highlighting the technological viability of 2D materials [27-29] and their wide scale adoption in brain-inspired computing [30-34]." This is too vague, in my opinion. In fact, the adoption of 2-D materials in brain-inspired computing is not that wide-scale: Refs 30-34 all have S. Das as senior author, if I am not mistaken.  
+
+<|ref|>text<|/ref|><|det|>[[113, 637, 884, 815]]<|/det|>
+We are happy to add further references supporting our statement. Various memristors, transistors and other devices based on 2D materials have been explored for neuromorphic computing applications [20- 22]. 2D materials have also been explored for optoelectronic synapses enabled by their optically active monolayers [23, 24]. Also, note that the atomically thin 2D semiconductors allow geometric miniaturization of FETs without any loss of electrostatic integrity. The scalability of FETs is captured through \(\lambda_{\mathrm{SC}}\) , shown in Eq. R6, which represents the competition between gate and drain potential for control of the channel charge.  
+
+<|ref|>equation<|/ref|><|det|>[[113, 821, 877, 872]]<|/det|>
+\[\lambda_{\mathrm{SC}} = \sqrt{\frac{\epsilon_{\mathrm{s}}}{\epsilon_{\mathrm{ox}}}} t_{\mathrm{s}}t_{\mathrm{ox}} \quad (R6)\]
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 268]]<|/det|>
+Here, \(t_{\mathrm{s}}\) and \(t_{\mathrm{ox}}\) are the thicknesses and \(\epsilon_{\mathrm{s}}\) and \(\epsilon_{\mathrm{ox}}\) are the dielectric constants of the semiconducting channel and the insulating oxide, respectively. To avoid short channel effects, \(L_{\mathrm{CH}}\) has to be at least three times higher than the screening length, i.e. \(\mathrm{L}_{\mathrm{CH}} > 3\lambda_{\mathrm{SC}}\) [6]. The atomically thin semiconducting monolayers allows extreme scalability, as they offer \(t_{\mathrm{s}}\) lower than \(1 \mathrm{nm}\) . Among various semiconducting 2D materials, \(\mathrm{MoS}_2\) has gained the most attention owing to its dominant n- type transport, stability, and ease of high- quality growth [7- 10]. Hence, we have utilized \(\mathrm{MoS}_2\) memtransistors for the implementation of BNN.  
+
+<|ref|>text<|/ref|><|det|>[[115, 300, 752, 319]]<|/det|>
+We have added more references to support our statement in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 351, 884, 450]]<|/det|>
+2. The title of the article, "A Neural Network Accelerator to Avoid Inference Inaccuracy", does not seem related to the paper results. The authors do not present a neural network accelerator, and the paper never really talks about avoiding inference inaccuracy. This was a little surprising.  
+
+<|ref|>text<|/ref|><|det|>[[113, 482, 882, 528]]<|/det|>
+In line with the reviewer's suggestion, we have changed the title to "A Bayesian Neural Network to Quantify Inference Inaccuracy".  
+
+<|ref|>text<|/ref|><|det|>[[113, 560, 883, 632]]<|/det|>
+3. The authors should position this new work more clearly with regards to their paper "Gaussian synapses for probabilistic neural networks" (Ref 30), which has some similar keywords and ideas (but is different).  
+
+<|ref|>text<|/ref|><|det|>[[113, 664, 884, 898]]<|/det|>
+A major difference between our implementation of a probabilistic neural network (PNN) [25] and BNN is the nature of the Gaussian synapse. To implement the PNN, we design a Gaussian synapse can mimics the Gaussian function. Fig. R4a shows the schematic of the two transistor Gaussian synapse based on the integration of \(n\) - type \(\mathrm{MoS}_2\) and \(p\) - type BP back- gated FETs. Figure R4b also shows the equivalent circuit diagram for the Gaussian synapse, which simply consists of two variable resistors in series. The two variable resistors, i.e., \(R_{\mathrm{MoS}_2}\) and \(R_{\mathrm{BP}}\) correspond to the \(\mathrm{MoS}_2\) and BP FETs. Figure R4c shows the experimentally measured transfer characteristics i.e., \(I_{\mathrm{DS}}\) versus \(V_{\mathrm{GS}}\) of the Gaussian synapse for different \(V_{\mathrm{DS}}\) . Here, the total current flowing through the series combination of \(\mathrm{MoS}_2\) and BP FETs is measured. Clearly, the transfer characteristics
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[123, 92, 876, 416]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 424, 877, 525]]<|/det|>
+<center>Figure R4. a) Schematic of the GRNG-based synapse (left). The input to the synapse, \(V_{\mathrm{in}}\) is applied as \(+V_{\mathrm{in}}\) and \(-V_{\mathrm{in}}\) to the memtransistors, \(T_{+}\) and \(T_{- }\) with conductance \(G_{+}\) and \(G_{- }\) (modulated using \(V_{G_{+}}\) and \(V_{G_{- }}\) ), respectively. The effective conductance of this synapse is given by \(G_{eff} = G_{+} - G_{- }\) , allowing positive and negative conductance. On the right side is its implementation in LTSpice, where both memtransistors are implemented using resistors. b) Device-to-device variation across 40 MoS2 memtransistors for different program and erase voltages. Relationship between the change in the drain current \((\Delta I_{\mathrm{DS}})\) as a result of c) erase and d) program operation and the starting current \((I_{\mathrm{DS,start}})\) . e) Histogram of \(\Delta I_{\mathrm{DS}} / I_{\mathrm{DS,start}}\) for the erase operation, following a Gaussian distribution. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 541, 884, 666]]<|/det|>
+resemble a Gaussian function which can be modeled. The emergence of Gaussian transfer characteristics can be explained using the experimentally measured transfer characteristics of its constituents, i.e., the MoS2 FET and the BP FET, as shown in Figure R4d and Figure R4d, respectively. MoS2 FETs exhibit unipolar \(n\) - type characteristics, whereas, BP FETs are predominantly p- type with large work function contact metals such as Ni [7, 26- 28].  
+
+<|ref|>text<|/ref|><|det|>[[112, 670, 884, 909]]<|/det|>
+For the implementation of the BNN, we design synapses that can draw random numbers from a Gaussian distribution. Fig. R4e shows the design of our GRNG- based synapse with independent control over its mean \((\mu)\) and standard deviation \((\sigma)\) , using two MoS2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) . Here, the input to the synapse, \(V_{\mathrm{in}}\) is applied as \(+V_{\mathrm{in}}\) and \(- V_{\mathrm{in}}\) to \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) , respectively, as shown in Fig. R4e. The current at the output node, \(I_{\mathrm{out}}\) is then given by sum of currents through \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) i.e., \(I_{\mathrm{T_{+}}}\) and \(I_{\mathrm{T_{- }}}\) , according to the KCL. To control \(\mu_{G_{\mathrm{eff}}}\) and \(\sigma_{G_{\mathrm{eff}}}\) , \(\mathrm{T_{+}}\) is subjected to successive erase- program- read pulse cycles, while \(\mathrm{T_{- }}\) is programmed to a given state and subsequently only read, using the waveforms shown in Fig. R4f. This results in \(G_{+}\) being drawn from a Gaussian distribution, with \(\mu_{G_{+}} = 5 \mathrm{nS}\) and \(\sigma_{G_{+}} = 0.49 \mathrm{nS}\) i.e., \(G_{+} \sim N\) and \(G_{- }\) having a constant value of \(\approx\)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 298]]<|/det|>
+8.89 nS, as shown in Fig. R4f. \(G_{\mathrm{eff}}\) is expected to be drawn from a distribution with \(\sigma_{G_{\mathrm{eff}}} = \sigma_{G_{+}}\) and \(\mu_{G_{\mathrm{eff}}} = \mu_{G_{+}}\cdot G_{-}\) . This is confirmed by our measurements as shown in Fig. R4g, \(G_{\mathrm{eff}}\sim N(- 3.9,0.49)\) nS. As evident, while we have some similar keywords in both works, even the nature of synapse is very different between our PNN and BNN work. Additionally, PNNs and BNNs are both different kind of networks based on different algorithms for training such as expectation maximization and backpropagation, respectively with different network structure. In terms of the structure and operation, BNNs are closer to deep artificial neural networks, with the addition of synapses being represented by probability distributions.  
+
+<|ref|>text<|/ref|><|det|>[[113, 329, 883, 403]]<|/det|>
+4. The programming voltages are very high, similar to FLASH memory, and much higher than memristors. The authors should benchmark their device with alternative approaches (FLASH, non-2D memristors, phase change memory...).  
+
+<|ref|>text<|/ref|><|det|>[[113, 432, 884, 610]]<|/det|>
+We are happy to benchmark our devices with other alternative approaches. Table R1 shows the comparison in energy consumption between different memory technologies such as memristor, phase- change memory (PCM), NAND and our 2D memtransistors [29]. While the program/erase voltages and times for the transistor memory technologies are high, the energy consumption is similar as the programming current is much lower for the transistor technologies ( \(\approx 10^{- 11}\) A) compared to memristors and PCM. Also note that, while our demonstration uses 2D memtransistors, it can also be implemented using commercial NAND technologies.  
+
+<|ref|>table<|/ref|><|det|>[[113, 614, 794, 870]]<|/det|>
+
+<table><tr><td colspan="5">Table R1. Memory Technologies Energy Benchmark</td></tr><tr><td></td><td>Memristor</td><td>PCM</td><td>NAND</td><td>2D Memtransistor</td></tr><tr><td>Cell Elements</td><td>1T1R</td><td>1T1R</td><td>1T</td><td>1T</td></tr><tr><td>Read Time (ns)</td><td>&amp;lt;50</td><td>&amp;lt;60</td><td>&amp;lt;50</td><td>&amp;lt;50</td></tr><tr><td>Read Voltage (V)</td><td>&amp;lt;3</td><td>3</td><td>2</td><td>0</td></tr><tr><td>Program/Erase Time (ns)</td><td>&amp;lt;250</td><td>60</td><td>10⁶</td><td>10⁵</td></tr><tr><td>Program/Erase Voltage (V)</td><td>&amp;lt;3</td><td>3</td><td>15</td><td>13</td></tr><tr><td>Program/Erase Energy (fJ)</td><td>&amp;lt;50</td><td>6 × 10³</td><td>10</td><td>10</td></tr></table>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 606, 108]]<|/det|>
+We have included this table in the supplementary information.  
+
+<|ref|>text<|/ref|><|det|>[[113, 140, 884, 267]]<|/det|>
+5. I have concerns about the proposed crossbar architecture. I understand that for each presented input, the crossbar devices need to be reprogrammed multiple times to provide a distribution at the output (once per sample). Having to reprogram the crossbar numerous times to perform a single inference seems an enormous energy cost. Due to this concern, the paper should include an energy analysis with some benchmarks, in my opinion.  
+
+<|ref|>text<|/ref|><|det|>[[114, 297, 883, 344]]<|/det|>
+We thank the reviewer for their recommendation. We have now included an energy analysis. The total energy consumption of the BNN accelerator is given by Eq. R7.  
+
+<|ref|>equation<|/ref|><|det|>[[113, 373, 880, 500]]<|/det|>
+\[\begin{array}{l}{{E_{\mathrm{Total}}=E_{\mathrm{syn}}+E_{\mathrm{sense}}+E_{\mathrm{tanh}}+E_{\mathrm{P}}+E_{\mathrm{E}}}}\\ {{=\sum_{N}\bigl[\bigl(I_{\mathrm{T}_{+}}+I_{\mathrm{T}_{-}}\bigr)V_{\mathrm{in}}t\bigr]_{\mathrm{syn}}+\bigl[I_{\mathrm{S}}V_{\mathrm{S}}t\bigr]_{\mathrm{sense}}+\bigl[(I_{\mathrm{T}_{1}}+I_{\mathrm{T}_{2}})V_{\mathrm{DD}}t\bigr]_{\mathrm{tanh}}+E_{\mathrm{P}}+E_{\mathrm{E}}}}\\ {{E_{\mathrm{P}}+E_{\mathrm{E}}=\sum_{N-100}t(I_{\mathrm{P}}V_{\mathrm{P}}+I_{\mathrm{E}}V_{\mathrm{E}})+\sum_{100}\bigl[Q t(I_{\mathrm{P}}V_{\mathrm{P}}+I_{\mathrm{E}}V_{\mathrm{E}}) \bigr]_{\mathrm{P,E}}}}\end{array} \quad (R7)\]  
+
+<|ref|>text<|/ref|><|det|>[[112, 503, 884, 896]]<|/det|>
+Here, \(E_{\mathrm{Total}}\) represents the total energy consumption per test sample. \(E_{\mathrm{syn}}\) , \(E_{\mathrm{sense}}\) , and \(E_{\mathrm{tanh}}\) represents the total energy consumption by the synapses, sense resistors, and the tanh circuit, respectively, during the inference step. \(I_{\mathrm{S}}\) represents the current through the sense resistor and \(N\) is the total number of devices. \(E_{\mathrm{P}}\) and \(E_{\mathrm{E}}\) represents the energy consumption for programming and erasing operation on these components. \(I_{\mathrm{P}}\) and \(I_{\mathrm{E}}\) respectively, are the gate currents associated with programming and erase operations. Here, only half of the synaptic devices (100 devices) needs to be programmed and erased for \(Z = 100\) Monte Carlo samples. The rest of the synaptic devices, sense transistors, and tanh circuit components are programmed and erased once and subsequently only used for inference. For \(E_{\mathrm{P}}\) and \(E_{\mathrm{E}}\) evaluations, maximum \(I_{\mathrm{P}}\) , \(V_{\mathrm{P}}\) , \(I_{\mathrm{E}}\) , \(V_{\mathrm{E}}\) values expected are used. Using Eq. R7, \(E_{\mathrm{syn}}\) of \(18 \mathrm{nJ}\) , \(E_{\mathrm{sense}}\) of \(0.025 \mathrm{nJ}\) , and \(E_{\mathrm{tanh}}\) of \(0.012 \mathrm{nJ}\) are obtained from LTSpice simulations and \(E_{\mathrm{P}} + E_{\mathrm{E}}\) of \(0.34 \mathrm{nJ}\) is estimated for a total of \(Z = 100\) Monte Carlo samples, resulting in an \(E_{\mathrm{Total}}\) of \(18.37 \mathrm{nJ}\) . Note that while the program/erase voltages are high the energy consumption associated with them is low as the gate currents are on the order of pA and since the transistors are biased with a drain voltage of \(0 \mathrm{V}\) during programming/erasing, there is no drain current.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 555, 108]]<|/det|>
+We have included the energy analysis main manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 141, 882, 186]]<|/det|>
+## 6. These multiple programming operations also raise the issue of device endurance, which should be discussed.  
+
+<|ref|>text<|/ref|><|det|>[[114, 218, 884, 372]]<|/det|>
+The reviewer raises an important point. For the use of \(\mathrm{MoS_2}\) memtransistors for BNN, it's important to evaluate its endurance. Fig. R5 shows the endurance characteristics of a \(\mathrm{MoS_2}\) memtransistor for a total of 2000 cycles. Here, for each cycle the device is switched between a high current state and a low current state by using a program voltage \((V_{\mathrm{P}})\) of - 11 V and erase voltage \((V_{\mathrm{E}})\) of 11 V. The current values are read for \(V_{\mathrm{R}}\) of 0 V. Minimal degradation is observed after 2000 cycles and the distinction between the high and low states are maintained.  
+
+<|ref|>text<|/ref|><|det|>[[115, 404, 680, 422]]<|/det|>
+We have included the endurance characteristics in the main manuscript.  
+
+<|ref|>image<|/ref|><|det|>[[352, 474, 636, 646]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[127, 652, 617, 667]]<|/det|>
+<center>Figure R5. Endurance characteristics of an \(\mathrm{MoS_2}\) memtransistor for 2000 cycles. </center>  
+
+<|ref|>text<|/ref|><|det|>[[114, 700, 884, 798]]<|/det|>
+7. Also, if the synapses need to be reprogrammed each sample, we need additional memory arrays storing the mean value and standard deviations of each weight. This is an important cost, and this would also limit the energy efficiency of the authors' approach, as important data movement will be involved.  
+
+<|ref|>text<|/ref|><|det|>[[115, 830, 883, 876]]<|/det|>
+We are happy to provide further clarification. We propose to use a scheme where all devices are erased with a high \(V_{\mathrm{E}}\) of 13 V. Following that, each device must be programmed with the separate
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 882, 135]]<|/det|>
+\(V_{\mathrm{P}}\) values to set them to their expected states. Hence, we agree that we will need to store the \(V_{\mathrm{P}}\) values to obtain the required mean and standard deviation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 168, 580, 186]]<|/det|>
+We have included this clarification in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 220, 883, 264]]<|/det|>
+8. A significant concern is that Fig 5 shows no Bayesian result: we only see the final accuracy, but no distributions.  
+
+<|ref|>image<|/ref|><|det|>[[123, 300, 870, 599]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 610, 877, 678]]<|/det|>
+<center>Figure R6. a) Training and testing curves for 300 epochs of the BNN constructed to classify the PIMA Indians diabetes dataset. b) Accuracy and predicative accuracy as function of model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. c) Total entropy, aleatoric entropy and epistemic entropy as a function of model variation. d) Accuracy and predictive accuracy as a function of input variation. e) Total entropy, aleatoric entropy and epistemic entropy as a function of input variation. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 697, 884, 876]]<|/det|>
+We are happy to include more Bayesian results. We have now implemented a BNN classifier to classify the PIMA Indian diabetes dataset. This dataset consists of nine parameters such as number of pregnancies, glucose levels, insulin levels, body mass index, and age. To classify this dataset, we use a fully connected \(8 \times 10 \times 2\) BNN i.e., it has an input layer with 8 neurons, one hidden layer with 10 neurons, and an output layer with 2 neurons. The dataset with 767 instances is divided into 720 for training and 47 for testing. Bayes by Backprop algorithm, with a Gaussian prior is used to train the synaptic weight distributions [13, 14]. The BNN is trained off- chip for 300 epochs as
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 883, 135]]<|/det|>
+shown in Fig. R6a, to obtain train accuracy of \(75.41\%\) and test accuracy of \(80.85\%\) . Similar accuracy numbers have been reported in prior works [15- 17].  
+
+<|ref|>text<|/ref|><|det|>[[113, 141, 883, 188]]<|/det|>
+Following Eq. R8, the output of the BNN accelerator is sampled \(Z = 100\) times to obtain a predictive distribution, and its mean is used to make the classification.  
+
+<|ref|>equation<|/ref|><|det|>[[115, 192, 880, 247]]<|/det|>
+\[p(y^{*}|x^{*},D)\approx \frac{1}{Z}\sum_{z = 1}^{Z}p(y^{*}|x^{*},W^{t}) \quad [R8]\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 253, 883, 328]]<|/det|>
+Here, \(x^{*}\) and \(y^{*}\) are the test input and output, respectively, D is the training data, and \(W^{z}\) represents the \(z^{\mathrm{th}}\) Monte Carlo weight sample. The softmax of predictive distribution can be used to calculate the uncertainty in classification or entropy given by Eq. R9. [18, 19].  
+
+<|ref|>equation<|/ref|><|det|>[[113, 330, 880, 389]]<|/det|>
+\[H(p(\hat{y}^{*}|x^{*},D)) = -\sum_{j = 1}^{J}p_{j}(\hat{y}^{*}|x^{*},D)*\log \left(p_{j}(\hat{y}^{*}|x^{*},D)\right) \quad [R9]\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 395, 883, 469]]<|/det|>
+Here, \(\hat{y}^{*}\) is the softmax output and \(J\) is the number of output classes. The entropy can be decomposed into epistemic entropy i.e., uncertainty in model and aleatoric entropy i.e., uncertainty in data as shown in Eq. R10.  
+
+<|ref|>equation<|/ref|><|det|>[[113, 475, 880, 500]]<|/det|>
+\[H(p(\hat{y}^{*}|x^{*},D)) = \Pi \big(p(\hat{y}^{*}|x^{*},D)\big) + E_{W\sim q(W;\theta)}[H(p(\hat{y}^{*}|x^{*},W))] \quad [R10]\]  
+
+<|ref|>text<|/ref|><|det|>[[113, 504, 883, 604]]<|/det|>
+Here, on the right- hand side, the first term represents epistemic entropy and the second term represents aleatoric entropy. Aleatoric entropy is the average entropy for fixed weights and hence the uncertainty arises from the data. Epistemic entropy can be obtained by subtracting aleatoric entropy from total entropy, following Eq. R10.  
+
+<|ref|>text<|/ref|><|det|>[[112, 609, 884, 902]]<|/det|>
+The BNN accelerator to classify PIMA Indian diabetes dataset is evaluated using LTSpice simulations. The synapses consisting of 2 memtransistors, \(\mathrm{T_{+}}\) and \(\mathrm{T_{- }}\) are modelled using resistors. Here, \(\mathrm{T_{+}}\) is modelled as a resistor which draws from Gaussian conductance distribution given by \(G_{+}\sim N(\mu_{G_{+}},\sigma_{G_{+}})\) and \(\mathrm{T_{- }}\) is modelled as a resistor set to a constant conductance of \(G_{- }\) . The sense transistor is modelled using a sense resistor. The neuron is implemented using the combination of an \(n\) - type FET and a \(p\) - type FET. Using the BNN accelerator, we are able replicate the test accuracy of \(80.85\%\) . In order to evaluate the effect of device- to- device variation in memtransistors, we implement up to \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the neuron: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) - type FET and the \(p\) - type FET, and the conductance of the sense resistor \(G_{S}\) . These parameters are drawn from Gaussian distributions where the mean is given by expected parameter value and
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 87, 884, 476]]<|/det|>
+standard deviation of up to \(10\%\) is considered. Fig. R6b shows the effect of device- to- device (model) variation on the test accuracy and predictive accuracy. The predictive accuracy demonstrates how well the predictions of the expected correct classes are made. Here, the BNN is simulated and averaged over 5 runs. While, we observe a decrease in the test accuracy, it is not seen to significantly impact the operation of the BNN and an accuracy of \(\approx 60\%\) is maintained for \(10\%\) variation. In a BNN, we can use entropy estimation and entropy decomposition to quantify uncertainty and to find its source, respectively. Fig. R6c shows the total entropy, aleatoric entropy, and epistemic entropy as a function of model variation, calculated using Eq. R4 and Eq. R5. Here, we would expect the aleatoric uncertainty to remain unchanged and total entropy to increase. However, their extraction is impacted by the increased model variation. Nevertheless, as expected the epistemic entropy increases as the model variation increases. As shown in Fig. R6d and Fig. R6e, an increase in the input variation results in the degradation of accuracy, along with an increase in the total and aleatoric entropy, while a constant epistemic entropy is maintained, as expected. Hence, in addition to the estimation of total entropy using a BNN, with uncertainty decomposition various sources of entropy/uncertainty can be identified.  
+
+<|ref|>text<|/ref|><|det|>[[115, 507, 547, 526]]<|/det|>
+We have added this discussion in the main manuscript.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 560, 883, 606]]<|/det|>
+## 9. Also, why are Bayesian neural networks useful for this example? I think that conventional neural networks get excellent accuracy on this task.  
+
+<|ref|>text<|/ref|><|det|>[[115, 636, 884, 736]]<|/det|>
+We thank the reviewer for their suggestion. As discussed in detail for the previous question, we have now implemented BNN classification on the PIMA Indian diabetes dataset, where we have a test accuracy of \(80.85\%\) . This dataset has typically demonstrated similar accuracy numbers as evident from prior works [15- 17], and hence serves better to demonstrate uncertainty estimation.  
+
+<|ref|>text<|/ref|><|det|>[[115, 767, 883, 813]]<|/det|>
+We have now implemented BNN to classify the PIMA Indian dataset and revised the main manuscript.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 88, 883, 161]]<|/det|>
+10. The degradation in accuracy between software and LTSpice simulation is quite severe, even for a very simple task, due to the neuron behavior. This is an important limitation. Can this problem be fixed?  
+
+<|ref|>image<|/ref|><|det|>[[122, 195, 874, 344]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 356, 877, 412]]<|/det|>
+<center>Figure R7. a) Schematic of circuit for the modified tanh activation function using a \(n\) -type \(\mathrm{MoS}_2\) memtransistor (T1) and a \(V\) -doped \(p\) -type WSe2 memtransistor (T2), where the input voltage \((V_{S})\) is applied to the gate terminal of T1 and T2. b) The transfer characteristics of the circuit (solid line) i.e., output voltage \((V_{O})\) versus \(V_{S}\) , closely models the tanh activation function (dotted line). a) Schematic of circuit for the modified sigmoid activation function and its b) transfer characteristics. </center>  
+
+<|ref|>text<|/ref|><|det|>[[112, 444, 885, 889]]<|/det|>
+We thank the reviewer for raising this question. As the reviewer rightly points out, when the tanh activation function is modelled using a resistor and an \(n\) - type FET, there is an asymmetric transfer function, which leads to a degradation in the accuracy. To avoid this, we have now implemented the tanh activation function using an \(n\) - type FET and a \(p\) - type FET, which enables us to replicate the test accuracy of \(80.85\%\) in LTSpice simulations. For the tanh activation function, we demonstrate a circuit for a modified tanh (m- tanh) activation function using a \(n\) - type \(\mathrm{MoS}_2\) memtransistors (T1) and a V- doped \(p\) - type WSe2 memtransistor (T2) as shown in Fig. R7a. The transfer function of the circuit i.e., output voltage \((V_{O})\) versus input voltage \((V_{S})\) closely follows the tanh activation function as shown in Fig. R7b. The maximum of the m- tanh activation function is determined by the drain voltage \((V_{\mathrm{DD}})\) . Here, \(V_{S}\) is applied to the gate of T1 and T2. T1 and T2 are programmed to ensure that the m- tanh function passes through the origin. Note that when \(V_{S} = - 2 \mathrm{V}\) , T1 operates in the off- state and T2 operates in the on- state, resulting in \(V_{O} = - V_{\mathrm{DD}}\) , whereas for \(V_{S} = 2 \mathrm{V}\) , T1 operates in the on- state and becomes more conductive than T2, which results in \(V_{O} = V_{\mathrm{DD}}\) . While this output characteristics is well- known [3, 12], to the best of our knowledge it has not been used to implement activation functions, as the activation functions are typically implemented using look- up- tables [3]. Additionally, modified sigmoid activation function can be realized by applying \(0 \mathrm{V}\) to the drain terminal of T1, as shown in Fig. R7c and Fig. R7d.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 135]]<|/det|>
+We have revised the implementation of the activation function and added this discussion in the main manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[114, 166, 883, 213]]<|/det|>
+11. Fig 5g with \(0\%\) variation shows a test accuracy that is practically \(100\%\) . However, in the text, the accuracy is said to be \(93.78\%\) . This is a major concern.  
+
+<|ref|>text<|/ref|><|det|>[[114, 244, 884, 369]]<|/det|>
+We are happy to provide further clarification. Here, the degradation of accuracy was a result of the asymmetry in the tanh activation function implemented using and \(n\) - type FET and a resistor. By implementing a neuron with a symmetric tanh activation function, using the integration of an \(n\) - type FET and a \(p\) - type FET enabled us to replicate the test accuracy of \(80.85\%\) in LTSpice simulations.  
+
+<|ref|>text<|/ref|><|det|>[[114, 400, 884, 476]]<|/det|>
+12. Does Fig 5g include the effects of variability of the neuron devices? I understand that it does not, and that would be a problem, as this variability might have worse effects than the one of synapses.  
+
+<|ref|>text<|/ref|><|det|>[[114, 506, 884, 664]]<|/det|>
+The reviewer raises an excellent point. Hence, we have now included the effect of neuron variation as well. In order to evaluate the effect of device- to- device variation in memtransistors, we implement up to \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the neuron: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) - type FET and the \(p\) - type FET, and the conductance of the sense resistor \(G_{S}\) . These parameters are drawn from Gaussian distributions where the mean is given by expected parameter value and  
+
+<|ref|>image<|/ref|><|det|>[[223, 685, 787, 840]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[122, 844, 872, 900]]<|/det|>
+<center>Figure R8. a) Accuracy and predicative accuracy as function of synaptic variation. Here, the effect of variation in synaptic devices and sense resistors is demonstrated. b) Accuracy and predicative accuracy as function of total model variation. Here, the effect of variation in synaptic devices, sense resistors, and activation function is demonstrated. As evident, the inclusion of activation function does not significantly impact the accuracy. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 87, 884, 240]]<|/det|>
+standard deviation of up to \(10\%\) is considered. To evaluate the added impact of neuron variation, Fig. R8a shows the effect of device- to- device variation when only the weights and sense resistors are considered and Fig. R8b shows the effect of device- to- device variation when weights, sense resistors and neurons are considered. Here, to obtain each data point, the BNN is simulated and averaged over 5 runs. We can clearly see that the variation in neurons does not significantly impact the accuracy.  
+
+<|ref|>text<|/ref|><|det|>[[114, 272, 707, 291]]<|/det|>
+In error analysis, we have now included the effect neuron variation as well.  
+
+<|ref|>text<|/ref|><|det|>[[114, 324, 883, 369]]<|/det|>
+13. The caption of Figure 5 lacks details. The methods section should include the methods associated with Figure 5.  
+
+<|ref|>text<|/ref|><|det|>[[114, 401, 883, 447]]<|/det|>
+We have now included an extensive discussion in the Methods section to provide further details on Fig. 5.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[114, 90, 210, 107]]<|/det|>
+## References
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[111, 88, 888, 870]]<|/det|>
+[12] A. Sebastian, S. Das, and S. Das, "An Annealing Accelerator for Ising Spin Systems Based on In- Memory Complementary 2D FETs," \*Adv Mater\*, vol. 34, p. e2107076, Jan 2022. [13] J. Antoran, "Bayesian- Neural- Networks," 2019. [14] C. Blundell, J. Cornebise, K. Kavukcuoglu, and D. Wierstra, "Weight Uncertainty in Neural Networks," \*ArXiv\*, vol. abs/1505.05424, 2015. [15] Q. Zou, K. Qu, Y. Luo, D. Yin, Y. Ju, and H. Tang, "Predicting Diabetes Mellitus With Machine Learning Techniques," \*Front Genet\*, vol. 9, p. 515, 2018. [16] R. Vaishali, R. Sasikala, S. Ramasubbareddy, S. Remya, and S. Nalluri, "Genetic algorithm based feature selection and MOE Fuzzy classification algorithm on Pima Indians Diabetes dataset," pp. 1- 5, 2017. [17] R. Z. Islamic, "Diagnosis of Diabetes in Female Population of Pima Indian Heritage with Ensemble of BP Neural Network and SVM," 2012. [18] Y. Kwon, J.- H. Won, B. J. Kim, and M. C. Paik, "Uncertainty quantification using Bayesian neural networks in classification: Application to biomedical image segmentation," \*Computational Statistics & Data Analysis\*, vol. 142, p. 106816, 2020. [19] A. Kendall and Y. Gal, "What uncertainties do we need in Bayesian deep learning for computer vision?," presented at the Proceedings of the 31st International Conference on Neural Information Processing Systems, Long Beach, California, USA, 2017. [20] T. J. Ko, H. Li, S. A. Mofid, C. Yoo, E. Okogbue, S. S. Han, et al., "Two- Dimensional Near- Atom- Thickness Materials for Emerging Neuromorphic Devices and Applications," \*iScience\*, vol. 23, p. 101676, Nov 20 2020. [21] K. C. Kwon, J. H. Baek, K. Hong, S. Y. Kim, and H. W. Jang, "Memristive Devices Based on Two- Dimensional Transition Metal Chalcogenides for Neuromorphic Computing," \*Nanomicro Lett\*, vol. 14, p. 58, Feb 5 2022. [22] J. Bian, Z. Cao, and P. Zhou, "Neuromorphic computing: Devices, hardware, and system application facilitated by two- dimensional materials," \*Applied Physics Reviews\*, vol. 8, p. 041313, 2021. [23] L. Mennel, J. Symonowicz, S. Wachter, D. K. Polyushkin, A. J. Molina- Mendoza, and T. Mueller, "Ultrafast machine vision with 2D material neural network image sensors," \*Nature\*, vol. 579, pp. 62- 66, Mar 2020.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[112, 88, 886, 450]]<|/det|>
+[24] P. Wu, T. He, H. Zhu, Y. Wang, Q. Li, Z. Wang, et al., "Next- generation machine vision systems incorporating two- dimensional materials: Progress and perspectives," InfoMat, vol. 4, 2021. [25] A. Sebastian, A. Pannone, S. Subbulakshmi Radhakrishnan, and S. Das, "Gaussian synapses for probabilistic neural networks," Nat Commun, vol. 10, p. 4199, Sep 13 2019. [26] S. Das, M. Demarteau, and A. Roelofs, "Ambipolar phosphorene field effect transistor," ACS nano, vol. 8, pp. 11730- 11738, 2014. [27] S. Das, J. A. Robinson, M. Dubey, H. Terrones, and M. Terrones, "Beyond Graphene: Progress in Novel Two- Dimensional Materials and van der Waals Solids," Annual Review of Materials Research, Vol 45, vol. 45, pp. 1- 27, 2015. [28] D. S. Schulman, A. J. Arnold, and S. Das, "Contact engineering for 2D materials and devices," Chem Soc Rev, Mar 2 2018. [29] O. Kavehei, "Memristive devices and circuits for computing, memory, and neuromorphic applications," 2012.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 145, 404, 161]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 204, 868, 277]]<|/det|>
+I have reviewed the response and new manuscript, including the new tanh circuit. This version has addressed many of my previous concerns in their detailed response. However, I still would consider the following two issues with the manuscript, in determining if it is impactful enough to publish in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[117, 319, 880, 540]]<|/det|>
+The most significant issue with the paper is that there does not seem to be an advantage to using a MoS2 charge trapping device for this work. The same body of work could have been demonstrated using measured data from a SONOS, TANOS, or other modern, commercial device (perhaps a floating gate cell) – almost certainly with more consistent and stable electrical behavior due to maturity (including consistent random noise). I agree with the standard scaling argument for MoS2 transistors (as logic devices) that the authors presented in their response, but this scaling argument is not really relevant to how the device is being used in the paper. Hence, this MoS2 is not needed to achieve the functionality demonstrated in the paper. My general impression is that the paper is implying that this novel (but previously demonstrated) MoS2 device is enabling this new functionality – the paper must be clear that a new device is - not- required to achieve the presented functionality. Hence, it also follows that the novelty of the paper must be judged on the new circuit and application concepts, and whether these results constitute publication in Nature Comm.  
+
+<|ref|>text<|/ref|><|det|>[[117, 582, 878, 710]]<|/det|>
+In general, I did find the circuit and relevant mathematical analysis presented in Fig 5 and the relevant equations compelling and well thought out, albeit an unusual circuit. I would comment that one significant challenge of this configuration will be that it is very dependent on the programming accuracy of the resistance of the three devices at the end/bottom of each column, and programming error of these "neurons" will lead to significantly greater accuracy degradation than the same error would cause in the synapses. Put slightly differently, any noise or variation in those column end devices will essentially have equal footing to the sum of all the noise from synapses in the column.  
+
+<|ref|>text<|/ref|><|det|>[[119, 814, 404, 830]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 874, 810, 908]]<|/det|>
+The authors have very significantly improved the quality of the manuscript. I still have a few comments.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 113, 860, 186]]<|/det|>
+- The endurance of 2000 is in fact low, as each inference will consume many of these cycles. Also, the endurance experiment only validates that the ON and OFF states can be distinguished, not the fact the probabilistic synapse behavior is retained. This experiment is important, I think. The endurance issue should also be clearly highlighted as one of the major limitations of the approach.  
+
+<|ref|>text<|/ref|><|det|>[[118, 229, 866, 283]]<|/det|>
+- The title of the paper is still not adapted. All Bayesian neural networks quantify accuracy, and this particular aspect is not very developed in the manuscript. A more relevant title could be something like "Two-dimensional materials-based synapses for Bayesian neural network".  
+
+<|ref|>text<|/ref|><|det|>[[118, 326, 848, 361]]<|/det|>
+- The comparison with Ref 30 that the authors gave me in their answer should be included in the manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 439, 108]]<|/det|>
+## Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[114, 115, 884, 214]]<|/det|>
+I have reviewed the response and new manuscript, including the new tanh circuit. This version has addressed many of my previous concerns in their detailed response. However, I still would consider the following two issues with the manuscript, in determining if it is impactful enough to publish in Nature Communications.  
+
+<|ref|>text<|/ref|><|det|>[[115, 246, 883, 291]]<|/det|>
+We are happy to know that we have addressed your previous concerns. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+<|ref|>text<|/ref|><|det|>[[113, 323, 884, 658]]<|/det|>
+1. The most significant issue with the paper is that there does not seem to be an advantage to using a MoS2 charge trapping device for this work. The same body of work could have been demonstrated using measured data from a SONOS, TANOS, or other modern, commercial device (perhaps a floating gate cell) – almost certainly with more consistent and stable electrical behavior due to maturity (including consistent random noise). I agree with the standard scaling argument for MoS2 transistors (as logic devices) that the authors presented in their response, but this scaling argument is not really relevant to how the device is being used in the paper. Hence, this MoS2 is not needed to achieve the functionality demonstrated in the paper. My general impression is that the paper is implying that this novel (but previously demonstrated) MoS2 device is enabling this new functionality – the paper must be clear that a new device is -not- required to achieve the presented functionality. Hence, it also follows that the novelty of the paper must be judged on the new circuit and application concepts, and whether these results constitute publication in Nature Comm.  
+
+<|ref|>text<|/ref|><|det|>[[113, 688, 884, 892]]<|/det|>
+We are happy to provide further clarification. We included the following statements in the main manuscript to clarify that MoS2 memristors can be replaced with devices with charge- trap memory. "Here, the random distributions are observed as an effect of random nature of charge trapping which is typically observed in charge- trap memory devices [1, 2]. Hence, these memtransistors can also be replaced with standard three- terminal charge- trap flash memories such as TaN- Al2O3- Si3N4- SiO2- Si (TANOS) and Si- SiO2- Si3N4- SiO2- Si (SONOS) [3- 5]." Note that MoS2 is being used here as it has emerged as a potential alternative to Si in recent years. Hence, MoS2 and other 2D materials have been used as channel material for devices with floating- gate to demonstrate
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[114, 89, 883, 135]]<|/det|>
+non- volatile memory [6- 9]. Hence, we have utilized MoS₂ memtransistors for the implementation of BNN.  
+
+<|ref|>text<|/ref|><|det|>[[113, 165, 884, 370]]<|/det|>
+2. In general, I did find the circuit and relevant mathematical analysis presented in Fig 5 and the relevant equations compelling and well thought out, albeit an unusual circuit. I would comment that one significant challenge of this configuration will be that it is very dependent on the programming accuracy of the resistance of the three devices at the end/bottom of each column, and programming error of these "neurons" will lead to significantly greater accuracy degradation than the same error would cause in the synapses. Put slightly differently, any noise or variation in those column end devices will essentially have equal footing to the sum of all the noise from synapses in the column.  
+
+<|ref|>text<|/ref|><|det|>[[113, 405, 884, 560]]<|/det|>
+We are happy to know that the reviewer finds the circuit and mathematical analysis compelling and well thought out. We agree that evaluating the programming error introduced by the neuron must be analyzed. In order to evaluate the effect of device- to- device variation in memtransistors, we implement \(10\%\) variation for the parameters of the synapse: \(\mu_{G_{+}}\) , \(\sigma_{G_{+}}\) , and \(G_{- }\) , which results in a corresponding variation in \(G_{eff}\) , the parameters of the neuron: \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) , and for the \(n\) -  
+
+<|ref|>text<|/ref|><|det|>[[113, 567, 373, 771]]<|/det|>
+drawn from Gaussian distributions where the mean is given by expected parameter value and standard deviation of \(10\%\) is considered. We separately evaluate the effect of variation in synapses and neurons on the classification  
+
+<|ref|>image<|/ref|><|det|>[[400, 570, 880, 707]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[395, 713, 872, 771]]<|/det|>
+<center>Figure R1. a) Accuracy and predicative accuracy as function of variation in synapse. b) Accuracy and predicative accuracy as function of variation in neuron. The variation in neuron includes variation in the circuit for activation function and the sense resistor. </center>  
+
+<|ref|>text<|/ref|><|det|>[[113, 778, 883, 823]]<|/det|>
+accuracy. As evident from Fig. R1, variation in synapse is much more detrimental to the performance of the BNN compared to the variation in neuron.  
+
+<|ref|>text<|/ref|><|det|>[[113, 856, 883, 901]]<|/det|>
+In the revised manuscript, we have clarified that variation in synapses is more detrimental to the operation of the BNN.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 90, 439, 108]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 115, 883, 161]]<|/det|>
+The authors have very significantly improved the quality of the manuscript. I still have a few comments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 194, 883, 240]]<|/det|>
+We are happy to know that the reviewer finds that the manuscript has improved. A point- by- point response to the comments raised by the reviewer can be found below.  
+
+<|ref|>text<|/ref|><|det|>[[114, 271, 884, 399]]<|/det|>
+1. The endurance of 2000 is in fact low, as each inference will consume many of these cycles. Also, the endurance experiment only validates that the ON and OFF states can be distinguished, not the fact the probabilistic synapse behavior is retained. This experiment is important, I think. The endurance issue should also be clearly highlighted as one of the major limitations of the approach.  
+
+<|ref|>text<|/ref|><|det|>[[113, 428, 884, 686]]<|/det|>
+The reviewer raises an important point regarding low endurance and retention of synaptic behavior. For the use of \(\mathrm{MoS_2}\) memtransistors as a synapse for a BNN, it's important to ensure that the random number generation is maintained across multiple cycles. Hence using \(\mathrm{MoS_2}\) memtransistors, we demonstrate the endurance characteristics/random number generation for 20,000 cycles in Fig. R2a. The gate is subjected to successive erase- program- read pulses, and hence each cycle shown in Fig. R2a includes the effect of a program and erase operation. Here, multiple current levels consisting of high and low current states (achieved with programming voltages, \(V_{\mathrm{P}}\) of - 8.5 V and - 8 V, respectively) for different drain- to- source voltages ( \(V_{\mathrm{DS}}\) ) are demonstrated. Additionally, to demonstrate the retention of the random number generation, the moving mean \((\mu)\) and moving standard deviation \((\sigma)\) for these states are demonstrated in Fig. R2b  
+
+<|ref|>image<|/ref|><|det|>[[123, 699, 870, 848]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[125, 858, 875, 890]]<|/det|>
+<center>Figure R2. a) Endurance characteristics of an \(\mathrm{MoS_2}\) memtransistor demonstrating various conductance states for a total of 20,000 cycles. b) Moving mean and c) moving standard deviation for these different states. </center>
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 215]]<|/det|>
+and Fig. R2c. The moving \(\mu\) and \(\sigma\) are obtained across 100 samples at a time. Clearly the random number generation is stable, and the expected synaptic behaviors of increasing \(\mu\) and \(\sigma\) as a function of \(V_{\mathrm{DS}}\) and \(|V_{\mathrm{P}}|\) is retained after numerous measurements. However, we agree with the reviewer that the endurance must be improved for \(\mathrm{MoS}_2\) memtransistors and we have clarified this in the main text.  
+
+<|ref|>text<|/ref|><|det|>[[113, 241, 819, 261]]<|/det|>
+We have included this discussion in the main manuscript and supplementary information.  
+
+<|ref|>text<|/ref|><|det|>[[113, 293, 884, 391]]<|/det|>
+2. The title of the paper is still not adapted. All Bayesian neural networks quantify accuracy, and this particular aspect is not very developed in the manuscript. A more relevant title could be something like "Two-dimensional materials-based synapses for Bayesian neural network".  
+
+<|ref|>text<|/ref|><|det|>[[114, 422, 883, 469]]<|/det|>
+We thank the reviewer for their suggestion. We have now changed the title to "Two- dimensional materials- based synapses for Bayesian neural network".  
+
+<|ref|>text<|/ref|><|det|>[[113, 501, 883, 548]]<|/det|>
+3. The comparison with Ref 30 that the authors gave me in their answer should be included in the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[113, 580, 525, 599]]<|/det|>
+We have included this discussion in the manuscript.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[114, 90, 210, 108]]<|/det|>
+## References  
+
+<|ref|>text<|/ref|><|det|>[[110, 113, 888, 741]]<|/det|>
+[1] L. Danial, V. Gupta, E. Pikhay, Y. Roizin, and S. Kvatinsky, "Modeling a Floating- Gate Memristive Device for Computer Aided Design of Neuromorphic Computing," pp. 472- 477, 2020. [2] C. Monzio Compagnoni, R. Gusmeroli, A. S. Spinelli, and A. Visconti, "Analytical Model for the Electron- Injection Statistics During Programming of Nanoscale and Flash Memories," IEEE Transactions on Electron Devices, vol. 55, pp. 3192- 3199, 2008. [3] J. K. Han, J. Oh, G. J. Yun, D. Yoo, M. S. Kim, J. M. Yu, et al., "Cointegration of single- transistor neurons and synapses by nanoscale CMOS fabrication for highly scalable neuromorphic hardware," Sci Adv, vol. 7, Aug 2021. [4] S. Hwang, J. Yu, G. H. Lee, M. S. Song, J. Chang, K. K. Min, et al., "Capacitor- Based Synaptic Devices for Hardware Spiking Neural Networks," IEEE Electron Device Letters, vol. 43, pp. 549- 552, 2022. [5] T. P. Xiao, C. H. Bennett, B. Feinberg, S. Agarwal, and M. J. Marinella, "Analog architectures for neural network acceleration based on non- volatile memory," Applied Physics Reviews, vol. 7, p. 031301, 2020. [6] D. Li, M. Chen, Z. Sun, P. Yu, Z. Liu, P. M. Ajayan, et al., "Two- dimensional non- volatile programmable p- n junctions," Nat Nanotechnol, vol. 12, pp. 901- 906, Sep 2017. [7] E. Zhang, W. Wang, C. Zhang, Y. Jin, G. Zhu, Q. Sun, et al., "Tunable charge- trap memory based on few- layer MoS2," ACS Nano, vol. 9, pp. 612- 9, Jan 27 2015. [8] C. Liu, X. Yan, J. Wang, S. Ding, P. Zhou, and D. W. Zhang, "Eliminating Overerase Behavior by Designing Energy Band in High- Speed Charge- Trap Memory Based on WSe2," Small, vol. 13, May 2017. [9] Q. Feng, F. Yan, W. Luo, and K. Wang, "Charge trap memory based on few- layer black phosphorus," Nanoscale, vol. 8, pp. 2686- 92, Feb 7 2016.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 145, 404, 160]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 204, 878, 535]]<|/det|>
+I appreciate the authors work modeling the neural net error as a function of variation in the neuron and synapses and found that an important addition to the manuscript. However, in the model of neuron variability vs accuracy, I have found an inconsistency that I've been unable to reconcile. The finding of Fig R1b shows the sense conductance Gs (which should be specific to each column, as Gsj) can vary by \(1000\%\) with a negligible impact on accuracy - which is a very surprising finding and merits a detailed explanation. Mathematically, for this to be true, the Gs term in the denominator of the first summation term in eqn 4a must be negligible compared to the other term (sum) in the denominator. However, this contradicts the original purpose of Gsj correction - which was to correct the error associated with each column, stated as "Gsj(k) is used to make sure that each column of the crossbar array has the same \(\gamma (k)\) ". If Gsj, which represents the inaccuracy of each column, is negligible, then why is this compensation scheme even needed? The proposal of the compensation term implies that Gs is large enough to significantly affect the denominator of 4a. But the plot in R1b implies that Gs has a negligible affect on the accuracy. Given the strong effect of Vthn and Vthp on the output, a similar argument could be made for the very small Vt variation which was varied in fig R1b. Perhaps more details on how specifically the \(10\%\) variation was modeled would clarify this result. In any case, I strongly recommend checking and clarifying this before the article is published - specifically assert whether Gs term has a non- negligible effect on the accuracy and is needed for column compensation, or if Gs is negligible and does not need to be adjusted between columns.  
+
+<|ref|>text<|/ref|><|det|>[[118, 608, 404, 623]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 668, 514, 683]]<|/det|>
+The authors have addressed my last comments well.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 89, 365, 108]]<|/det|>
+## COMMENTS TO AUTHOR:  
+
+<|ref|>text<|/ref|><|det|>[[115, 123, 440, 143]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[112, 152, 886, 727]]<|/det|>
+I appreciate the authors work modeling the neural net error as a function of variation in the neuron and synapses and found that an important addition to the manuscript. However, in the model of neuron variability vs accuracy, I have found an inconsistency that I've been unable to reconcile. The finding of Fig R1b shows the sense conductance Gs (which should be specific to each column, as Gsj) can vary by \(1000\%\) with a negligible impact on accuracy - which is a very surprising finding and merits a detailed explanation. Mathematically, for this to be true, the Gs term in the denominator of the first summation term in eqn 4a must be negligible compared to the other term (sum) in the denominator. However, this contradicts the original purpose of Gsj correction - which was to correct the error associated with each column, stated as "Gsj(k) is used to make sure that each column of the crossbar array has the same \(\gamma (k)\) ." If Gsj, which represents the inaccuracy of each column, is negligible, then why is this compensation scheme even needed? The proposal of the compensation term implies that Gs is large enough to significantly affect the denominator of 4a. But the plot in R1b implies that Gs has a negligible affect on the accuracy. Given the strong effect of Vthn and Vthp on the output, a similar argument could be made for the very small Vt variation which was varied in fig R1b. Perhaps more details on how specifically the \(10\%\) variation was modeled would clarify this result. In any case, I strongly recommend checking and clarifying this before the article is published - specifically assert whether Gs term has a non- negligible effect on the accuracy and is needed for column compensation, or if Gs is negligible and does not need to be adjusted between columns.  
+
+<|ref|>text<|/ref|><|det|>[[114, 760, 884, 850]]<|/det|>
+We are happy to know that the reviewer appreciates our response. We are happy to provide further clarification on the relevance of the sense transistor. Note that the primary purpose of the sense transistor is to perform a current- to- voltage conversion. Here, the current through a column
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 883, 144]]<|/det|>
+(corresponding to the dot- product of input and weights) is converted to a proportional voltage given by Eq. R1.  
+
+<|ref|>equation<|/ref|><|det|>[[113, 158, 880, 214]]<|/det|>
+\[V_{\mathrm{S}j}^{(k)} = \frac{\sum_{i = 1}^{M}V_{i}^{(k)}G_{ij}^{(k)}}{G_{\mathrm{S}j}^{(k)} + \sum_{i = 1}^{M}G_{ij}^{(k)}} \quad (R1)\]  
+
+<|ref|>text<|/ref|><|det|>[[112, 228, 885, 641]]<|/det|>
+Here, \(V_{i}^{(k)}\) is the input voltage, \(G_{ij}^{(k)}\) is the conductance of synapse, and \(G_{\mathrm{S}j}^{(k)}\) is the conductance of the sense transistor for the \(i^{\mathrm{th}}\) row, \(j^{\mathrm{th}}\) column and \(k^{\mathrm{th}}\) layer. The current- to- voltage conversion allows us to seamlessly integrate the circuit for m- tanh activation function into the crossbar array, as the input to the m- tanh activation function is a voltage. The secondary purpose of the sense transistor is to ensure that the nonideality in Eq. R1 i.e., the denominator term is consistent between different columns. Note, that for our implementation, the difference in \(\sum_{i = 1}^{M}G_{ij}^{(k)}\) between different columns is not significant and hence the \(G_{\mathrm{S}j}^{(k)}\) correction in our case is small in magnitude. Additionally, note that each synapse is represented by two devices instead one and hence their combined conductance leads to a much higher magnitude of \(\sum_{i = 1}^{M}G_{ij}^{(k)}\) compared to \(G_{\mathrm{S}j}^{(k)}\) . In a different network implementation, where \(\sum_{i = 1}^{M}G_{ij}^{(k)}\) is significantly different between columns, the impact error in \(G_{\mathrm{S}j}^{(k)}\) on the classification accuracy will be higher.  
+
+<|ref|>text<|/ref|><|det|>[[112, 656, 885, 891]]<|/det|>
+We believe that the impact of variation the tanh activation (implemented through variation in \(V_{\mathrm{TH},n}\) and \(V_{\mathrm{TH},p}\) ) is low due to the binary nature of the classification problem that we have chosen. For a multiclass classification problem, the impact of the variation in activation function for is expected to be larger. In fact, in the initial version of the manuscript, we inspected a multiclass problem, where the impact of variation in the activation function was seen to be higher. To evaluate the effect of model variation, the parameters are drawn from Gaussian distributions where the mean is given by expected parameter value and standard deviation of up to \(10\%\) is considered.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[113, 88, 884, 179]]<|/det|>
+This is achieved using gaussian random number generation in Python, with the expected parameter value and percentage of standard deviation. We have verified our implementation to ensure that the variation in the parameters is modelled correctly.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[119, 144, 404, 160]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 204, 871, 277]]<|/det|>
+I thank the authors for clarifying the previous response. The authors made a clarification that seems to suggest a change in the manuscript text, but I did not see a change, and hence the statement in current version of the manuscript is not completely correct. I would recommend fixing this before publication.  
+
+<|ref|>text<|/ref|><|det|>[[118, 320, 876, 392]]<|/det|>
+Specifically, the authors have written in the rebuttal: "We believe that the impact of variation the tanh activation (implemented through variation in VVTH, nn and VVTH, pp) is low due to the binary nature of the classification problem that we have chosen. For a multiclass classification problem, the impact of the variation in activation function for is expected to be larger."  
+
+<|ref|>text<|/ref|><|det|>[[118, 435, 875, 527]]<|/det|>
+This would render the statement in the manuscript incorrect: "In model variation, it was determined that the variation is synapses is more determinantal to the performance of BNN compared to variations in the neuron." From the rebuttal statement, the authors appear to believe this is not always true, but depends on the nature of the classification problem. I would recommend the authors correct this before publication.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[115, 89, 365, 108]]<|/det|>
+## COMMENTS TO AUTHOR:  
+
+<|ref|>text<|/ref|><|det|>[[115, 124, 439, 143]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[113, 157, 884, 283]]<|/det|>
+I thank the authors for clarifying the previous response. The authors made a clarification that seems to suggest a change in the manuscript text, but I did not see a change, and hence the statement in the current version of the manuscript is not completely correct. I would recommend fixing this before publication.  
+
+<|ref|>text<|/ref|><|det|>[[113, 297, 884, 457]]<|/det|>
+Specifically, the authors have written in the rebuttal: "We believe that the impact of variation the tanh activation (implemented through variation in VTH, nn and VTH, pp) is low due to the binary nature of the classification problem that we have chosen. For a multiclass classification problem, the impact of the variation in activation function for is expected to be larger."  
+
+<|ref|>text<|/ref|><|det|>[[113, 471, 884, 632]]<|/det|>
+This would render the statement in the manuscript incorrect: "In model variation, it was determined that the variation is synapses is more determinantal to the performance of BNN compared to variations in the neuron." From the rebuttal statement, the authors appear to believe this is not always true, but depends on the nature of the classification problem. I would recommend the authors correct this before publication.  
+
+<|ref|>text<|/ref|><|det|>[[113, 672, 880, 693]]<|/det|>
+We thank the reviewer for this suggestion. We have made this clarification in the main manuscript.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__44e0278cb1f5606a8d5bc36783fcb7feb6bb37b64915576d51c9046779a60453/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__44e0278cb1f5606a8d5bc36783fcb7feb6bb37b64915576d51c9046779a60453/supplementary_0_Peer Review File__44e0278cb1f5606a8d5bc36783fcb7feb6bb37b64915576d51c9046779a60453.mmd

@@ -0,0 +1,717 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+Rbpms2 promotes female fate upstream of the nutrient sensing Gator2 complex component Mios  
+
+![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 by Wilson et al., sheds light on the mechanism by which RbmpS2 acts as a necessary factor for oocyte development and subsequent female sex differentiation in the zebrafish. The authors show that RbmpS2 likely represses male differentiation factors while simultaneously promoting translation of ribosomal biogenesis factors and nucleolar amplification, which are critical for proper oocyte progression. Ultimately, the authors put forth an interesting model where RbmpS2 acts as a fate switch upstream of a nutrient sensing pathway, ultimately necessary for "pushing" a bipotential gonocyte towards an oocyte fate. In my opinion, this is a strong manuscript and the authors' claims are well supported. The experiments performed here are a logical progression from the previous work published by this group. The results flesh out the previous story and integrate nutrient signaling into the pathway. The findings will be a valuable contribution to the field of sex determination and germ cell biology.  
+
+Overall, the manuscript is well written. However, it was difficult to follow the logic in the section " Mios is required for oocyte differentiation independent of double strand break repair". Perhaps the authors could clarify this section by directly stating what the hypothesis is that they are trying to test. The manuscript would benefit from a clear 'working model' section in the discussion to bring everything together. The authors could consider incorporating something like the legend in figure 7C into the actual text.  
+
+Specific Comments:  
+
+- Define DMs at the beginning of the manuscript.- Why is Pol II diminished in mios-/-? Could the authors please discuss this?- It is difficult to see p-Ubft localized to nucleoli in Fig.4 h,I (much less to a distinct compartment)- a higher magnification would be helpful.  
+
+Reviewer #2 (Remarks to the Author):  
+
+- What are the noteworthy results?  
+
+- The identification of mRNAs that are bound by RbpmS2 proteins in zebrafish. This is important because RbpmS2 is required for the gonad to maintain an ovary fate and some of these transcripts may mediate those fate choices.  
+
+- RbpmS2 is required for nucleoli to proliferate during meiosis in oocytes.  
+
+- RbpmS2 acts via mTorc1 signaling.  
+
+- RbpmS2 is epistatic to mios.  
+
+- Without RbpmS2, testis factors are not repressed.  
+
+- Proper ribosome function is essential for ovary fate in zebrafish.  
+
+- Will the work be of significance to the field and related fields?  
+
+- yes.  
+
+How does it compare to the established literature?  
+
+Microscopy is very well done. The use of immunofluorescence is beautiful. Nice electron microscopy. Good use of various mutants and inducing new ones when they needed to. Good development of transgenic animals for specific purposes.
+
+<--- Page Split --->
+
+If the work is not original, please provide relevant references. The work is original.  
+
+- Does the work support the conclusions and claims, or is additional evidence needed? Some of the stated conclusions are logical inferences but are not the only possible explanations for the results. These are mentioned in the detailed reviewer comments and can be fixed in a revision.  
+
+- Are there any flaws in the data analysis, interpretation and conclusions? Some conclusions fail to take into consideration alternative interpretations as noted in detailed comments.  
+
+- Do these prohibit publication or require revision? The authors can easily revise the document to fix the perceived problems.  
+
+- Is the methodology sound? Does the work meet the expected standards in your field? Yes. The methodology in general is sound. Two exceptions – first, the numbers of individuals checked for sex ratios in Figs. 5, and especially 6 and 7 are spectacularly low and no statistics are given, and second, no statistical analysis (unless I missed it) was performed to be sure that the identified mRNAs bound to RbpmS2 and not to the control are statistically robust. Nothing analogous to DESeq2 statistical output that one would see in an RNA-seq experiment although the methods are conceptually similar, seeing if the number of RNA molecules from a gene are statistically different in one situation vs. another.  
+
+- Is there enough detail provided in the methods for the work to be reproduced? Yes.  
+
+## Specific comments  
+
+Fig. 1A. The normal expression profiles of the two genes used for the transgenic analysis, rpbms2b and buc, need to be shown in Fig. 1B to compare to the other markers. Probably mios also because it plays such a pivotal role in the story.  
+
+Also, the expression pattern of rbpms2a and rbpms2b are essential with respect to cell type and time of expression, and their expression patterns in the scRNA-seq results.  
+
+P5. Define: DMs for non- zebrafish people. The introduction or start of the results should point out that zebrafish has two copies of this gene and the authors knocked out both.  
+
+P5. Was buc:mApple specifically expressed in oocytes according to microscopic observations?  
+
+P5. Tell the reader somewhere early at what specific stage in oogenesis rbpms2a and rbpms2b begin to be expressed. Especially relevant is when relative to the expression of the oocyte- specific gene buc, which provided the promotor for the transgene.  
+
+Likewise, readers need to know the expression pattern of rbpms2a and b in tests to understand the system. The human protein atlas shows that RBPMs2 is expressed in both spermatogonia and oocytes, but higher in oocytes. https://www.proteinatlas.org/ENSG00000166831- RBPMs2/single+cell+type. Maybe it is also expressed in zebrafish spermatocytes. Public data might be available for this so new experiments would need to be done if the lab doesn't have the data already.  
+
+P5 The transgenic/immunoprecipitation method is a good one to identify interacting RNAs and the Fig. 1A does a good job of visually explaining the strategy.  
+
+P5. "The remaining 52 rboRNAs that did not map to the 40 dpf ovary dataset are likely transcripts expressed in later stage oocytes present in the fully mature adult ovary."  
+
+Yes, probably so, and that idea can be tested just by showing that they are or aren't previously characterized as late oocyte proteins from prior literature or public RNA- seq data.  
+
+Fig. 1 "(d) Volcano plot of differential gene expression in 21 dpf wildtype and rbpms2 DMs from bulk RNA sequencing."  
+
+This legend doesn't say what organs were taken, ovaries, testes, or whether these are whole animals. P6 'Therefore, we conclude that 1) RbpmS2 likely functions to repress translation of rbtRNAs
+
+<--- Page Split --->
+
+expressed in early gonocytes in ovaries and 2) promote translation of oocyte factors,'  
+
+Is this a conclusion or a hypothesis? From the data presented, it seems like these are logical hypotheses stemming from those data, but we can't actually conclude these two points from the presented data because translation was not specifically tested and other mechanisms are possible. P6 'we analyzed GSEA enrichment plots'. Tell reader what these plots are, briefly.  
+
+P7 'Rbpm s2 mediates a binary fate- switch by repressing testis factors and promoting nucleolar formation and oocyte development via mTOR signaling'  
+
+The results do show that Rbpm s2 promotes nucleolar formation, but the link to repressing testis factors is not really shown.  
+
+Fig. 2. Nice immunestaining images. It would be good for the legend to tell reader what the red nuage is.  
+
+Fig. 2 'wild- type (rbpm s2aae30; rbpm s2bsa9329 HM; n=4) and rbpm s2 DM'  
+
+Would be easier to understand if instead of HM it said double heterozygous and instead of DM it said double homozygous mutants.  
+
+pUbfl: Shouldn't the figure legend use p- Ubt like the text instead of p- Ubf1?  
+
+P7 'we reanalyzed transmission electron microscopy (TEM) images of rbpm s2 DM oocytes'  
+
+Were these reanalyzed from a previous publication? If so, give citation here.  
+
+P7 'wild- type prophase I oocytes had greater than four nucleoli per cell at 35 dpf (Figure 2b, d) whereas most rbpm s2 DM oocytes had fewer than 4 nucleoli (Figure 2c, d).'  
+
+Fig. 2a shows a great variation in nucleolus number for cells in different stages of prophase I in wildtypes. Which phases were present at 35dpf and which stage in meiosis were the cells in that were counted?  
+
+P8 'In rbpm s2 DMs, while RNA pol I localized to the cytoplasm and nucleoli of mitotic germ cells, it was observed throughout oocyte nuclei and overlapped with RNA pol II nuclear localization (Figure 2f).'  
+
+Draw a conclusion here for that finding.  
+
+P8 'may do so by regulating translation of these factors to promote'  
+
+Conclusion is appropriate from the data, but I had to go back and reread the sentence to be sure of the antecedent to 'these'. It would help to say 'may do so by regulating translation of ribosome biogenesis factors to promote'.  
+
+P8. 'In wild-type ovaries, localization of phosphorylated p70-S6K (p-Ps6k), a kinase directly phosphorylated by the active form of mTOR1'  
+
+Do you think that readers should know that zebrafish has two copies of the relevant gene and that its official names are rps6kb1a and rps6kb1b?  
+
+Fig. 3 '(c) Table of mTOR1- related components with Rbpm s2 binding sites in their 3' and/or 5' UTRs.' Are these demonstrated Rbpm s2 binding sites? Or are they sites predicted from some kind of consensus sequence?  
+
+P9 'wildtype, p- Ps6k was not detected in DM oocytes (Figure 3b). This observation suggests that dysregulated mTOR1 signaling in the absence of Rbpm s2 contributes to impaired oogenesis.'  
+
+The data at this point show that Rbpm s2 contributes to impaired mTOR1 signaling. It would take other experiments to show that it is the mTOR1 signaling that contributes to impaired oogenesis. It could be that it is one of the other things that's disrupted, like ribosome biogenesis, that contributes to impaired oogenesis and that the impaired mTOR1 signaling indeed happens but it's not the main reason oogenesis fails.  
+
+P9 'Missing oocyte (Mios), contains 4 Rbpm s2 binding sites in its 3' UTR'  
+
+Again, are these proven binding sites? Or just predicted based on sequence?  
+
+P10 'Analysis of miosms20 and miossa22946 heterozygous and mutant progeny'  
+
+Unclear if 'heterozygous' here means miosms20/miossa22946 trans heterozygous or miosms20/mios+ and miossa22946/mios+.  
+
+P9 'Several in- frame deletions were recovered including miosms20, an 11bp deletion allele that leads to a frameshift'  
+
+Sentence says miosms20 is an in- frame deletion that is an 11bp deletion. But 11 is not divisible by 3 so how is this an in- frame deletion? Maybe it's a wording problem with the sentence.  
+
+P10. 'miosms20 and miossa22946 mutants and miosms20/sa22946 compound heterozygous develop
+
+<--- Page Split --->
+
+functional testes and differentiate as males, exclusively'  
+
+Previously in the text, heterozygotes for a mutated allele were also called mutants. So here, text should specify whether 'miosms20 and miossa22946 mutants' means heterozygotes for each or homozygotes for each allele. The 'compound heterozygotes' phrase is clear.  
+
+P10. \`investigation at earlier timepoints,\`  
+
+Earlier than what? From the writing, earlier than the stage indicated by 'Additionally, gonad morphological...' But the reader is not told what stage that is.  
+
+Fig. 4a Legend doesn't tell us that the asterisk mean stop for miossa/sa.  
+
+P10 \`number of Fibrillarin but these puncta were much smaller'  
+
+Change to: number of Fibrillarin puncta, but these were much smaller.  
+
+P10 \`between wild-type and mios- /- oocytes (Figure 4h- i). In wildtype,'  
+
+Be consistent with 'wildtype', If you prefer to use 'wildtype' as one word when it is a noun, then you should use 'wildtype' as an adjective, e.g., 'this fish is a wildtype and this is a wildtype fish.' Only if you use 'wild type' as a noun should you use 'wild- type' as an adjective; e.g., 'this fish is a wild type and this is a wild- type fish'.  
+
+Fig. 5 Legend. It's unclear what 'specified cells' means in 'p- Ps6k and DAPI localization in specified cells'. Does it mean specified as germ cells vs. somatic gonadal cells? Specified as oocytes vs. spermatocytes? Or something else?  
+
+Fig. 5 Legend: 'Scale bar for (a,b) are 50 \(\mu M\) . Should be either scale bars are or scale bar is.  
+
+P10 \`We generated two alleles, ms49 and ms64, of the mTORca transgene.'  
+
+Text should tell us what the sequence difference is between these two alleles.  
+
+P11. \`more total eggs than wildtype non- transgenic siblings'  
+
+This 'wildtype' usage as an adjective is correct if text consistently uses 'wildtype' as a noun to mean a genotype.  
+
+Fig. 5e. Numbers are quite small, six or seven or eight animals used to determine sex ratios. The conclusions are probably accurate, but numbers are quite small and no statistics are given to confirm that different genotypes are statistically different.  
+
+P11 \`In fertility assays,'  
+
+Tell here briefly how these were done. Single pair matings? Wildtype AB or TU males? How many males per female? If different male genotypes were used, as the Methods section says, maybe some of the variation could be in the males.  
+
+Fig. 6 legend. 'HH, HW, and MH'  
+
+I couldn't find a definition of these abbreviations.  
+
+Fig. 5 legend. 'Number of fish screened are indicated'  
+
+Number is indicated. Or Numbers are indicated.  
+
+P13. \`Germline expression of Rhebca did not disrupt sex determination in wild- type genotypes nor prevent oocyte loss in mios- /- (Figure'  
+
+What does the text mean by 'wild- type genotypes'? If these animals are transgenic and expressing a constitutive allele of mouse Rheb, clearly they are not wildtype.  
+
+Fig. 7a, b. Numbers are extraordinarily small. I'm not sure that finding 3 males means that no females will be found. After all, there are lots of human families with three sons, but that doesn't mean that their next child won't be a daughter. No statistics are given to give confidence that the conclusions about sex ratios are valid.  
+
+Fig. 7c. Label the horizontal axis. Should the vertical axis have an arrowhead at the top?  
+
+P13. \`vertebrate specific RNAbp,'  
+
+Spell out RNA binding protein here for clarity.  
+
+P13 \`Rbpms2 translationally represses testis- associated factors'  
+
+Remind reader here of the specific data that shows that translation of factors specific for testis development is repressed but translation of factors specific for ovary development are not repressed. This seems to be an important point of the paper and the data to make it need to be clearly stated and that a translation difference between the two gonad sexes is the main issue and not some other difference that results in different amounts of message or protein made.  
+
+The problem is that, if multitudes of ribosomes and high translation rates are required for oocytes to develop but spermatocytes can get by with fewer ribosomes and lower translation rates, that then
+
+<--- Page Split --->
+
+oogenesis will fail, and a failure of oogenesis by any of a variety of mechanisms, like blocking meiosis, also leads to testis development.  
+
+P13 'we demonstrate that the nutrient sensing arm of the mTORc1 pathway is uniquely required for oocyte progression and sustained oogenesis'  
+
+The result and conclusion are good, but could the effect be due to nutrient sensing difficulties in the soma - intestine, liver that makes vitellogenin, brain cells making gonadotrophs, etc. - rather than in the oocytes themselves? Is the effect cell autonomous to germ cells? Experiments can be designed to answer that question, but they are time consuming and difficult and publication shouldn't be held up for that, but the text should acknowledge that possibility.  
+
+P14. 'Additionally, RNA pol II, which we show is present in nuclei of wild-type zebrafish oocytes up to diplotene arrest, is also intact in rbpmS2 DM oocytes suggesting that RbpmS2 likely regulates sexual differentiation through translational control.'  
+
+The result shows that RbpmS2 doesn't regulate sex differentiation by controlling the location or amount of RNA pol II. But that result doesn't show that Rbms2 control is likely translational. It could also function on which genes are transcribed rather than just being in the nucleus at normal amounts, or on differential message stability, which also wouldn't necessarily change Rpol location and quantity. Then, the text can go on and rule out alternative explanations for the cited result.  
+
+P14. 'Specifically, several rbRNAs are associated with testis functions and their expression is limited to the early, undifferentiated cell types of the 40 dpf ovary.'  
+
+Do we know that those early, undifferentiated cell types at 40dpf are not already determined to be oocytes? I don't think the text told us when sex determination occurs in zebrafish. Is it before 40dpf? Also, the single cell transcriptomics for ovary should be contrasted to the single cell transcriptomics for testes of the same age to draw adequate conclusions.  
+
+P14 'This restricted expression is consistent with our hypothesis that RbpmS2 suppresses testis factors'  
+
+Yes, it is consistent, but other possibilities exist too. Factors other than RbpmS2 could be responsible for decreasing the number of testis associated RNAs after the undifferentiated cell type stage. And if there are no testis mRNAs, then they would not be there for RbpmS2 to bind whether or not RbpmS2 is responsible for them disappearing so only ovary associated mRNAs would be present for RbpmS2 to bind after ovary commitment. Some other ovary- promoting factor could be suppressing testis associated gene transcription or transcript stability early and so there are no testis associated mRNAs left for RbpmS2 to bind at later stages.  
+
+Do we know that RbpmS2 is expressed before the sex determination stage? P15 'Notably, in zebrafish it has been shown that demethylation and amplification of an rDNA locus at the end of chromosome four (femrDNA) strongly correlates with female sexdetermination and differentiation56.'  
+
+Could it be that the femrDNA and RbpmS2 act on sex determination in exactly the same way, by preventing the formation of the many ribosomes that are needed to make a mature oocyte? Can you rule out the model that 1. ribosomes are essential for making a mature oocyte; 2. without mature oocytes, some regulatory mechanism is disturbed that normally would prevent the ovary from transitioning to testis; 3. That there are several genes that are independently necessary, each in a different way, for making ribosomes functional, including both femrDNA and Rboms2.  
+
+P16. Did the zebrafish spo11 mutations block DSBs and meiosis? The text told us that homozygous mothers gave embryos that didn't do well but was it shown definitively that this was because DSBs didn't occur?  
+
+P16. 'inhibition of mTORC1 by conditional knockout'  
+
+Tell reader what was the condition was in the conditional ko. Probably it was a cell- type specific, presumably oocyte specific ko, but in a couple of words the text would avoid making the reader go to the original paper just to find out.  
+
+P16. 'which in oocytes is orchestrated in part by mTORc1 signaling'  
+
+Do we know that that mTORc1 signaling is due to action in oocytes or is it non- autonomous due to effects in other cell types or in organs other than the gonad, like liver or gonadotrophin secreting brain cells?  
+
+P16. 'using Crispr- Cas9 mutagenesis as in as detailed below 62'
+
+<--- Page Split --->
+
+Fix wording.  
+
+P21. 'Sequencing data was aligned'  
+
+Either sequencing data were aligned or sequencing datum was aligned.  
+
+P21 'using the Illumina website'  
+
+Text could give here the URL.  
+
+P21. 'Only RNAs that appeared in both the cross-linked and uncrosslinked mApple- Rbpm s2, but not in controls were considered Rbpm s2 target RNAs.'  
+
+Good to have four replicates. Was there a minimum read count for concluding that a gene's transcripts were bound? Or was a single read sufficient to mean that a gene's transcripts were pulled down? Was there an adjusted p- value used to identify differentially pulled down transcripts between experimental and control?  
+
+P22 '(mios wildtype'  
+
+It would be less ambiguous to say homozygous mios wildtype, because it could have meant phenotypically wildtype, which would include heterozygotes. Likewise for homozygous mutants.  
+
+Reviewer #3 (Remarks to the Author):  
+
+Previous results implicate the RNA- binding protein Rbpm s2 in ovary fate during zebrafish development. This work identifies RNA targets of Rbpm s2 and, using a variety of approaches, supports a model whereby Rbpm s2 promotes nucleolar amplification via TORC1, a step that supports oogenesis. In particular, the work examines the role of the GATOR2 component Mios in nucleolar development in oocytes, and its dependence on TOR to promote oogenesis independent of the TSC Rheb arm of TORC1. Nutrient availability had been implicated in oogenesis in other species, as well as in zebrafish sex determination. Impact of the work is high, for those interested in TOR signaling, germline development and the role of nutrients in oogenesis.  
+
+Several points listed below to clarify the results, strengthen the conclusions, and improve accessibility to a wider audience.  
+
+1. p.6-7: Refer to transcript abundance differences, rather than "stability". Elaborate on proposed feedback. 
+2. Throughout manuscript: Define all terms and abbreviations (e.g., DM, H, M, HH, HM, MW, MH) 
+3. Figure 2: define white arrows in legend. 
+4. Figure 3: add quantification to support conclusion of loss of Ps6k localization in rbpm s2 DM; loss in mios mutant looks more convincing, but also needs quantification. 
+5. Figure 4g: add quantification to support differences in size of puncta. Presumably mios oocyte analysis was on the early arrested oocytes? Please clarify. Temper statement that mios is "required" for nucleologenesis since fibrillar staining shows nucleoli. 
+6. Figure 5: change or clarify labeling of x axes in legend (H, M, etc., change to genotype; state meaning of + and - for transgene). 
+7. Figure 7: correct the arrow from Gator1 to TORC as it should be negative. 
+8. Title of final results section is misleading. State more clearly result concerning amino acid sensing (GATOR-dependent) versus TSC/Rheb arm. 
+9. Supplemental Figure 6: stated phenotypes difficult to see; add zoom box.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The manuscript by Wilson et al., sheds light on the mechanism by which Rbpmps2 acts as a necessary factor for oocyte development and subsequent female sex differentiation in the zebrafish. The authors show that Rbpmps2 likely represses male differentiation factors while simultaneously promoting translation of ribosomal biogenesis factors and nucleolar amplification, which are critical for proper oocyte progression. Ultimately, the authors put forth an interesting model where Rbpmps2 acts as a fate switch upstream of a nutrient sensing pathway, ultimately necessary for "pushing" a bipotential gonocyte towards an oocyte fate. In my opinion, this is a strong manuscript and the authors' claims are well supported. The experiments performed here are a logical progression from the previous work published by this group. The results flesh out the previous story and integrate nutrient signaling into the pathway. The findings will be a valuable contribution to the field of sex determination and germ cell biology.  
+
+Overall, the manuscript is well written. However, it was difficult to follow the logic in the section " Mios is required for oocyte differentiation independent of double strand break repair". Perhaps the authors could clarify this section by directly stating what the hypothesis is that they are trying to test.  
+
+We have clarified the introduction in this section to make it clearer that we sought to determine if suppression of DSB formation could restore oogenesis in our mios- fish as it does in Drosophila mio- . We find that this is not the case and therefore conclude that this mechanism is not conserved in our vertebrate system.  
+
+The manuscript would benefit from a clear 'working model' section in the discussion to bring everything together. The authors could consider incorporating something like the legend in figure 7C into the actual text.  
+
+We thank the reviewer for this suggestion. We have added more detail to the discussion to more clearly state our overarching findings and hypotheses.  
+
+Specific Comments:  
+
+- Define DMs at the beginning of the manuscript.  
+
+We thank all reviewers for bringing this to our attention have defined rbpmps2 DMs.  
+
+- Why is Pol II diminished in mios-/-? Could the authors please discuss this?  
+
+We noticed variation from sample to sample such that some oocytes had lower levels of RNA pol II in their nuclei as compared to other cells, but this was not different between genotypes. We have corrected the brightness/contrast of the image to accurately show the RNA pol II intensity.  
+
+- It is difficult to see p-Ubft localized to nucleoli in Fig.4 h,I (much less to a distinct compartment)- a higher magnification would be helpful.  
+
+We have modified the inset to show the individual p-Ubft and RNA pol II channels and have outlined the RNA pol II excluded region to make the lack of the granular compartment in mios mutants clearer.  
+
+Reviewer #2 (Remarks to the Author):  
+
+- What are the noteworthy results?  
+
+- The identification of mRNAs that are bound by Rbpmps2 proteins in zebrafish. This is important because Rbpmps2 is required for the gonad to maintain an ovary fate and some of these transcripts may mediate those fate choices.  
+
+- Rbpmps2 is required for nucleoli to proliferate during meiosis in oocytes.  
+
+- Rbpmps2 acts via mTOR1 signaling.  
+
+- Rbpmps2 is epistatic to mios.
+
+<--- Page Split --->
+
+- Without Rbpm s2, testis factors are not repressed.- Proper ribosome function is essential for ovary fate in zebrafish.  
+
+- Will the work be of significance to the field and related fields?- yes.  
+
+How does it compare to the established literature?  
+
+Microscopy is very well done. The use of immunofluorescence is beautiful. Nice electron microscopy. Good use of various mutants and inducing new ones when they needed to. Good development of transgenic animals for specific purposes.  
+
+If the work is not original, please provide relevant references. The work is original.  
+
+- Does the work support the conclusions and claims, or is additional evidence needed?- Some of the stated conclusions are logical inferences but are not the only possible explanations for the results. These are mentioned in the detailed reviewer comments and can be fixed in a revision.  
+
+- Are there any flaws in the data analysis, interpretation and conclusions?- Some conclusions fail to take into consideration alternative interpretations as noted in detailed comments.  
+
+- Do these prohibit publication or require revision?- The authors can easily revise the document to fix the perceived problems.  
+
+- Is the methodology sound?- Does the work meet the expected standards in your field?- Yes. The methodology in general is sound. Two exceptions - first, the numbers of individuals checked for sex ratios in Figs. 5, and especially 6 and 7 are spectacularly low and no statistics are given, and second, no statistical analysis (unless I missed it) was performed to be sure that the identified mRNAs bound to Rbpm s2 and not to the control are statistically robust. Nothing analogous to DESeq2 statistical output that one would see in an RNA-seq experiment although the methods are conceptually similar, seeing if the number of RNA molecules from a gene are statistically different in one situation vs. another.  
+
+Yes. The methodology in general is sound. Two exceptions - first, the numbers of individuals checked for sex ratios in Figs. 5, and especially 6 and 7 are spectacularly low and no statistics are given, and second, no statistical analysis (unless I missed it) was performed to be sure that the identified mRNAs bound to Rbpm s2 and not to the control are statistically robust. Nothing analogous to DESeq2 statistical output that one would see in an RNA-seq experiment although the methods are conceptually similar, seeing if the number of RNA molecules from a gene are statistically different in one situation vs. another.  
+
+We have addressed the sex ratio numbers for experiments in figures 5- 7 and have performed corresponding statistical analyses. We observed 3 female mios mutants carrying the ms44 Tg[ziwi:Rheb \(^{ca}\) ; cmlc2:mCherry] transgenic insertion from a single cross out of 3 independent crosses that were evaluated for this allele. As female mutants were not observed among progeny from 2 other sibling crosses carrying the same transgenic insertion, we conclude that the suppression by Rheb \(^{ca}\) was incompletely penetrant within the ms44 line. Furthermore, because suppression was not observed in 2 other independent alleles (representing different insertion events) we conclude that overexpression of Rheb \(^{ca}\) and therefore activation through this arm of the mTorc1 signaling pathway is not sufficient to restore oogenesis in mios \(^{c}\) fish.  
+
+We have added additional language to the RNAseq methods section to explain our rboRNA selection methodology. Specifically, we took an "all or none" approach - we only counted RNAs as bound if they were present in the mApple- Rbpm s2 crosslinked and uncrosslinked samples and absent in mApple crosslinked and/or uncrosslinked controls. Therefore, enrichment statistics were not necessary for these RNAs. It is true that additional targets may be found if we were to perform a less stringent enrichment analysis.  
+
+- Is there enough detail provided in the methods for the work to be reproduced?- Yes.  
+
+Specific comments  
+
+Fig. 1A. The normal expression profiles of the two genes used for the transgenic analysis, rbpm s2b and buc, need to be shown in Fig. 1B to compare to the other markers. Probably mios also because it plays such a pivotal role in the story.  
+
+Also, the expression pattern of rbpm s2a and rbpm s2b are essential with respect to cell type and time of expression, and their expression patterns in the scRNA- seq results.
+
+<--- Page Split --->
+
+We have revised the text to clarify the expression of RbpmS2 and Buckyball proteins. Previous work from our lab and others have demonstrated that both RNAs and proteins are present at the same time and in the early oocyte- specific structure, the Balbiani body. To facilitate comparisons of their RNA expression profiles, we have added the relevant RNA expression UMAPs to supplemental Figure 1, which demonstrates their highly similar expression profiles in the early ovary.  
+
+P5. Define: DMs for non- zebrafish people. The introduction or start of the results should point out that zebrafish has two copies of this gene and the authors knocked out both.  
+
+We thank the reviewer for noting this and have clarified our language to address the presence of the two genes and have defined "DM" in this context.  
+
+P5. Was buc:mApple specifically expressed in oocytes according to microscopic observations?  
+
+This promoter has been confirmed to express specifically in early- stage oocytes and drives mApple expression in oocytes specifically, as we have previously shown in Kaufman et al. PLoS Genetics 2014 (https://doi.org/10.1371/journal.pgen.1007489).  
+
+P5. Tell the reader somewhere early at what specific stage in oogenesis rbpmS2a and rbpmS2b begin to be expressed. Especially relevant is when relative to the expression of the oocyte- specific gene buc, which provided the promotor for the transgene.  
+
+We have clarified our language to make it clear that our previous work has shown that these proteins are both expressed at the same time and localize the Balbiani body of prophase I oocytes.  
+
+Likewise, readers need to know the expression pattern of rbpmS2a and b in testis to understand the system. The human protein atlas shows that RBPMs2 is expressed in both spermatogonia and oocytes, but higher in oocytes. https://www.proteinatlas.org/ENSG00000166831- RBPMs2/single+cell+type. Maybe it is also expressed in zebrafish spermatocytes. Public data might be available for this so new experiments would need to be done if the lab doesn't have the data already.  
+
+We have previously shown in Romano et al. Development 2020 (https://journals.biologists.com/dev/article/147/18/dev190942/225837/Loss- of- dmrt1- restores- zebrafish- female- fates- in) that RbpmS2 is only detected in ovaries and we have clarified the text to make this clear.  
+
+P5 The transgenic/immunoprecipitation method is a good one to identify interacting RNAs and the Fig. 1A does a good job of visually explaining the strategy.  
+
+We thank the reviewer for this positive feedback \(\circledcirc\)  
+
+P5. "The remaining 52 rboRNAs that did not map to the 40 dpf ovary dataset are likely transcripts expressed in later stage oocytes present in the fully mature adult ovary."  
+
+Yes, probably so, and that idea can be tested just by showing that they are or aren't previously characterized as late oocyte proteins from prior literature or public RNA- seq data.  
+
+Due to the cell size limitation of the FACS sorting performed on the zebrafish 40 dpf ovary, we could not analyze oocytes larger than \(\sim 70 \mu \mathrm{m}\) . We have gone through the originally reported 52 RNAs that were not found in the 40 dpf dataset and have determined by searching updated gene names that only 7 of the 52 do not map to the 40 dpf dataset. We have searched for these 7 genes in other public oocyte datasets (https://www.nature.com/articles/s41586- 022- 04918- 4) and have determined that all 7 are found in PGCs, oogonia, and oocytes of mouse and human fetal ovaries. Additionally, we find that 5 are present in an early zebrafish embryo dataset (https://www.science.org/doi/10.1126/science.aar3131), specifically at stages around when ZGA occurs, suggesting that these 4 RNAs are present in late oocyte stages. Only 2 could not be found in available databases or expression analyses. Figure 1 was modified based on these analyses. Specifically,
+
+<--- Page Split --->
+
+the panel of RNAs that were not enriched (Titled "None") was removed and the figure was rearranged to accommodate this change.  
+
+Fig. 1 "(d) Volcano plot of differential gene expression in 21 dpf wildtype and rbpms2 DMs from bulk RNA sequencing." This legend doesn't say what organs were taken, ovaries, testes, or whether these are whole animals.  
+
+We have revised the figure legend to state that the bulk RNA sequencing was performed on bipotential gonads of wildtype and rbpms2 DM 21 dpf fish.  
+
+P6 'Therefore, we conclude that 1) Rbpms2 likely functions to repress translation of rbtRNAs expressed in early gonocytes in ovaries and 2) promote translation of oocyte factors,' Is this a conclusion or a hypothesis? From the data presented, it seems like these are logical hypotheses stemming from those data, but we can't actually conclude these two points from the presented data because translation was not specifically tested and other mechanisms are possible.  
+
+We have revised the text to clarify our language and to include the possibility of alternative mechanisms.  
+
+P6 'we analyzed GSEA enrichment plots'. Tell reader what these plots are, briefly.  
+
+We have added an explanatory statement to make this clear.  
+
+P7 'Rbpm2 mediates a binary fate- switch by repressing testis factors and promoting nucleolar formation and oocyte development via mTOR signaling'  
+
+The results do show that Rbpm2 promotes nucleolar formation, but the link to repressing testis factors is not really shown.  
+
+We have expanded the discussion to include qualifying language to make clear that the proposed functions of Rbpm2 as a translational repressor of testis RNAs and promoter of RNAs related to nucleoli formation and ribosome biogenesis are hypothesized functions given our results in this work and in the published literature.  
+
+Fig. 2. Nice immunestaining images. It would be good for the legend to tell reader what the red nuage is.  
+
+We have adjusted the images of the Mios protein localization to the cytoplasm of oocytes and have added magnified images of early oocyte stages. Additionally, we have added fluorescence quantification of Mios in wildtype and rbpms2 DM oocytes to further support our claim that Mios protein is lacking in rbpms2 DMs. In reference to the red nuage of figure 2, we are unclear what exactly is being noted here and have assumed this comment was regarding the Mios staining in figure 3.  
+
+Fig. 2 'wild- type (rbpms2aae30; rbpms2bsa9329 HM; n=4) and rbpms2 DM' Would be easier to understand if instead of HM it said double heterozygous and instead of DM it said double homozygous mutants.  
+
+We have clarified the genotype information in each figure legend.  
+
+pUbf1: Shouldn't the figure legend use p- Ubf1 like the text instead of p- Ubf1?  
+
+We have corrected the discrepancy and have corrected the gene name to ubtf (Ubtf).  
+
+P7 'we reanalyzed transmission electron microscopy (TEM) images of rbpms2 DM oocytes' Were these reanalyzed from a previous publication? If so, give citation here.  
+
+We thank the reviewer for noting this and have updated the text with the corresponding reference.  
+
+P7 'wild- type prophase I oocytes had greater than four nucleoli per cell at 35 dpf (Figure 2b, d) whereas most rbpms2 DM oocytes had fewer than 4 nucleoli (Figure 2c, d).'
+
+<--- Page Split --->
+
+Fig. 2a shows a great variation in nucleolus number for cells in different stages of prophase I in wildtypes. Which phases were present at 35dpf and which stage in meiosis were the cells in that were counted?  
+
+rbpms2 DM oocytes do not appear to reach diplotene but do progress through the early stages of prophase I as judged by Balbiani body (Bb) formation, so we analyzed early Bb stage wildtype and rbpms2 DM cells that were in the stages of prophase I prior to arrest. We have highlighted a representative cell in the wildtype image to clarify this point.  
+
+P8 'In rbpms2 DMs, while RNA pol I localized to the cytoplasm and nucleoli of mitotic germ cells, it was observed throughout oocyte nuclei and overlapped with RNA pol II nuclear localization (Figure 2f).' Draw a conclusion here for that finding.  
+
+We have expanded upon this finding, noting that while RNA pol I localizes to 1- 2 nucleoli in rbpms2 DM oocytes, it also remains distributed throughout the oocyte nucleus (in contrast to wildtype) and suggests nucleolar seeding and consequently rRNA transcription is diminished.  
+
+P8 'may do so by regulating translation of these factors to promote' Conclusion is appropriate from the data, but I had to go back and reread the sentence to be sure of the antecedent to 'these'. It would help to say 'may do so by regulating translation of ribosome biogenesis factors to promote'.  
+
+We thank the reviewer for their suggestion and have revised the sentence accordingly.  
+
+P8. 'In wild-type ovaries, localization of phosphorylated p70-S6K (p-Ps6k), a kinase directly phosphorylated by the active form of mTorc1' Do you think that readers should know that zebrafish has two copies of the relevant gene and that its official names are rps6kb1a and rps6kb1b?  
+
+We have analyzed the amino sequences of the Rps6kb1a and Rps6kb1b zebrafish proteins against the human Rps6k1b and Rps6k2b proteins. Both zebrafish proteins are annotated as orthologs of the human Rps6k1b. Aligning the zebrafish proteins to both human proteins shows that the epitope target of the antibody is conserved in all cases, indicating that the antibody we used likely recognizes both zebrafish proteins. Further, single cell data indicates rps6kb1a is lowly expressed throughout the ovary (highest expression is in postmeiotic cells) and rps6kb1b is expressed throughout but is highest in GSCs/TA and oocyte cells. We have revised the text as recommended to clarify that there are two proteins in zebrafish.  
+
+Fig. 3 '(c) Table of mTorc1- related components with Rbpmss2 binding sites in their 3' and/or 5' UTRs.' Are these demonstrated Rbpmss2 binding sites? Or are they sites predicted from some kind of consensus sequence?  
+
+We have clarified that these bindings sites were predicted based on the published binding motif for the related family member Rbpmss, as the exact binding motif of Rbpmss2 has not been precisely mapped. We have revised the text to include the citation to the work that characterized the Rbpmss binding motif.  
+
+P9 'wildtype, p- Ps6k was not detected in DM oocytes (Figure 3b). This observation suggests that dysregulated mTorc1 signaling in the absence of Rbpmss2 contributes to impaired oogenesis.' The data at this point show that Rbpmss2 contributes to impaired mTorc1 signaling. It would take other experiments to show that it is the mTorc1 signaling that contributes to impaired oogenesis. It could be that it is one of the other things that's disrupted, like ribosome biogenesis, that contributes to impaired oogenesis and that the impaired mTorc1 signaling indeed happens but it's not the main reason oogenesis fails.  
+
+We have revised the text to clarify that loss of Rbpmss2 leads to dysregulation of mTorc1 signaling, which may contribute to or be a consequence of defects in ribosome biogenesis and oogenesis.  
+
+P9 'Missing oocyte (Mios), contains 4 Rbpmss2 binding sites in its 3' UTR' Again, are these proven binding sites? Or just predicted based on sequence?
+
+<--- Page Split --->
+
+These binding sites are predicted based on the validated Rbpmns binding sites as described above.  
+
+P10 'Analysis of miosms20 and miossa22946 heterozygous and mutant progeny' Unclear if 'heterozygous' here means miosms20/miossa22946 trans heterozygote or miosms20/mios+ and miossa22946/mios+.  
+
+We have revised the text to clarify that the fish noted here are single heterozygotes or mutants for the indicated alleles.  
+
+P9 'Several in- frame deletions were recovered including miosms20, an 11bp deletion allele that leads to a frameshift' Sentence says miosms20 is an in- frame deletion that is an 11bp deletion. But 11 is not divisible by 3 so how is this an in- frame deletion? Maybe it's a wording problem with the sentence.  
+
+Thank you for pointing this out. We did not word this well. We have revised the text to clarify that we found in- frame deletions as well as nonsense mutations including the ms20 allele.  
+
+P10. 'miosms20 and miossa22946 mutants and miosms20/sa22946 compound heterozygotes develop functional testes and differentiate as males, exclusively'  
+
+Previously in the text, heterozygotes for a mutated allele were also called mutants. So here, text should specify whether 'miosms20 and miossa22946 mutants' means heterozygotes for each or homozygotes for each allele. The 'compound heterozygotes' phrase is clear.  
+
+We have revised the text and legends to indicate what "heterozygote" and "mutant" mean in the given contexts.  
+
+P10. 'investigation at earlier timepoints,'  
+
+Earlier than what? From the writing, earlier than the stage indicated by 'Additionally, gonad morphological...'. But the reader is not told what stage that is.  
+
+We have revised the wording here to make clear that we also analyzed these fish prior to testis development.  
+
+Fig. 4a Legend doesn't tell us that the asterisk mean stop for miossa/sa. We have defined this in the revised figure legend.  
+
+P10 'number of Fibrillarin but these puncta were much smaller' Change to: number of Fibrillarin puncta, but these were much smaller.  
+
+We thank the reviewer for this suggestion and have made the correction.  
+
+P10 'between wild-type and mios-/- oocytes (Figure 4h- i). In wildtype,' Be consistent with 'wildtype', If you prefer to use 'wildtype' as one word when it is a noun, then you should use 'wildtype' as an adjective, e.g., 'this fish is a wildtype and this is a wildtype fish.' Only if you use 'wild type' as a noun should you use 'wild-type' as an adjective; e.g., 'this fish is a wild type and this is a wild-type fish'.  
+
+We have revised the text to use "wildtype" for nouns and adjectives.  
+
+Fig. 5 Legend. It's unclear what 'specified cells' means in 'p- Ps6k and DAPI localization in specified cells'. Does it mean specified as germ cells vs. somatic gonadal cells? Specified as oocytes vs. spermatocytes? Or something else?  
+
+We have revised the wording here to "indicated cells" rather than specified.  
+
+Fig. 5 Legend: 'Scale bar for (a,b) are 50 μM'. Should be either scale bars are or scale bar is.  
+
+We have made this correction.
+
+<--- Page Split --->
+
+P10 'We generated two alleles, ms49 and ms64, of the mTORca transgene.' Text should tell us what the sequence difference is between these two alleles.  
+
+We have revised the sentence to indicate that these alleles are the same transgene and that the two alleles only differ by independent insertion events in the genome.  
+
+P11. 'more total eggs than wildtype non- transgenic siblings' This 'wildtype' usage as an adjective is correct if text consistently uses 'wildtype' as a noun to mean a genotype.  
+
+We have made this correction.  
+
+Fig. 5e. Numbers are quite small, six or seven or eight animals used to determine sex ratios. The conclusions are probably accurate, but numbers are quite small and no statistics are given to confirm that different genotypes are statistically different.  
+
+We have updated all sex ratio graphs with corresponding statistics and increased numbers as indicated above.  
+
+P11 'In fertility assays,'  
+
+Tell here briefly how these were done. Single pair matings? Wildtype AB or TU males? How many males per female? If different male genotypes were used, as the Methods section says, maybe some of the variation could be in the males.  
+
+We have revised the methods section to indicate that these crosses were single paired matings, and to indicate that the wildtype fish were SATs. The females analyzed were not crossed to the same male every instance. Although we cannot exclude that the males did contribute to the results we see, but we did not see significant differences among females (of the same mutant and transgenic genotypes) from mating to mating, but differences were seen between the mios mutant \(\mathsf{Tg + }\) females compared to nonmutant \(\mathsf{Tg + }\) and \(\mathsf{Tg - }\) female siblings. Thus, we conclude that the effects seen are due to the transgene in the context of lack of Mios. Additionally, we did compare the \(\mathsf{Tg + }\) and \(\mathsf{Tg - }\) wildtype and heterozygous siblings to each other and found that these siblings, as expected, were not statistically different from each other, suggesting that the male- to- male variation didn't contribute significantly to the results seen.  
+
+We have also updated the text with a brief description of how the fertility assay was conducted.  
+
+Fig. 6 legend. 'HH, HW, and MH' I couldn't find a definition of these abbreviations.  
+
+We have included these definitions in the revised manuscript.  
+
+Fig. 5 legend. 'Number of fish screened are indicated' Number is indicated. Or Numbers are indicated.  
+
+We have made this correction.  
+
+P13. 'Germline expression of Rhebca did not disrupt sex determination in wild-type genotypes nor prevent oocyte loss in mios-/- (Figure'  
+
+What does the text mean by 'wild- type genotypes'? If these animals are transgenic and expressing a constitutive allele of mouse Rheb, clearly they are not wildtype.  
+
+We have revised the language here.  
+
+Fig. 7a, b. Numbers are extraordinarily small. I'm not sure that finding 3 males means that no females will be found. After all, there are lots of human families with three sons, but that doesn't mean that their next child won't be a daughter. No statistics are given to give confidence that the conclusions about sex ratios are valid.
+
+<--- Page Split --->
+
+We have increased the numbers of animals examined for the \(ms44\) and \(ms47\) alleles. We observed 3 female mios mutants carrying the \(ms44\) Tg[ziwi:Rheb \(^{ca}\) ; cmlc2:mCherry] transgenic insertion from a single cross out of 3 independent crosses that were evaluated for this allele. As female mutants were not observed among progeny from 2 other sibling crosses carrying the same transgenic insertion, we conclude that the suppression by Rheb \(^{ca}\) was incompletely penetrant within the \(ms44\) line. The data for these crosses are included as Supp. Figure 8 in the revised manuscript. Furthermore, because suppression was not observed in 2 other independent alleles (representing different insertion events) we conclude that overexpression of Rheb \(^{ca}\) and therefore activation through this arm of the mTorc1 signaling pathway is not sufficient to restore oogenesis in mios \(^{c}\) fish.  
+
+Fig. 7c. Label the horizontal axis. Should the vertical axis have an arrowhead at the top?  
+
+We have corrected this in the revised manuscript.  
+
+P13. 'vertebrate specific RNAbp,' Spell out RNA binding protein here for clarity.  
+
+We have revised this as recommended.  
+
+P13 'Rbpms2 translationally represses testis- associated factors' Remind reader here of the specific data that shows that translation of factors specific for testis development is repressed but translation of factors specific for ovary development are not repressed. This seems to be an important point of the paper and the data to make it need to be clearly stated and that a translation difference between the two gonad sexes is the main issue and not some other difference that results in different amounts of message or protein made. The problem is that, if multitudes of ribosomes and high translation rates are required for oocytes to develop but spermatocytes can get by with fewer ribosomes and lower translation rates, that then oogenesis will fail, and a failure of oogenesis by any of a variety of mechanisms, like blocking meiosis, also leads to testis development.  
+
+We have tried to emphasize these points in the appropriate sections of the revised text and hope it is more clear now.  
+
+P13 'we demonstrate that the nutrient sensing arm of the mTorc1 pathway is uniquely required for oocyte progression and sustained oogenesis 'The result and conclusion are good, but could the effect be due to nutrient sensing difficulties in the soma - intestine, liver that makes vitellogenin, brain cells making gonadotrophis, etc. - rather than in the oocytes themselves? Is the effect cell autonomous to germ cells? Experiments can be designed to answer that question, but they are time consuming and difficult and publication shouldn't be held up for that, but the text should acknowledge that possibility.  
+
+We thank the reviewer for pointing out this alternative. Because we are able to restore oogenesis by expressing mTOR \(^{ca}\) only in germ cells under the germline- specific promoter, zivi, this indicates that restoring activity in the germ cells, including oocytes, is sufficient. If required in the somatic cells this activity is likely regulated independent of Mios.  
+
+P14. 'Additionally, RNA pol II, which we show is present in nuclei of wild-type zebrafish oocytes up to diplotene arrest, is also intact in rbpms2 DM oocytes suggesting that RbpmS2 likely regulates sexual differentiation through translational control.'  
+
+The result shows that RbpmS2 doesn't regulate sex differentiation by controlling the location or amount of RNA pol II. But that result doesn't show that Rbms2 control is likely translational. It could also function on which genes are transcribed rather than just being in the nucleus at normal amounts, or on differential message stability, which also wouldn't necessarily change Rpol location and quantity. Then, the text can go on and rule out alternative explanations for the cited result.  
+
+Our conclusion that these oocytes are not failing due to transcriptional dysregulation is based on the RNAseq data showing that neither 1) RNA levels of rboRNAs nor 2) recruitment of the RNA pol II machinery to the
+
+<--- Page Split --->
+
+nucleus change between wildtype and rbpm s2 DM fish. Therefore, we conclude that Rbpm s2 likely regulates translation of rboRNAs since RNA abundance appears unperturbed. For example, because we did not see significant changes in mios abundance between wildtype and rbpm s2 DMs, but we did see that rbpm s2 DM oocytes have significantly less Mios protein, this suggests, and we hypothesize that Rbpm s2 promotes its translation.  
+
+P14. 'Specifically, several rboRNAs are associated with testis functions and their expression is limited to the early, undifferentiated cell types of the 40 dpf ovary.'  
+
+Do we know that those early, undifferentiated cell types at 40dpf are not already determined to be oocytes? I don't think the text told us when sex determination occurs in zebrafish. Is it before 40dpf? Also, the single cell transcriptomics for ovary should be contrasted to the single cell transcriptomics for testes of the same age to draw adequate conclusions.  
+
+Because these cells underwent single cell sequencing, we can say by the RNAs present that these cells are still bipotential and have the capacity to differentiate into oocytes or spermatocytes. Because Rbpm s2 protein is detected in the ovary, but not the testis we assume that rbtRNAs in the testis are available for translation in the absence of Rbpm s2. We have revised the text to clarify the significance of the observation that rbtRNAs are restricted to early, undifferentiated cells of the ovary. We have also revised the text to better explain sex determination and differentiation in zebrafish as well as the plasticity of the zebrafish ovary.  
+
+P14 'This restricted expression is consistent with our hypothesis that Rbpm s2 suppresses testis factors' Yes, it is consistent, but other possibilities exist too. Factors other than Rbpm s2 could be responsible for decreasing the number of testis associated RNAs after the undifferentiated cell type stage.  
+
+We thank the reviewer for pointing this out. We have revised the text to include the possibility that Rbpm s2 may directly or indirectly influence rboRNAs and aspects of oogenesis through yet to be determined binding partners.  
+
+And if there are no testis mRNAs, then they would not be there for Rbpm s2 to bind whether or not Rbpm s2 is responsible for them disappearing so only ovary associated mRNAs would be present for Rbpm s2 to bind after ovary commitment.  
+
+Because we identified the rbtRNAs in wildtype oocytes, this does indicate that in a wildtype context the testis RNAs are present and available for Rbpm s2 to bind and regulate, even after ovary commitment. Continuous RNA expression and maintenance in a repressed state is common in oocytes and other several cell types, and we think this state is likely key to plasticity of the ovary. Accordingly, maintaining but repressing these testis RNAs in early germ cells of the ovary, as we think Rbpm s2 does, would be important for sustained oocyte development and maintenance of the ovary.  
+
+Some other ovary- promoting factor could be suppressing testis associated gene transcription or transcript stability early and so there are no testis associated mRNAs left for Rbpm s2 to bind at later stages.  
+
+We agree that other ovary- promoting factors likely suppress transcription of testis associated genes in differentiated oocytes at later stages. However, because rbtRNAs were found bound to Rbpm s2 in adult ovary, and because their abundance did not change in early oocytes of rbpm s2 DMs, and because loss of rbpm s2 results in testis development, we think Rbpm s2 translationally represses these testis RNAs in early oocytes. Because rbpm s2 DM oocytes do not make it past prophase I, we cannot make any claims about the role of Rbpm s2 and its potential interacting partners in later oocyte stages. In the revised discussion, we have added the possibility that loss of Rbpm s2 may have direct and indirect effects on oogenesis.  
+
+Do we know that Rbpm s2 is expressed before the sex determination stage?  
+
+Our bulk sequence analysis indicates that rbpm s2 transcripts are present in 21 day and 28 day gonads, stages prior to sex determination. Single cell analysis indicates that rbpm s2 transcripts are detected in some mitotic cells but are most highly expressed in early meiotic cells and differentiating oocytes (we have added UMAP plots of rbpm s2a and rbpm s2b expression in the 40 dpf). However, Rbpm s2 protein is not detected in mitotic
+
+<--- Page Split --->
+
+germ cells, but is present in meiotic cells and differentiating oocytes, indicating that rbpms2 expression and activity increase as the cells differentiate.  
+
+P15 'Notably, in zebrafish it has been shown that demethylation and amplification of an rDNA locus at the end of chromosome four (femrDNA) strongly correlates with female sex determination and differentiation56.' Could it be that the femrDNA and RbpmS2 act on sex determination in exactly the same way, by preventing the formation of the many ribosomes that are needed to make a mature oocyte?  
+
+We agree that both femrDNA and RbpmS2 could both converge on regulation of ribosomes. Because rDNAs are required for ribosome nucleation and because femrDNA is unmethylated in ovary, femrDNA would promote increased nucleation of nucleoli in oocytes, a prerequisite for Poll recruitment. Because we see reduced Poll in rbpmS2 DMs, it is tempting to speculate that RbpmS2 might indirectly regulate femrDNA accessibility and are keen to test this possibility. Given the importance of rDNA for nucleation, we hypothesize that RbpmS2 would be positively influencing expression from the femrDNA locus to promote ribosome biogenesis through direct or indirect mechanisms or act downstream of femrDNA modification. However, we feel that testing these notions, although exciting, is beyond the scope of this work.  
+
+Can you rule out the model that 1. ribosomes are essential for making a mature oocyte;  
+
+We agree that ribosomes are essential for making mature oocytes.  
+
+2. without mature oocytes, some regulatory mechanism is disturbed that normally would prevent the ovary from transitioning to testis;  
+
+We agree that mature oocytes are important; however, in the absence of RbpmS2 oogenesis is perturbed at stages before mature oocytes are present in wild-type; thus, we think it plays a central role in differentiation.  
+
+3. That there are several genes that are independently necessary, each in a different way, for making ribosomes functional, including both femrDNA and Rboms2.  
+
+Without additional experiments, we cannot exclude that regulation of femrDNA is independent of RbpmS2. Further, although likely important, femrDNA modification has yet to be shown to be essential in functional assays. As mentioned above, our findings that ribosome biogenesis may be impaired by reduced RNA pol I recruitment, we hypothesize that promoting femrDNA locus accessibility (and thus nucleolar nucleation) could require RbpmS2 through direct or indirect mechanisms.  
+
+P16. Did the zebrafish spo11 mutations block DSBs and meiosis? The text told us that homozygous mothers gave embryos that didn't do well but was it shown definitively that this was because DSBs didn't occur?  
+
+The allele we utilized has been previously published and has shown to prevent DSBs in zebrafish testes, and to cause aneuploidy in oocytes. According, it is likely that DSBs are correspondingly blocked in zebrafish oocytes.  
+
+P16. 'inhibition of mTORC1 by conditional knockout'  
+
+Tell reader what was the condition was in the conditional ko. Probably it was a cell- type specific, presumably oocyte specific ko, but in a couple of words the text would avoid making the reader go to the original paper just to find out.  
+
+We have revised the text to clarify that this conditional knockout was in the mouse testis.  
+
+P16. 'which in oocytes is orchestrated in part by mTOR1 signaling'  
+
+Do we know that that mTOR1 signaling is due to action in oocytes or is it non- autonomous due to effects in other cell types or in organs other than the gonad, like liver or gonadotrophin secreting brain cells?
+
+<--- Page Split --->
+
+As mentioned above, because mTOR<sup>ca</sup> expression in germ cells alone (expressed using the germ cell specific ziwipromoter) was sufficient to restore oogenesis and support female sex differentiation, we hypothesize that this activity is essential in oocytes.  
+
+P16. 'using Crispr- Cas9 mutagenesis as in as detailed below 62' Fix wording.  
+
+We have made this correction.  
+
+P21. 'Sequencing data was aligned' Either sequencing data were aligned or sequencing datum was aligned.  
+
+We have corrected this in the revised manuscript.  
+
+P21 'using the Illumina website' Text could give here the URL.  
+
+The URL, https://basespace.illumina.com, has been added as recommended.  
+
+P21. 'Only RNAs that appeared in both the cross-linked and uncrosslinked mApple- RbpmS2, but not in controls were considered RbpmS2 target RNAs.'  
+
+Good to have four replicates. Was there a minimum read count for concluding that a gene's transcripts were bound? Or was a single read sufficient to mean that a gene's transcripts were pulled down? Was there an adjusted p- value used to identify differentially pulled down transcripts between experimental and control?  
+
+We have added additional language to the RNAseq methods section to explain our rboRNA selection methodology. Specifically, we took an "all or none" approach - we only counted RNAs as bound if they were present in the mApple- RbpmS2 crosslinked and uncrosslinked samples and absent in mApple crosslinked and/or uncrosslinked controls.  
+
+P22 '(mios wildtype'  
+
+It would be less ambiguous to say homozygous mios wildtype, because it could have meant phenotypically wildtype, which would include heterozygotes. Likewise for homozygous mutants.  
+
+We have stated all of the genotypes used because we evaluated effects of the transgene on heterozygous and mutant fish compared to their homozygous wildtype siblings.  
+
+Reviewer #3 (Remarks to the Author):  
+
+Previous results implicate the RNA- binding protein RbpmS2 in ovary fate during zebrafish development. This work identifies RNA targets of RbpmS2 and, using a variety of approaches, supports a model whereby RbpmS2 promotes nucleolar amplification via TORC1, a step that supports oogenesis. In particular, the work examines the role of the GATOR2 component Mios in nucleolar development in oocytes, and its dependence on TOR to promote oogenesis independent of the TSC Rheb arm of TORC1. Nutrient availability had been implicated in oogenesis in other species, as well as in zebrafish sex determination. Impact of the work is high, for those interested in TOR signaling, germline development and the role of nutrients in oogenesis.  
+
+Several points listed below to clarify the results, strengthen the conclusions, and improve accessibility to a wider audience.  
+
+1. p.6-7: Refer to transcript abundance differences, rather than "stability". Elaborate on proposed feedback.  
+
+We thank the reviewer for the suggestion and have revised the text accordingly.  
+
+2. Throughout manuscript: Define all terms and abbreviations (e.g., DM, H, M, HH, HM, MW, MH)
+
+<--- Page Split --->
+
+We thank the reviewer for noting this and have defined abbreviations in the revised manuscript.  
+
+3. Figure 2: define white arrows in legend.  
+
+We have defined this in the revised manuscript.  
+
+4. Figure 3: add quantification to support conclusion of loss of Ps6k localization in rbpm s2 DM; loss in mios mutant looks more convincing, but also needs quantification.  
+
+We thank the reviewer for this suggestion and have quantified the p- Ps6k and Mios in rbpm s2 DMs and added plots to Figure 3.  
+
+In the revised text, figures 3 and 5 were modified from an overview image of p- Ps6k in wildtype and rbpm s2 DMs or mios cells to magnified images of the nuclear localization in mitotic/early meiotic cells and oocytes to make the findings clearer. Overview images corresponding to those used for the magnified views are now in Supplemental Figure 2.  
+
+In figure 3, the overall brightness of the channels in the panels of the Mios staining was adjusted, and the adjacent panels that showed only Mios and DAPI were replaced with insets to make the phenotypes clearer to viewers. Further, quantification of Mios protein in wildtype and rbpm s2 DMs was added for consistency with the other quantifications performed (this data is in Figure 3 and Supp. Figure 6). In addition, the RNA names for Ps6k to rps6kb1a and rps6kb1b were corrected (they were originally written as rps6kb1a and rps6kb1b).  
+
+5. Figure 4g: add quantification to support differences in size of puncta. Presumably mios oocyte analysis was on the early arrested oocytes? Please clarify. Temper statement that mios is "required" for nucleologenesis since fibrillarin staining shows nucleoli.  
+
+We have added the quantification of the fibrillarin puncta to Figure 4 and have added quantification of the nuclear sizes of the cells used for the analysis to supplemental figure 6. Further, we have revised the language from "required for nucleologenesis" to "nuclear maturation" based on the data. In Figure 4, the overall brightness of the Fibrillarin staining was increased to make the phenotypes clearer to viewers.  
+
+In addition, we corrected the x- axes labels for plots Supp. Figure 4 g- h and the y- axis of Supp. Figure 6d due to scaling errors. These changes did not change the representation of the data.  
+
+6. Figure 5: change or clarify labeling of x axes in legend (H, M, etc., change to genotype; state meaning of + and - for transgene).  
+
+We have revised the use of H, M, etc. to the \(+ / +\) , \(+ / -\) , etc. convention. We have also revised the figure legend to clarify which fish are transgenic and non-transgenic.  
+
+7. Figure 7: correct the arrow from Gator1 to TORC as it should be negative.  
+
+Thank you. This has been corrected in the revised manuscript.  
+
+8. Title of final results section is misleading. State more clearly result concerning amino acid sensing (GATOR-dependent) versus TSC/Rheb arm.  
+
+We have edited this title and revised the concluding statement regarding the unique Gator2-Mios mediated activation of mTOR1 signaling in zebrafish oocytes.  
+
+9. Supplemental Figure 6: stated phenotypes difficult to see; add zoom box. We thank the reviewer for this feedback and have made the recommended changes.
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The authors have responded carefully to reviewers' concerns. The work will be a significant contribution to the field.  
+
+In the abstract, it is a bit of a stretch to state that the mammalian gonad is initially an ovary. It would be more accurate to state that is has a minor ovarian bias.  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have responded to all suggestions.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The revisions have satisfactorily addressed my previous comments.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__44e0278cb1f5606a8d5bc36783fcb7feb6bb37b64915576d51c9046779a60453/supplementary_0_Peer Review File__44e0278cb1f5606a8d5bc36783fcb7feb6bb37b64915576d51c9046779a60453_det.mmd

@@ -0,0 +1,1029 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 508, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 139]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 155, 808, 211]]<|/det|>
+Rbpms2 promotes female fate upstream of the nutrient sensing Gator2 complex component Mios  
+
+<|ref|>image<|/ref|><|det|>[[95, 732, 262, 780]]<|/det|>  
+
+<|ref|>text<|/ref|><|det|>[[271, 732, 879, 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, 88, 305, 103]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 118, 403, 133]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 147, 876, 312]]<|/det|>
+The manuscript by Wilson et al., sheds light on the mechanism by which RbmpS2 acts as a necessary factor for oocyte development and subsequent female sex differentiation in the zebrafish. The authors show that RbmpS2 likely represses male differentiation factors while simultaneously promoting translation of ribosomal biogenesis factors and nucleolar amplification, which are critical for proper oocyte progression. Ultimately, the authors put forth an interesting model where RbmpS2 acts as a fate switch upstream of a nutrient sensing pathway, ultimately necessary for "pushing" a bipotential gonocyte towards an oocyte fate. In my opinion, this is a strong manuscript and the authors' claims are well supported. The experiments performed here are a logical progression from the previous work published by this group. The results flesh out the previous story and integrate nutrient signaling into the pathway. The findings will be a valuable contribution to the field of sex determination and germ cell biology.  
+
+<|ref|>text<|/ref|><|det|>[[115, 327, 876, 416]]<|/det|>
+Overall, the manuscript is well written. However, it was difficult to follow the logic in the section " Mios is required for oocyte differentiation independent of double strand break repair". Perhaps the authors could clarify this section by directly stating what the hypothesis is that they are trying to test. The manuscript would benefit from a clear 'working model' section in the discussion to bring everything together. The authors could consider incorporating something like the legend in figure 7C into the actual text.  
+
+<|ref|>text<|/ref|><|det|>[[116, 432, 263, 447]]<|/det|>
+Specific Comments:  
+
+<|ref|>text<|/ref|><|det|>[[115, 461, 871, 521]]<|/det|>
+- Define DMs at the beginning of the manuscript.- Why is Pol II diminished in mios-/-? Could the authors please discuss this?- It is difficult to see p-Ubft localized to nucleoli in Fig.4 h,I (much less to a distinct compartment)- a higher magnification would be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[116, 566, 403, 581]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 625, 375, 639]]<|/det|>
+- What are the noteworthy results?  
+
+<|ref|>text<|/ref|><|det|>[[115, 640, 852, 670]]<|/det|>
+- The identification of mRNAs that are bound by RbpmS2 proteins in zebrafish. This is important because RbpmS2 is required for the gonad to maintain an ovary fate and some of these transcripts may mediate those fate choices.  
+
+<|ref|>text<|/ref|><|det|>[[115, 688, 659, 702]]<|/det|>
+- RbpmS2 is required for nucleoli to proliferate during meiosis in oocytes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 703, 386, 716]]<|/det|>
+- RbpmS2 acts via mTorc1 signaling.  
+
+<|ref|>text<|/ref|><|det|>[[115, 718, 338, 730]]<|/det|>
+- RbpmS2 is epistatic to mios.  
+
+<|ref|>text<|/ref|><|det|>[[115, 732, 496, 745]]<|/det|>
+- Without RbpmS2, testis factors are not repressed.  
+
+<|ref|>text<|/ref|><|det|>[[115, 747, 603, 760]]<|/det|>
+- Proper ribosome function is essential for ovary fate in zebrafish.  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 586, 789]]<|/det|>
+- Will the work be of significance to the field and related fields?  
+
+<|ref|>text<|/ref|><|det|>[[115, 791, 159, 804]]<|/det|>
+- yes.  
+
+<|ref|>text<|/ref|><|det|>[[115, 819, 491, 833]]<|/det|>
+How does it compare to the established literature?  
+
+<|ref|>text<|/ref|><|det|>[[115, 834, 860, 879]]<|/det|>
+Microscopy is very well done. The use of immunofluorescence is beautiful. Nice electron microscopy. Good use of various mutants and inducing new ones when they needed to. Good development of transgenic animals for specific purposes.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 575, 120]]<|/det|>
+If the work is not original, please provide relevant references. The work is original.  
+
+<|ref|>text<|/ref|><|det|>[[115, 134, 883, 180]]<|/det|>
+- Does the work support the conclusions and claims, or is additional evidence needed? Some of the stated conclusions are logical inferences but are not the only possible explanations for the results. These are mentioned in the detailed reviewer comments and can be fixed in a revision.  
+
+<|ref|>text<|/ref|><|det|>[[115, 194, 822, 238]]<|/det|>
+- Are there any flaws in the data analysis, interpretation and conclusions? Some conclusions fail to take into consideration alternative interpretations as noted in detailed comments.  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 666, 283]]<|/det|>
+- Do these prohibit publication or require revision? The authors can easily revise the document to fix the perceived problems.  
+
+<|ref|>text<|/ref|><|det|>[[115, 298, 881, 416]]<|/det|>
+- Is the methodology sound? Does the work meet the expected standards in your field? Yes. The methodology in general is sound. Two exceptions – first, the numbers of individuals checked for sex ratios in Figs. 5, and especially 6 and 7 are spectacularly low and no statistics are given, and second, no statistical analysis (unless I missed it) was performed to be sure that the identified mRNAs bound to RbpmS2 and not to the control are statistically robust. Nothing analogous to DESeq2 statistical output that one would see in an RNA-seq experiment although the methods are conceptually similar, seeing if the number of RNA molecules from a gene are statistically different in one situation vs. another.  
+
+<|ref|>text<|/ref|><|det|>[[115, 431, 714, 461]]<|/det|>
+- Is there enough detail provided in the methods for the work to be reproduced? Yes.  
+
+<|ref|>sub_title<|/ref|><|det|>[[115, 477, 256, 491]]<|/det|>
+## Specific comments  
+
+<|ref|>text<|/ref|><|det|>[[115, 492, 875, 536]]<|/det|>
+Fig. 1A. The normal expression profiles of the two genes used for the transgenic analysis, rpbms2b and buc, need to be shown in Fig. 1B to compare to the other markers. Probably mios also because it plays such a pivotal role in the story.  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 872, 566]]<|/det|>
+Also, the expression pattern of rbpms2a and rbpms2b are essential with respect to cell type and time of expression, and their expression patterns in the scRNA-seq results.  
+
+<|ref|>text<|/ref|><|det|>[[115, 566, 875, 596]]<|/det|>
+P5. Define: DMs for non- zebrafish people. The introduction or start of the results should point out that zebrafish has two copies of this gene and the authors knocked out both.  
+
+<|ref|>text<|/ref|><|det|>[[115, 596, 812, 611]]<|/det|>
+P5. Was buc:mApple specifically expressed in oocytes according to microscopic observations?  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 879, 656]]<|/det|>
+P5. Tell the reader somewhere early at what specific stage in oogenesis rbpms2a and rbpms2b begin to be expressed. Especially relevant is when relative to the expression of the oocyte- specific gene buc, which provided the promotor for the transgene.  
+
+<|ref|>text<|/ref|><|det|>[[115, 657, 880, 730]]<|/det|>
+Likewise, readers need to know the expression pattern of rbpms2a and b in tests to understand the system. The human protein atlas shows that RBPMs2 is expressed in both spermatogonia and oocytes, but higher in oocytes. https://www.proteinatlas.org/ENSG00000166831- RBPMs2/single+cell+type. Maybe it is also expressed in zebrafish spermatocytes. Public data might be available for this so new experiments would need to be done if the lab doesn't have the data already.  
+
+<|ref|>text<|/ref|><|det|>[[115, 745, 875, 775]]<|/det|>
+P5 The transgenic/immunoprecipitation method is a good one to identify interacting RNAs and the Fig. 1A does a good job of visually explaining the strategy.  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 844, 805]]<|/det|>
+P5. "The remaining 52 rboRNAs that did not map to the 40 dpf ovary dataset are likely transcripts expressed in later stage oocytes present in the fully mature adult ovary."  
+
+<|ref|>text<|/ref|><|det|>[[115, 805, 825, 835]]<|/det|>
+Yes, probably so, and that idea can be tested just by showing that they are or aren't previously characterized as late oocyte proteins from prior literature or public RNA- seq data.  
+
+<|ref|>text<|/ref|><|det|>[[115, 835, 866, 865]]<|/det|>
+Fig. 1 "(d) Volcano plot of differential gene expression in 21 dpf wildtype and rbpms2 DMs from bulk RNA sequencing."  
+
+<|ref|>text<|/ref|><|det|>[[115, 866, 872, 896]]<|/det|>
+This legend doesn't say what organs were taken, ovaries, testes, or whether these are whole animals. P6 'Therefore, we conclude that 1) RbpmS2 likely functions to repress translation of rbtRNAs
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 752, 105]]<|/det|>
+expressed in early gonocytes in ovaries and 2) promote translation of oocyte factors,'  
+
+<|ref|>text<|/ref|><|det|>[[115, 105, 850, 163]]<|/det|>
+Is this a conclusion or a hypothesis? From the data presented, it seems like these are logical hypotheses stemming from those data, but we can't actually conclude these two points from the presented data because translation was not specifically tested and other mechanisms are possible. P6 'we analyzed GSEA enrichment plots'. Tell reader what these plots are, briefly.  
+
+<|ref|>text<|/ref|><|det|>[[115, 164, 825, 193]]<|/det|>
+P7 'Rbpm s2 mediates a binary fate- switch by repressing testis factors and promoting nucleolar formation and oocyte development via mTOR signaling'  
+
+<|ref|>text<|/ref|><|det|>[[115, 194, 833, 222]]<|/det|>
+The results do show that Rbpm s2 promotes nucleolar formation, but the link to repressing testis factors is not really shown.  
+
+<|ref|>text<|/ref|><|det|>[[115, 223, 875, 252]]<|/det|>
+Fig. 2. Nice immunestaining images. It would be good for the legend to tell reader what the red nuage is.  
+
+<|ref|>text<|/ref|><|det|>[[115, 253, 688, 268]]<|/det|>
+Fig. 2 'wild- type (rbpm s2aae30; rbpm s2bsa9329 HM; n=4) and rbpm s2 DM'  
+
+<|ref|>text<|/ref|><|det|>[[115, 268, 866, 297]]<|/det|>
+Would be easier to understand if instead of HM it said double heterozygous and instead of DM it said double homozygous mutants.  
+
+<|ref|>text<|/ref|><|det|>[[115, 297, 692, 312]]<|/det|>
+pUbfl: Shouldn't the figure legend use p- Ubt like the text instead of p- Ubf1?  
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 785, 327]]<|/det|>
+P7 'we reanalyzed transmission electron microscopy (TEM) images of rbpm s2 DM oocytes'  
+
+<|ref|>text<|/ref|><|det|>[[115, 328, 682, 342]]<|/det|>
+Were these reanalyzed from a previous publication? If so, give citation here.  
+
+<|ref|>text<|/ref|><|det|>[[115, 343, 825, 373]]<|/det|>
+P7 'wild- type prophase I oocytes had greater than four nucleoli per cell at 35 dpf (Figure 2b, d) whereas most rbpm s2 DM oocytes had fewer than 4 nucleoli (Figure 2c, d).'  
+
+<|ref|>text<|/ref|><|det|>[[115, 374, 872, 416]]<|/det|>
+Fig. 2a shows a great variation in nucleolus number for cells in different stages of prophase I in wildtypes. Which phases were present at 35dpf and which stage in meiosis were the cells in that were counted?  
+
+<|ref|>text<|/ref|><|det|>[[115, 417, 852, 461]]<|/det|>
+P8 'In rbpm s2 DMs, while RNA pol I localized to the cytoplasm and nucleoli of mitotic germ cells, it was observed throughout oocyte nuclei and overlapped with RNA pol II nuclear localization (Figure 2f).'  
+
+<|ref|>text<|/ref|><|det|>[[115, 462, 410, 476]]<|/det|>
+Draw a conclusion here for that finding.  
+
+<|ref|>text<|/ref|><|det|>[[115, 477, 616, 492]]<|/det|>
+P8 'may do so by regulating translation of these factors to promote'  
+
+<|ref|>text<|/ref|><|det|>[[115, 492, 860, 536]]<|/det|>
+Conclusion is appropriate from the data, but I had to go back and reread the sentence to be sure of the antecedent to 'these'. It would help to say 'may do so by regulating translation of ribosome biogenesis factors to promote'.  
+
+<|ref|>text<|/ref|><|det|>[[115, 536, 800, 566]]<|/det|>
+P8. 'In wild-type ovaries, localization of phosphorylated p70-S6K (p-Ps6k), a kinase directly phosphorylated by the active form of mTOR1'  
+
+<|ref|>text<|/ref|><|det|>[[115, 566, 870, 596]]<|/det|>
+Do you think that readers should know that zebrafish has two copies of the relevant gene and that its official names are rps6kb1a and rps6kb1b?  
+
+<|ref|>text<|/ref|><|det|>[[115, 596, 866, 641]]<|/det|>
+Fig. 3 '(c) Table of mTOR1- related components with Rbpm s2 binding sites in their 3' and/or 5' UTRs.' Are these demonstrated Rbpm s2 binding sites? Or are they sites predicted from some kind of consensus sequence?  
+
+<|ref|>text<|/ref|><|det|>[[115, 641, 835, 670]]<|/det|>
+P9 'wildtype, p- Ps6k was not detected in DM oocytes (Figure 3b). This observation suggests that dysregulated mTOR1 signaling in the absence of Rbpm s2 contributes to impaired oogenesis.'  
+
+<|ref|>text<|/ref|><|det|>[[115, 670, 879, 744]]<|/det|>
+The data at this point show that Rbpm s2 contributes to impaired mTOR1 signaling. It would take other experiments to show that it is the mTOR1 signaling that contributes to impaired oogenesis. It could be that it is one of the other things that's disrupted, like ribosome biogenesis, that contributes to impaired oogenesis and that the impaired mTOR1 signaling indeed happens but it's not the main reason oogenesis fails.  
+
+<|ref|>text<|/ref|><|det|>[[115, 745, 654, 760]]<|/det|>
+P9 'Missing oocyte (Mios), contains 4 Rbpm s2 binding sites in its 3' UTR'  
+
+<|ref|>text<|/ref|><|det|>[[115, 760, 688, 775]]<|/det|>
+Again, are these proven binding sites? Or just predicted based on sequence?  
+
+<|ref|>text<|/ref|><|det|>[[115, 776, 714, 790]]<|/det|>
+P10 'Analysis of miosms20 and miossa22946 heterozygous and mutant progeny'  
+
+<|ref|>text<|/ref|><|det|>[[115, 791, 875, 820]]<|/det|>
+Unclear if 'heterozygous' here means miosms20/miossa22946 trans heterozygous or miosms20/mios+ and miossa22946/mios+.  
+
+<|ref|>text<|/ref|><|det|>[[115, 821, 870, 850]]<|/det|>
+P9 'Several in- frame deletions were recovered including miosms20, an 11bp deletion allele that leads to a frameshift'  
+
+<|ref|>text<|/ref|><|det|>[[115, 851, 872, 880]]<|/det|>
+Sentence says miosms20 is an in- frame deletion that is an 11bp deletion. But 11 is not divisible by 3 so how is this an in- frame deletion? Maybe it's a wording problem with the sentence.  
+
+<|ref|>text<|/ref|><|det|>[[115, 881, 872, 895]]<|/det|>
+P10. 'miosms20 and miossa22946 mutants and miosms20/sa22946 compound heterozygous develop
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 532, 105]]<|/det|>
+functional testes and differentiate as males, exclusively'  
+
+<|ref|>text<|/ref|><|det|>[[115, 106, 840, 150]]<|/det|>
+Previously in the text, heterozygotes for a mutated allele were also called mutants. So here, text should specify whether 'miosms20 and miossa22946 mutants' means heterozygotes for each or homozygotes for each allele. The 'compound heterozygotes' phrase is clear.  
+
+<|ref|>text<|/ref|><|det|>[[115, 151, 410, 164]]<|/det|>
+P10. \`investigation at earlier timepoints,\`  
+
+<|ref|>text<|/ref|><|det|>[[115, 165, 797, 195]]<|/det|>
+Earlier than what? From the writing, earlier than the stage indicated by 'Additionally, gonad morphological...' But the reader is not told what stage that is.  
+
+<|ref|>text<|/ref|><|det|>[[115, 195, 655, 209]]<|/det|>
+Fig. 4a Legend doesn't tell us that the asterisk mean stop for miossa/sa.  
+
+<|ref|>text<|/ref|><|det|>[[115, 210, 586, 223]]<|/det|>
+P10 \`number of Fibrillarin but these puncta were much smaller'  
+
+<|ref|>text<|/ref|><|det|>[[115, 224, 639, 238]]<|/det|>
+Change to: number of Fibrillarin puncta, but these were much smaller.  
+
+<|ref|>text<|/ref|><|det|>[[115, 239, 644, 253]]<|/det|>
+P10 \`between wild-type and mios- /- oocytes (Figure 4h- i). In wildtype,'  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 861, 312]]<|/det|>
+Be consistent with 'wildtype', If you prefer to use 'wildtype' as one word when it is a noun, then you should use 'wildtype' as an adjective, e.g., 'this fish is a wildtype and this is a wildtype fish.' Only if you use 'wild type' as a noun should you use 'wild- type' as an adjective; e.g., 'this fish is a wild type and this is a wild- type fish'.  
+
+<|ref|>text<|/ref|><|det|>[[115, 313, 852, 357]]<|/det|>
+Fig. 5 Legend. It's unclear what 'specified cells' means in 'p- Ps6k and DAPI localization in specified cells'. Does it mean specified as germ cells vs. somatic gonadal cells? Specified as oocytes vs. spermatocytes? Or something else?  
+
+<|ref|>text<|/ref|><|det|>[[115, 358, 805, 373]]<|/det|>
+Fig. 5 Legend: 'Scale bar for (a,b) are 50 \(\mu M\) . Should be either scale bars are or scale bar is.  
+
+<|ref|>text<|/ref|><|det|>[[115, 373, 680, 387]]<|/det|>
+P10 \`We generated two alleles, ms49 and ms64, of the mTORca transgene.'  
+
+<|ref|>text<|/ref|><|det|>[[115, 388, 692, 401]]<|/det|>
+Text should tell us what the sequence difference is between these two alleles.  
+
+<|ref|>text<|/ref|><|det|>[[115, 403, 562, 416]]<|/det|>
+P11. \`more total eggs than wildtype non- transgenic siblings'  
+
+<|ref|>text<|/ref|><|det|>[[115, 417, 872, 446]]<|/det|>
+This 'wildtype' usage as an adjective is correct if text consistently uses 'wildtype' as a noun to mean a genotype.  
+
+<|ref|>text<|/ref|><|det|>[[115, 447, 870, 491]]<|/det|>
+Fig. 5e. Numbers are quite small, six or seven or eight animals used to determine sex ratios. The conclusions are probably accurate, but numbers are quite small and no statistics are given to confirm that different genotypes are statistically different.  
+
+<|ref|>text<|/ref|><|det|>[[115, 492, 290, 505]]<|/det|>
+P11 \`In fertility assays,'  
+
+<|ref|>text<|/ref|><|det|>[[115, 507, 866, 551]]<|/det|>
+Tell here briefly how these were done. Single pair matings? Wildtype AB or TU males? How many males per female? If different male genotypes were used, as the Methods section says, maybe some of the variation could be in the males.  
+
+<|ref|>text<|/ref|><|det|>[[115, 552, 352, 565]]<|/det|>
+Fig. 6 legend. 'HH, HW, and MH'  
+
+<|ref|>text<|/ref|><|det|>[[115, 567, 483, 580]]<|/det|>
+I couldn't find a definition of these abbreviations.  
+
+<|ref|>text<|/ref|><|det|>[[115, 581, 515, 595]]<|/det|>
+Fig. 5 legend. 'Number of fish screened are indicated'  
+
+<|ref|>text<|/ref|><|det|>[[115, 597, 475, 610]]<|/det|>
+Number is indicated. Or Numbers are indicated.  
+
+<|ref|>text<|/ref|><|det|>[[115, 611, 845, 640]]<|/det|>
+P13. \`Germline expression of Rhebca did not disrupt sex determination in wild- type genotypes nor prevent oocyte loss in mios- /- (Figure'  
+
+<|ref|>text<|/ref|><|det|>[[115, 641, 866, 670]]<|/det|>
+What does the text mean by 'wild- type genotypes'? If these animals are transgenic and expressing a constitutive allele of mouse Rheb, clearly they are not wildtype.  
+
+<|ref|>text<|/ref|><|det|>[[115, 671, 878, 729]]<|/det|>
+Fig. 7a, b. Numbers are extraordinarily small. I'm not sure that finding 3 males means that no females will be found. After all, there are lots of human families with three sons, but that doesn't mean that their next child won't be a daughter. No statistics are given to give confidence that the conclusions about sex ratios are valid.  
+
+<|ref|>text<|/ref|><|det|>[[115, 730, 775, 744]]<|/det|>
+Fig. 7c. Label the horizontal axis. Should the vertical axis have an arrowhead at the top?  
+
+<|ref|>text<|/ref|><|det|>[[115, 746, 358, 759]]<|/det|>
+P13. \`vertebrate specific RNAbp,'  
+
+<|ref|>text<|/ref|><|det|>[[115, 760, 455, 774]]<|/det|>
+Spell out RNA binding protein here for clarity.  
+
+<|ref|>text<|/ref|><|det|>[[115, 775, 589, 789]]<|/det|>
+P13 \`Rbpms2 translationally represses testis- associated factors'  
+
+<|ref|>text<|/ref|><|det|>[[115, 790, 866, 864]]<|/det|>
+Remind reader here of the specific data that shows that translation of factors specific for testis development is repressed but translation of factors specific for ovary development are not repressed. This seems to be an important point of the paper and the data to make it need to be clearly stated and that a translation difference between the two gonad sexes is the main issue and not some other difference that results in different amounts of message or protein made.  
+
+<|ref|>text<|/ref|><|det|>[[115, 864, 866, 894]]<|/det|>
+The problem is that, if multitudes of ribosomes and high translation rates are required for oocytes to develop but spermatocytes can get by with fewer ribosomes and lower translation rates, that then
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 875, 120]]<|/det|>
+oogenesis will fail, and a failure of oogenesis by any of a variety of mechanisms, like blocking meiosis, also leads to testis development.  
+
+<|ref|>text<|/ref|><|det|>[[115, 121, 857, 150]]<|/det|>
+P13 'we demonstrate that the nutrient sensing arm of the mTORc1 pathway is uniquely required for oocyte progression and sustained oogenesis'  
+
+<|ref|>text<|/ref|><|det|>[[115, 151, 875, 225]]<|/det|>
+The result and conclusion are good, but could the effect be due to nutrient sensing difficulties in the soma - intestine, liver that makes vitellogenin, brain cells making gonadotrophs, etc. - rather than in the oocytes themselves? Is the effect cell autonomous to germ cells? Experiments can be designed to answer that question, but they are time consuming and difficult and publication shouldn't be held up for that, but the text should acknowledge that possibility.  
+
+<|ref|>text<|/ref|><|det|>[[115, 225, 870, 270]]<|/det|>
+P14. 'Additionally, RNA pol II, which we show is present in nuclei of wild-type zebrafish oocytes up to diplotene arrest, is also intact in rbpmS2 DM oocytes suggesting that RbpmS2 likely regulates sexual differentiation through translational control.'  
+
+<|ref|>text<|/ref|><|det|>[[115, 270, 875, 344]]<|/det|>
+The result shows that RbpmS2 doesn't regulate sex differentiation by controlling the location or amount of RNA pol II. But that result doesn't show that Rbms2 control is likely translational. It could also function on which genes are transcribed rather than just being in the nucleus at normal amounts, or on differential message stability, which also wouldn't necessarily change Rpol location and quantity. Then, the text can go on and rule out alternative explanations for the cited result.  
+
+<|ref|>text<|/ref|><|det|>[[115, 344, 870, 374]]<|/det|>
+P14. 'Specifically, several rbRNAs are associated with testis functions and their expression is limited to the early, undifferentiated cell types of the 40 dpf ovary.'  
+
+<|ref|>text<|/ref|><|det|>[[115, 374, 879, 433]]<|/det|>
+Do we know that those early, undifferentiated cell types at 40dpf are not already determined to be oocytes? I don't think the text told us when sex determination occurs in zebrafish. Is it before 40dpf? Also, the single cell transcriptomics for ovary should be contrasted to the single cell transcriptomics for testes of the same age to draw adequate conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[115, 433, 825, 462]]<|/det|>
+P14 'This restricted expression is consistent with our hypothesis that RbpmS2 suppresses testis factors'  
+
+<|ref|>text<|/ref|><|det|>[[115, 463, 875, 567]]<|/det|>
+Yes, it is consistent, but other possibilities exist too. Factors other than RbpmS2 could be responsible for decreasing the number of testis associated RNAs after the undifferentiated cell type stage. And if there are no testis mRNAs, then they would not be there for RbpmS2 to bind whether or not RbpmS2 is responsible for them disappearing so only ovary associated mRNAs would be present for RbpmS2 to bind after ovary commitment. Some other ovary- promoting factor could be suppressing testis associated gene transcription or transcript stability early and so there are no testis associated mRNAs left for RbpmS2 to bind at later stages.  
+
+<|ref|>text<|/ref|><|det|>[[115, 568, 878, 627]]<|/det|>
+Do we know that RbpmS2 is expressed before the sex determination stage? P15 'Notably, in zebrafish it has been shown that demethylation and amplification of an rDNA locus at the end of chromosome four (femrDNA) strongly correlates with female sexdetermination and differentiation56.'  
+
+<|ref|>text<|/ref|><|det|>[[115, 627, 868, 717]]<|/det|>
+Could it be that the femrDNA and RbpmS2 act on sex determination in exactly the same way, by preventing the formation of the many ribosomes that are needed to make a mature oocyte? Can you rule out the model that 1. ribosomes are essential for making a mature oocyte; 2. without mature oocytes, some regulatory mechanism is disturbed that normally would prevent the ovary from transitioning to testis; 3. That there are several genes that are independently necessary, each in a different way, for making ribosomes functional, including both femrDNA and Rboms2.  
+
+<|ref|>text<|/ref|><|det|>[[115, 717, 856, 761]]<|/det|>
+P16. Did the zebrafish spo11 mutations block DSBs and meiosis? The text told us that homozygous mothers gave embryos that didn't do well but was it shown definitively that this was because DSBs didn't occur?  
+
+<|ref|>text<|/ref|><|det|>[[115, 762, 501, 776]]<|/det|>
+P16. 'inhibition of mTORC1 by conditional knockout'  
+
+<|ref|>text<|/ref|><|det|>[[115, 777, 870, 820]]<|/det|>
+Tell reader what was the condition was in the conditional ko. Probably it was a cell- type specific, presumably oocyte specific ko, but in a couple of words the text would avoid making the reader go to the original paper just to find out.  
+
+<|ref|>text<|/ref|><|det|>[[115, 821, 608, 835]]<|/det|>
+P16. 'which in oocytes is orchestrated in part by mTORc1 signaling'  
+
+<|ref|>text<|/ref|><|det|>[[115, 836, 880, 879]]<|/det|>
+Do we know that that mTORc1 signaling is due to action in oocytes or is it non- autonomous due to effects in other cell types or in organs other than the gonad, like liver or gonadotrophin secreting brain cells?  
+
+<|ref|>text<|/ref|><|det|>[[115, 880, 590, 895]]<|/det|>
+P16. 'using Crispr- Cas9 mutagenesis as in as detailed below 62'
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 90, 207, 104]]<|/det|>
+Fix wording.  
+
+<|ref|>text<|/ref|><|det|>[[115, 105, 380, 120]]<|/det|>
+P21. 'Sequencing data was aligned'  
+
+<|ref|>text<|/ref|><|det|>[[115, 120, 644, 135]]<|/det|>
+Either sequencing data were aligned or sequencing datum was aligned.  
+
+<|ref|>text<|/ref|><|det|>[[115, 136, 352, 150]]<|/det|>
+P21 'using the Illumina website'  
+
+<|ref|>text<|/ref|><|det|>[[115, 150, 333, 164]]<|/det|>
+Text could give here the URL.  
+
+<|ref|>text<|/ref|><|det|>[[115, 165, 875, 195]]<|/det|>
+P21. 'Only RNAs that appeared in both the cross-linked and uncrosslinked mApple- Rbpm s2, but not in controls were considered Rbpm s2 target RNAs.'  
+
+<|ref|>text<|/ref|><|det|>[[115, 195, 878, 254]]<|/det|>
+Good to have four replicates. Was there a minimum read count for concluding that a gene's transcripts were bound? Or was a single read sufficient to mean that a gene's transcripts were pulled down? Was there an adjusted p- value used to identify differentially pulled down transcripts between experimental and control?  
+
+<|ref|>text<|/ref|><|det|>[[115, 254, 266, 268]]<|/det|>
+P22 '(mios wildtype'  
+
+<|ref|>text<|/ref|><|det|>[[115, 268, 825, 298]]<|/det|>
+It would be less ambiguous to say homozygous mios wildtype, because it could have meant phenotypically wildtype, which would include heterozygotes. Likewise for homozygous mutants.  
+
+<|ref|>text<|/ref|><|det|>[[115, 448, 402, 462]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 476, 881, 596]]<|/det|>
+Previous results implicate the RNA- binding protein Rbpm s2 in ovary fate during zebrafish development. This work identifies RNA targets of Rbpm s2 and, using a variety of approaches, supports a model whereby Rbpm s2 promotes nucleolar amplification via TORC1, a step that supports oogenesis. In particular, the work examines the role of the GATOR2 component Mios in nucleolar development in oocytes, and its dependence on TOR to promote oogenesis independent of the TSC Rheb arm of TORC1. Nutrient availability had been implicated in oogenesis in other species, as well as in zebrafish sex determination. Impact of the work is high, for those interested in TOR signaling, germline development and the role of nutrients in oogenesis.  
+
+<|ref|>text<|/ref|><|det|>[[115, 610, 872, 640]]<|/det|>
+Several points listed below to clarify the results, strengthen the conclusions, and improve accessibility to a wider audience.  
+
+<|ref|>text<|/ref|><|det|>[[111, 654, 872, 878]]<|/det|>
+1. p.6-7: Refer to transcript abundance differences, rather than "stability". Elaborate on proposed feedback. 
+2. Throughout manuscript: Define all terms and abbreviations (e.g., DM, H, M, HH, HM, MW, MH) 
+3. Figure 2: define white arrows in legend. 
+4. Figure 3: add quantification to support conclusion of loss of Ps6k localization in rbpm s2 DM; loss in mios mutant looks more convincing, but also needs quantification. 
+5. Figure 4g: add quantification to support differences in size of puncta. Presumably mios oocyte analysis was on the early arrested oocytes? Please clarify. Temper statement that mios is "required" for nucleologenesis since fibrillar staining shows nucleoli. 
+6. Figure 5: change or clarify labeling of x axes in legend (H, M, etc., change to genotype; state meaning of + and - for transgene). 
+7. Figure 7: correct the arrow from Gator1 to TORC as it should be negative. 
+8. Title of final results section is misleading. State more clearly result concerning amino acid sensing (GATOR-dependent) versus TSC/Rheb arm. 
+9. Supplemental Figure 6: stated phenotypes difficult to see; add zoom box.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[57, 42, 264, 60]]<|/det|>
+## REVIEWER COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[58, 75, 360, 92]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[55, 107, 937, 268]]<|/det|>
+The manuscript by Wilson et al., sheds light on the mechanism by which Rbpmps2 acts as a necessary factor for oocyte development and subsequent female sex differentiation in the zebrafish. The authors show that Rbpmps2 likely represses male differentiation factors while simultaneously promoting translation of ribosomal biogenesis factors and nucleolar amplification, which are critical for proper oocyte progression. Ultimately, the authors put forth an interesting model where Rbpmps2 acts as a fate switch upstream of a nutrient sensing pathway, ultimately necessary for "pushing" a bipotential gonocyte towards an oocyte fate. In my opinion, this is a strong manuscript and the authors' claims are well supported. The experiments performed here are a logical progression from the previous work published by this group. The results flesh out the previous story and integrate nutrient signaling into the pathway. The findings will be a valuable contribution to the field of sex determination and germ cell biology.  
+
+<|ref|>text<|/ref|><|det|>[[56, 282, 937, 333]]<|/det|>
+Overall, the manuscript is well written. However, it was difficult to follow the logic in the section " Mios is required for oocyte differentiation independent of double strand break repair". Perhaps the authors could clarify this section by directly stating what the hypothesis is that they are trying to test.  
+
+<|ref|>text<|/ref|><|det|>[[56, 346, 925, 396]]<|/det|>
+We have clarified the introduction in this section to make it clearer that we sought to determine if suppression of DSB formation could restore oogenesis in our mios- fish as it does in Drosophila mio- . We find that this is not the case and therefore conclude that this mechanism is not conserved in our vertebrate system.  
+
+<|ref|>text<|/ref|><|det|>[[57, 410, 923, 444]]<|/det|>
+The manuscript would benefit from a clear 'working model' section in the discussion to bring everything together. The authors could consider incorporating something like the legend in figure 7C into the actual text.  
+
+<|ref|>text<|/ref|><|det|>[[57, 458, 925, 492]]<|/det|>
+We thank the reviewer for this suggestion. We have added more detail to the discussion to more clearly state our overarching findings and hypotheses.  
+
+<|ref|>text<|/ref|><|det|>[[57, 506, 218, 523]]<|/det|>
+Specific Comments:  
+
+<|ref|>text<|/ref|><|det|>[[57, 538, 444, 556]]<|/det|>
+- Define DMs at the beginning of the manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[57, 570, 711, 588]]<|/det|>
+We thank all reviewers for bringing this to our attention have defined rbpmps2 DMs.  
+
+<|ref|>text<|/ref|><|det|>[[57, 602, 662, 620]]<|/det|>
+- Why is Pol II diminished in mios-/-? Could the authors please discuss this?  
+
+<|ref|>text<|/ref|><|det|>[[56, 634, 904, 684]]<|/det|>
+We noticed variation from sample to sample such that some oocytes had lower levels of RNA pol II in their nuclei as compared to other cells, but this was not different between genotypes. We have corrected the brightness/contrast of the image to accurately show the RNA pol II intensity.  
+
+<|ref|>text<|/ref|><|det|>[[57, 698, 901, 732]]<|/det|>
+- It is difficult to see p-Ubft localized to nucleoli in Fig.4 h,I (much less to a distinct compartment)- a higher magnification would be helpful.  
+
+<|ref|>text<|/ref|><|det|>[[57, 745, 923, 779]]<|/det|>
+We have modified the inset to show the individual p-Ubft and RNA pol II channels and have outlined the RNA pol II excluded region to make the lack of the granular compartment in mios mutants clearer.  
+
+<|ref|>text<|/ref|><|det|>[[57, 793, 360, 810]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[57, 826, 333, 842]]<|/det|>
+- What are the noteworthy results?  
+
+<|ref|>text<|/ref|><|det|>[[57, 842, 925, 890]]<|/det|>
+- The identification of mRNAs that are bound by Rbpmps2 proteins in zebrafish. This is important because Rbpmps2 is required for the gonad to maintain an ovary fate and some of these transcripts may mediate those fate choices.  
+
+<|ref|>text<|/ref|><|det|>[[57, 890, 633, 907]]<|/det|>
+- Rbpmps2 is required for nucleoli to proliferate during meiosis in oocytes.  
+
+<|ref|>text<|/ref|><|det|>[[57, 907, 348, 923]]<|/det|>
+- Rbpmps2 acts via mTOR1 signaling.  
+
+<|ref|>text<|/ref|><|det|>[[57, 923, 293, 939]]<|/det|>
+- Rbpmps2 is epistatic to mios.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 44, 576, 78]]<|/det|>
+- Without Rbpm s2, testis factors are not repressed.- Proper ribosome function is essential for ovary fate in zebrafish.  
+
+<|ref|>text<|/ref|><|det|>[[56, 92, 555, 125]]<|/det|>
+- Will the work be of significance to the field and related fields?- yes.  
+
+<|ref|>text<|/ref|><|det|>[[56, 140, 459, 157]]<|/det|>
+How does it compare to the established literature?  
+
+<|ref|>text<|/ref|><|det|>[[56, 157, 937, 206]]<|/det|>
+Microscopy is very well done. The use of immunofluorescence is beautiful. Nice electron microscopy. Good use of various mutants and inducing new ones when they needed to. Good development of transgenic animals for specific purposes.  
+
+<|ref|>text<|/ref|><|det|>[[56, 220, 542, 253]]<|/det|>
+If the work is not original, please provide relevant references. The work is original.  
+
+<|ref|>text<|/ref|><|det|>[[56, 267, 936, 317]]<|/det|>
+- Does the work support the conclusions and claims, or is additional evidence needed?- Some of the stated conclusions are logical inferences but are not the only possible explanations for the results. These are mentioned in the detailed reviewer comments and can be fixed in a revision.  
+
+<|ref|>text<|/ref|><|det|>[[56, 331, 900, 365]]<|/det|>
+- Are there any flaws in the data analysis, interpretation and conclusions?- Some conclusions fail to take into consideration alternative interpretations as noted in detailed comments.  
+
+<|ref|>text<|/ref|><|det|>[[56, 379, 646, 413]]<|/det|>
+- Do these prohibit publication or require revision?- The authors can easily revise the document to fix the perceived problems.  
+
+<|ref|>text<|/ref|><|det|>[[56, 427, 750, 444]]<|/det|>
+- Is the methodology sound?- Does the work meet the expected standards in your field?- Yes. The methodology in general is sound. Two exceptions - first, the numbers of individuals checked for sex ratios in Figs. 5, and especially 6 and 7 are spectacularly low and no statistics are given, and second, no statistical analysis (unless I missed it) was performed to be sure that the identified mRNAs bound to Rbpm s2 and not to the control are statistically robust. Nothing analogous to DESeq2 statistical output that one would see in an RNA-seq experiment although the methods are conceptually similar, seeing if the number of RNA molecules from a gene are statistically different in one situation vs. another.  
+
+<|ref|>text<|/ref|><|det|>[[56, 448, 928, 541]]<|/det|>
+Yes. The methodology in general is sound. Two exceptions - first, the numbers of individuals checked for sex ratios in Figs. 5, and especially 6 and 7 are spectacularly low and no statistics are given, and second, no statistical analysis (unless I missed it) was performed to be sure that the identified mRNAs bound to Rbpm s2 and not to the control are statistically robust. Nothing analogous to DESeq2 statistical output that one would see in an RNA-seq experiment although the methods are conceptually similar, seeing if the number of RNA molecules from a gene are statistically different in one situation vs. another.  
+
+<|ref|>text<|/ref|><|det|>[[56, 555, 930, 684]]<|/det|>
+We have addressed the sex ratio numbers for experiments in figures 5- 7 and have performed corresponding statistical analyses. We observed 3 female mios mutants carrying the ms44 Tg[ziwi:Rheb \(^{ca}\) ; cmlc2:mCherry] transgenic insertion from a single cross out of 3 independent crosses that were evaluated for this allele. As female mutants were not observed among progeny from 2 other sibling crosses carrying the same transgenic insertion, we conclude that the suppression by Rheb \(^{ca}\) was incompletely penetrant within the ms44 line. Furthermore, because suppression was not observed in 2 other independent alleles (representing different insertion events) we conclude that overexpression of Rheb \(^{ca}\) and therefore activation through this arm of the mTorc1 signaling pathway is not sufficient to restore oogenesis in mios \(^{c}\) fish.  
+
+<|ref|>text<|/ref|><|det|>[[56, 699, 920, 780]]<|/det|>
+We have added additional language to the RNAseq methods section to explain our rboRNA selection methodology. Specifically, we took an "all or none" approach - we only counted RNAs as bound if they were present in the mApple- Rbpm s2 crosslinked and uncrosslinked samples and absent in mApple crosslinked and/or uncrosslinked controls. Therefore, enrichment statistics were not necessary for these RNAs. It is true that additional targets may be found if we were to perform a less stringent enrichment analysis.  
+
+<|ref|>text<|/ref|><|det|>[[56, 795, 689, 828]]<|/det|>
+- Is there enough detail provided in the methods for the work to be reproduced?- Yes.  
+
+<|ref|>text<|/ref|><|det|>[[56, 843, 210, 859]]<|/det|>
+Specific comments  
+
+<|ref|>text<|/ref|><|det|>[[56, 859, 920, 909]]<|/det|>
+Fig. 1A. The normal expression profiles of the two genes used for the transgenic analysis, rbpm s2b and buc, need to be shown in Fig. 1B to compare to the other markers. Probably mios also because it plays such a pivotal role in the story.  
+
+<|ref|>text<|/ref|><|det|>[[56, 908, 880, 940]]<|/det|>
+Also, the expression pattern of rbpm s2a and rbpm s2b are essential with respect to cell type and time of expression, and their expression patterns in the scRNA- seq results.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 58, 927, 141]]<|/det|>
+We have revised the text to clarify the expression of RbpmS2 and Buckyball proteins. Previous work from our lab and others have demonstrated that both RNAs and proteins are present at the same time and in the early oocyte- specific structure, the Balbiani body. To facilitate comparisons of their RNA expression profiles, we have added the relevant RNA expression UMAPs to supplemental Figure 1, which demonstrates their highly similar expression profiles in the early ovary.  
+
+<|ref|>text<|/ref|><|det|>[[56, 155, 937, 190]]<|/det|>
+P5. Define: DMs for non- zebrafish people. The introduction or start of the results should point out that zebrafish has two copies of this gene and the authors knocked out both.  
+
+<|ref|>text<|/ref|><|det|>[[56, 203, 937, 237]]<|/det|>
+We thank the reviewer for noting this and have clarified our language to address the presence of the two genes and have defined "DM" in this context.  
+
+<|ref|>text<|/ref|><|det|>[[58, 250, 812, 270]]<|/det|>
+P5. Was buc:mApple specifically expressed in oocytes according to microscopic observations?  
+
+<|ref|>text<|/ref|><|det|>[[56, 282, 937, 333]]<|/det|>
+This promoter has been confirmed to express specifically in early- stage oocytes and drives mApple expression in oocytes specifically, as we have previously shown in Kaufman et al. PLoS Genetics 2014 (https://doi.org/10.1371/journal.pgen.1007489).  
+
+<|ref|>text<|/ref|><|det|>[[56, 346, 915, 397]]<|/det|>
+P5. Tell the reader somewhere early at what specific stage in oogenesis rbpmS2a and rbpmS2b begin to be expressed. Especially relevant is when relative to the expression of the oocyte- specific gene buc, which provided the promotor for the transgene.  
+
+<|ref|>text<|/ref|><|det|>[[56, 410, 930, 445]]<|/det|>
+We have clarified our language to make it clear that our previous work has shown that these proteins are both expressed at the same time and localize the Balbiani body of prophase I oocytes.  
+
+<|ref|>text<|/ref|><|det|>[[56, 458, 930, 540]]<|/det|>
+Likewise, readers need to know the expression pattern of rbpmS2a and b in testis to understand the system. The human protein atlas shows that RBPMs2 is expressed in both spermatogonia and oocytes, but higher in oocytes. https://www.proteinatlas.org/ENSG00000166831- RBPMs2/single+cell+type. Maybe it is also expressed in zebrafish spermatocytes. Public data might be available for this so new experiments would need to be done if the lab doesn't have the data already.  
+
+<|ref|>text<|/ref|><|det|>[[56, 554, 936, 604]]<|/det|>
+We have previously shown in Romano et al. Development 2020 (https://journals.biologists.com/dev/article/147/18/dev190942/225837/Loss- of- dmrt1- restores- zebrafish- female- fates- in) that RbpmS2 is only detected in ovaries and we have clarified the text to make this clear.  
+
+<|ref|>text<|/ref|><|det|>[[56, 618, 936, 653]]<|/det|>
+P5 The transgenic/immunoprecipitation method is a good one to identify interacting RNAs and the Fig. 1A does a good job of visually explaining the strategy.  
+
+<|ref|>text<|/ref|><|det|>[[56, 667, 473, 685]]<|/det|>
+We thank the reviewer for this positive feedback \(\circledcirc\)  
+
+<|ref|>text<|/ref|><|det|>[[56, 700, 936, 732]]<|/det|>
+P5. "The remaining 52 rboRNAs that did not map to the 40 dpf ovary dataset are likely transcripts expressed in later stage oocytes present in the fully mature adult ovary."  
+
+<|ref|>text<|/ref|><|det|>[[56, 732, 927, 765]]<|/det|>
+Yes, probably so, and that idea can be tested just by showing that they are or aren't previously characterized as late oocyte proteins from prior literature or public RNA- seq data.  
+
+<|ref|>text<|/ref|><|det|>[[56, 780, 936, 925]]<|/det|>
+Due to the cell size limitation of the FACS sorting performed on the zebrafish 40 dpf ovary, we could not analyze oocytes larger than \(\sim 70 \mu \mathrm{m}\) . We have gone through the originally reported 52 RNAs that were not found in the 40 dpf dataset and have determined by searching updated gene names that only 7 of the 52 do not map to the 40 dpf dataset. We have searched for these 7 genes in other public oocyte datasets (https://www.nature.com/articles/s41586- 022- 04918- 4) and have determined that all 7 are found in PGCs, oogonia, and oocytes of mouse and human fetal ovaries. Additionally, we find that 5 are present in an early zebrafish embryo dataset (https://www.science.org/doi/10.1126/science.aar3131), specifically at stages around when ZGA occurs, suggesting that these 4 RNAs are present in late oocyte stages. Only 2 could not be found in available databases or expression analyses. Figure 1 was modified based on these analyses. Specifically,
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 44, 888, 78]]<|/det|>
+the panel of RNAs that were not enriched (Titled "None") was removed and the figure was rearranged to accommodate this change.  
+
+<|ref|>text<|/ref|><|det|>[[56, 90, 896, 142]]<|/det|>
+Fig. 1 "(d) Volcano plot of differential gene expression in 21 dpf wildtype and rbpms2 DMs from bulk RNA sequencing." This legend doesn't say what organs were taken, ovaries, testes, or whether these are whole animals.  
+
+<|ref|>text<|/ref|><|det|>[[56, 155, 936, 190]]<|/det|>
+We have revised the figure legend to state that the bulk RNA sequencing was performed on bipotential gonads of wildtype and rbpms2 DM 21 dpf fish.  
+
+<|ref|>text<|/ref|><|det|>[[56, 203, 920, 285]]<|/det|>
+P6 'Therefore, we conclude that 1) Rbpms2 likely functions to repress translation of rbtRNAs expressed in early gonocytes in ovaries and 2) promote translation of oocyte factors,' Is this a conclusion or a hypothesis? From the data presented, it seems like these are logical hypotheses stemming from those data, but we can't actually conclude these two points from the presented data because translation was not specifically tested and other mechanisms are possible.  
+
+<|ref|>text<|/ref|><|det|>[[61, 298, 888, 317]]<|/det|>
+We have revised the text to clarify our language and to include the possibility of alternative mechanisms.  
+
+<|ref|>text<|/ref|><|det|>[[58, 330, 710, 348]]<|/det|>
+P6 'we analyzed GSEA enrichment plots'. Tell reader what these plots are, briefly.  
+
+<|ref|>text<|/ref|><|det|>[[56, 363, 541, 380]]<|/det|>
+We have added an explanatory statement to make this clear.  
+
+<|ref|>text<|/ref|><|det|>[[56, 394, 930, 428]]<|/det|>
+P7 'Rbpm2 mediates a binary fate- switch by repressing testis factors and promoting nucleolar formation and oocyte development via mTOR signaling'  
+
+<|ref|>text<|/ref|><|det|>[[56, 427, 925, 460]]<|/det|>
+The results do show that Rbpm2 promotes nucleolar formation, but the link to repressing testis factors is not really shown.  
+
+<|ref|>text<|/ref|><|det|>[[56, 474, 934, 524]]<|/det|>
+We have expanded the discussion to include qualifying language to make clear that the proposed functions of Rbpm2 as a translational repressor of testis RNAs and promoter of RNAs related to nucleoli formation and ribosome biogenesis are hypothesized functions given our results in this work and in the published literature.  
+
+<|ref|>text<|/ref|><|det|>[[60, 538, 888, 556]]<|/det|>
+Fig. 2. Nice immunestaining images. It would be good for the legend to tell reader what the red nuage is.  
+
+<|ref|>text<|/ref|><|det|>[[56, 570, 936, 652]]<|/det|>
+We have adjusted the images of the Mios protein localization to the cytoplasm of oocytes and have added magnified images of early oocyte stages. Additionally, we have added fluorescence quantification of Mios in wildtype and rbpms2 DM oocytes to further support our claim that Mios protein is lacking in rbpms2 DMs. In reference to the red nuage of figure 2, we are unclear what exactly is being noted here and have assumed this comment was regarding the Mios staining in figure 3.  
+
+<|ref|>text<|/ref|><|det|>[[56, 666, 920, 716]]<|/det|>
+Fig. 2 'wild- type (rbpms2aae30; rbpms2bsa9329 HM; n=4) and rbpms2 DM' Would be easier to understand if instead of HM it said double heterozygous and instead of DM it said double homozygous mutants.  
+
+<|ref|>text<|/ref|><|det|>[[56, 730, 576, 748]]<|/det|>
+We have clarified the genotype information in each figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[56, 761, 666, 780]]<|/det|>
+pUbf1: Shouldn't the figure legend use p- Ubf1 like the text instead of p- Ubf1?  
+
+<|ref|>text<|/ref|><|det|>[[56, 794, 733, 812]]<|/det|>
+We have corrected the discrepancy and have corrected the gene name to ubtf (Ubtf).  
+
+<|ref|>text<|/ref|><|det|>[[56, 826, 784, 860]]<|/det|>
+P7 'we reanalyzed transmission electron microscopy (TEM) images of rbpms2 DM oocytes' Were these reanalyzed from a previous publication? If so, give citation here.  
+
+<|ref|>text<|/ref|><|det|>[[56, 873, 833, 892]]<|/det|>
+We thank the reviewer for noting this and have updated the text with the corresponding reference.  
+
+<|ref|>text<|/ref|><|det|>[[56, 905, 925, 939]]<|/det|>
+P7 'wild- type prophase I oocytes had greater than four nucleoli per cell at 35 dpf (Figure 2b, d) whereas most rbpms2 DM oocytes had fewer than 4 nucleoli (Figure 2c, d).'
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 44, 896, 77]]<|/det|>
+Fig. 2a shows a great variation in nucleolus number for cells in different stages of prophase I in wildtypes. Which phases were present at 35dpf and which stage in meiosis were the cells in that were counted?  
+
+<|ref|>text<|/ref|><|det|>[[56, 92, 937, 157]]<|/det|>
+rbpms2 DM oocytes do not appear to reach diplotene but do progress through the early stages of prophase I as judged by Balbiani body (Bb) formation, so we analyzed early Bb stage wildtype and rbpms2 DM cells that were in the stages of prophase I prior to arrest. We have highlighted a representative cell in the wildtype image to clarify this point.  
+
+<|ref|>text<|/ref|><|det|>[[56, 172, 880, 221]]<|/det|>
+P8 'In rbpms2 DMs, while RNA pol I localized to the cytoplasm and nucleoli of mitotic germ cells, it was observed throughout oocyte nuclei and overlapped with RNA pol II nuclear localization (Figure 2f).' Draw a conclusion here for that finding.  
+
+<|ref|>text<|/ref|><|det|>[[56, 236, 889, 285]]<|/det|>
+We have expanded upon this finding, noting that while RNA pol I localizes to 1- 2 nucleoli in rbpms2 DM oocytes, it also remains distributed throughout the oocyte nucleus (in contrast to wildtype) and suggests nucleolar seeding and consequently rRNA transcription is diminished.  
+
+<|ref|>text<|/ref|><|det|>[[56, 300, 928, 365]]<|/det|>
+P8 'may do so by regulating translation of these factors to promote' Conclusion is appropriate from the data, but I had to go back and reread the sentence to be sure of the antecedent to 'these'. It would help to say 'may do so by regulating translation of ribosome biogenesis factors to promote'.  
+
+<|ref|>text<|/ref|><|det|>[[57, 380, 744, 397]]<|/det|>
+We thank the reviewer for their suggestion and have revised the sentence accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[56, 411, 916, 476]]<|/det|>
+P8. 'In wild-type ovaries, localization of phosphorylated p70-S6K (p-Ps6k), a kinase directly phosphorylated by the active form of mTorc1' Do you think that readers should know that zebrafish has two copies of the relevant gene and that its official names are rps6kb1a and rps6kb1b?  
+
+<|ref|>text<|/ref|><|det|>[[56, 491, 936, 603]]<|/det|>
+We have analyzed the amino sequences of the Rps6kb1a and Rps6kb1b zebrafish proteins against the human Rps6k1b and Rps6k2b proteins. Both zebrafish proteins are annotated as orthologs of the human Rps6k1b. Aligning the zebrafish proteins to both human proteins shows that the epitope target of the antibody is conserved in all cases, indicating that the antibody we used likely recognizes both zebrafish proteins. Further, single cell data indicates rps6kb1a is lowly expressed throughout the ovary (highest expression is in postmeiotic cells) and rps6kb1b is expressed throughout but is highest in GSCs/TA and oocyte cells. We have revised the text as recommended to clarify that there are two proteins in zebrafish.  
+
+<|ref|>text<|/ref|><|det|>[[56, 618, 896, 667]]<|/det|>
+Fig. 3 '(c) Table of mTorc1- related components with Rbpmss2 binding sites in their 3' and/or 5' UTRs.' Are these demonstrated Rbpmss2 binding sites? Or are they sites predicted from some kind of consensus sequence?  
+
+<|ref|>text<|/ref|><|det|>[[56, 682, 933, 730]]<|/det|>
+We have clarified that these bindings sites were predicted based on the published binding motif for the related family member Rbpmss, as the exact binding motif of Rbpmss2 has not been precisely mapped. We have revised the text to include the citation to the work that characterized the Rbpmss binding motif.  
+
+<|ref|>text<|/ref|><|det|>[[56, 745, 936, 840]]<|/det|>
+P9 'wildtype, p- Ps6k was not detected in DM oocytes (Figure 3b). This observation suggests that dysregulated mTorc1 signaling in the absence of Rbpmss2 contributes to impaired oogenesis.' The data at this point show that Rbpmss2 contributes to impaired mTorc1 signaling. It would take other experiments to show that it is the mTorc1 signaling that contributes to impaired oogenesis. It could be that it is one of the other things that's disrupted, like ribosome biogenesis, that contributes to impaired oogenesis and that the impaired mTorc1 signaling indeed happens but it's not the main reason oogenesis fails.  
+
+<|ref|>text<|/ref|><|det|>[[57, 856, 930, 889]]<|/det|>
+We have revised the text to clarify that loss of Rbpmss2 leads to dysregulation of mTorc1 signaling, which may contribute to or be a consequence of defects in ribosome biogenesis and oogenesis.  
+
+<|ref|>text<|/ref|><|det|>[[56, 905, 673, 937]]<|/det|>
+P9 'Missing oocyte (Mios), contains 4 Rbpmss2 binding sites in its 3' UTR' Again, are these proven binding sites? Or just predicted based on sequence?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[60, 59, 840, 77]]<|/det|>
+These binding sites are predicted based on the validated Rbpmns binding sites as described above.  
+
+<|ref|>text<|/ref|><|det|>[[57, 91, 907, 140]]<|/det|>
+P10 'Analysis of miosms20 and miossa22946 heterozygous and mutant progeny' Unclear if 'heterozygous' here means miosms20/miossa22946 trans heterozygote or miosms20/mios+ and miossa22946/mios+.  
+
+<|ref|>text<|/ref|><|det|>[[57, 156, 872, 189]]<|/det|>
+We have revised the text to clarify that the fish noted here are single heterozygotes or mutants for the indicated alleles.  
+
+<|ref|>text<|/ref|><|det|>[[57, 204, 920, 268]]<|/det|>
+P9 'Several in- frame deletions were recovered including miosms20, an 11bp deletion allele that leads to a frameshift' Sentence says miosms20 is an in- frame deletion that is an 11bp deletion. But 11 is not divisible by 3 so how is this an in- frame deletion? Maybe it's a wording problem with the sentence.  
+
+<|ref|>text<|/ref|><|det|>[[57, 283, 925, 316]]<|/det|>
+Thank you for pointing this out. We did not word this well. We have revised the text to clarify that we found in- frame deletions as well as nonsense mutations including the ms20 allele.  
+
+<|ref|>text<|/ref|><|det|>[[57, 331, 875, 365]]<|/det|>
+P10. 'miosms20 and miossa22946 mutants and miosms20/sa22946 compound heterozygotes develop functional testes and differentiate as males, exclusively'  
+
+<|ref|>text<|/ref|><|det|>[[57, 365, 940, 413]]<|/det|>
+Previously in the text, heterozygotes for a mutated allele were also called mutants. So here, text should specify whether 'miosms20 and miossa22946 mutants' means heterozygotes for each or homozygotes for each allele. The 'compound heterozygotes' phrase is clear.  
+
+<|ref|>text<|/ref|><|det|>[[60, 428, 936, 445]]<|/det|>
+We have revised the text and legends to indicate what "heterozygote" and "mutant" mean in the given contexts.  
+
+<|ref|>text<|/ref|><|det|>[[57, 460, 370, 476]]<|/det|>
+P10. 'investigation at earlier timepoints,'  
+
+<|ref|>text<|/ref|><|det|>[[57, 477, 923, 509]]<|/det|>
+Earlier than what? From the writing, earlier than the stage indicated by 'Additionally, gonad morphological...'. But the reader is not told what stage that is.  
+
+<|ref|>text<|/ref|><|det|>[[60, 524, 923, 541]]<|/det|>
+We have revised the wording here to make clear that we also analyzed these fish prior to testis development.  
+
+<|ref|>text<|/ref|><|det|>[[57, 556, 630, 589]]<|/det|>
+Fig. 4a Legend doesn't tell us that the asterisk mean stop for miossa/sa. We have defined this in the revised figure legend.  
+
+<|ref|>text<|/ref|><|det|>[[57, 604, 614, 637]]<|/det|>
+P10 'number of Fibrillarin but these puncta were much smaller' Change to: number of Fibrillarin puncta, but these were much smaller.  
+
+<|ref|>text<|/ref|><|det|>[[57, 652, 635, 668]]<|/det|>
+We thank the reviewer for this suggestion and have made the correction.  
+
+<|ref|>text<|/ref|><|det|>[[57, 683, 940, 748]]<|/det|>
+P10 'between wild-type and mios-/- oocytes (Figure 4h- i). In wildtype,' Be consistent with 'wildtype', If you prefer to use 'wildtype' as one word when it is a noun, then you should use 'wildtype' as an adjective, e.g., 'this fish is a wildtype and this is a wildtype fish.' Only if you use 'wild type' as a noun should you use 'wild-type' as an adjective; e.g., 'this fish is a wild type and this is a wild-type fish'.  
+
+<|ref|>text<|/ref|><|det|>[[57, 763, 595, 779]]<|/det|>
+We have revised the text to use "wildtype" for nouns and adjectives.  
+
+<|ref|>text<|/ref|><|det|>[[57, 794, 922, 843]]<|/det|>
+Fig. 5 Legend. It's unclear what 'specified cells' means in 'p- Ps6k and DAPI localization in specified cells'. Does it mean specified as germ cells vs. somatic gonadal cells? Specified as oocytes vs. spermatocytes? Or something else?  
+
+<|ref|>text<|/ref|><|det|>[[57, 858, 653, 875]]<|/det|>
+We have revised the wording here to "indicated cells" rather than specified.  
+
+<|ref|>text<|/ref|><|det|>[[57, 891, 797, 907]]<|/det|>
+Fig. 5 Legend: 'Scale bar for (a,b) are 50 μM'. Should be either scale bars are or scale bar is.  
+
+<|ref|>text<|/ref|><|det|>[[57, 922, 300, 938]]<|/det|>
+We have made this correction.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 60, 676, 92]]<|/det|>
+P10 'We generated two alleles, ms49 and ms64, of the mTORca transgene.' Text should tell us what the sequence difference is between these two alleles.  
+
+<|ref|>text<|/ref|><|det|>[[57, 108, 911, 141]]<|/det|>
+We have revised the sentence to indicate that these alleles are the same transgene and that the two alleles only differ by independent insertion events in the genome.  
+
+<|ref|>text<|/ref|><|det|>[[57, 156, 866, 205]]<|/det|>
+P11. 'more total eggs than wildtype non- transgenic siblings' This 'wildtype' usage as an adjective is correct if text consistently uses 'wildtype' as a noun to mean a genotype.  
+
+<|ref|>text<|/ref|><|det|>[[58, 220, 300, 236]]<|/det|>
+We have made this correction.  
+
+<|ref|>text<|/ref|><|det|>[[57, 251, 930, 300]]<|/det|>
+Fig. 5e. Numbers are quite small, six or seven or eight animals used to determine sex ratios. The conclusions are probably accurate, but numbers are quite small and no statistics are given to confirm that different genotypes are statistically different.  
+
+<|ref|>text<|/ref|><|det|>[[60, 315, 933, 332]]<|/det|>
+We have updated all sex ratio graphs with corresponding statistics and increased numbers as indicated above.  
+
+<|ref|>text<|/ref|><|det|>[[58, 347, 240, 363]]<|/det|>
+P11 'In fertility assays,'  
+
+<|ref|>text<|/ref|><|det|>[[57, 364, 923, 411]]<|/det|>
+Tell here briefly how these were done. Single pair matings? Wildtype AB or TU males? How many males per female? If different male genotypes were used, as the Methods section says, maybe some of the variation could be in the males.  
+
+<|ref|>text<|/ref|><|det|>[[56, 426, 935, 571]]<|/det|>
+We have revised the methods section to indicate that these crosses were single paired matings, and to indicate that the wildtype fish were SATs. The females analyzed were not crossed to the same male every instance. Although we cannot exclude that the males did contribute to the results we see, but we did not see significant differences among females (of the same mutant and transgenic genotypes) from mating to mating, but differences were seen between the mios mutant \(\mathsf{Tg + }\) females compared to nonmutant \(\mathsf{Tg + }\) and \(\mathsf{Tg - }\) female siblings. Thus, we conclude that the effects seen are due to the transgene in the context of lack of Mios. Additionally, we did compare the \(\mathsf{Tg + }\) and \(\mathsf{Tg - }\) wildtype and heterozygous siblings to each other and found that these siblings, as expected, were not statistically different from each other, suggesting that the male- to- male variation didn't contribute significantly to the results seen.  
+
+<|ref|>text<|/ref|><|det|>[[57, 571, 810, 588]]<|/det|>
+We have also updated the text with a brief description of how the fertility assay was conducted.  
+
+<|ref|>text<|/ref|><|det|>[[57, 603, 441, 636]]<|/det|>
+Fig. 6 legend. 'HH, HW, and MH' I couldn't find a definition of these abbreviations.  
+
+<|ref|>text<|/ref|><|det|>[[57, 651, 545, 667]]<|/det|>
+We have included these definitions in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[57, 682, 483, 714]]<|/det|>
+Fig. 5 legend. 'Number of fish screened are indicated' Number is indicated. Or Numbers are indicated.  
+
+<|ref|>text<|/ref|><|det|>[[57, 730, 300, 746]]<|/det|>
+We have made this correction.  
+
+<|ref|>text<|/ref|><|det|>[[57, 761, 901, 789]]<|/det|>
+P13. 'Germline expression of Rhebca did not disrupt sex determination in wild-type genotypes nor prevent oocyte loss in mios-/- (Figure'  
+
+<|ref|>text<|/ref|><|det|>[[57, 790, 860, 823]]<|/det|>
+What does the text mean by 'wild- type genotypes'? If these animals are transgenic and expressing a constitutive allele of mouse Rheb, clearly they are not wildtype.  
+
+<|ref|>text<|/ref|><|det|>[[57, 840, 346, 856]]<|/det|>
+We have revised the language here.  
+
+<|ref|>text<|/ref|><|det|>[[57, 872, 925, 921]]<|/det|>
+Fig. 7a, b. Numbers are extraordinarily small. I'm not sure that finding 3 males means that no females will be found. After all, there are lots of human families with three sons, but that doesn't mean that their next child won't be a daughter. No statistics are given to give confidence that the conclusions about sex ratios are valid.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[55, 43, 936, 199]]<|/det|>
+We have increased the numbers of animals examined for the \(ms44\) and \(ms47\) alleles. We observed 3 female mios mutants carrying the \(ms44\) Tg[ziwi:Rheb \(^{ca}\) ; cmlc2:mCherry] transgenic insertion from a single cross out of 3 independent crosses that were evaluated for this allele. As female mutants were not observed among progeny from 2 other sibling crosses carrying the same transgenic insertion, we conclude that the suppression by Rheb \(^{ca}\) was incompletely penetrant within the \(ms44\) line. The data for these crosses are included as Supp. Figure 8 in the revised manuscript. Furthermore, because suppression was not observed in 2 other independent alleles (representing different insertion events) we conclude that overexpression of Rheb \(^{ca}\) and therefore activation through this arm of the mTorc1 signaling pathway is not sufficient to restore oogenesis in mios \(^{c}\) fish.  
+
+<|ref|>text<|/ref|><|det|>[[57, 226, 760, 244]]<|/det|>
+Fig. 7c. Label the horizontal axis. Should the vertical axis have an arrowhead at the top?  
+
+<|ref|>text<|/ref|><|det|>[[57, 254, 451, 270]]<|/det|>
+We have corrected this in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[57, 284, 419, 318]]<|/det|>
+P13. 'vertebrate specific RNAbp,' Spell out RNA binding protein here for clarity.  
+
+<|ref|>text<|/ref|><|det|>[[58, 333, 373, 350]]<|/det|>
+We have revised this as recommended.  
+
+<|ref|>text<|/ref|><|det|>[[57, 365, 940, 510]]<|/det|>
+P13 'Rbpms2 translationally represses testis- associated factors' Remind reader here of the specific data that shows that translation of factors specific for testis development is repressed but translation of factors specific for ovary development are not repressed. This seems to be an important point of the paper and the data to make it need to be clearly stated and that a translation difference between the two gonad sexes is the main issue and not some other difference that results in different amounts of message or protein made. The problem is that, if multitudes of ribosomes and high translation rates are required for oocytes to develop but spermatocytes can get by with fewer ribosomes and lower translation rates, that then oogenesis will fail, and a failure of oogenesis by any of a variety of mechanisms, like blocking meiosis, also leads to testis development.  
+
+<|ref|>text<|/ref|><|det|>[[57, 525, 904, 558]]<|/det|>
+We have tried to emphasize these points in the appropriate sections of the revised text and hope it is more clear now.  
+
+<|ref|>text<|/ref|><|det|>[[57, 572, 894, 670]]<|/det|>
+P13 'we demonstrate that the nutrient sensing arm of the mTorc1 pathway is uniquely required for oocyte progression and sustained oogenesis 'The result and conclusion are good, but could the effect be due to nutrient sensing difficulties in the soma - intestine, liver that makes vitellogenin, brain cells making gonadotrophis, etc. - rather than in the oocytes themselves? Is the effect cell autonomous to germ cells? Experiments can be designed to answer that question, but they are time consuming and difficult and publication shouldn't be held up for that, but the text should acknowledge that possibility.  
+
+<|ref|>text<|/ref|><|det|>[[57, 685, 923, 750]]<|/det|>
+We thank the reviewer for pointing out this alternative. Because we are able to restore oogenesis by expressing mTOR \(^{ca}\) only in germ cells under the germline- specific promoter, zivi, this indicates that restoring activity in the germ cells, including oocytes, is sufficient. If required in the somatic cells this activity is likely regulated independent of Mios.  
+
+<|ref|>text<|/ref|><|det|>[[57, 764, 936, 813]]<|/det|>
+P14. 'Additionally, RNA pol II, which we show is present in nuclei of wild-type zebrafish oocytes up to diplotene arrest, is also intact in rbpms2 DM oocytes suggesting that RbpmS2 likely regulates sexual differentiation through translational control.'  
+
+<|ref|>text<|/ref|><|det|>[[57, 813, 936, 893]]<|/det|>
+The result shows that RbpmS2 doesn't regulate sex differentiation by controlling the location or amount of RNA pol II. But that result doesn't show that Rbms2 control is likely translational. It could also function on which genes are transcribed rather than just being in the nucleus at normal amounts, or on differential message stability, which also wouldn't necessarily change Rpol location and quantity. Then, the text can go on and rule out alternative explanations for the cited result.  
+
+<|ref|>text<|/ref|><|det|>[[57, 907, 926, 941]]<|/det|>
+Our conclusion that these oocytes are not failing due to transcriptional dysregulation is based on the RNAseq data showing that neither 1) RNA levels of rboRNAs nor 2) recruitment of the RNA pol II machinery to the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 44, 920, 125]]<|/det|>
+nucleus change between wildtype and rbpm s2 DM fish. Therefore, we conclude that Rbpm s2 likely regulates translation of rboRNAs since RNA abundance appears unperturbed. For example, because we did not see significant changes in mios abundance between wildtype and rbpm s2 DMs, but we did see that rbpm s2 DM oocytes have significantly less Mios protein, this suggests, and we hypothesize that Rbpm s2 promotes its translation.  
+
+<|ref|>text<|/ref|><|det|>[[56, 139, 920, 172]]<|/det|>
+P14. 'Specifically, several rboRNAs are associated with testis functions and their expression is limited to the early, undifferentiated cell types of the 40 dpf ovary.'  
+
+<|ref|>text<|/ref|><|det|>[[56, 172, 925, 236]]<|/det|>
+Do we know that those early, undifferentiated cell types at 40dpf are not already determined to be oocytes? I don't think the text told us when sex determination occurs in zebrafish. Is it before 40dpf? Also, the single cell transcriptomics for ovary should be contrasted to the single cell transcriptomics for testes of the same age to draw adequate conclusions.  
+
+<|ref|>text<|/ref|><|det|>[[56, 251, 930, 348]]<|/det|>
+Because these cells underwent single cell sequencing, we can say by the RNAs present that these cells are still bipotential and have the capacity to differentiate into oocytes or spermatocytes. Because Rbpm s2 protein is detected in the ovary, but not the testis we assume that rbtRNAs in the testis are available for translation in the absence of Rbpm s2. We have revised the text to clarify the significance of the observation that rbtRNAs are restricted to early, undifferentiated cells of the ovary. We have also revised the text to better explain sex determination and differentiation in zebrafish as well as the plasticity of the zebrafish ovary.  
+
+<|ref|>text<|/ref|><|det|>[[57, 363, 888, 411]]<|/det|>
+P14 'This restricted expression is consistent with our hypothesis that Rbpm s2 suppresses testis factors' Yes, it is consistent, but other possibilities exist too. Factors other than Rbpm s2 could be responsible for decreasing the number of testis associated RNAs after the undifferentiated cell type stage.  
+
+<|ref|>text<|/ref|><|det|>[[57, 426, 915, 475]]<|/det|>
+We thank the reviewer for pointing this out. We have revised the text to include the possibility that Rbpm s2 may directly or indirectly influence rboRNAs and aspects of oogenesis through yet to be determined binding partners.  
+
+<|ref|>text<|/ref|><|det|>[[57, 490, 936, 540]]<|/det|>
+And if there are no testis mRNAs, then they would not be there for Rbpm s2 to bind whether or not Rbpm s2 is responsible for them disappearing so only ovary associated mRNAs would be present for Rbpm s2 to bind after ovary commitment.  
+
+<|ref|>text<|/ref|><|det|>[[57, 555, 936, 652]]<|/det|>
+Because we identified the rbtRNAs in wildtype oocytes, this does indicate that in a wildtype context the testis RNAs are present and available for Rbpm s2 to bind and regulate, even after ovary commitment. Continuous RNA expression and maintenance in a repressed state is common in oocytes and other several cell types, and we think this state is likely key to plasticity of the ovary. Accordingly, maintaining but repressing these testis RNAs in early germ cells of the ovary, as we think Rbpm s2 does, would be important for sustained oocyte development and maintenance of the ovary.  
+
+<|ref|>text<|/ref|><|det|>[[60, 667, 900, 699]]<|/det|>
+Some other ovary- promoting factor could be suppressing testis associated gene transcription or transcript stability early and so there are no testis associated mRNAs left for Rbpm s2 to bind at later stages.  
+
+<|ref|>text<|/ref|><|det|>[[57, 714, 930, 826]]<|/det|>
+We agree that other ovary- promoting factors likely suppress transcription of testis associated genes in differentiated oocytes at later stages. However, because rbtRNAs were found bound to Rbpm s2 in adult ovary, and because their abundance did not change in early oocytes of rbpm s2 DMs, and because loss of rbpm s2 results in testis development, we think Rbpm s2 translationally represses these testis RNAs in early oocytes. Because rbpm s2 DM oocytes do not make it past prophase I, we cannot make any claims about the role of Rbpm s2 and its potential interacting partners in later oocyte stages. In the revised discussion, we have added the possibility that loss of Rbpm s2 may have direct and indirect effects on oogenesis.  
+
+<|ref|>text<|/ref|><|det|>[[57, 841, 660, 858]]<|/det|>
+Do we know that Rbpm s2 is expressed before the sex determination stage?  
+
+<|ref|>text<|/ref|><|det|>[[57, 873, 936, 939]]<|/det|>
+Our bulk sequence analysis indicates that rbpm s2 transcripts are present in 21 day and 28 day gonads, stages prior to sex determination. Single cell analysis indicates that rbpm s2 transcripts are detected in some mitotic cells but are most highly expressed in early meiotic cells and differentiating oocytes (we have added UMAP plots of rbpm s2a and rbpm s2b expression in the 40 dpf). However, Rbpm s2 protein is not detected in mitotic
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 44, 915, 77]]<|/det|>
+germ cells, but is present in meiotic cells and differentiating oocytes, indicating that rbpms2 expression and activity increase as the cells differentiate.  
+
+<|ref|>text<|/ref|><|det|>[[56, 91, 937, 156]]<|/det|>
+P15 'Notably, in zebrafish it has been shown that demethylation and amplification of an rDNA locus at the end of chromosome four (femrDNA) strongly correlates with female sex determination and differentiation56.' Could it be that the femrDNA and RbpmS2 act on sex determination in exactly the same way, by preventing the formation of the many ribosomes that are needed to make a mature oocyte?  
+
+<|ref|>text<|/ref|><|det|>[[56, 170, 931, 300]]<|/det|>
+We agree that both femrDNA and RbpmS2 could both converge on regulation of ribosomes. Because rDNAs are required for ribosome nucleation and because femrDNA is unmethylated in ovary, femrDNA would promote increased nucleation of nucleoli in oocytes, a prerequisite for Poll recruitment. Because we see reduced Poll in rbpmS2 DMs, it is tempting to speculate that RbpmS2 might indirectly regulate femrDNA accessibility and are keen to test this possibility. Given the importance of rDNA for nucleation, we hypothesize that RbpmS2 would be positively influencing expression from the femrDNA locus to promote ribosome biogenesis through direct or indirect mechanisms or act downstream of femrDNA modification. However, we feel that testing these notions, although exciting, is beyond the scope of this work.  
+
+<|ref|>text<|/ref|><|det|>[[58, 315, 752, 332]]<|/det|>
+Can you rule out the model that 1. ribosomes are essential for making a mature oocyte;  
+
+<|ref|>text<|/ref|><|det|>[[57, 347, 586, 364]]<|/det|>
+We agree that ribosomes are essential for making mature oocytes.  
+
+<|ref|>text<|/ref|><|det|>[[56, 379, 936, 412]]<|/det|>
+2. without mature oocytes, some regulatory mechanism is disturbed that normally would prevent the ovary from transitioning to testis;  
+
+<|ref|>text<|/ref|><|det|>[[57, 427, 916, 460]]<|/det|>
+We agree that mature oocytes are important; however, in the absence of RbpmS2 oogenesis is perturbed at stages before mature oocytes are present in wild-type; thus, we think it plays a central role in differentiation.  
+
+<|ref|>text<|/ref|><|det|>[[57, 475, 866, 508]]<|/det|>
+3. That there are several genes that are independently necessary, each in a different way, for making ribosomes functional, including both femrDNA and Rboms2.  
+
+<|ref|>text<|/ref|><|det|>[[56, 523, 925, 603]]<|/det|>
+Without additional experiments, we cannot exclude that regulation of femrDNA is independent of RbpmS2. Further, although likely important, femrDNA modification has yet to be shown to be essential in functional assays. As mentioned above, our findings that ribosome biogenesis may be impaired by reduced RNA pol I recruitment, we hypothesize that promoting femrDNA locus accessibility (and thus nucleolar nucleation) could require RbpmS2 through direct or indirect mechanisms.  
+
+<|ref|>text<|/ref|><|det|>[[57, 618, 920, 652]]<|/det|>
+P16. Did the zebrafish spo11 mutations block DSBs and meiosis? The text told us that homozygous mothers gave embryos that didn't do well but was it shown definitively that this was because DSBs didn't occur?  
+
+<|ref|>text<|/ref|><|det|>[[56, 666, 930, 715]]<|/det|>
+The allele we utilized has been previously published and has shown to prevent DSBs in zebrafish testes, and to cause aneuploidy in oocytes. According, it is likely that DSBs are correspondingly blocked in zebrafish oocytes.  
+
+<|ref|>text<|/ref|><|det|>[[57, 730, 470, 747]]<|/det|>
+P16. 'inhibition of mTORC1 by conditional knockout'  
+
+<|ref|>text<|/ref|><|det|>[[56, 747, 936, 795]]<|/det|>
+Tell reader what was the condition was in the conditional ko. Probably it was a cell- type specific, presumably oocyte specific ko, but in a couple of words the text would avoid making the reader go to the original paper just to find out.  
+
+<|ref|>text<|/ref|><|det|>[[57, 810, 759, 827]]<|/det|>
+We have revised the text to clarify that this conditional knockout was in the mouse testis.  
+
+<|ref|>text<|/ref|><|det|>[[56, 842, 582, 858]]<|/det|>
+P16. 'which in oocytes is orchestrated in part by mTOR1 signaling'  
+
+<|ref|>text<|/ref|><|det|>[[57, 859, 904, 891]]<|/det|>
+Do we know that that mTOR1 signaling is due to action in oocytes or is it non- autonomous due to effects in other cell types or in organs other than the gonad, like liver or gonadotrophin secreting brain cells?
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[56, 44, 937, 92]]<|/det|>
+As mentioned above, because mTOR<sup>ca</sup> expression in germ cells alone (expressed using the germ cell specific ziwipromoter) was sufficient to restore oogenesis and support female sex differentiation, we hypothesize that this activity is essential in oocytes.  
+
+<|ref|>text<|/ref|><|det|>[[56, 108, 570, 140]]<|/det|>
+P16. 'using Crispr- Cas9 mutagenesis as in as detailed below 62' Fix wording.  
+
+<|ref|>text<|/ref|><|det|>[[57, 156, 300, 172]]<|/det|>
+We have made this correction.  
+
+<|ref|>text<|/ref|><|det|>[[56, 187, 627, 220]]<|/det|>
+P21. 'Sequencing data was aligned' Either sequencing data were aligned or sequencing datum was aligned.  
+
+<|ref|>text<|/ref|><|det|>[[56, 235, 451, 251]]<|/det|>
+We have corrected this in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[56, 267, 309, 299]]<|/det|>
+P21 'using the Illumina website' Text could give here the URL.  
+
+<|ref|>text<|/ref|><|det|>[[57, 315, 677, 332]]<|/det|>
+The URL, https://basespace.illumina.com, has been added as recommended.  
+
+<|ref|>text<|/ref|><|det|>[[56, 347, 936, 380]]<|/det|>
+P21. 'Only RNAs that appeared in both the cross-linked and uncrosslinked mApple- RbpmS2, but not in controls were considered RbpmS2 target RNAs.'  
+
+<|ref|>text<|/ref|><|det|>[[56, 380, 920, 428]]<|/det|>
+Good to have four replicates. Was there a minimum read count for concluding that a gene's transcripts were bound? Or was a single read sufficient to mean that a gene's transcripts were pulled down? Was there an adjusted p- value used to identify differentially pulled down transcripts between experimental and control?  
+
+<|ref|>text<|/ref|><|det|>[[56, 443, 920, 507]]<|/det|>
+We have added additional language to the RNAseq methods section to explain our rboRNA selection methodology. Specifically, we took an "all or none" approach - we only counted RNAs as bound if they were present in the mApple- RbpmS2 crosslinked and uncrosslinked samples and absent in mApple crosslinked and/or uncrosslinked controls.  
+
+<|ref|>text<|/ref|><|det|>[[57, 523, 216, 539]]<|/det|>
+P22 '(mios wildtype'  
+
+<|ref|>text<|/ref|><|det|>[[57, 540, 910, 572]]<|/det|>
+It would be less ambiguous to say homozygous mios wildtype, because it could have meant phenotypically wildtype, which would include heterozygotes. Likewise for homozygous mutants.  
+
+<|ref|>text<|/ref|><|det|>[[57, 587, 936, 620]]<|/det|>
+We have stated all of the genotypes used because we evaluated effects of the transgene on heterozygous and mutant fish compared to their homozygous wildtype siblings.  
+
+<|ref|>text<|/ref|><|det|>[[57, 635, 360, 652]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[56, 666, 940, 780]]<|/det|>
+Previous results implicate the RNA- binding protein RbpmS2 in ovary fate during zebrafish development. This work identifies RNA targets of RbpmS2 and, using a variety of approaches, supports a model whereby RbpmS2 promotes nucleolar amplification via TORC1, a step that supports oogenesis. In particular, the work examines the role of the GATOR2 component Mios in nucleolar development in oocytes, and its dependence on TOR to promote oogenesis independent of the TSC Rheb arm of TORC1. Nutrient availability had been implicated in oogenesis in other species, as well as in zebrafish sex determination. Impact of the work is high, for those interested in TOR signaling, germline development and the role of nutrients in oogenesis.  
+
+<|ref|>text<|/ref|><|det|>[[57, 795, 904, 827]]<|/det|>
+Several points listed below to clarify the results, strengthen the conclusions, and improve accessibility to a wider audience.  
+
+<|ref|>text<|/ref|><|det|>[[57, 842, 905, 859]]<|/det|>
+1. p.6-7: Refer to transcript abundance differences, rather than "stability". Elaborate on proposed feedback.  
+
+<|ref|>text<|/ref|><|det|>[[57, 874, 691, 891]]<|/det|>
+We thank the reviewer for the suggestion and have revised the text accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[57, 906, 832, 923]]<|/det|>
+2. Throughout manuscript: Define all terms and abbreviations (e.g., DM, H, M, HH, HM, MW, MH)
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[57, 43, 813, 61]]<|/det|>
+We thank the reviewer for noting this and have defined abbreviations in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[57, 75, 392, 92]]<|/det|>
+3. Figure 2: define white arrows in legend.  
+
+<|ref|>text<|/ref|><|det|>[[57, 108, 436, 125]]<|/det|>
+We have defined this in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[57, 139, 911, 174]]<|/det|>
+4. Figure 3: add quantification to support conclusion of loss of Ps6k localization in rbpm s2 DM; loss in mios mutant looks more convincing, but also needs quantification.  
+
+<|ref|>text<|/ref|><|det|>[[57, 187, 933, 222]]<|/det|>
+We thank the reviewer for this suggestion and have quantified the p- Ps6k and Mios in rbpm s2 DMs and added plots to Figure 3.  
+
+<|ref|>text<|/ref|><|det|>[[57, 235, 933, 304]]<|/det|>
+In the revised text, figures 3 and 5 were modified from an overview image of p- Ps6k in wildtype and rbpm s2 DMs or mios cells to magnified images of the nuclear localization in mitotic/early meiotic cells and oocytes to make the findings clearer. Overview images corresponding to those used for the magnified views are now in Supplemental Figure 2.  
+
+<|ref|>text<|/ref|><|det|>[[57, 314, 940, 402]]<|/det|>
+In figure 3, the overall brightness of the channels in the panels of the Mios staining was adjusted, and the adjacent panels that showed only Mios and DAPI were replaced with insets to make the phenotypes clearer to viewers. Further, quantification of Mios protein in wildtype and rbpm s2 DMs was added for consistency with the other quantifications performed (this data is in Figure 3 and Supp. Figure 6). In addition, the RNA names for Ps6k to rps6kb1a and rps6kb1b were corrected (they were originally written as rps6kb1a and rps6kb1b).  
+
+<|ref|>text<|/ref|><|det|>[[57, 428, 933, 480]]<|/det|>
+5. Figure 4g: add quantification to support differences in size of puncta. Presumably mios oocyte analysis was on the early arrested oocytes? Please clarify. Temper statement that mios is "required" for nucleologenesis since fibrillarin staining shows nucleoli.  
+
+<|ref|>text<|/ref|><|det|>[[57, 489, 936, 570]]<|/det|>
+We have added the quantification of the fibrillarin puncta to Figure 4 and have added quantification of the nuclear sizes of the cells used for the analysis to supplemental figure 6. Further, we have revised the language from "required for nucleologenesis" to "nuclear maturation" based on the data. In Figure 4, the overall brightness of the Fibrillarin staining was increased to make the phenotypes clearer to viewers.  
+
+<|ref|>text<|/ref|><|det|>[[57, 581, 925, 616]]<|/det|>
+In addition, we corrected the x- axes labels for plots Supp. Figure 4 g- h and the y- axis of Supp. Figure 6d due to scaling errors. These changes did not change the representation of the data.  
+
+<|ref|>text<|/ref|><|det|>[[57, 644, 925, 679]]<|/det|>
+6. Figure 5: change or clarify labeling of x axes in legend (H, M, etc., change to genotype; state meaning of + and - for transgene).  
+
+<|ref|>text<|/ref|><|det|>[[57, 705, 925, 739]]<|/det|>
+We have revised the use of H, M, etc. to the \(+ / +\) , \(+ / -\) , etc. convention. We have also revised the figure legend to clarify which fish are transgenic and non-transgenic.  
+
+<|ref|>text<|/ref|><|det|>[[57, 752, 664, 770]]<|/det|>
+7. Figure 7: correct the arrow from Gator1 to TORC as it should be negative.  
+
+<|ref|>text<|/ref|><|det|>[[57, 784, 554, 802]]<|/det|>
+Thank you. This has been corrected in the revised manuscript.  
+
+<|ref|>text<|/ref|><|det|>[[57, 815, 933, 849]]<|/det|>
+8. Title of final results section is misleading. State more clearly result concerning amino acid sensing (GATOR-dependent) versus TSC/Rheb arm.  
+
+<|ref|>text<|/ref|><|det|>[[57, 863, 910, 897]]<|/det|>
+We have edited this title and revised the concluding statement regarding the unique Gator2-Mios mediated activation of mTOR1 signaling in zebrafish oocytes.  
+
+<|ref|>text<|/ref|><|det|>[[57, 911, 730, 945]]<|/det|>
+9. Supplemental Figure 6: stated phenotypes difficult to see; add zoom box. We thank the reviewer for this feedback and have made the recommended changes.
+
+<--- Page Split --->
+<|ref|>sub_title<|/ref|><|det|>[[116, 89, 321, 105]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 119, 404, 134]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 148, 795, 179]]<|/det|>
+The authors have responded carefully to reviewers' concerns. The work will be a significant contribution to the field.  
+
+<|ref|>text<|/ref|><|det|>[[115, 193, 875, 224]]<|/det|>
+In the abstract, it is a bit of a stretch to state that the mammalian gonad is initially an ovary. It would be more accurate to state that is has a minor ovarian bias.  
+
+<|ref|>text<|/ref|><|det|>[[116, 253, 404, 268]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 283, 467, 298]]<|/det|>
+The authors have responded to all suggestions.  
+
+<|ref|>text<|/ref|><|det|>[[116, 327, 404, 342]]<|/det|>
+Reviewer #3 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 357, 617, 372]]<|/det|>
+The revisions have satisfactorily addressed my previous comments.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__450135318d771e1ad02d941c44ec73a4211958ceb573b2e79287eceef29d0695/images_list.json

@@ -0,0 +1,25 @@
+[
+  {
+    "type": "image",
+    "img_path": "images/Figure_1.jpg",
+    "caption": "Figure 1 for reviewer #1 (Figure 4e in the manuscript) Genome browser visualization of PFA ependymoma-specific DNA loops that associate two PFA enhancers (E1 and E2) with the ARL4C gene. Tracks for RNA-seq, H3K27ac, CTCF and Hi-C derived DNA loops are obtained by merging PFA (9EP1,9EP9, 7EP18) or ZFTA (11EP22, 4EP53, 7EP41), respectively.",
+    "footnote": [],
+    "bbox": [],
+    "page_idx": 0
+  },
+  {
+    "type": "image",
+    "img_path": "images/Figure_2.jpg",
+    "caption": "Figure 2 for reviewer #1. Boxplot showing NELFB gene expression in PFA and ZFTA tumors (RNA-seq gene expression data for \\(n = 117\\) ependymoma tumors). The center line, box limits, whiskers and points indicate the median, upper/lower quartiles, \\(1.5 \\times\\) interquartile range and outliers, respectively. DEG limma p-val.: 1.107e-07.",
+    "footnote": [],
+    "bbox": [
+      [
+        373,
+        94,
+        616,
+        280
+      ]
+    ],
+    "page_idx": 4
+  }
+]

peer_reviews/supplementary_0_Peer Review File__450135318d771e1ad02d941c44ec73a4211958ceb573b2e79287eceef29d0695/supplementary_0_Peer Review File__450135318d771e1ad02d941c44ec73a4211958ceb573b2e79287eceef29d0695.mmd

@@ -0,0 +1,174 @@
+
+# nature portfolio  
+
+Peer Review File  
+
+3D genome mapping identifies subgroup- specific chromosome conformations and tumor- dependency genes in ependymoma  
+
+![](images/Figure_1.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 --->
+
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications. Mentions of prior referee reports have been redacted. Mentions of the other journal have been redacted.  
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+## reviewer 1  
+
+There are interesting aspect of the revised manuscript. However, it still suffers from over- interpretation and claims of novelty (see below in bold and a point- by- point reply)  
+
+[redacted]  
+
+Reviewer 1 reply: I am not convinced about the usage of "novel" in structural variants (title) - and remain unconvinced about the (likely co- incidental) SVs near LAMC1. Moreover, without further functional analysis, these gene- regulatory dependencies remain putative  
+
+[redacted]  
+
+Reviewer 1 reply: The approach of requiring both Hi- C and WGS for SV calling is somewhat unintuitive. Hi- C has advantages over WGS and vice- versa. Line 143 suggests that Hi- C was used for SV calling, which was verified by WGS- based SV calling (line 150)  
+
+[redacted]  
+
+Reviewer 1 reply: Ok  
+
+[redacted]  
+
+Reviewer 1 reply: Ok  
+
+[redacted]  
+
+Reviewer 1 reply: What is the p- value provided in Fig 3c based on? Specific test should be performed comparing specific paired bins  
+
+[redacted]  
+
+Reviewer 1 reply: Ok  
+
+[redacted]  
+
+Reviewer 1 reply: I remain unconvinced that LAMC1 is not merely a dosage- gene. Fig S5h shows an expression increase from \(\sim 25\) RPKM to \(\sim 37\) RPKM of LAMC1 ( \(\sim 1.5\) - fold) in 1q- gain samples, which matches exactly a duplication, leading to an additional DNA copy of the
+
+<--- Page Split --->
+
+gene.  
+
+[redacted]  
+
+Reviewer 1 reply: I appreciate this analysis. Can the authors also please include the CTCF ChIP- seq data in Fig 5e.  
+
+The authors should perform the same analysis on ZFTA samples to demonstrate specificity. The expression and chromatin landscape appear similar for PFA and ZFTA.  
+
+In this respect, I find it unlikely that there are no ZFTA loops at all in the genomic view. If this is the case, the authors should state this explicitly.  
+
+A new set of analysis has been performed on NELFB. Why did the authors not also delete the CTCF binding site in ZFTA - again, to show specificity. In addition to the chromatin landscape, the expression of NELFB is almost similar for PFA and ZFTA (Fig S6o)  
+
+## reviewer 3  
+
+[redacted]  
+
+reviewer 3 reply: The fact that these are rare tumors does not merit circumstantial evidence and there is not sufficient evidence for any of these SVs in driving LAMC1 expression changes. LAMC1 is unconvincing  
+
+Reviewer #2 (Remarks to the Author):  
+
+The authors have sufficiently addressed my concerns. I still find the scope of this paper to be overly large. But I am supportive of publication at this point.
+
+<--- Page Split --->
+
+# Point-by-point replies to review Nature Communications [redacted]  
+
+## ## Reviewer 1  
+
+There are interesting aspect of the revised manuscript. However, it still suffers from overinterpretation and claims of novelty (see below in bold and a point- by- point reply).  
+
+[redacted]  
+
+Reviewer 1 reply: I am not convinced about the usage of "novel" in structural variants (title) – and remain unconvinced about the (likely co- incidental) SVs near LAMC1. Moreover, without further functional analysis, these gene- regulatory dependencies remain putative  
+
+Reply to the reviewer: Based on the reviewers' comments, we have now changed the title as follows:  
+
+New title: 3D genome mapping identifies subgroup- specific chromosome conformations and tumor- dependency genes in ependymoma.  
+
+[redacted]  
+
+Reviewer 1 reply: Ok  
+
+[redacted]  
+
+Reviewer 1 reply: The approach of requiring both Hi- C and WGS for SV calling is somewhat unintuitive. Hi- C has advantages over WGS and vice- versa. Line 143 suggests that Hi- C was used for SV calling, which was verified by WGS- based SV calling (line 150)  
+
+Reply to the reviewer: We agree that HiC has advantages in SV calling and may be able to detect SVs that are invisible or difficult to call in standard WGS. In this study, we present all SVs called by HiC as Supplementary Tables, however, we decided to focus on the reconstruction of neo- TADs and integration of the gene regulatory environment near SV breakpoints for selected SVs that we were able to validate by WGS.  
+
+[redacted]  
+
+Reviewer 1 reply: Ok  
+
+[redacted]  
+
+Reviewer 1 reply: Ok  
+
+[redacted]  
+
+Reviewer 1 reply: What is the p- value provided in Fig 3c based on? Specific test should be performed comparing specific paired bins
+
+<--- Page Split --->
+
+Reply to the reviewer: The corresponding \(p\) - value is representing the difference between paired bins and was computed using the R package DiffLoop (PMID: 29028898). We now name this tool in the extended legend of Figure 3c.  
+
+[redacted]  
+
+Reviewer 1 reply: Ok  
+
+[redacted]  
+
+Reviewer 1 reply: I remain unconvinced that LAMC1 is not merely a dosage- gene. Fig S5h shows an expression increase from \(\sim 25\) RPKM to \(\sim 37\) RPKM of LAMC1 (\~1.5- fold) in 1q- gain samples, which matches exactly a duplication, leading to an additional DNA copy of the gene.  
+
+Reply to the reviewer: Based on the reviewer's comment, we have removed all LAMC1- related material from the main Figures. We have moved the corresponding Figures to the Supplementary Material and stress in the updated manuscript that a functional relationship between SVs and LAMC1 transcription was not demonstrated.  
+
+[redacted]  
+
+Reviewer 1 reply: I appreciate this analysis. Can the authors also please include the CTCF ChIP- seq data in Fig 5e.  
+
+Reply to the reviewer: We have now added the CTCF ChIP- seq data tracks into the updated Figure 4e (previously Figure 5, Figure 1 for reviewer #1 below). The CTCF ChIP- seq data at the functionally tested CTCF binding site is already available, together with the WGBS data, in Fig 4f.  
+
+![](images/Figure_2.jpg)
+
+<center>Figure 1 for reviewer #1 (Figure 4e in the manuscript) Genome browser visualization of PFA ependymoma-specific DNA loops that associate two PFA enhancers (E1 and E2) with the ARL4C gene. Tracks for RNA-seq, H3K27ac, CTCF and Hi-C derived DNA loops are obtained by merging PFA (9EP1,9EP9, 7EP18) or ZFTA (11EP22, 4EP53, 7EP41), respectively. </center>  
+
+The authors should perform the same analysis on ZFTA samples to demonstrate specificity.
+
+<--- Page Split --->
+
+Reply to the reviewer: We initially performed the same analysis in ZFTA tumors, but despite similar chromosomal landscapes, we did not identify genomic regions with hypermethylation and CTCF loss in ZFTA that fit our filtering criteria (as also described in the Methods section). We now state this explicitly in the manuscript Methods part accordingly.  
+
+New text: "The DMRs were overlapped with CTCF differential peaks (min adj. p- value 0.05) resulting in 1254 pairs. The analysis for gene selection was performed on hypermethylated CTCF loci for PFA (n=966) and ZFTA (n=43) using two different approaches. From both of them no results were found for ZFTA, but only for PFA."  
+
+The expression and chromatin landscape appear similar for PFA and ZFTAs. In this respect, I find it unlikely that there are no ZFTA loops at all in the genomic view. If this is the case, the authors should state this explicitly.  
+
+Reply to the reviewer: The Figure showed only DNA loops between the E1 enhancer and the ARL4C gene that span the CTCF binding site, which is present in ZFTA tumors but replaced by DNA methylation in PFA tumors. Since these long- range DNA loops are only present in PFA ependymoma tumors, the ZFTA HiC track was empty. Based on the reviewers' comment, we now include all DNA loops passing \(q\) - value 0.05 limit without additional filtering (Figure 1 for reviewer #1, Figure 4e in the manuscript).  
+
+A new set of analysis has been performed on NELFB. Why did the authors not also delete the CTCF binding site in ZFTA - again, to show specificity. In addition to the chromatin landscape, the expression of NELFB is almost similar for PFA and ZFTA (Fig S6o).  
+
+Reply to the reviewer: These experiments were actually already performed, confirming the effect for NELFB: the target CTCF loci was removed in ZFTA cell line (Suppl. Fig. 7l- p) and the expression of the gene strongly increased (Suppl. Fig. 7n). We have not performed the experiment in PFA ependymoma, as the CTCF binding site was replaced by hypermethylation in this subtype.  
+
+Indeed, the difference in mean values is rather small between PFA and ZFTA, but strong variance is also observed. This could be an impact of array limitations as well as variance among PFA subgroups. Nevertheless from additional inspection of NELFB expression in ependymoma RNA- seq data from INFORM data cohort (PMID: 34373263), more evident effect was observed as shown in the figure below.
+
+<--- Page Split --->
+![PLACEHOLDER_6_0]
+
+<center>Figure 2 for reviewer #1. Boxplot showing NELFB gene expression in PFA and ZFTA tumors (RNA-seq gene expression data for \(n = 117\) ependymoma tumors). The center line, box limits, whiskers and points indicate the median, upper/lower quartiles, \(1.5 \times\) interquartile range and outliers, respectively. DEG limma p-val.: 1.107e-07. </center>  
+
+## ## reviewer 3  
+
+[redacted]  
+
+reviewer 3 reply: The fact that these are rare tumors does not merit circumstantial evidence and there is not sufficient evidence for any of these SVs in driving LAMC1 expression changes. LAMC1 is unconvincing  
+
+Reply to the reviewer: Based on the reviewers' comment, we have removed all LAMC1- related material from the main Figures. We have moved the corresponding Figures to the Supplementary Material and stress in the updated manuscript that a functional relationship between SVs and LAMC1 transcription was not demonstrated.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+The authors have sufficiently addressed my concerns. I still find the scope of this paper to be overly large. But I am supportive of publication at this point.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__450135318d771e1ad02d941c44ec73a4211958ceb573b2e79287eceef29d0695/supplementary_0_Peer Review File__450135318d771e1ad02d941c44ec73a4211958ceb573b2e79287eceef29d0695_det.mmd

@@ -0,0 +1,242 @@
+<|ref|>title<|/ref|><|det|>[[99, 40, 506, 90]]<|/det|>
+# nature portfolio  
+
+<|ref|>text<|/ref|><|det|>[[106, 110, 373, 140]]<|/det|>
+Peer Review File  
+
+<|ref|>text<|/ref|><|det|>[[108, 162, 894, 220]]<|/det|>
+3D genome mapping identifies subgroup- specific chromosome conformations and tumor- dependency genes in ependymoma  
+
+<|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|>text<|/ref|><|det|>[[115, 90, 857, 157]]<|/det|>
+Editorial Note: This manuscript has been previously reviewed at another journal that is not operating a transparent peer review scheme. This document only contains reviewer comments and rebuttal letters for versions considered at Nature Communications. Mentions of prior referee reports have been redacted. Mentions of the other journal have been redacted.  
+
+<|ref|>sub_title<|/ref|><|det|>[[116, 196, 320, 211]]<|/det|>
+## REVIEWERS' COMMENTS  
+
+<|ref|>text<|/ref|><|det|>[[116, 226, 404, 241]]<|/det|>
+Reviewer #1 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[116, 256, 223, 271]]<|/det|>
+## reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[115, 281, 785, 311]]<|/det|>
+There are interesting aspect of the revised manuscript. However, it still suffers from over- interpretation and claims of novelty (see below in bold and a point- by- point reply)  
+
+<|ref|>text<|/ref|><|det|>[[116, 325, 195, 341]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[115, 350, 870, 412]]<|/det|>
+Reviewer 1 reply: I am not convinced about the usage of "novel" in structural variants (title) - and remain unconvinced about the (likely co- incidental) SVs near LAMC1. Moreover, without further functional analysis, these gene- regulatory dependencies remain putative  
+
+<|ref|>text<|/ref|><|det|>[[116, 425, 195, 441]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[115, 465, 875, 511]]<|/det|>
+Reviewer 1 reply: The approach of requiring both Hi- C and WGS for SV calling is somewhat unintuitive. Hi- C has advantages over WGS and vice- versa. Line 143 suggests that Hi- C was used for SV calling, which was verified by WGS- based SV calling (line 150)  
+
+<|ref|>text<|/ref|><|det|>[[116, 525, 195, 541]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[116, 565, 290, 580]]<|/det|>
+Reviewer 1 reply: Ok  
+
+<|ref|>text<|/ref|><|det|>[[116, 594, 195, 610]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[116, 634, 290, 650]]<|/det|>
+Reviewer 1 reply: Ok  
+
+<|ref|>text<|/ref|><|det|>[[116, 664, 195, 680]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[115, 704, 863, 735]]<|/det|>
+Reviewer 1 reply: What is the p- value provided in Fig 3c based on? Specific test should be performed comparing specific paired bins  
+
+<|ref|>text<|/ref|><|det|>[[116, 749, 195, 764]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[116, 788, 290, 804]]<|/det|>
+Reviewer 1 reply: Ok  
+
+<|ref|>text<|/ref|><|det|>[[116, 818, 195, 833]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[115, 857, 880, 904]]<|/det|>
+Reviewer 1 reply: I remain unconvinced that LAMC1 is not merely a dosage- gene. Fig S5h shows an expression increase from \(\sim 25\) RPKM to \(\sim 37\) RPKM of LAMC1 ( \(\sim 1.5\) - fold) in 1q- gain samples, which matches exactly a duplication, leading to an additional DNA copy of the
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[115, 91, 161, 104]]<|/det|>
+gene.  
+
+<|ref|>text<|/ref|><|det|>[[115, 120, 194, 134]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[115, 135, 866, 164]]<|/det|>
+Reviewer 1 reply: I appreciate this analysis. Can the authors also please include the CTCF ChIP- seq data in Fig 5e.  
+
+<|ref|>text<|/ref|><|det|>[[115, 164, 870, 190]]<|/det|>
+The authors should perform the same analysis on ZFTA samples to demonstrate specificity. The expression and chromatin landscape appear similar for PFA and ZFTA.  
+
+<|ref|>text<|/ref|><|det|>[[115, 191, 866, 222]]<|/det|>
+In this respect, I find it unlikely that there are no ZFTA loops at all in the genomic view. If this is the case, the authors should state this explicitly.  
+
+<|ref|>text<|/ref|><|det|>[[115, 237, 856, 281]]<|/det|>
+A new set of analysis has been performed on NELFB. Why did the authors not also delete the CTCF binding site in ZFTA - again, to show specificity. In addition to the chromatin landscape, the expression of NELFB is almost similar for PFA and ZFTA (Fig S6o)  
+
+<|ref|>text<|/ref|><|det|>[[115, 333, 224, 346]]<|/det|>
+## reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[115, 370, 194, 385]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[115, 410, 880, 456]]<|/det|>
+reviewer 3 reply: The fact that these are rare tumors does not merit circumstantial evidence and there is not sufficient evidence for any of these SVs in driving LAMC1 expression changes. LAMC1 is unconvincing  
+
+<|ref|>text<|/ref|><|det|>[[115, 486, 404, 500]]<|/det|>
+Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[115, 515, 860, 545]]<|/det|>
+The authors have sufficiently addressed my concerns. I still find the scope of this paper to be overly large. But I am supportive of publication at this point.
+
+<--- Page Split --->
+<|ref|>title<|/ref|><|det|>[[118, 85, 641, 100]]<|/det|>
+# Point-by-point replies to review Nature Communications [redacted]  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 131, 228, 145]]<|/det|>
+## ## Reviewer 1  
+
+<|ref|>text<|/ref|><|det|>[[118, 162, 808, 194]]<|/det|>
+There are interesting aspect of the revised manuscript. However, it still suffers from overinterpretation and claims of novelty (see below in bold and a point- by- point reply).  
+
+<|ref|>text<|/ref|><|det|>[[118, 224, 200, 240]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 254, 880, 302]]<|/det|>
+Reviewer 1 reply: I am not convinced about the usage of "novel" in structural variants (title) – and remain unconvinced about the (likely co- incidental) SVs near LAMC1. Moreover, without further functional analysis, these gene- regulatory dependencies remain putative  
+
+<|ref|>text<|/ref|><|det|>[[118, 316, 878, 332]]<|/det|>
+Reply to the reviewer: Based on the reviewers' comments, we have now changed the title as follows:  
+
+<|ref|>text<|/ref|><|det|>[[118, 346, 878, 378]]<|/det|>
+New title: 3D genome mapping identifies subgroup- specific chromosome conformations and tumor- dependency genes in ependymoma.  
+
+<|ref|>text<|/ref|><|det|>[[118, 395, 200, 410]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 425, 280, 440]]<|/det|>
+Reviewer 1 reply: Ok  
+
+<|ref|>text<|/ref|><|det|>[[118, 457, 200, 472]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 502, 880, 550]]<|/det|>
+Reviewer 1 reply: The approach of requiring both Hi- C and WGS for SV calling is somewhat unintuitive. Hi- C has advantages over WGS and vice- versa. Line 143 suggests that Hi- C was used for SV calling, which was verified by WGS- based SV calling (line 150)  
+
+<|ref|>text<|/ref|><|det|>[[118, 565, 880, 644]]<|/det|>
+Reply to the reviewer: We agree that HiC has advantages in SV calling and may be able to detect SVs that are invisible or difficult to call in standard WGS. In this study, we present all SVs called by HiC as Supplementary Tables, however, we decided to focus on the reconstruction of neo- TADs and integration of the gene regulatory environment near SV breakpoints for selected SVs that we were able to validate by WGS.  
+
+<|ref|>text<|/ref|><|det|>[[118, 660, 200, 675]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 692, 280, 707]]<|/det|>
+Reviewer 1 reply: Ok  
+
+<|ref|>text<|/ref|><|det|>[[118, 723, 200, 738]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 755, 280, 770]]<|/det|>
+Reviewer 1 reply: Ok  
+
+<|ref|>text<|/ref|><|det|>[[118, 787, 200, 802]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 819, 880, 851]]<|/det|>
+Reviewer 1 reply: What is the p- value provided in Fig 3c based on? Specific test should be performed comparing specific paired bins
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 85, 880, 133]]<|/det|>
+Reply to the reviewer: The corresponding \(p\) - value is representing the difference between paired bins and was computed using the R package DiffLoop (PMID: 29028898). We now name this tool in the extended legend of Figure 3c.  
+
+<|ref|>text<|/ref|><|det|>[[118, 148, 200, 164]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 180, 280, 196]]<|/det|>
+Reviewer 1 reply: Ok  
+
+<|ref|>text<|/ref|><|det|>[[118, 212, 200, 227]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 243, 880, 292]]<|/det|>
+Reviewer 1 reply: I remain unconvinced that LAMC1 is not merely a dosage- gene. Fig S5h shows an expression increase from \(\sim 25\) RPKM to \(\sim 37\) RPKM of LAMC1 (\~1.5- fold) in 1q- gain samples, which matches exactly a duplication, leading to an additional DNA copy of the gene.  
+
+<|ref|>text<|/ref|><|det|>[[118, 323, 880, 387]]<|/det|>
+Reply to the reviewer: Based on the reviewer's comment, we have removed all LAMC1- related material from the main Figures. We have moved the corresponding Figures to the Supplementary Material and stress in the updated manuscript that a functional relationship between SVs and LAMC1 transcription was not demonstrated.  
+
+<|ref|>text<|/ref|><|det|>[[118, 403, 200, 419]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 434, 880, 467]]<|/det|>
+Reviewer 1 reply: I appreciate this analysis. Can the authors also please include the CTCF ChIP- seq data in Fig 5e.  
+
+<|ref|>text<|/ref|><|det|>[[118, 482, 880, 531]]<|/det|>
+Reply to the reviewer: We have now added the CTCF ChIP- seq data tracks into the updated Figure 4e (previously Figure 5, Figure 1 for reviewer #1 below). The CTCF ChIP- seq data at the functionally tested CTCF binding site is already available, together with the WGBS data, in Fig 4f.  
+
+<|ref|>image<|/ref|><|det|>[[125, 552, 880, 808]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[130, 810, 866, 850]]<|/det|>
+<center>Figure 1 for reviewer #1 (Figure 4e in the manuscript) Genome browser visualization of PFA ependymoma-specific DNA loops that associate two PFA enhancers (E1 and E2) with the ARL4C gene. Tracks for RNA-seq, H3K27ac, CTCF and Hi-C derived DNA loops are obtained by merging PFA (9EP1,9EP9, 7EP18) or ZFTA (11EP22, 4EP53, 7EP41), respectively. </center>  
+
+<|ref|>text<|/ref|><|det|>[[118, 861, 817, 878]]<|/det|>
+The authors should perform the same analysis on ZFTA samples to demonstrate specificity.
+
+<--- Page Split --->
+<|ref|>text<|/ref|><|det|>[[118, 84, 880, 148]]<|/det|>
+Reply to the reviewer: We initially performed the same analysis in ZFTA tumors, but despite similar chromosomal landscapes, we did not identify genomic regions with hypermethylation and CTCF loss in ZFTA that fit our filtering criteria (as also described in the Methods section). We now state this explicitly in the manuscript Methods part accordingly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 163, 880, 228]]<|/det|>
+New text: "The DMRs were overlapped with CTCF differential peaks (min adj. p- value 0.05) resulting in 1254 pairs. The analysis for gene selection was performed on hypermethylated CTCF loci for PFA (n=966) and ZFTA (n=43) using two different approaches. From both of them no results were found for ZFTA, but only for PFA."  
+
+<|ref|>text<|/ref|><|det|>[[118, 243, 880, 291]]<|/det|>
+The expression and chromatin landscape appear similar for PFA and ZFTAs. In this respect, I find it unlikely that there are no ZFTA loops at all in the genomic view. If this is the case, the authors should state this explicitly.  
+
+<|ref|>text<|/ref|><|det|>[[118, 307, 880, 402]]<|/det|>
+Reply to the reviewer: The Figure showed only DNA loops between the E1 enhancer and the ARL4C gene that span the CTCF binding site, which is present in ZFTA tumors but replaced by DNA methylation in PFA tumors. Since these long- range DNA loops are only present in PFA ependymoma tumors, the ZFTA HiC track was empty. Based on the reviewers' comment, we now include all DNA loops passing \(q\) - value 0.05 limit without additional filtering (Figure 1 for reviewer #1, Figure 4e in the manuscript).  
+
+<|ref|>text<|/ref|><|det|>[[118, 418, 880, 467]]<|/det|>
+A new set of analysis has been performed on NELFB. Why did the authors not also delete the CTCF binding site in ZFTA - again, to show specificity. In addition to the chromatin landscape, the expression of NELFB is almost similar for PFA and ZFTA (Fig S6o).  
+
+<|ref|>text<|/ref|><|det|>[[118, 499, 880, 562]]<|/det|>
+Reply to the reviewer: These experiments were actually already performed, confirming the effect for NELFB: the target CTCF loci was removed in ZFTA cell line (Suppl. Fig. 7l- p) and the expression of the gene strongly increased (Suppl. Fig. 7n). We have not performed the experiment in PFA ependymoma, as the CTCF binding site was replaced by hypermethylation in this subtype.  
+
+<|ref|>text<|/ref|><|det|>[[118, 578, 880, 658]]<|/det|>
+Indeed, the difference in mean values is rather small between PFA and ZFTA, but strong variance is also observed. This could be an impact of array limitations as well as variance among PFA subgroups. Nevertheless from additional inspection of NELFB expression in ependymoma RNA- seq data from INFORM data cohort (PMID: 34373263), more evident effect was observed as shown in the figure below.
+
+<--- Page Split --->
+<|ref|>image<|/ref|><|det|>[[373, 94, 616, 280]]<|/det|>
+<|ref|>image_caption<|/ref|><|det|>[[133, 287, 863, 327]]<|/det|>
+<center>Figure 2 for reviewer #1. Boxplot showing NELFB gene expression in PFA and ZFTA tumors (RNA-seq gene expression data for \(n = 117\) ependymoma tumors). The center line, box limits, whiskers and points indicate the median, upper/lower quartiles, \(1.5 \times\) interquartile range and outliers, respectively. DEG limma p-val.: 1.107e-07. </center>  
+
+<|ref|>sub_title<|/ref|><|det|>[[118, 386, 225, 400]]<|/det|>
+## ## reviewer 3  
+
+<|ref|>text<|/ref|><|det|>[[118, 434, 199, 450]]<|/det|>
+[redacted]  
+
+<|ref|>text<|/ref|><|det|>[[118, 466, 880, 514]]<|/det|>
+reviewer 3 reply: The fact that these are rare tumors does not merit circumstantial evidence and there is not sufficient evidence for any of these SVs in driving LAMC1 expression changes. LAMC1 is unconvincing  
+
+<|ref|>text<|/ref|><|det|>[[118, 530, 880, 593]]<|/det|>
+Reply to the reviewer: Based on the reviewers' comment, we have removed all LAMC1- related material from the main Figures. We have moved the corresponding Figures to the Supplementary Material and stress in the updated manuscript that a functional relationship between SVs and LAMC1 transcription was not demonstrated.  
+
+<|ref|>sub_title<|/ref|><|det|>[[119, 625, 411, 640]]<|/det|>
+## Reviewer #2 (Remarks to the Author):  
+
+<|ref|>text<|/ref|><|det|>[[118, 657, 880, 689]]<|/det|>
+The authors have sufficiently addressed my concerns. I still find the scope of this paper to be overly large. But I am supportive of publication at this point.
+
+<--- Page Split --->

peer_reviews/supplementary_0_Peer Review File__45142058222c4ee18640002c7658fa05a38dbf6116a638ff9ce04deab9bdf06e/images_list.json

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

peer_reviews/supplementary_0_Peer Review File__45142058222c4ee18640002c7658fa05a38dbf6116a638ff9ce04deab9bdf06e/supplementary_0_Peer Review File__45142058222c4ee18640002c7658fa05a38dbf6116a638ff9ce04deab9bdf06e.mmd

@@ -0,0 +1,550 @@
+
+## REVIEWER COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+The distribution, metabolism, and importance of 2AEP in marine ecosystems have long been recognised by the scientific community. Nevertheless, the ecological and biogeochemical role of this natural occurring phosphonate has just begun to be understood. Despite our advance knowledge on the subject, many features on the metabolism of 2AEP remain still undiscovered. For example, the ability of 2AEP- degrading microorganisms to import this phosphonate into the cell, that, however, lack the specific transporters for phosphonates. Therefore, the ability to expand our knowledge on the acquisition and catabolism of phosphonates by microorganisms is a crucial step to comprehend the dynamics of natural occurrence phosphonates in nature fully. Your manuscript describes, in a very structured study and with a well- performed methodology, the discovery of these novel bacterial transporter systems, which will contribute to close the existing gaps in the known metabolism of 2AEP and the role of organophosphonates in the biogeochemical cycle of phosphorus. Therefore, your contribution to this specific topic of interest, which has attracted the attention of many respected scientists since the identification of phosphonates in the seawater column in the early 2000s, will be universally recognised. However, I must say that the current manuscript's title is quite ambitious and does not reflect the extension of the experimental results presented in this work. Therefore, more experimental work is needed at the field level to prove the statement.  
+
+Throughout the manuscript, there are a couple of specific gaps in the general knowledge and state of the art about the metabolism of 2AEP that I would like to address in order to increase the quality of your manuscript:  
+
+The catabolism of 2AEP by microorganisms is particularly interesting as it can be achieved by different pathways, such as 2- aminoethylphosphonate dioxygenase (PhnY\\*), phosphonoacetaldehyde hydrolase (phosphonatase - PhnX), and phosphonoacetate hydrolase (PhnA). Interestingly, the operons that contain the genes encoding for either PhnX or PhnA usually contain genes that encode for a 2- AEP transaminase (PhnW) and/or a phosphonoacetaldehyde dehydrogenase (PhnY), which are accessory proteins that participate on the degradation of 2AEP (Agarwal et al. 2014). However, it must be considered that these operons are known to be regulated under either the PhoBR regulatory system related to the Pho regulon or by a LysR- like transcriptional activator responsive to 2AEP, which enable the microorganisms to activate both pathways independently of ambient Pi concentrations (Kulakova et al., 2001, 2009, Cooley et al. 2011, Borisova et al. 2011). Therefore, it is highly recommended that authors should analyse the flanking regions of the operons found in Pseudomonas putida BIRD- 1, Stappia stellulata DMS 5886, and other Pi- insensitive 2AEP- degrading microorganisms to determine if these species possess a LysR type of regulator, as it is known that the majority of phnX operons present in Pseudomonas and phnA in alpha- proteobacteria are associated with this substrate- inducible type or regulator (Villarreal- Chiu et al. 2012).  
+
+The phylum proteobacteria is known to excel in accumulating metabolic pathways to degrade natural and xenobiotic compounds, including phosphonates. The existence of different phosphonate degradation pathways or multiple copies of phosphonate degradation genes in members of this genus is well known (Enterobacter aerogenes: Lee et al. 1992, Salmonella typhimurium: Jiang et al. 1995, Pseudomonas stutzeri: White and Metcalf 2004, Mesorhiobium loti: Huang et al. 2005). This concept is valid for the presence of multiple copies of the phosphonate transporter gene phnD (Prochlorococcus marinus: Feingersch et al. 2012). However, it has been reported that a vast number of bacteria that possessed genes associated with any known phosphonate degradation pathway showed no evidence of possessing the PhnCDE route of phosphonate acquisition (Villarreal- Chiu et al. 2012), which can contribute to exalt the importance of this manuscript. On this regard, it must be mentioned that PhnD has been demonstrated to show the highest affinity to 2AEP. However, this protein shows high specificity to other natural occurring phosphonates as well (Rizk et al. 2006). Therefore, authors must demonstrate those novel proteins AepX and AepP are specific to 2AEP, as stated in the manuscript. Otherwise, authors should confirm that these proteins exhibit a group
+
+<--- Page Split --->
+
+specificity to phosphonates, similarly as PhnD.  
+
+Microorganisms have evolved several mechanisms to acquire and conserve P as an adaptation to cope with the ever- changing P concentrations in the environment. These mechanisms are known to be related not just with the Pi levels of the surrounding environment (Pho regulon: Kononova and Nesmeyanova, 2002), but also with nitrogen, iron, oxygen or even light (Dyhrman et al. 2007). A clear example of these mechanisms of acquisition and conservation of P as a response to P and/or N levels, is the ability of microorganisms to accumulate polyhydroxyalkanoates (PHA: Valentino et al. 2015). These are a group of natural polymers accumulated intracellularly as a carbon and energy storage material (Tobin and O'Connor, 2005). PHA production and accumulation has been demonstrated to be widespread among marine microorganisms (Uwamori- Takahashi et al. 2017, Ganapathy et al., 2018). This information is relevant for the present work, as experimentally, PHA production and accumulation by microorganisms can be detected by an increase of turbidity in the culture. This occurs as the PHA internal granules tend to increase against time in response to a nutrient limitation in the presence of excess carbon. This increase of turbidity in the culture can produce false- positive results due to the inability to discriminate between PHA accumulation and biomass production (Acosta- Cortés et al. 2019). On this regard, it is possible that this phenomenon is occurring on the marine bacterial strains tested for 2AEP supplemented as N source, as the utilisation of 2AEP as P is significantly higher to the exceptionally low OD540 obtained for 2AEP cultures supplemented as N source (Table S1). An easy and rapid method to discern the presence of PHA is by visualising it under fluorescent microscopy using a Nile red or Nile blue A 1% staining (Giin- Yu et al. 2014). Demonstrating that Alliroseovarius strains do not accumulate PHA when grown on 2AEP supplied as N source, would significantly improve the hypothesis of the existence of a new catabolic pathway for 2AEP.  
+
+The bioinformatic analysis of the abundance of phosphonate metabolic pathways among microorganisms have been widely studied (Huang et al. 2005, Villarreal- Chiu et al. 2012). With their new approach, it is recommended that authors should combine their results on the distribution of aepX and aepP across bacterial taxa and the results of previous studies on the distribution of catabolic pathways. Many potential new transporters or catabolic pathways may be found with this analysis. On the other hand, while authors' approach on the transcript abundance of aepX on metagenomic data is valid, it is essential to compare these results with a housekeeping gene in order to normalise the relative abundance of all compared genes against the number of resultants obtained for a housekeeping gene, generally, recA, used as a single- copy- per- genome control in distribution and prevalence analyses (Moran et al., 2004).  
+
+Finally, references 75, 76, and 77 are missing from the reference list.  
+
+References used in this revision: Agarwal et al. 2014: Chem Biol, 21; DOI: 10.1016/j.chembiol.2013.11.006 Kulakova et al., 2001: J Bacteriol, 183: 3268- 3275. DOI: 10.1128/JB.183.11.3268- 3275.2001 Kulakova et al., 2009: Microb Biotechnol, 2: 234- 240 DOI: 10.1111/j.1751- 7915.2008.00082. x Cooley et al. 2011: Microbiology, 80: 335- 340; DOI: 10.1134/S0026261711030076 Borisova et al. 2011: J Biol Chem, 286: 22283- 22290; DOI: 10.1074/jbc.M111.237735 Villarreal- Chiu et al. 2012: Front Microbiol, 3: 19; DOI: 10.3389/fmicb.2012.00019 Lee et al. 1992: J Bacteriol, 174: 2501- 2510; DOI: 10.1128/jb.174.8.2501- 2510.1992 Jiang et al. 1995: J Bacteriol, 177: 6411- 6421; DOI: 10.1128/jb.177.22.6411- 6421.1995 White and Metcalf 2004: J Bacteriol, 186: 4730- 4739; DOI: 10.1128/jb.186.14.4730- 4739.2004 Huang et al. 2005: J Mol Evol, 61: 682- 690; DOI: 10.1007/s00239- 004- 0349- 4 Feingersch et al. 2012: ISME J, 6: 827- 834. DOI: 10.1038/ismej.2011.149 Rizk et al. 2006: Protein Sci, 15: 1745- 1751. DOI: 10.1110/ps.062135206 Kononova and Nesmeyanova, 2002: Biochemistry, 67: 184- 95; DOI: 10.1023/A:1014409929875 Dyhrman et al. 2007: Oceanography, 20: 110- 116; DOI: 10.5670/oceanog.2007.54
+
+<--- Page Split --->
+
+Valentino et al. 2015: Water Res, 77: 49- 63; DOI: 10.1016/j.watres.2015.03.016Tobin and O'Connor, 2005: FEMS Microbiol Lett, 253: 111- 118; DOI: 10.1016/j.femsle.2005.09.025Uwamori- Takahashi et al. 2017: Bioeng, 4:60; DOI: 10.3390/bioengineering4030060Ganapathy et al. 2018: Int J Biol Macromol, 111: 102- 108; DOI: 10.1016/j.ijbiomac.2017.12.155Acosta- Cortés et al. 2019: ISME J, 13: 1497- 1505; DOI: 10.1038/s41396- 019- 0366- 3Giin- Yu et al. 2014: Polymers, 6: 706- 754; DOI: 10.3390/polym6030706Moran et al. 2004: Nature, 432: 910- 913. DOI: 10.1038/nature03170.  
+
+Reviewer #2 (Remarks to the Author):  
+
+This manuscript describes the discovery and characterisation of several new transporters for 2- aminoethylphosphonate (2AEP), coupled with bioinformatic studies of their prevalence and correlation with environmental factors and phosphonate catabolism genes. The characterisation of the transporters themselves is interesting, but potentially more interesting is the demonstration that these are likely to contribute to phosphate- insensitive metabolism of 2AEP in the marine environment: while this catabolism has been previously shown in isolates, there wasn't clear data on the extent of it in the environment. This is likely to expand the interest in the manuscript beyond microbiologists and into a much broader biogeochemistry/oceanography community, and further supports the recent re- evaluation of the role of reduced phosphorus in the oceans. The experimental approach in the work seems appropriate and is well detailed, and the conclusions are justified. Overall I think this manuscript would make an important contribution to the field.  
+
+I have a few core suggestions:  
+
+The authors investigate the relationship between the new transporters and PhnA/X/J, and use this to infer that 2AEP- specific mineralisation is a more prevalent process than non- specific phosphonate metabolism via C- P lyase. Is there a reason that PhnZ was omitted from this comparison? It is present in Figure 1 but not examined further.  
+
+On line 60 phosphonatases are described as "phosphonate degradation systems", and this word is used throughout the text to refer to the action of PhnA, PhnX, or C- P lyase. The term phosphonatase was coined by La Nauze et al. (Biochimica et Biophysica Acta Enzymology, 212[2], 1970) as the trivial name for phosphonoacetaldehyde hydrolase/PhnX specifically, and largely isn't used to mean phosphonate degradation enzymes generally (e.g. see reviews by Horsman and Zechel [2017, Chemical Reviews, 117(8)], McGrath et al. [2013, Nature Reviews Microbiology, 412], or Peck and van der Donk [2013, Current Opinion in Chemical Biology, 17], which all use phosphonatase exclusively to refer to PhnX). It would be more in keeping with the wider phosphonate literature to avoid that term when describing PhnA/C- P lyase.  
+
+Line 213 states "aepX transcription was 40- fold and 350- fold greater than phnD in the epipelagic and mesopelagic, respectively (Fig 5B)." The 40- fold number is fine, but Fig. 5B doesn't seem to show a 350- fold difference in aepX and phnD in the mesopelagic, it looks closer to \(\sim 130\) - fold (0.3 vs 40). Is this value correct, or am I misinterpreting something?  
+
+The sentence beginning on Line 217 discusses metagenome/transcriptome data for phnA, J and X, but doesn't refer to Figure S8 where this data is shown.  
+
+Line 33 states "Collectively, our data identifies a mechanism responsible for the oxidative step in the marine phosphorus redox cycle". Slightly nitpicky, but saying "the oxidative step" implies that this is the only oxidation reaction which phosphorus goes through in the ocean, which isn't correct: aside from phosphonate oxidation systems there are separate enzymes for phosphite and hypophosphite
+
+<--- Page Split --->
+
+oxidation present in marine organisms (e.g. Martinez et al., Environmental Microbiology, 2011, 14(6), pp1363- 77). Perhaps rephrase this along the lines of "for a major oxidation process"?  
+
+Line 216: "The majority of aepX sequences were related to the cosmopolitan Alphaproteobacteria and Deltaproteobacteria (Fig 4)." Figure 4 appears to show IM- RGC sequences are mostly in the Alpha and Gammaproteobacteria, not the Deltas?  
+
+Figure S8F appears to be missing metatranscriptomic data for the Southern Ocean. Figure 5C/D also seem to be missing matching MG/MT data for certain areas (e.g. aepP data for the Southern Ocean in 5C, aepP MT data for the SAO and MS in 5C, all MT data for SO in 5D). Are these occluded by other symbols or are they missing from the figures/datasets?  
+
+Other comments:  
+
+Line 55: When arguing that 2AEP is absent from HWM DOP the authors cite article 17 (Sosa et al.). The article in question doesn't state that 2AEP wasn't detected in their sample, only that MPn and 2- HEP were present as the major components along with "minor unidentified phosphonates" (which may have included 2AEP). It may be better to rephrase this sentence to suggest that 2AEP is either absent or present in significantly lower proportions than other phosphonates, thus the ubiquitous synthesis genes would suggest preferential degradation.  
+
+In Figure 2A, the colour of the PhnY symbol in the key appears to be darker than the colour used in the diagram itself.  
+
+Line 163: "To confirm that Stappia AepXVW" should read "To confirm that Stappia stellulata AepXVW"  
+
+Line 100: "BIRD- 1 was capable of growth on 1.5 mM 2AEP as either a sole N, P, or N and P source, the latter resulting in mineralisation of Pi which was subsequently exported from the cell". Was P export from the cell measured when 2AEP was provided as the sole N source? Previous literature (e.g. citation 22/Chin et al.) would suggest that P export should be observed here as well, and it isn't clear from the text if this didn't occur or just wasn't measured.  
+
+Fig. S3: What is the significance of the dashed line at an abundance of 27? This should be stated in the figure legend.  
+
+Figure 5's legend is missing a statement identifying the error bars.  
+
+Line 316 states "Co- culture experiments were carried out according to the protocol described in60". It's not clear to me where in the manuscript co- cultures were performed/described, certainly not in the context of the work performed in citation 60. Could the authors clarify this?  
+
+Reviewer #3 (Remarks to the Author):  
+
+The manuscript "Aminophosphonate mineralisation is a major step in the global oceanic phosphorus redox cycle" uses a combination of involved gene knockout/complementation culture experiments, proteomics, and other multi- OMIC analyses to identify and characterize multiple bacterial aminoethylphosphonate (AEP) transporters. Through their careful analyses they evaluate what controls the expression of these transporters, the fact that the different transporters appear to be involved with either using AEP as a nitrogen (N) or phosphorus (P) source, as well as their ubiquity in both published genomes and global ocean datasets. This is an impressive piece of work and of critical importance to the oceanographic community as it relates to the availability of two essential nutrients
+
+<--- Page Split --->
+
+for phytoplankton growth.  
+
+That said, I do have several comments on the current manuscript's structure that I feel need to be addressed before publication. The first has to do with the title. The work only focused on AEP, not aminophosphonates in general. While AEP is an example of an aminophosphonate this may be somewhat nit- picking, but I think that Aminoethylphosphonate should replace aminophosphonate in the title. Also, the article focuses much of its discussion on the role in the phosphorus redox cycle, but it's clear from their work that bacteria can use AEP as an N source (and when doing so can release inorganic P), would it be more appropriate to have the title and abstract indicate AEP may be an important N source as well. Global modeling efforts suggest a much larger region of the ocean is N limited than P limited. It is interesting that in meeting N demands, bacteria may release bioavailable inorganic P.  
+
+As to the structure of the manuscript. There is an excessive use of abbreviations. I understand that the authors have limited space to tell a large story, but it makes the manuscript very hard to read. Some of this may be alleviated by having the figures (especially figure 1 which has all of the gene abbreviations) closer to the text. The gene name abbreviations are obviously necessary, as are many others, but in thinking about the fact that Nature Communications readership is more broad, it might behoove the authors to question whether it is necessary to introduce the abbreviation for something like MPn and SBP, which are only used a few times in the manuscript and are not common abbreviations outside of a small subset of the field. Also, there's definitely cases where an abbreviation is used that hasn't been introduced and a whole word is used when an abbreviation is in use (and not just at the start of a sentence). I have some other general comments. In the results/discussion of the ubiquity of these transporters in the global datasets, it looks like there's a difference in AepX types (BIRD- 1 and Stappia), it looks like it is the Stappia variant that is more abundant in the open ocean. AepX was Pi sensitive in BIRD- 1 and BIRD- 1 also had AepP, which if I interpret the results correctly was much rarer in both published genomes and the TARA dataset. Did you explore via MAGs or IMG if the organisms that had AepP had a similar variant of AepX as BIRD- 1? It seems like an interesting angle to pursue. When discussing the MG and MT data, all focus is on the role of Pi in potentially controlling abundance/expression. AEP is a potential N source. Considering there are some weak but significant inverse correlations with R\* for gene abundance (though not expression), it seems like this should at least be explored in the discussion. Finally, I would suggest some alterations to the bold statement in the summary paragraph (lines 293- 294). I didn't see MG/MT data for AepV and AepW (though I might have missed this), so I would limit this to AepX. Also, given that this work has expanded the known number of phosphonate transporters by 3 and there still could be others out there, I would qualify that AepX is the most abundant of the known phosphonate transporters.  
+
+Before I list the final more copy- edit style comments, I want to reiterate that I think the author make a compelling case that this is an important new discovery that will require a re- assessment of the phosphorus (and nitrogen) cycle in the surface ocean.  
+
+Other specific comments:  
+
+Line 47- 49: The way this is phrased, it almost suggests that this is only important in response to increased anthropogenic P loading.  
+
+Line 53: Pi has not been introduced yet.  
+
+Line 58- 60: The opening sentence of this paragraph is really awkwardly written and hard to follow.  
+
+Line 67: after strains, add "of bacteria" (assuming that's what you mean) Line 78: whose? Sentence is confusing as written. For clarity, replace 'whose" with "with an" and insert a "that" after abundance
+
+<--- Page Split --->
+
+Line 80: PhnWAY not introduced, add "that" after fact  
+
+Line 82- 83: This statement about transporters being superb molecular tools for investigating in- situ cycling of metabolites, while being a true statement, seems out of place/unconnected.  
+
+Lines 94- 97: This sentence is very confusing as written with too many clauses so that it's hard to sort out which clause is related to which statement.  
+
+Line 101: Other statements reference Pi being exported whenever AEP is the N source (i.e., AEP as just N source or as N and P source). What that not the case in BIRD- 1 (or did the authors not check for Pi release when BIRD- 1 was growing on AEP as an N source with added Pi? This should be made clearer (as should the later statements).  
+
+Line 179: N has been introduced as abbreviation for nitrogen  
+
+Line 180: What or where is table S3? Is it the "dataset" that is attached? If so, please make that clear and add a table legend so that the reader knows what they're looking at.  
+
+Line 188: Statement as written implies that this is absent from the Rhodobacteriaceae. Do you mean in addition to?  
+
+Line 316: What do you mean by co- culture experiments? Please provide a bit more detail. Line 325: Table S4 is missing strain information. Also, as a general comment, I assume that all of the strains used in this work were axenic. The authors should explicitly state this somewhere.  
+
+Methods: General comment. There are a lot of abbreviations in the methods of things that are somewhat standard/well known to a biologist, but probably should still be spelled out given the audience for Nature Communications (HEPES, LDS, PCR).  
+
+Figure 3 Legend: This legend is hard to follow and doesn't explain all of the components of the figure. What do the colors mean? What do the dashed lines mean?
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+We thank all three reviewers for their expert remarks. We have now performed several new experiments and analyses in accordance with their comments which we hope meets with their approval and allows acceptance of this manuscript in Nature Communications, as follows:  
+
+1) Ligand-binding affinity assays of recombinant S. stellulata AepX 
+2) Enumeration of cells for Rosoebacter strains growing on 2AEP as either an N or P source. 
+3) Pi efflux quantification in P. putida BIRD-1 cultures where 2AEP is a sole N source 
+4) Bioinformatics analysis of the distribution of PhnZ, the key marker of the 2AEP oxidative degradation pathway. 
+5) Bioinformatics analysis of the LysR-type regulators located adjacent to various 2AEP utilisation gene clusters.  
+
+Please note that in addition to the point- by- point responses, we have slightly amended the methods section. This now includes methods for quantifying extracellular phosphate as well as the production, purification and characterisation of recombinant AepX cloned from Stappia stellulata and associated ligand binding assays. We have also moved some of the details on protocols to the Supplementary Materials in order to save space.  
+
+We have also amended the abstract to include the new binding affinity data and other suggestions, whilst adhering to the 150 max word limit.  
+
+Reviewer #1 (Remarks to the Author):  
+
+The distribution, metabolism, and importance of 2AEP in marine ecosystems have long been recognised by the scientific community. Nevertheless, the ecological and biogeochemical role of this natural occurring phosphonate has just begun to be understood. Despite our advance knowledge on the subject, many features on the metabolism of 2AEP remain still undiscovered. For example, the ability of 2AEP- degrading microorganisms to import this phosphonate into the cell, that, however, lack the specific transporters for phosphonates. Therefore, the ability to expand our knowledge on the acquisition and catabolism of phosphonates by microorganisms is a crucial step to comprehend the dynamics of natural occurrence phosphonates in nature fully. Your manuscript describes, in a very structured study and with a well- performed methodology, the discovery of these novel bacterial transporter systems, which will contribute to close the existing gaps in the known metabolism of 2AEP and the role of organophosphonates in the biogeochemical cycle of phosphorus. Therefore, your contribution to this specific topic of interest, which has attracted the attention of many respected scientists since the identification of phosphonates in the seawater column in the early 2000s, will be universally recognised. However, I must say that the current manuscript's title is quite ambitious and does not reflect the extension of the experimental results presented in this work. Therefore, more experimental work is needed at the field level to prove the statement.  
+
+We acknowledge that our manuscript lacks environmental field data, and have thus amended the title accordingly to highlight this fact:  
+
+'Transporter characterisation reveals aminoethylphosphonate mineralisation as a key step in the marine phosphorus redox cycle'  
+
+We have largely retained the latter part of the original title since we do believe that our new data on the binding affinity of AepX combined with our originally presented omics data presents a very compelling argument for the high turnover of 2AEP in seawater and hence a key step in the marine phosphorus redox cycle. Clearly, there is now an urgent need to use refined analytical techniques to obtain field- based process data.
+
+<--- Page Split --->
+
+Throughout the manuscript, there are a couple of specific gaps in the general knowledge and state of the art about the metabolism of 2AEP that I would like to address in order to increase the quality of your manuscript:  
+
+The catabolism of 2AEP by microorganisms is particularly interesting as it can be achieved by different pathways, such as 2- aminoethylphosphonate dioxygenase (PhnY\\*), phosphonoacetaldehyde hydrolase (phosphonatase - PhnX), and phosphonoacetate hydrolase (PhnA). Interestingly, the operons that contain the genes encoding for either PhnX or PhnA usually contain genes that encode for a 2- AEP transaminase (PhnW) and/or a phosphonoacetaldehyde dehydrogenase (PhnY), which are accessory proteins that participate on the degradation of 2AEP (Agarwal et al. 2014). However, it must be considered that these operons are known to be regulated under either the PhoBR regulatory system related to the Pho regulon or by a LysR- like transcriptional activator responsive to 2AEP, which enable the microorganisms to activate both pathways independently of ambient Pi concentrations (Kulakova et al., 2001, 2009, Cooley et al. 2011, Borisova et al. 2011).  
+
+Therefore, it is highly recommended that authors should analyse the flanking regions of the operons found in Pseudomonas putida BIRD- 1, Stappia stellulata DMS 5886, and other Pi- insensitive 2AEP- degrading microorganisms to determine if these species possess a LysR type of regulator, as it is known that the majority of phnX operons present in Pseudomonas and phnA in alpha- proteobacteria are associated with this substrate- inducible type or regulator (Villarreal- Chiu et al. 2012).  
+
+We thank the reviewer for this insightful comment. The LysR- family of regulators are actually highlighted in Figure 2A. However, we acknowledge that the link between our labelled 'AepR' and LysR- like was not clearly presented. We do have a large amount of data on the regulation of the 2AEP transporters and PhnWX in P. putida. We initially had some of this data presented in this manuscript. However, the regulation is very complex and involves the three master regulators of the C, N and P stress responses, CbrAB, NtrBC and PhoBR, respectively. Therefore, we removed this for simplicity since we are performing a more comprehensive analysis of the regulation, which is the focus of a follow- on paper.  
+
+However, we have now added a phylogenetic comparison of the LysR- like regulators found in 2AEP operons (Figure 2C). This demonstrates that P. putida possess three distinct LysR- like regulators for AepXWV (AepR- XWV), AepP (AepR- P) and PhnWX (Aep- WX) that are divergent from either PhnR or PalR. S. stellulata and S. melilotti possess just the one upstream of aepXWV- phnWAY, which is most similar to the AepP- associated LysR- type. Thus, it is likely that these regulators have evolved to perform divergent functions. To our knowledge, given the divergence between PhnR and aminophosphonate operon LysR- type regulators, the only direct experimental evidence regarding their function comes from Martinez et al. 2010, where a lysR gene was shown to be essential for complementing an E. coli ΔphnHIJKLMNOP mutant with a Pseudomonas PhnWX system. We note, though, that substrate induction via these LysR- like regulators in no way precludes regulation through other mechanisms – indeed the data we refer to above demonstrates this.  
+
+We have added a paragraph starting line 163:  
+
+'In many 2AEP gene clusters, we identified LysR- type regulators, which we refer to as AepR. A homolog of AepR has been shown to be essential for complementation of an Escherichia coli ΔphnHIJKLMNOP mutant with a Pseudomonas PhnWX \(^{25}\) , implying substrate inducible regulation. Additionally, PhnA activity has been shown to be induced by 2AEP in a marine Falsirhodobacter isolate even under nutrient replete conditions \(^{22}\) , though unfortunately no sequenced Falsirhodobacter strain possesses an aminophosphonate operon so it is not clear if AepR is responsible for this regulation. BIRD- 1 and other Pseudomonas, whose 2AEP
+
+<--- Page Split --->
+
+operons are often fragmented throughout the genome, possess up to three distinct genes encoding LysR- type regulators (Fig 2A). In contrast, most Alphaproteobacteria only possess a single gene, located upstream of the aepXvW- phnWAY operon. These newly identified forms are phylogenetically distinct from either the archetypal PhnR or PalR found in P. fluorescens sp. 23F and Variovorax sp. PAL2, respectively (Fig 2C). The three BIRD- 1 LysR- like forms were clearly distinct from each other with the AepP- associated form being closely related to the single LysR- type regulator found in Alphaproteobacteria.  
+
+As well as a sentence starting at line 192:  
+
+'These data are consistent with, the hypothesis that S. stellulata 2AEP catabolism is regulated by a LysR- type regulator solely through substrate- induction, which will be investigated in a future study.'  
+
+The phylum proteobacteria is known to excel in accumulating metabolic pathways to degrade natural and xenobiotic compounds, including phosphonates. The existence of different phosphonate degradation pathways or multiple copies of phosphonate degradation genes in members of this genus is well known (Enterobacter aerogenes: Lee et al. 1992, Salmonella typhimurium: Jiang et al. 1995, Pseudomonas stutzeri: White and Metcalf 2004, Mesorhizobium loti: Huang et al. 2005). This concept is valid for the presence of multiple copies of the phosphonate transporter gene phnD (Prochlorococcus marinus: Feingersch et al. 2012). However, it has been reported that a vast number of bacteria that possessed genes associated with any known phosphonate degradation pathway showed no evidence of possessing the PhnCDE route of phosphonate acquisition (Villarreal- Chiu et al. 2012), which can contribute to exalt the importance of this manuscript. On this regard, it must be mentioned that PhnD has been demonstrated to show the highest affinity to 2AEP. However, this protein shows high specificity to other natural occurring phosphonates as well (Rizk et al. 2006). Therefore, authors must demonstrate those novel proteins AepX and AepP are specific to 2AEP, as stated in the manuscript. Otherwise, authors should confirm that these proteins exhibit a group- specificity to phosphonates, similarly as PhnD.  
+
+Given this referee's very interesting comment regarding the binding specificity of our newly characterised transporters we have now performed ligand- binding interaction assays for the Stappia stellulata AepX using microscale thermophoresis (MST). Our data (Table 1, Figure S6) highlights a key finding revealing that AepX has very high affinity for 2AEP ( \(K_{\mathrm{d}}\) 23±4 nM) compared to a much lower affinity for ethylphosphate (145±15 \(\mu \mathrm{M}\) ), methylphosphonate ( \(K_{\mathrm{d}}\) 3.4±0.28 mM) and aminomethylphosphonate (4.4±0.8 mM). This finding is consistent with the association of aepXvW with 2AEP- specific degradation genes. Thus, AepX differs from the E. coli PhnD in so much that it has much greater specificity towards a single phosphonate (2AEP).  
+
+We have not provided any data for AepP for two reasons: 1) AepP is a membrane protein that is not trivial to over- express, purify and subsequently assay, issues which are only magnified by the current pandemic. 2) Importantly, this transporter is far less abundant in the ocean and thus its characterisation would not significantly affect the narrative of this paper, unlike AepX.  
+
+We have now added the AepX binding data to the results in the S. stellulata proteomics section starting line 195 and in the abstract:  
+
+'Finally, we determined the substrate specificity of recombinant S. stellulata AepX towards 2AEP and other (alkyl)phosphonates using microscale thermophoresis \(^{53,54}\) . Unlike the relatively promiscuous phosphonate binding protein PhnD \(^{43}\) , AepX was highly specific for 2AEP with a \(K_{\mathrm{d}}\) in the nanomolar range (Table 1, Supplementary Fig. 7), consistent with the observation that aepXvW is typically co- localised with either phnWX or phnWAY that encode 2AEP- specific degradation systems (Fig. 2A).'
+
+<--- Page Split --->
+
+'Unlike the archetypal phosphonate binding protein, PhnD, AepX has high specificity for 2AEP (Stappia stellulata AepX \(K_{\mathrm{d}}\) \(23\pm 4\) nM; methylphosphonate \(K_{\mathrm{d}}\) \(3.4\pm 0.3\) mM).'  
+
+We have also added the following sentences in the discussion starting line 296:  
+
+'To date, whilst several ABC transporters show preference for methylphosphonate and phosphite \(^{53,67}\) , or bind a range of phosphonates with low micromolar or less \(K_{\mathrm{d}}^{43}\) , no ABC transporter showing a strong preference for 2AEP has been identified. Here, we revealed AepX appears to be highly specific for 2AEP and has substantially lower affinity for methylphosphonate or ethylphosphonate than PhnD \(^{43}\) . The occurrence of aepXVW adjacent to putative phosphonate catabolic genes, and the characterised PbfA \(^{37}\) , does suggest some degree of promiscuous binding, albeit likely to related aminophosphonates. Thus, the molecular mechanisms governing the specificity of AepX towards aminophosphonates warrant further investigation.'  
+
+Microorganisms have evolved several mechanisms to acquire and conserve P as an adaptation to cope with the ever- changing P concentrations in the environment. These mechanisms are known to be related not just with the Pi levels of the surrounding environment (Pho regulon: Kononova and Nesmeyanova, 2002), but also with nitrogen, iron, oxygen or even light (Dyrman et al. 2007). A clear example of these mechanisms of acquisition and conservation of P as a response to P and/or N levels, is the ability of microorganisms to accumulate polyhydroxyalkanoates (PHA: Valentino et al. 2015). These are a group of natural polymers accumulated intracellularly as a carbon and energy storage material (Tobin and O'Connor, 2005). PHA production and accumulation has been demonstrated to be widespread among marine microorganisms (Uwamori- Takahashi et al. 2017, Ganapathy et al., 2018). This information is relevant for the present work, as experimentally, PHA production and accumulation by microorganisms can be detected by an increase of turbidity in the culture. This occurs as the PHA internal granules tend to increase against time in response to a nutrient limitation in the presence of excess carbon. This increase of turbidity in the culture can produce false- positive results due to the inability to discriminate between PHA accumulation and biomass production (Acosta- Cortés et al. 2019). On this regard, it is possible that this phenomenon is occurring on the marine bacterial strains tested for 2AEP supplemented as N source, as the utilisation of 2AEP as P is significantly higher to the exceptionally low OD540 obtained for 2AEP cultures supplemented as N source (Table S1). An easy and rapid method to discern the presence of PHA is by visualising it under fluorescent microscopy using a Nile red or Nile blue A 1% staining (Giin- Yu et al. 2014). Demonstrating that Alliroseovarius strains do not accumulate PHA when grown on 2AEP supplied as N source, would significantly improve the hypothesis of the existence of a new catabolic pathway for 2AEP.  
+
+We thank the reviewer for their thoughts here and have now determined biomass increase by enumerating growth via colony forming units (c.f.u). This clearly demonstrates that cells are actively catabolising and assimilating the nitrogen to facilitate proliferation as opposed to simply accumulating excess carbon (Figure S5).  
+
+The lower biomass observed when \(1.5 \mathrm{mM}\) 2AEP is used as the sole N source compared to when \(1.5 \mathrm{mM}\) 2AEP is used as the sole P source is due to the higher cellular demand for N relative to P. Given the 1:1 ratio of N:P in 2AEP, one would expect a lower final biomass. We have now added controls using \(1.5 \mathrm{mM}\) ammonium as the sole N source and you can clearly see comparable \(\mathrm{OD}_{540}\) and c.f.u counts between the controls and 2AEP- grown cells.  
+
+The bioinformatic analysis of the abundance of phosphonate metabolic pathways among microorganisms have been widely studied (Huang et al. 2005, Villarreal- Chiu et al. 2012). With
+
+<--- Page Split --->
+
+their new approach, it is recommended that authors should combine their results on the distribution of aepX and aepP across bacterial taxa and the results of previous studies on the distribution of catabolic pathways. Many potential new transporters or catabolic pathways may be found with this analysis.  
+
+We apologise that this was not clearer in our manuscript. Our analysis also included markers of the catabolic enzymes (phnA, phnJ and phnX). We have now added a line to reference Fig S8:  
+
+'The distribution of markers (phnJ, phnA, phnX) for the various phosphonate degradation systems were also analysed (Fig S8)'  
+
+We found that PhnA was the most abundant catabolic gene also. However, we did not identify any other putative transporters, other than AepXvW, when screening the neighbourhoods of some of these gene clusters. We also scrutinised the neighbourhoods of several AepX and AepP genes and did indeed identify several putative catabolic genes for phosphonate degradation, shown in Figure 4.  
+
+We have now amended the results section to acknowledge there were novel catabolic genes located in these AepX/AepP neighbourhoods:  
+
+'Many taxonomically divergent AepX ORFs were co- localised with ORFs encoding the various 2AEP degradation systems, the C- P lyase, or putative uncharacterised ORFs encoding potential novel phosphonate catabolic enzymes, supporting a role in 2AEP transport (Fig 4).'  
+
+Also:  
+
+'Again, for all of these strains ORFs encoding AepP were co- localised with ORFs encoding PhnWAY or PhnWX, or putative catabolic enzymes (Fig S6).'  
+
+Also:  
+
+'We confirmed that these abundant environmental sequences were also co- localised with characterised and putative phosphonate degradation genes (Fig 4).'  
+
+On the other hand, while authors' approach on the transcript abundance of aepX on metagenomic data is valid, it is essential to compare these results with a housekeeping gene in order to normalise the relative abundance of all compared genes against the number of resultants obtained for a housekeeping gene, generally, recA, used as a single- copy- per- genome control in distribution and prevalence analyses (Moran et al., 2004).  
+
+We fully agree. We think the reviewer has missed our statement concerning normalisation methods which can be found in either the figure legend or methods section. We have already normalised all marker gene data (MG or MT) to the median gene abundance (for MG) or transcript count (for MT) of ten housekeeping genes. Hence, there is a value greater than \(100\%\) in some MT samples. Indeed, this further demonstrates very high expression of AepX.  
+
+(Figure 5):  
+
+'Abundance (Log2 % abundance [gene or transcript] relative to the median abundance [gene or transcript] of 10 single copy core genes) of phnD, aepX, aepP in marine epipelagic (red) and mesopelagic (blue) waters, split by metagenome (MG) (A), and metatranscriptome (MT).'  
+
+Methods:
+
+<--- Page Split --->
+
+'Sequence abundances were expressed as the average percentage of genomes containing a copy by dividing the percentage of total mapped reads by the median abundance (as a percentage of total mapped reads) of 10 single- copy marker genes<sup>82</sup> for both MG and MT'  
+
+Finally, references 75, 76, and 77 are missing from the reference list. We apologise for this. These are now added.  
+
+References used in this revision:  
+
+We have now added the appropriate references suggested by yourself. Others we already had in the manuscript.  
+
+Agarwal et al. 2014: Chem Biol, 21; DOI: 10.1016/j.chembiol.2013.11.006 Included Kulakova et al., 2001: J Bacteriol, 183: 3268- 3275. DOI: 10.1128/JB.183.11.3268- 3275.2001 Included Kulakova et al., 2009: Microb Biotechnol, 2: 234- 240 DOI: 10.1111/j.1751- 7915.2008.00082. x Included Cooley et al. 2011: Microbiology, 80: 335- 340; DOI: 10.1134/S0026261711030076 Included Borisova et al. 2011: J Biol Chem, 286: 22283- 22290; DOI: 10.1074/jbc.M111.237735 Included Villarreal- Chiu et al. 2012: Front Microbiol, 3: 19; DOI: 10.3389/fmicb.2012.00019 Included Lee et al. 1992: J Bacteriol, 174: 2501- 2510; DOI: 10.1128/jb.174.8.2501- 2510.1992 Jiang et al. 1995: J Bacteriol, 177: 6411- 6421; DOI: 10.1128/jb.177.22.6411- 6421.1995 Included White and Metcalf 2004: J Bacteriol, 186: 4730- 4739; DOI: 10.1128/jb.186.14.4730- 4739.2004 Included Huang et al. 2005: J Mol Evol, 61: 682- 690; DOI: 10.1007/s00239- 004- 0349- 4 Feingersch et al. 2012: ISME J, 6: 827- 834. DOI: 10.1038/ismej.2011.149. Included Rizk et al. 2006: Protein Sci, 15: 1745- 1751. DOI: 10.1110/ps.062135206. Included Kononova and Nesmeyanova, 2002: Biochemistry, 67: 184- 95; DOI: 10.1023/A:1014409929875 Included Dyhrman et al. 2007: Oceanography, 20: 110- 116; DOI: 10.5670/oceanog.2007.54 Included Valentino et al. 2015: Water Res, 77: 49- 63; DOI: 10.1016/j.watres.2015.03.016 Tobin and O'Connor, 2005: FEMS Microbiol Lett, 253: 111- 118; DOI: 10.1016/j.femsle.2005.09.025 Uwamori- Takahashi et al. 2017: Bioeng, 4: 60; DOI: 10.3390/bioengineering4030060 Ganapathy et al. 2018: Int J Biol Macromol, 111: 102- 108; DOI: 10.1016/j.ijbiomac.2017.12.155 Acosta- Cortés et al. 2019: ISME J, 13: 1497- 1505; DOI: 10.1038/s41396- 019- 0366- 3 Giin- Yu et al. 2014: Polymers, 6: 706- 754; DOI: 10.3390/polym6030706 Moran et al. 2004: Nature, 432: 910- 913. DOI: 10.1038/nature03170.  
+
+## Reviewer #2 (Remarks to the Author):  
+
+This manuscript describes the discovery and characterisation of several new transporters for 2- aminoethylphosphonate (2AEP), coupled with bioinformatic studies of their prevalence and correlation with environmental factors and phosphonate catabolism genes. The characterisation of the transporters themselves is interesting, but potentially more interesting is the demonstration that these are likely to contribute to phosphate- insensitive metabolism of 2AEP in the marine environment: while this catabolism has been previously shown in isolates, there wasn't clear data on the extent of it in the environment. This is likely to expand the interest in the manuscript beyond microbiologists and into a much broader biogeochemistry/oceanography community, and further supports the recent re- evaluation of
+
+<--- Page Split --->
+
+the role of reduced phosphorus in the oceans. The experimental approach in the work seems appropriate and is well detailed, and the conclusions are justified.  
+
+Overall I think this manuscript would make an important contribution to the field.  
+
+I have a few core suggestions:  
+
+The authors investigate the relationship between the new transporters and PhnA/X/J, and use this to infer that 2AEP- specific mineralisation is a more prevalent process than non- specific phosphonate metabolism via C- P lyase. Is there a reason that PhnZ was omitted from this comparison? It is present in Figure 1 but not examined further.  
+
+We initially omitted PhnZ since there are a variety of forms. To address this comment we have now constructed a phylogeny of all PhnZ- like sequences (Fig S9). Characterised forms can be involved in either methylphosphonate or 2AEP catabolism and there is substantial sequence variation even within variants of known function. As such, our hmm detects all forms of PhnZ, including the N- trimethyl- 2- aminoethylphosphonate specific TmpB. In addition to these three forms, several PhnZ- like ORFs are frequently found associated with PhnWAY/XW and even C- P lyase operons, inflating the abundance of this gene, but with no real evidence as to their function. We also scrutinised the TARA dataset for PhnY\*, the first enzyme in this degradation pathway, but found it to be in very low abundance (Fig S10). This further indicates diversification in the function of PhnZ. All of this is really exciting, but beyond the scope of our current paper which specifically focuses on 2AEP metabolism.  
+
+We have added this sentence to the introduction starting line 54:  
+
+'The phosphonate dioxygenase (PhnY\*) phosphohydrolase (PhnZ) system has also been shown to degrade 2AEP<sup>33,34</sup>, though at least some homologs of this system are specific to (hydroxy-) methylphosphonate and cannot degrade 2AEP<sup>35,36</sup>.'  
+
+We have also added this small paragraph to the results section starting line 243:  
+
+'The phnZ marker is split into several subclades, with only the original PhnY\*Z specific for 2AEP (Fig S9). This 2AEP- specific form was found at very few sites (MG = 9; MT=2) and in low abundance in both the MG and MT (Fig S9). Homologs related to the two PhnZ clades associated with either methylphosphonate or N- Trimethyl- 2- aminoethylphosphonate degradation were found at comparable gene and transcript abundances to phnJ and significantly lower than phnA (Fig S9).'  
+
+On line 60 phosphonatases are described as "phosphonate degradation systems", and this word is used throughout the text to refer to the action of PhnA, PhnX, or C- P lyase. The term phosphonatase was coined by La Nauze et al. (Biochimica et Biophysica Acta Enzymology, 212[2], 1970) as the trivial name for phosphonoacetaldehyde hydrolase/PhnX specifically, and largely isn't used to mean phosphonate degradation enzymes generally (e.g. see reviews by Horsman and Zechel [2017, Chemical Reviews, 117(8)], McGrath et al. [2013, Nature Reviews Microbiology, 412], or Peck and van der Donk [2013, Current Opinion in Chemical Biology, 17], which all use phosphonatase exclusively to refer to PhnX). It would be more in keeping with the wider phosphonate literature to avoid that term when describing PhnA/C- P lyase.  
+
+Thank you for this comment and we agree. We initially thought it would simplify the terminology for a general audience, but fully accept it is better to stay in keeping with the current literature. We have corrected throughout the manuscript.  
+
+Line 213 states "aepX transcription was 40- fold and 350- fold greater than phnD in the epipelagic and mesopelagic, respectively (Fig 5B)." The 40- fold number is fine, but Fig. 5B doesn't seem to show a 350- fold difference in aepX and phnD in the mesopelagic, it looks closer to \(\sim 130\) - fold (0.3 vs 40). Is this value correct, or am I misinterpreting something?
+
+<--- Page Split --->
+
+Thank you kindly for spotting this. We have checked the raw data and it is \(\sim 140\) - fold (0.3 v 42).  
+
+Sentence changed to:  
+
+'40- fold and 140- fold greater than....'  
+
+The sentence beginning on Line 217 discusses metagenome/transcriptome data for phnA, J and X, but doesn't refer to Figure S8 where this data is shown.  
+
+Again, thank you very much for spotting this. We have now added the appropriate citation:  
+
+'...major oceanic process (Fig S8).'  
+
+Line 33 states "Collectively, our data identifies a mechanism responsible for the oxidative step in the marine phosphorus redox cycle". Slightly nitpicky, but saying "the oxidative step" implies that this is the only oxidation reaction which phosphorus goes through in the ocean, which isn't correct: aside from phosphonate oxidation systems there are separate enzymes for phosphite and hypophosphite oxidation present in marine organisms (e.g. Martinez et al., Environmental Microbiology, 2011, 14(6), pp1363- 77). Perhaps rephrase this along the lines of "for a major oxidation process"?  
+
+Thank you and agreed. Changed to:  
+
+'responsible for a major oxidation process in the...'  
+
+Line 216: "The majority of aepX sequences were related to the cosmopolitan Alphaproteobacteria and Deltaproteobacteria (Fig 4)." Figure 4 appears to show IM- RGC sequences are mostly in the Alpha and Gammaproteobacteria, not the Deltas?  
+
+This statement refers to sequences pulled out from the TARA oceans dataset, which are coloured blue in the figure. Some of the abundant OTUs, denoted by the outer ring bars are related to the SAR324 cluster.  
+
+Figure S8F appears to be missing metatranscriptomic data for the Southern Ocean. Figure 5C/D also seem to be missing matching MG/MT data for certain areas (e.g. aepP data for the Southern Ocean in 5C, aepP MT data for the SAO and MS in 5C, all MT data for SO in 5D). Are these occluded by other symbols or are they missing from the figures/dataset?  
+
+There is no mesopelagic MT site, and only a single mesopelagic MG site, in the Southern Ocean in the TARA dataset. If a gene is absent from an oceanic region it is not shown. If a gene is only detected at a single location within an oceanic region no error bars are shown. This only occurs for aepP, the least abundant transporter. We have amended the figure legends to make this clear:  
+
+'Note: AepP transcripts were not detected at numerous sites; represented by an omission of data points.'  
+
+Other comments:  
+
+Line 55: When arguing that 2AEP is absent from HWM DOP the authors cite article 17 (Sosa et al.). The article in question doesn't state that 2AEP wasn't detected in their sample, only that MPn and 2- HEP were present as the major components along with "minor unidentified phosphonates" (which may have included 2AEP). It may be better to rephrase this sentence to suggest that 2AEP is either absent or present in significantly lower proportions than other phosphonates, thus the ubiquitous synthesis genes would suggest preferential degradation.
+
+<--- Page Split --->
+
+We agree. Changed to:  
+
+'Notably, the fact that 2AEP is either absent from, or a minor component of, otherwise phosphonate rich high molecular weight dissolved organic matter (HMW DOM) \(^{16,17}\) .  
+
+Additionally, in the discussion:  
+
+'However, several studies have shown 2AEP is not detected as a significant component of 'semi- labile' DOM whilst alkylphosphonates tend to accumulate \(^{16,17,65}\) .  
+
+In Figure 2A, the colour of the PhnY symbol in the key appears to be darker than the colour used in the diagram itself.  
+
+Thank you for spotting this. This has now been changed.  
+
+Line 163: "To confirm that Stappia AepXVW" should read "To confirm that Stappia stellulata AepXVW"  
+
+Thank you. We have changed to:  
+
+'To confirm that S. stellulata AepXVW...'  
+
+Line 100: "BIRD- 1 was capable of growth on 1.5 mM 2AEP as either a sole N, P, or N and P source, the latter resulting in mineralisation of Pi which was subsequently exported from the cell". Was P export from the cell measured when 2AEP was provided as the sole N source? Previous literature (e.g. citation 22/Chin et al.) would suggest that P export should be observed here as well, and it isn't clear from the text if this didn't occur or just wasn't measured.  
+
+We have now repeated this experiment with P. putida to match the same data as we collected for S. stellulata. We have amended Figure S1 accordingly. The data now clearly shows mineralisation occurs even when exogenous phosphate is present.  
+
+We have amended the sentence in the results to make this statement clearer:  
+
+'BIRD- 1 was capable of growth on 1.5 mM 2AEP as either a sole P, N or N and P source, the latter two resulting in mineralisation of Pi which was subsequently exported from the cell (Fig 1C, Fig S1).'  
+
+Fig. S3: What is the significance of the dashed line at an abundance of 27? This should be stated in the figure legend.  
+
+We initially added this here as a visual aid. However, we have now removed this as there is nothing statistically relevant about this line.  
+
+Figure 5's legend is missing a statement identifying the error bars.  
+
+Thank you kindly for spotting this. We have now added: 'Error bars denote standard deviation of the mean.'  
+
+Line 316 states "Co- culture experiments were carried out according to the protocol described in60". It's not clear to me where in the manuscript co- cultures were performed/described, certainly not in the context of the work performed in citation 60. Could the authors clarify this?  
+
+Many apologies for this statement which is a relic of an earlier version of the manuscript. This sentence has now been removed from the revised manuscript.
+
+<--- Page Split --->
+
+## Reviewer #3 (Remarks to the Author):  
+
+The manuscript "Aminophosphonate mineralisation is a major step in the global oceanic phosphorus redox cycle" uses a combination of involved gene knockout/complementation culture experiments, proteomics, and other multi- OMIC analyses to identify and characterize multiple bacterial aminothylphosphonate (AEP) transporters. Through their careful analyses they evaluate what controls the expression of these transporters, the fact that the different transporters appear to be involved with either using AEP as a nitrogen (N) or phosphorus (P) source, as well as their ubiquity in both published genomes and global ocean datasets. This is an impressive piece of work and of critical importance to the oceanographic community as it relates to the availability of two essential nutrients for phytoplankton growth.  
+
+That said, I do have several comments on the current manuscript's structure that I feel need to be addressed before publication. The first has to do with the title. The work only focused on AEP, not aminophosphonates in general. While AEP is an example of an aminophosphonate this may be somewhat nit- picking, but I think that Aminothylphosphonate should replace aminophosphonate in the title. Also, the article focuses much of its discussion on the role in the phosphorus redox cycle, but it's clear from their work that bacteria can use AEP as an N source (and when doing so can release inorganic P), would it be more appropriate to have the title and abstract indicate AEP may be an important N source as well. Global modelling efforts suggest a much larger region of the ocean is N limited than P limited. It is interesting that in meeting N demands, bacteria may release bioavailable inorganic P.  
+
+Thank you for this comment. Indeed, in response to reviewer 1 we have obtained new data showing AepX binds to 2AEP with high specificity compared to ethylphosphonate, methylphosphonate or aminomethylphosphonate. As mentioned in our response to reviewer 1 we believe that our new data on the binding affinity of AepX combined with our originally presented omics data presents a very compelling argument for the high turnover of 2AEP in seawater and hence a key step in the marine phosphorus redox cycle. Hence, we feel justified to retain that aspect of the original title (but indeed changing aminophosphonate for aminomethylphosphonate).  
+
+Whilst we thank the referee for highlighting that 2AEP can be used as an N source we do not think it appropriate to add this aspect to the title given we have no data to compare its use with other organic N sources.  
+
+Our justification for focussing on the P redox cycle is because we directly compare all phosphonate utilisation genes which likely represent the major pathways for oxidation of reduced phosphorus. Given the very high expression of AepX and PhnA in seawater compared to PhnD, PhnJ and PhnZ, our data provides a clear mechanism for one of the major steps in this oxidative pathway. In contrast, we do not compare AepX expression to other N cycling genes. This is beyond the scope of the current manuscript and would require a very detailed analysis of all pathways linked to the degradation of e.g. methylated amines, quaternary amines, polyamines, amino acids, etc. Thus, at the moment we believe it is too speculative to state 2AEP mineralisation is an important part of the N cycle, or carbon for that matter. However, we do acknowledge that the role in N cycling should be explored in the discussion and our data does provide a ubiquitous molecular mechanism for the regeneration of ammonium. We have thus significantly amended the discussion section on 2AEP mineralisation as follows:  
+
+'Substrate inducible expression of catabolic genes targeting organic N molecules, irrespective of nutrient status, has previously been shown to drive mineralisation of N and cross feed into surrounding microbial cells<sup>68- 71</sup>. Indeed, ammonium mineralisation may also occur if 2AEP,
+
+<--- Page Split --->
+
+\((R)\) - 1- hydroxy- 2- aminoethylphosphonate or \((N)\) - trimethyl- 2- aminoethylphosphonate are also used as carbon and energy sources, similar to methylamines and quaternary amines \(^{68,69}\) . In agreement with Chin et al. \(^{22}\) , our proteomic data for S. stellulata and in situ environmental data strongly suggests PhnA and AepX are highly synthesised in a substrate- inducible manner, that would facilitate the remineralisation of labile inorganic N and P. Even if ammonium concentration does play a role in the occurrence and regulation of 2AEP degradation genes (i.e. 2AEP is primarily a N source), our combined data, clearly demonstrates the potential for the in situ mineralisation of semi- recalcitrant DOP into labile Pi, a mechanism which is important for maintaining biological production in Pi- deplete regions of the ocean \(^{72,73}\) . It should be noted that the Pseudomonas AepR located adjacent to AepP is most similar to marine AepR. This could explain why AepP was more abundant in the presence of 2AEP (Fig 1D), although the differences in growth rate and protein abundance suggest substrate induction is not the sole mechanism of regulation in P. putida BIRD- 1. '  
+
+In addition, we have also recognised the N cycle in the abstract and have also amended the abstract at line 26:  
+
+'2AEP may be an important source of regenerated phosphate and ammonium, which are required for oceanic primary production.'  
+
+As to the structure of the manuscript. There is an excessive use of abbreviations. I understand that the authors have limited space to tell a large story, but it makes the manuscript very hard to read. Some of this may be alleviated by having the figures (especially figure 1 which has all of the gene abbreviations) closer to the text. The gene name abbreviations are obviously necessary, as are many others, but in thinking about the fact that Nature Communications readership is more broad, it might behoove the authors to question whether it is necessary to introduce the abbreviation for something like MPn and SBP, which are only used a few times in the manuscript and are not common abbreviations outside of a small subset of the field.  
+
+Thank you and we do agree. We have changed many of these now to keep in full.  
+
+Also, there's definitely cases where an abbreviation is used that hasn't been introduced and a whole word is used when an abbreviation is in use (and not just at the start of a sentence).  
+
+Thank you for spotting. We have now checked the manuscript and amended where appropriate.  
+
+I have some other general comments. In the results/discussion of the ubiquity of these transporters in the global datasets, it looks like there's a difference in AepX types (BIRD- 1 and Stappia), it looks like it is the Stappia variant that is more abundant in the open ocean. AepX was Pi sensitive in BIRD- 1 and BIRD- 1 also had AepP, which if I interpret the results correctly was much rarer in both published genomes and the TARA dataset. Did you explore via MAGs or IMG if the organisms that had AepP had a similar variant of AepX as BIRD- 1? It seems like an interesting angle to pursue.  
+
+Thank you for this thoughtful comment. We found no correlation between the AepX variant and the possession of AepP. Only Pseudomonas have both AepX and AepP, but other AepP homologs are found in strains lacking AepX.  
+
+We have now added binding data and analysis of the LysR- type regulators as requested by reviewer 1. It has been noted that the LysR- type activator found in alphaproteobacterial AEP operons is more similar to the LysR- type regulator associated with AepP. This may point towards a divergent role for Pseudomonas- like compared to other AepX homologs. On the other hand, Betaproteobacteria in the same clade as the Pseudomonas possess a LysR- like regulator that is more similar to the AepXVW.
+
+<--- Page Split --->
+
+As stated to reviewer 1, we have two other papers detailing both structure- function relationships and regulation of these transporters, the latter of which is very complicated.  
+
+When discussing the MG and MT data, all focus is on the role of Pi in potentially controlling abundance/expression. AEP is a potential N source. Considering there are some weak but significant inverse correlations with \(\mathbb{R}^*\) for gene abundance (though not expression), it seems like this should at least be explored in the discussion.  
+
+We agree. Similar to your previous comment above, we have now amended the section discussing substrate- inducible expression to add more weight to the notion that 2AEP may result in remineralised ammonium and/or serve as a nitrogen source.  
+
+'Substrate inducible expression of catabolic genes targeting organic N molecules, irrespective of nutrient status, has previously been shown to drive mineralisation of N and cross feed into surrounding microbial cells<sup>68- 71</sup>. Indeed, ammonium mineralisation may also occur if 2AEP, (R)- 1-hydroxy- 2- aminoethylphosphonate or (N)- trimethyl- 2- aminoethylphosphonate are also used as carbon and energy sources, similar to methylamines and quaternary amines<sup>68,69</sup>. In agreement with Chin et al.<sup>22</sup>, our proteomic data for S. stellulata and in situ environmental data strongly suggests PhnA and AepX are highly expressed in a substrate- inducible manner, that would facilitate the remineralisation of labile inorganic N and P. Even if ammonium concentration does play a role in the occurrence and regulation of 2AEP degradation genes (i.e. 2AEP is primarily a N source), our combined data, clearly demonstrates the potential for the in situ mineralisation of semi- recalcitrant DOP into labile Pi, a mechanism which is important for maintaining biological production in Pi- deplete regions of the ocean<sup>72,73</sup>. It should be noted that the Pseudomonas AepR located adjacent to AepP is most similar to marine AepR. This could explain why AepP was more abundant in the presence of 2AEP (Fig 1D), although the differences in growth rate and protein abundance suggest substrate induction is not the sole mechanism of regulation in P. putida BIRD- 1.'  
+
+Finally, I would suggest some alterations to the bold statement in the summary paragraph (lines 293- 294).  
+
+We have amended the concluding summary statement as follows:  
+
+'One of these, AepX, is the most abundant and highly transcribed of the known phosphonate transporters in seawater. Therefore, 2AEP mineralisation may represent a major process in marine organic N and P cycles and a significant source of regenerated labile Pi for oceanic production.'  
+
+I didn't see MG/MT data for AepV and AepW (though I might have missed this), so I would limit this to AepX. Also, given that this work has expanded the known number of phosphonate transporters by 3 and there still could be others out there, I would qualify that AepX is the most abundant of the known phosphonate transporters.  
+
+Two valid comments. We have now changed accordingly.  
+
+Before I list the final more copy- edit style comments, I want to reiterate that I think the author make a compelling case that this is an important new discovery that will require a reassessment of the phosphorus (and nitrogen) cycle in the surface ocean.  
+
+Thank you very much.  
+
+Other specific comments:
+
+<--- Page Split --->
+
+Line 47- 49: The way this is phrased, it almost suggests that this is only important in response to increased anthropogenic P loading.  
+
+This has been changed to the following to emphasise that the comparison is with riverine input:  
+
+'Collectively, this synthesis drives a vast global oceanic phosphorus redox cycle with reduced phosphorus input in the surface ocean estimated to be an order of magnitude greater than (non- anthropogenic) riverine phosphorus input'9,  
+
+Line 53: Pi has not been introduced yet.  
+
+Thank you for spotting this. We have now changed this to make the sentence clearer:  
+
+'...sources of C and/or nitrogen (N) in the presence of inorganic phosphate (Pi), i.e. in a Pi- insensitive manner, has been neglected.'  
+
+Line 58- 60: The opening sentence of this paragraph is really awkwardly written and hard to follow.  
+
+Thank you. We have now changed to:  
+
+'Unlike the majority of C- O- P monoster bonds, the C- P bond requires specific enzymes to break it, such as the C- P lyase \(^{26,27}\) .'  
+
+Line 67: after strains, add "of bacteria" (assuming that's what you mean)  
+
+Thank you for this. Changed to:  
+
+'occurs in a few strains of bacteria related'  
+
+Line 78: whose? Sentence is confusing as written. For clarity, replace 'whose' with "with an" and insert a "that" after abundance  
+
+Changed accordingly  
+
+Line 80: PhnWAY not introduced, add "that" after fact  
+
+We have now fully introduced the degradation systems in the previous paragraph:  
+
+'Unlike the majority of C- O- P monoster bonds, the C- P bond requires specific enzymes to break it, such as the C- P lyase \(^{26,27}\) . Several 2AEP- specific phosphonate degradation systems have been characterised (Fig 1A), including the 2AEP transaminase (PhnW) - phosphonoacetaldehyde hydrolase (phosphonatase - PhnX) system \(^{28,29}\) and the PhnW - phosphonoacetaldehyde dehydrogenase (PhnY) - phosphonoacetate hydrolase (PhnA) system \(^{30- 32}\) . The phosphonate dioxygenase (PhnY\*) phosphohydrolase (PhnZ) system has also been shown to degrade 2AEP \(^{33,34}\) , though at least some homologs of this system are specific to (hydroxy-) methylphosphonate and cannot degrade 2AEP \(^{35,36}\) . In addition, a gene encoding a recently characterised (R)-1- hydroxy- 2- aminoethylphosphonate ammonia lyase (PbfA) is often found in phnWX and phnWAY operons \(^{37}\) , expanding the known repertoire of aminophosphonate degrading capabilities \(^{37}\) .
+
+<--- Page Split --->
+
+Line 82- 83: This statement about transporters being superb molecular tools for investigating in- situ cycling of metabolites, while being a true statement, seems out of place/unconnected.  
+
+Agreed. We have now moved this up to the start of the paragraph introducing phosphonate transport systems.  
+
+'When analytical methods are not sensitive enough to accurately quantify the concentration and turnover of specific environmental metabolites, screening for the expression of their respective uptake systems becomes an important tool in understanding their in situ cycling'47,40- 42.  
+
+The amended sentence now reads:  
+
+'Here, we sought to identify transporters required for 2AEP catabolism in environmental bacteria lacking PhnCDE or PhnSTU.'  
+
+Lines 94- 97: This sentence is very confusing as written with too many clauses so that it's hard to sort out which clause is related to which statement.  
+
+Thank you for this suggestion. We have now amended this as follows:  
+
+"In Pseudomonas putida BIRD- 1 (hereafter BIRD- 1), a periplasmic substrate binding protein associated with one of these putative transporters was induced under Pi- deplete growth conditions in a PhoBR- dependent manner"43. We hereafter refer to this substrate- binding protein (PPUBIRD1_4925) as 2- aminoethylphosphonate X (AepX). AepX belongs to the same family (pfam13343) as PhnS, iron and sulphate substrate- binding proteins but is....  
+
+Line 101: Other statements reference Pi being exported whenever AEP is the N source (i.e., AEP as just N source or as N and P source). What that not the case in BIRD- 1 (or did the authors not check for Pi release when BIRD- 1 was growing on AEP as an N source with added Pi? This should be made clearer (as should the later statements).  
+
+As per reviewer 2's comment, we have now performed this experiment in \(P\) . putida. We have amended the line in the results section as well as updating Figure S1.  
+
+'BIRD- 1 was capable of growth on 1.5 mM 2AEP as either a sole P, sole N or sole N and sole P source, the latter two resulting in mineralisation of excess Pi which was subsequently exported from the cell (Fig 1C, Fig S1).'  
+
+As well as:  
+
+'However, S. stellulata lacks AepP but is still capable of Pi- insensitive growth and Pi export when grown on 2AEP as a sole N or sole N and sole P source (Fig S5A).'  
+
+Line 179: N has been introduced as abbreviation for nitrogen  
+
+Thank you for spotting this. Amended accordingly  
+
+Line 180: What or where is table S3? Is it the "dataset" that is attached? If so, please make that clear and add a table legend so that the reader knows what they're looking at.  
+
+This is a table containing the raw data for the proteomics analysis of Stappia stellulata. We uploaded it as a separate excel file.
+
+<--- Page Split --->
+
+Line 188: Statement as written implies that this is absent from the Rhodobacteriaceae. Do you mean in addition to?  
+
+Agreed. Changed to:  
+
+'Alphaproteobacteria in addition to Rhodobacteraceae, marine Deltaproteobacteria,....'  
+
+Line 316: What do you mean by co- culture experiments? Please provide a bit more detail.  
+
+As spotted also by reviewer 2 this mention of co- cultures was a relic of an earlier version of the manuscript and has now been removed.  
+
+Line 325: Table S4 is missing strain information. Also, as a general comment, I assume that all of the strains used in this work were axenic. The authors should explicitly state this somewhere.  
+
+Thank you for this suggestion. Gene names and locus tags have been added to Table S4, along with the purpose of each primer. We have added the following sentence to the beginning of the methods section:  
+
+'All strains used in this work were axenic.'  
+
+Methods: General comment. There are a lot of abbreviations in the methods of things that are somewhat standard/well known to a biologist, but probably should still be spelled out given the audience for Nature Communications (HEPES, LDS, PCR).  
+
+Thank you for your suggestion. We have removed all unnecessary abbreviations from the entire manuscript.  
+
+Figure 3 Legend: This legend is hard to follow and doesn't explain all of the components of the figure. What do the colors mean? What do the dashed lines mean?  
+
+Thank for this comment. We have now added the following sentences at the end of the figure legend.  
+
+'Vertical dashed lines represent an \(\mathrm{LFQ}_{\mathrm{Log2}}\) difference \(>3\) or \(< 3\) . The horizontal dashed line illustrates a cut off for a significant Q value \((p< 0.05)\) . Sky blue represents protein showing no significant difference between treatments. Peach and red indicate proteins significantly changing in abundance \(< 3\) - fold or \(>3\) - fold, respectively.'
+
+<--- Page Split --->
+
+## REVIEWERS' COMMENTS  
+
+Reviewer #1 (Remarks to the Author):  
+
+I want to congratulate and thank the authors for considering the knowledge and experience of the reviewers on this subject, who I believe will be satisfied with the effort made. The manuscript and the study were strengthened by the comments and suggestions made and the additional experiments that were carried out.  
+
+Although some concepts remain to be confirmed in the field, I believe that this document lays a solid foundation for further understanding the role of phosphonates in the oceanic phosphorus cycle.  
+
+Therefore, I would recommend the publication of this manuscript.  
+
+Reviewer #2 (Remarks to the Author):  
+
+I'd like to thank the authors for their detailed responses and clarifications, and especially for the additional labwork. I'm fully satisfied that the queries I raised have been addressed, and hope to see the paper published in its final form soon.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have appropriately addressed my comments in my previous review. The additional data presented in response to my and other reviewer comments have improved an already impressive manuscript. I did note some weirdness with bolding of certain words that were not consistent and had not been in the previous version of the manuscript. I'm not sure if those were just there because of the way they were tracking changes. Also the newly added sentence at the bottom of page 2 (lines 39- 41) has abbreviation issues, but all other abbreviation issues appear to have been fixed. I especially appreciate the inclusion of the full names of various 2- AEP phosphonate degradation systems that was added to the introduction.
+
+<--- Page Split --->
+
+## REVIEWER COMMENTS  
+
+Once again, we thank the reviewers for their time helping us improve our manuscript.  
+
+Reviewer #1 (Remarks to the Author):  
+
+I want to congratulate and thank the authors for considering the knowledge and experience of the reviewers on this subject, who I believe will be satisfied with the effort made. The manuscript and the study were strengthened by the comments and suggestions made and the additional experiments that were carried out.  
+
+Although some concepts remain to be confirmed in the field, I believe that this document lays a solid foundation for further understanding the role of phosphonates in the oceanic phosphorus cycle.  
+
+Therefore, I would recommend the publication of this manuscript.  
+
+Reviewer #2 (Remarks to the Author):  
+
+I'd like to thank the authors for their detailed responses and clarifications, and especially for the additional labwork. I'm fully satisfied that the queries I raised have been addressed, and hope to see the paper published in its final form soon.  
+
+Reviewer #3 (Remarks to the Author):  
+
+The authors have appropriately addressed my comments in my previous review. The additional data presented in response to my and other reviewer comments have improved an already impressive manuscript. I did note some weirdness with bolding of certain words that were not consistent and had not been in the previous version of the manuscript. I'm not sure if those were just there because of the way they were tracking changes. Also the newly added sentence at the bottom of page 2 (lines 39- 41) has abbreviation issues, but all other abbreviation issues appear to have been fixed. I especially appreciate the inclusion of the full names of various 2- AEP phosphonate degradation systems that was added to the introduction.  
+
+We have amended line 39- 41 and two other instances where phosphorus was spell in full, instead of P. These are marked in the final word document with a blue P.  
+
+Perhaps the bolding of certain words was a formatting error when converting to .pdf? We have thoroughly checked the word document and cannot find any reference to this.
+
+<--- Page Split --->